Ubuntu Manpages

collectd.conf

Configuration for the system statistics collection daemon collectd

  BaseDir "/path/to/data/"
  PIDFile "/path/to/pidfile/collectd.pid"
  Server  "123.123.123.123" 12345
  
  LoadPlugin cpu
  LoadPlugin load
  
  <LoadPlugin df>
    Interval 3600
  </LoadPlugin>
  
  LoadPlugin ping
  <Plugin ping>
    Host "example.org"
    Host "provider.net"
  </Plugin>

This config file controls how the system statistics collection daemon collectd behaves. The most significant option is LoadPlugin, which controls which plugins to load. These plugins ultimately define collectd's behavior.

The syntax of this config file is similar to the config file of the famous Apache webserver. Each line contains either an option (a key and a list of one or more values) or a section-start or -end. Empty lines and everything after a non-quoted hash-symbol ("#") is ignored. Keys are unquoted strings, consisting only of alphanumeric characters and the underscore ("_") character. Keys are handled case insensitive by collectd itself and all plugins included with it. Values can either be an unquoted string, a quoted string (enclosed in double-quotes) a number or a boolean expression. Unquoted strings consist of only alphanumeric characters and underscores ("_") and do not need to be quoted. Quoted strings are enclosed in double quotes ("""). You can use the backslash character ("\") to include double quotes as part of the string. Numbers can be specified in decimal and floating point format (using a dot "." as decimal separator), hexadecimal when using the "0x" prefix and octal with a leading zero (0). Boolean values are either true or false.

Lines may be wrapped by using "\" as the last character before the newline. This allows long lines to be split into multiple lines. Quoted strings may be wrapped as well. However, those are treated special in that whitespace at the beginning of the following lines will be ignored, which allows for nicely indenting the wrapped lines.

The configuration is read and processed in order, i.e. from top to bottom. So the plugins are loaded in the order listed in this config file. It is a good idea to load any logging plugins first in order to catch messages from plugins during configuration. Also, the "LoadPlugin" option must occur before the appropriate "<Plugin ...>" block.

Sets the base directory. This is the directory beneath all RRD-files are created. Possibly more subdirectories are created. This is also the working directory for the daemon.
Loads the plugin Plugin. This is required to load plugins, unless the AutoLoadPlugin option is enabled (see below). Without any loaded plugins, collectd will be mostly useless.

Only the first LoadPlugin statement or block for a given plugin name has any effect. This is useful when you want to split up the configuration into smaller files and want each file to be "self contained", i.e. it contains a Plugin block and then appropriate LoadPlugin statement. The downside is that if you have multiple conflicting LoadPlugin blocks, e.g. when they specify different intervals, only one of them (the first one encountered) will take effect and all others will be silently ignored.

LoadPlugin may either be a simple configuration statement or a block with additional options, affecting the behavior of LoadPlugin. A simple statement looks like this:

 LoadPlugin "cpu"

Options inside a LoadPlugin block can override default settings and influence the way plugins are loaded, e.g.:

 <LoadPlugin perl>
   Globals true
   Interval 60
 </LoadPlugin>

The following options are valid inside LoadPlugin blocks:

If enabled, collectd will export all global symbols of the plugin (and of all libraries loaded as dependencies of the plugin) and, thus, makes those symbols available for resolving unresolved symbols in subsequently loaded plugins if that is supported by your system.

This is useful (or possibly even required), e.g., when loading a plugin that embeds some scripting language into the daemon (e.g. the Perl and Python plugins). Scripting languages usually provide means to load extensions written in C. Those extensions require symbols provided by the interpreter, which is loaded as a dependency of the respective collectd plugin. See the documentation of those plugins (e.g., collectd-perl(5) or collectd-python(5)) for details.

By default, this is disabled. As a special exception, if the plugin name is either "perl" or "python", the default is changed to enabled in order to keep the average user from ever having to deal with this low level linking stuff.

Sets a plugin-specific interval for collecting metrics. This overrides the global Interval setting. If a plugin provides own support for specifying an interval, that setting will take precedence.
When set to false (the default), each plugin needs to be loaded explicitly, using the LoadPlugin statement documented above. If a <Plugin ...> block is encountered and no configuration handling callback for this plugin has been registered, a warning is logged and the block is ignored.

When set to true, explicit LoadPlugin statements are not required. Each <Plugin ...> block acts as if it was immediately preceded by a LoadPlugin statement. LoadPlugin statements are still required for plugins that don't provide any configuration, e.g. the Load plugin.

If Path points to a file, includes that file. If Path points to a directory, recursively includes all files within that directory and its subdirectories. If the "wordexp" function is available on your system, shell-like wildcards are expanded before files are included. This means you can use statements like the following: 

  Include "/etc/collectd.d/*.conf"

Starting with version 5.3, this may also be a block in which further options affecting the behavior of Include may be specified. The following option is currently allowed:

  <Include "/etc/collectd.d">
    Filter "*.conf"
  </Include>
If the "fnmatch" function is available on your system, a shell-like wildcard pattern may be specified to filter which files to include. This may be used in combination with recursively including a directory to easily be able to arbitrarily mix configuration files and other documents (e.g. README files). The given example is similar to the first example above but includes all files matching "*.conf" in any subdirectory of "/etc/collectd.d": 

  Include "/etc/collectd.d" "*.conf"

If more than one files are included by a single Include option, the files will be included in lexicographical order (as defined by the "strcmp" function). Thus, you can e. g. use numbered prefixes to specify the order in which the files are loaded.

To prevent loops and shooting yourself in the foot in interesting ways the nesting is limited to a depth of 8 levels, which should be sufficient for most uses. Since symlinks are followed it is still possible to crash the daemon by looping symlinks. In our opinion significant stupidity should result in an appropriate amount of pain.

It is no problem to have a block like "<Plugin foo>" in more than one file, but you cannot include files from within blocks.

Sets where to write the PID file to. This file is overwritten when it exists and deleted when the program is stopped. Some init-scripts might override this setting using the -P command-line option.
Path to the plugins (shared objects) of collectd.
Set one or more files that contain the data-set descriptions. See types.db(5) for a description of the format of this file.
Configures the interval in which to query the read plugins. Obviously smaller values lead to a higher system load produced by collectd, while higher values lead to more coarse statistics.

Warning: You should set this once and then never touch it again. If you do, you will have to delete all your RRD files or know some serious RRDtool magic! (Assuming you're using the RRDtool or RRDCacheD plugin.)

Consider a value list "missing" when no update has been read or received for Iterations iterations. By default, collectd considers a value list missing when no update has been received for twice the update interval. Since this setting uses iterations, the maximum allowed time without update depends on the Interval information contained in each value list. This is used in the Threshold configuration to dispatch notifications about missing values, see collectd-threshold(5) for details.
Number of threads to start for reading plugins. The default value is 5, but you may want to increase this if you have more than five plugins that take a long time to read. Mostly those are plugins that do network-IO. Setting this to a value higher than the number of registered read callbacks is not recommended.
Number of threads to start for dispatching value lists to write plugins. The default value is 5, but you may want to increase this if you have more than five plugins that may take relatively long to write to.
Metrics are read by the read threads and then put into a queue to be handled by the write threads. If one of the write plugins is slow (e.g. network timeouts, I/O saturation of the disk) this queue will grow. In order to avoid running into memory issues in such a case, you can limit the size of this queue.

By default, there is no limit and memory may grow indefinitely. This is most likely not an issue for clients, i.e. instances that only handle the local metrics. For servers it is recommended to set this to a non-zero value, though.

You can set the limits using WriteQueueLimitHigh and WriteQueueLimitLow. Each of them takes a numerical argument which is the number of metrics in the queue. If there are HighNum metrics in the queue, any new metrics will be dropped. If there are less than LowNum metrics in the queue, all new metrics will be enqueued. If the number of metrics currently in the queue is between LowNum and HighNum, the metric is dropped with a probability that is proportional to the number of metrics in the queue (i.e. it increases linearly until it reaches 100%.)

If WriteQueueLimitHigh is set to non-zero and WriteQueueLimitLow is unset, the latter will default to half of WriteQueueLimitHigh.

If you do not want to randomly drop values when the queue size is between LowNum and HighNum, set If WriteQueueLimitHigh and WriteQueueLimitLow to same value.

Sets the hostname that identifies a host. If you omit this setting, the hostname will be determined using the gethostname(2) system call.
If Hostname is determined automatically this setting controls whether or not the daemon should try to figure out the "fully qualified domain name", FQDN. This is done using a lookup of the name returned by "gethostname". This option is enabled by default.
Configure the name of the "pre-cache chain" and the "post-cache chain". Please see "FILTER CONFIGURATION" below on information on chains and how these setting change the daemon's behavior.

Some plugins may register own options. These options must be enclosed in a "Plugin"-Section. Which options exist depends on the plugin used. Some plugins require external configuration, too. The "apache plugin", for example, required "mod_status" to be configured in the webserver you're going to collect data from. These plugins are listed below as well, even if they don't require any configuration within collectd's configuration file.

A list of all plugins and a short summary for each plugin can be found in the README file shipped with the sourcecode and hopefully binary packets as well.

The Aggregation plugin makes it possible to aggregate several values into one using aggregation functions such as sum, average, min and max. This can be put to a wide variety of uses, e.g. average and total CPU statistics for your entire fleet.

The grouping is powerful but, as with many powerful tools, may be a bit difficult to wrap your head around. The grouping will therefore be demonstrated using an example: The average and sum of the CPU usage across all CPUs of each host is to be calculated.

To select all the affected values for our example, set "Plugin cpu" and "Type cpu". The other values are left unspecified, meaning "all values". The Host, Plugin, PluginInstance, Type and TypeInstance options work as if they were specified in the "WHERE" clause of an "SELECT" SQL statement.

  Plugin "cpu"
  Type "cpu"

Although the Host, PluginInstance (CPU number, i.e. 0, 1, 2, ...) and TypeInstance (idle, user, system, ...) fields are left unspecified in the example, the intention is to have a new value for each host / type instance pair. This is achieved by "grouping" the values using the "GroupBy" option. It can be specified multiple times to group by more than one field.

  GroupBy "Host"
  GroupBy "TypeInstance"

We do neither specify nor group by plugin instance (the CPU number), so all metrics that differ in the CPU number only will be aggregated. Each aggregation needs at least one such field, otherwise no aggregation would take place.

The full example configuration looks like this:

 <Plugin "aggregation">
   <Aggregation>
     Plugin "cpu"
     Type "cpu"
     
     GroupBy "Host"
     GroupBy "TypeInstance"
     
     CalculateSum true
     CalculateAverage true
   </Aggregation>
 </Plugin>

There are a couple of limitations you should be aware of:

  • The Type cannot be left unspecified, because it is not reasonable to add apples to oranges. Also, the internal lookup structure won't work if you try to group by type.
  • There must be at least one unspecified, ungrouped field. Otherwise nothing will be aggregated.

As you can see in the example above, each aggregation has its own Aggregation block. You can have multiple aggregation blocks and aggregation blocks may match the same values, i.e. one value list can update multiple aggregations. The following options are valid inside Aggregation blocks:

Selects the value lists to be added to this aggregation. Type must be a valid data set name, see types.db(5) for details.

If the string starts with and ends with a slash ("/"), the string is interpreted as a regular expression. The regex flavor used are POSIX extended regular expressions as described in regex(7). Example usage:

 Host "/^db[0-9]\\.example\\.com$/"
Group valued by the specified field. The GroupBy option may be repeated to group by multiple fields.
Sets the appropriate part of the identifier to the provided string.

The PluginInstance should include the placeholder "%{aggregation}" which will be replaced with the aggregation function, e.g. "average". Not including the placeholder will result in duplication warnings and/or messed up values if more than one aggregation function are enabled.

The following example calculates the average usage of all "even" CPUs:

 <Plugin "aggregation">
   <Aggregation>
     Plugin "cpu"
     PluginInstance "/[0,2,4,6,8]$/"
     Type "cpu"
     
     SetPlugin "cpu"
     SetPluginInstance "even-%{aggregation}"
     
     GroupBy "Host"
     GroupBy "TypeInstance"
     
     CalculateAverage true
   </Aggregation>
 </Plugin>

This will create the files:

  • foo.example.com/cpu-even-average/cpu-idle
  • foo.example.com/cpu-even-average/cpu-system
  • foo.example.com/cpu-even-average/cpu-user
  • ...
Boolean options for enabling calculation of the number of value lists, their sum, average, minimum, maximum and / or standard deviation. All options are disabled by default.

The AMQMP plugin can be used to communicate with other instances of collectd or third party applications using an AMQP message broker. Values are sent to or received from the broker, which handles routing, queueing and possibly filtering or messages.

 <Plugin "amqp">
   # Send values to an AMQP broker
   <Publish "some_name">
     Host "localhost"
     Port "5672"
     VHost "/"
     User "guest"
     Password "guest"
     Exchange "amq.fanout"
 #   ExchangeType "fanout"
 #   RoutingKey "collectd"
 #   Persistent false
 #   Format "command"
 #   StoreRates false
 #   GraphitePrefix "collectd."
 #   GraphiteEscapeChar "_"
   </Publish>
   
   # Receive values from an AMQP broker
   <Subscribe "some_name">
     Host "localhost"
     Port "5672"
     VHost "/"
     User "guest"
     Password "guest"
     Exchange "amq.fanout"
 #   ExchangeType "fanout"
 #   Queue "queue_name"
 #   RoutingKey "collectd.#"
   </Subscribe>
 </Plugin>

The plugin's configuration consists of a number of Publish and Subscribe blocks, which configure sending and receiving of values respectively. The two blocks are very similar, so unless otherwise noted, an option can be used in either block. The name given in the blocks starting tag is only used for reporting messages, but may be used to support flushing of certain Publish blocks in the future.

Hostname or IP-address of the AMQP broker. Defaults to the default behavior of the underlying communications library, rabbitmq-c, which is "localhost".
Service name or port number on which the AMQP broker accepts connections. This argument must be a string, even if the numeric form is used. Defaults to "5672".
Name of the virtual host on the AMQP broker to use. Defaults to "/".
Credentials used to authenticate to the AMQP broker. By default "guest"/"guest" is used.
In Publish blocks, this option specifies the exchange to send values to. By default, "amq.fanout" will be used.

In Subscribe blocks this option is optional. If given, a binding between the given exchange and the queue is created, using the routing key if configured. See the Queue and RoutingKey options below.

If given, the plugin will try to create the configured exchange with this type after connecting. When in a Subscribe block, the queue will then be bound to this exchange.
Configures the queue name to subscribe to. If no queue name was configures explicitly, a unique queue name will be created by the broker.
In Publish blocks, this configures the routing key to set on all outgoing messages. If not given, the routing key will be computed from the identifier of the value. The host, plugin, type and the two instances are concatenated together using dots as the separator and all containing dots replaced with slashes. For example "collectd.host/example/com.cpu.0.cpu.user". This makes it possible to receive only specific values using a "topic" exchange.

In Subscribe blocks, configures the routing key used when creating a binding between an exchange and the queue. The usual wildcards can be used to filter messages when using a "topic" exchange. If you're only interested in CPU statistics, you could use the routing key "collectd.*.cpu.#" for example.

Selects the delivery method to use. If set to true, the persistent mode will be used, i.e. delivery is guaranteed. If set to false (the default), the transient delivery mode will be used, i.e. messages may be lost due to high load, overflowing queues or similar issues.
Selects the format in which messages are sent to the broker. If set to Command (the default), values are sent as "PUTVAL" commands which are identical to the syntax used by the Exec and UnixSock plugins. In this case, the "Content-Type" header field will be set to "text/collectd".

If set to JSON, the values are encoded in the JavaScript Object Notation, an easy and straight forward exchange format. The "Content-Type" header field will be set to "application/json".

If set to Graphite, values are encoded in the Graphite format, which is "<metric> <value> <timestamp>\n". The "Content-Type" header field will be set to "text/graphite".

A subscribing client should use the "Content-Type" header field to determine how to decode the values. Currently, the AMQP plugin itself can only decode the Command format.

Determines whether or not "COUNTER", "DERIVE" and "ABSOLUTE" data sources are converted to a rate (i.e. a "GAUGE" value). If set to false (the default), no conversion is performed. Otherwise the conversion is performed using the internal value cache.

Please note that currently this option is only used if the Format option has been set to JSON.

A prefix can be added in the metric name when outputting in the Graphite format. It's added before the Host name. Metric name will be "<prefix><host><postfix><plugin><type><name>"
A postfix can be added in the metric name when outputting in the Graphite format. It's added after the Host name. Metric name will be "<prefix><host><postfix><plugin><type><name>"
Specify a character to replace dots (.) in the host part of the metric name. In Graphite metric name, dots are used as separators between different metric parts (host, plugin, type). Default is "_" (Underscore).

To configure the "apache"-plugin you first need to configure the Apache webserver correctly. The Apache-plugin "mod_status" needs to be loaded and working and the "ExtendedStatus" directive needs to be enabled. You can use the following snipped to base your Apache config upon:

  ExtendedStatus on
  <IfModule mod_status.c>
    <Location /mod_status>
      SetHandler server-status
    </Location>
  </IfModule>

Since its "mod_status" module is very similar to Apache's, lighttpd is also supported. It introduces a new field, called "BusyServers", to count the number of currently connected clients. This field is also supported.

The configuration of the Apache plugin consists of one or more "<Instance />" blocks. Each block requires one string argument as the instance name. For example:

 <Plugin "apache">
   <Instance "www1">
     URL "http://www1.example.com/mod_status?auto"
   </Instance>
   <Instance "www2">
     URL "http://www2.example.com/mod_status?auto"
   </Instance>
 </Plugin>

The instance name will be used as the plugin instance. To emulate the old (version 4) behavior, you can use an empty string (""). In order for the plugin to work correctly, each instance name must be unique. This is not enforced by the plugin and it is your responsibility to ensure it.

The following options are accepted within each Instance block:

Sets the URL of the "mod_status" output. This needs to be the output generated by "ExtendedStatus on" and it needs to be the machine readable output generated by appending the "?auto" argument. This option is mandatory.
Optional user name needed for authentication.
Optional password needed for authentication.
Enable or disable peer SSL certificate verification. See <http://curl.haxx.se/docs/sslcerts.html> for details. Enabled by default.
Enable or disable peer host name verification. If enabled, the plugin checks if the "Common Name" or a "Subject Alternate Name" field of the SSL certificate matches the host name provided by the URL option. If this identity check fails, the connection is aborted. Obviously, only works when connecting to a SSL enabled server. Enabled by default.
File that holds one or more SSL certificates. If you want to use HTTPS you will possibly need this option. What CA certificates come bundled with "libcurl" and are checked by default depends on the distribution you use.

Hostname of the host running apcupsd. Defaults to localhost. Please note that IPv6 support has been disabled unless someone can confirm or decline that apcupsd can handle it.
TCP-Port to connect to. Defaults to 3551.
If set to true, the time reported in the "timeleft" metric will be converted to seconds. This is the recommended setting. If set to false, the default for backwards compatibility, the time will be reported in minutes.

This plugin collects the value of the available sensors in an Aquaero 5 board. Aquaero 5 is a water-cooling controller board, manufactured by Aqua Computer GmbH <http://www.aquacomputer.de/>, with a USB2 connection for monitoring and configuration. The board can handle multiple temperature sensors, fans, water pumps and water level sensors and adjust the output settings such as fan voltage or power used by the water pump based on the available inputs using a configurable controller included in the board. This plugin collects all the available inputs as well as some of the output values chosen by this controller. The plugin is based on the libaquaero5 library provided by aquatools-ng.

Device path of the Aquaero 5's USB HID (human interface device), usually in the form "/dev/usb/hiddevX". If this option is no set the plugin will try to auto-detect the Aquaero 5 USB device based on vendor-ID and product-ID.

This plugin collects information about an Ascent server, a free server for the "World of Warcraft" game. This plugin gathers the information by fetching the XML status page using "libcurl" and parses it using "libxml2".

The configuration options are the same as for the "apache" plugin above:

Sets the URL of the XML status output.
Optional user name needed for authentication.
Optional password needed for authentication.
Enable or disable peer SSL certificate verification. See <http://curl.haxx.se/docs/sslcerts.html> for details. Enabled by default.
Enable or disable peer host name verification. If enabled, the plugin checks if the "Common Name" or a "Subject Alternate Name" field of the SSL certificate matches the host name provided by the URL option. If this identity check fails, the connection is aborted. Obviously, only works when connecting to a SSL enabled server. Enabled by default.
File that holds one or more SSL certificates. If you want to use HTTPS you will possibly need this option. What CA certificates come bundled with "libcurl" and are checked by default depends on the distribution you use.

Starting with BIND 9.5.0, the most widely used DNS server software provides extensive statistics about queries, responses and lots of other information. The bind plugin retrieves this information that's encoded in XML and provided via HTTP and submits the values to collectd.

To use this plugin, you first need to tell BIND to make this information available. This is done with the "statistics-channels" configuration option:

 statistics-channels {
   inet localhost port 8053;
 };

The configuration follows the grouping that can be seen when looking at the data with an XSLT compatible viewer, such as a modern web browser. It's probably a good idea to make yourself familiar with the provided values, so you can understand what the collected statistics actually mean.

Synopsis:

 <Plugin "bind">
   URL "http://localhost:8053/"
   ParseTime       false
   OpCodes         true
   QTypes          true
 
   ServerStats     true
   ZoneMaintStats  true
   ResolverStats   false
   MemoryStats     true
 
   <View "_default">
     QTypes        true
     ResolverStats true
     CacheRRSets   true
 
     Zone "127.in-addr.arpa/IN"
   </View>
 </Plugin>

The bind plugin accepts the following configuration options:

URL from which to retrieve the XML data. If not specified, "http://localhost:8053/" will be used.
When set to true, the time provided by BIND will be parsed and used to dispatch the values. When set to false, the local time source is queried.

This setting is set to true by default for backwards compatibility; setting this to false is recommended to avoid problems with timezones and localization.

When enabled, statistics about the "OpCodes", for example the number of "QUERY" packets, are collected.

Default: Enabled.

When enabled, the number of incoming queries by query types (for example "A", "MX", "AAAA") is collected.

Default: Enabled.

Collect global server statistics, such as requests received over IPv4 and IPv6, successful queries, and failed updates.

Default: Enabled.

Collect zone maintenance statistics, mostly information about notifications (zone updates) and zone transfers.

Default: Enabled.

Collect resolver statistics, i. e. statistics about outgoing requests (e. g. queries over IPv4, lame servers). Since the global resolver counters apparently were removed in BIND 9.5.1 and 9.6.0, this is disabled by default. Use the ResolverStats option within a View "_default" block instead for the same functionality.

Default: Disabled.

Collect global memory statistics.

Default: Enabled.

Collect statistics about a specific "view". BIND can behave different, mostly depending on the source IP-address of the request. These different configurations are called "views". If you don't use this feature, you most likely are only interested in the "_default" view.

Within a <View name> block, you can specify which information you want to collect about a view. If no View block is configured, no detailed view statistics will be collected.

If enabled, the number of outgoing queries by query type (e. g. "A", "MX") is collected.

Default: Enabled.

Collect resolver statistics, i. e. statistics about outgoing requests (e. g. queries over IPv4, lame servers).

Default: Enabled.

If enabled, the number of entries ("RR sets") in the view's cache by query type is collected. Negative entries (queries which resulted in an error, for example names that do not exist) are reported with a leading exclamation mark, e. g. "!A".

Default: Enabled.

When given, collect detailed information about the given zone in the view. The information collected if very similar to the global ServerStats information (see above).

You can repeat this option to collect detailed information about multiple zones.

By default no detailed zone information is collected.

This plugin collects the CPU user/system time for each cgroup by reading the cpuacct.stat files in the first cpuacct-mountpoint (typically /sys/fs/cgroup/cpu.cpuacct on machines using systemd).

Select cgroup based on the name. Whether only matching cgroups are collected or if they are ignored is controlled by the IgnoreSelected option; see below.
Invert the selection: If set to true, all cgroups except the ones that match any one of the criteria are collected. By default only selected cgroups are collected if a selection is made. If no selection is configured at all, all cgroups are selected.

This plugin doesn't have any options. It reads /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq (for the first CPU installed) to get the current CPU frequency. If this file does not exist make sure cpufreqd (<http://cpufreqd.sourceforge.net/>) or a similar tool is installed and an "cpu governor" (that's a kernel module) is loaded.

Set the directory to store CSV-files under. Per default CSV-files are generated beneath the daemon's working directory, i. e. the BaseDir. The special strings stdout and stderr can be used to write to the standard output and standard error channels, respectively. This, of course, only makes much sense when collectd is running in foreground- or non-daemon-mode.
If set to true, convert counter values to rates. If set to false (the default) counter values are stored as is, i. e. as an increasing integer number.

The curl plugin uses the libcurl (<http://curl.haxx.se/>) to read web pages and the match infrastructure (the same code used by the tail plugin) to use regular expressions with the received data.

The following example will read the current value of AMD stock from Google's finance page and dispatch the value to collectd.

  <Plugin curl>
    <Page "stock_quotes">
      URL "http://finance.google.com/finance?q=NYSE%3AAMD"
      User "foo"
      Password "bar"
      <Match>
        Regex "<span +class=\"pr\"[^>]*> *([0-9]*\\.[0-9]+) *</span>"
        DSType "GaugeAverage"
        # Note: `stock_value' is not a standard type.
        Type "stock_value"
        Instance "AMD"
      </Match>
    </Page>
  </Plugin>

In the Plugin block, there may be one or more Page blocks, each defining a web page and one or more "matches" to be performed on the returned data. The string argument to the Page block is used as plugin instance.

The following options are valid within Page blocks:

URL of the web site to retrieve. Since a regular expression will be used to extract information from this data, non-binary data is a big plus here ;)
Username to use if authorization is required to read the page.
Password to use if authorization is required to read the page.
Enable or disable peer SSL certificate verification. See <http://curl.haxx.se/docs/sslcerts.html> for details. Enabled by default.
Enable or disable peer host name verification. If enabled, the plugin checks if the "Common Name" or a "Subject Alternate Name" field of the SSL certificate matches the host name provided by the URL option. If this identity check fails, the connection is aborted. Obviously, only works when connecting to a SSL enabled server. Enabled by default.
File that holds one or more SSL certificates. If you want to use HTTPS you will possibly need this option. What CA certificates come bundled with "libcurl" and are checked by default depends on the distribution you use.
A HTTP header to add to the request. Multiple headers are added if this option is specified more than once.
Specifies that the HTTP operation should be a POST instead of a GET. The complete data to be posted is given as the argument. This option will usually need to be accompanied by a Header option to set an appropriate "Content-Type" for the post body (e.g. to "application/x-www-form-urlencoded").
Measure response time for the request. If this setting is enabled, Match blocks (see below) are optional. Disabled by default.
<Match>
One or more Match blocks that define how to match information in the data returned by "libcurl". The "curl" plugin uses the same infrastructure that's used by the "tail" plugin, so please see the documentation of the "tail" plugin below on how matches are defined. If the MeasureResponseTime option is set to true, Match blocks are optional.

The curl_json plugin collects values from JSON data to be parsed by libyajl (<http://www.lloydforge.org/projects/yajl/>) retrieved via either libcurl (<http://curl.haxx.se/>) or read directly from a unix socket. The former can be used, for example, to collect values from CouchDB documents (which are stored JSON notation), and the latter to collect values from a uWSGI stats socket.

The following example will collect several values from the built-in "_stats" runtime statistics module of CouchDB (<http://wiki.apache.org/couchdb/Runtime_Statistics>).

  <Plugin curl_json>
    <URL "http://localhost:5984/_stats">
      Instance "httpd"
      <Key "httpd/requests/count">
        Type "http_requests"
      </Key>
      <Key "httpd_request_methods/*/count">
        Type "http_request_methods"
      </Key>
      <Key "httpd_status_codes/*/count">
        Type "http_response_codes"
      </Key>
    </URL>
  </Plugin>

This example will collect data directly from a uWSGI "Stats Server" socket.

  <Plugin curl_json>
    <Sock "/var/run/uwsgi.stats.sock">
      Instance "uwsgi"
      <Key "workers/*/requests">
        Type "http_requests"
      </Key>
      <Key "workers/*/apps/*/requests">
        Type "http_requests"
      </Key>
    </Sock>
  </Plugin>

In the Plugin block, there may be one or more URL blocks, each defining a URL to be fetched via HTTP (using libcurl) or Sock blocks defining a unix socket to read JSON from directly. Each of these blocks may have one or more Key blocks.

The Key string argument must be in a path format. Each component is used to match the key from a JSON map or the index of an JSON array. If a path component of a Key is a * wildcard, the values for all map keys or array indices will be collectd.

The following options are valid within URL blocks:

Sets the plugin instance to Instance.
These options behave exactly equivalent to the appropriate options of the cURL plugin. Please see there for a detailed description.

The following options are valid within Key blocks:

Sets the type used to dispatch the values to the daemon. Detailed information about types and their configuration can be found in types.db(5). This option is mandatory.
Type-instance to use. Defaults to the current map key or current string array element value.

The curl_xml plugin uses libcurl (<http://curl.haxx.se/>) and libxml2 (<http://xmlsoft.org/>) to retrieve XML data via cURL.

 <Plugin "curl_xml">
   <URL "http://localhost/stats.xml">
     Host "my_host"
     Instance "some_instance"
     User "collectd"
     Password "thaiNg0I"
     VerifyPeer true
     VerifyHost true
     CACert "/path/to/ca.crt"
     <XPath "table[@id=\"magic_level\"]/tr">
       Type "magic_level"
       #InstancePrefix "prefix-"
       InstanceFrom "td[1]"
       ValuesFrom "td[2]/span[@class=\"level\"]"
     </XPath>
   </URL>
 </Plugin>

In the Plugin block, there may be one or more URL blocks, each defining a URL to be fetched using libcurl. Within each URL block there are options which specify the connection parameters, for example authentication information, and one or more XPath blocks.

Each XPath block specifies how to get one type of information. The string argument must be a valid XPath expression which returns a list of "base elements". One value is dispatched for each "base element". The type instance and values are looked up using further XPath expressions that should be relative to the base element.

Within the URL block the following options are accepted:

Use Name as the host name when submitting values. Defaults to the global host name setting.
Use Instance as the plugin instance when submitting values. Defaults to an empty string (no plugin instance).
If an XPath expression references namespaces, they must be specified with this option. Prefix is the "namespace prefix" used in the XML document. URL is the "namespace name", an URI reference uniquely identifying the namespace. The option can be repeated to register multiple namespaces.

Examples:

  Namespace "s" "http://schemas.xmlsoap.org/soap/envelope/"
  Namespace "m" "http://www.w3.org/1998/Math/MathML"
These options behave exactly equivalent to the appropriate options of the cURL plugin. Please see there for a detailed description.
<XPath XPath-expression>
Within each URL block, there must be one or more XPath blocks. Each XPath block specifies how to get one type of information. The string argument must be a valid XPath expression which returns a list of "base elements". One value is dispatched for each "base element".

Within the XPath block the following options are accepted:

Specifies the Type used for submitting patches. This determines the number of values that are required / expected and whether the strings are parsed as signed or unsigned integer or as double values. See types.db(5) for details. This option is required.
Prefix the type instance with InstancePrefix. The values are simply concatenated together without any separator. This option is optional.
Specifies a XPath expression to use for determining the type instance. The XPath expression must return exactly one element. The element's value is then used as type instance, possibly prefixed with InstancePrefix (see above).

This value is required. As a special exception, if the "base XPath expression" (the argument to the XPath block) returns exactly one argument, then this option may be omitted.

Specifies one or more XPath expression to use for reading the values. The number of XPath expressions must match the number of data sources in the type specified with Type (see above). Each XPath expression must return exactly one element. The element's value is then parsed as a number and used as value for the appropriate value in the value list dispatched to the daemon.

This plugin uses the dbi library (<http://libdbi.sourceforge.net/>) to connect to various databases, execute SQL statements and read back the results. dbi is an acronym for "database interface" in case you were wondering about the name. You can configure how each column is to be interpreted and the plugin will generate one or more data sets from each row returned according to these rules.

Because the plugin is very generic, the configuration is a little more complex than those of other plugins. It usually looks something like this:

  <Plugin dbi>
    <Query "out_of_stock">
      Statement "SELECT category, COUNT(*) AS value FROM products WHERE in_stock = 0 GROUP BY category"
      # Use with MySQL 5.0.0 or later
      MinVersion 50000
      <Result>
        Type "gauge"
        InstancePrefix "out_of_stock"
        InstancesFrom "category"
        ValuesFrom "value"
      </Result>
    </Query>
    <Database "product_information">
      Driver "mysql"
      DriverOption "host" "localhost"
      DriverOption "username" "collectd"
      DriverOption "password" "aZo6daiw"
      DriverOption "dbname" "prod_info"
      SelectDB "prod_info"
      Query "out_of_stock"
    </Database>
  </Plugin>

The configuration above defines one query with one result and one database. The query is then linked to the database with the Query option within the <Database> block. You can have any number of queries and databases and you can also use the Include statement to split up the configuration file in multiple, smaller files. However, the <Query> block must precede the <Database> blocks, because the file is interpreted from top to bottom!

The following is a complete list of options:

Query blocks

Query blocks define SQL statements and how the returned data should be interpreted. They are identified by the name that is given in the opening line of the block. Thus the name needs to be unique. Other than that, the name is not used in collectd.

In each Query block, there is one or more Result blocks. Result blocks define which column holds which value or instance information. You can use multiple Result blocks to create multiple values from one returned row. This is especially useful, when queries take a long time and sending almost the same query again and again is not desirable.

Example:

  <Query "environment">
    Statement "select station, temperature, humidity from environment"
    <Result>
      Type "temperature"
      # InstancePrefix "foo"
      InstancesFrom "station"
      ValuesFrom "temperature"
    </Result>
    <Result>
      Type "humidity"
      InstancesFrom "station"
      ValuesFrom "humidity"
    </Result>
  </Query>

The following options are accepted:

Sets the statement that should be executed on the server. This is not interpreted by collectd, but simply passed to the database server. Therefore, the SQL dialect that's used depends on the server collectd is connected to.

The query has to return at least two columns, one for the instance and one value. You cannot omit the instance, even if the statement is guaranteed to always return exactly one line. In that case, you can usually specify something like this:

  Statement "SELECT \"instance\", COUNT(*) AS value FROM table"

(That works with MySQL but may not be valid SQL according to the spec. If you use a more strict database server, you may have to select from a dummy table or something.)

Please note that some databases, for example Oracle, will fail if you include a semicolon at the end of the statement.

Only use this query for the specified database version. You can use these options to provide multiple queries with the same name but with a slightly different syntax. The plugin will use only those queries, where the specified minimum and maximum versions fit the version of the database in use.

The database version is determined by "dbi_conn_get_engine_version", see the libdbi documentation <http://libdbi.sourceforge.net/docs/programmers-guide/reference-conn.html#DBI-CONN-GET-ENGINE-VERSION> for details. Basically, each part of the version is assumed to be in the range from 00 to 99 and all dots are removed. So version "4.1.2" becomes "40102", version "5.0.42" becomes "50042".

Warning: The plugin will use all matching queries, so if you specify multiple queries with the same name and overlapping ranges, weird stuff will happen. Don't to it! A valid example would be something along these lines:

  MinVersion 40000
  MaxVersion 49999
  ...
  MinVersion 50000
  MaxVersion 50099
  ...
  MinVersion 50100
  # No maximum

In the above example, there are three ranges that don't overlap. The last one goes from version "5.1.0" to infinity, meaning "all later versions". Versions before "4.0.0" are not specified.

The type that's used for each line returned. See types.db(5) for more details on how types are defined. In short: A type is a predefined layout of data and the number of values and type of values has to match the type definition.

If you specify "temperature" here, you need exactly one gauge column. If you specify "if_octets", you will need two counter columns. See the ValuesFrom setting below.

There must be exactly one Type option inside each Result block.

Prepends prefix to the type instance. If InstancesFrom (see below) is not given, the string is simply copied. If InstancesFrom is given, prefix and all strings returned in the appropriate columns are concatenated together, separated by dashes ("-").
Specifies the columns whose values will be used to create the "type-instance" for each row. If you specify more than one column, the value of all columns will be joined together with dashes ("-") as separation characters.

The plugin itself does not check whether or not all built instances are different. It's your responsibility to assure that each is unique. This is especially true, if you do not specify InstancesFrom: You have to make sure that only one row is returned in this case.

If neither InstancePrefix nor InstancesFrom is given, the type-instance will be empty.

Names the columns whose content is used as the actual data for the data sets that are dispatched to the daemon. How many such columns you need is determined by the Type setting above. If you specify too many or not enough columns, the plugin will complain about that and no data will be submitted to the daemon.

The actual data type in the columns is not that important. The plugin will automatically cast the values to the right type if it know how to do that. So it should be able to handle integer an floating point types, as well as strings (if they include a number at the beginning).

There must be at least one ValuesFrom option inside each Result block.

Database blocks

Database blocks define a connection to a database and which queries should be sent to that database. Since the used "dbi" library can handle a wide variety of databases, the configuration is very generic. If in doubt, refer to libdbi's documentation - we stick as close to the terminology used there.

Each database needs a "name" as string argument in the starting tag of the block. This name will be used as "PluginInstance" in the values submitted to the daemon. Other than that, that name is not used.

Specifies the driver to use to connect to the database. In many cases those drivers are named after the database they can connect to, but this is not a technical necessity. These drivers are sometimes referred to as "DBD", DataBase Driver, and some distributions ship them in separate packages. Drivers for the "dbi" library are developed by the libdbi-drivers project at <http://libdbi-drivers.sourceforge.net/>.

You need to give the driver name as expected by the "dbi" library here. You should be able to find that in the documentation for each driver. If you mistype the driver name, the plugin will dump a list of all known driver names to the log.

Sets driver-specific options. What option a driver supports can be found in the documentation for each driver, somewhere at <http://libdbi-drivers.sourceforge.net/>. However, the options "host", "username", "password", and "dbname" seem to be de facto standards.

DBDs can register two types of options: String options and numeric options. The plugin will use the "dbi_conn_set_option" function when the configuration provides a string and the "dbi_conn_require_option_numeric" function when the configuration provides a number. So these two lines will actually result in different calls being used:

  DriverOption "Port" 1234      # numeric
  DriverOption "Port" "1234"    # string

Unfortunately, drivers are not too keen to report errors when an unknown option is passed to them, so invalid settings here may go unnoticed. This is not the plugin's fault, it will report errors if it gets them from the library / the driver. If a driver complains about an option, the plugin will dump a complete list of all options understood by that driver to the log. There is no way to programatically find out if an option expects a string or a numeric argument, so you will have to refer to the appropriate DBD's documentation to find this out. Sorry.

In some cases, the database name you connect with is not the database name you want to use for querying data. If this option is set, the plugin will "select" (switch to) that database after the connection is established.
Associates the query named QueryName with this database connection. The query needs to be defined before this statement, i. e. all query blocks you want to refer to must be placed above the database block you want to refer to them from.
Sets the host field of value lists to Hostname when dispatching values. Defaults to the global hostname setting.

Select partitions based on the devicename.
Select partitions based on the mountpoint.
Select partitions based on the filesystem type.
Invert the selection: If set to true, all partitions except the ones that match any one of the criteria are collected. By default only selected partitions are collected if a selection is made. If no selection is configured at all, all partitions are selected.
Report using the device name rather than the mountpoint. i.e. with this false, (the default), it will report a disk as "root", but with it true, it will be "sda1" (or whichever).
Enables or disables reporting of free, reserved and used inodes. Defaults to inode collection being disabled.

Enable this option if inodes are a scarce resource for you, usually because many small files are stored on the disk. This is a usual scenario for mail transfer agents and web caches.

Enables or disables reporting of free, used and used disk space in 1K-blocks. Defaults to true.
Enables or disables reporting of free, used and used disk space in percentage. Defaults to false.

This is useful for deploying collectd on the cloud, where machines with different disk size may exist. Then it is more practical to configure thresholds based on relative disk size.

The "disk" plugin collects information about the usage of physical disks and logical disks (partitions). Values collected are the number of octets written to and read from a disk or partition, the number of read/write operations issued to the disk and a rather complex "time" it took for these commands to be issued.

Using the following two options you can ignore some disks or configure the collection only of specific disks.

Select the disk Name. Whether it is collected or ignored depends on the IgnoreSelected setting, see below. As with other plugins that use the daemon's ignorelist functionality, a string that starts and ends with a slash is interpreted as a regular expression. Examples: 

  Disk "sdd"
  Disk "/hda[34]/"
Sets whether selected disks, i. e. the ones matches by any of the Disk statements, are ignored or if all other disks are ignored. The behavior (hopefully) is intuitive: If no Disk option is configured, all disks are collected. If at least one Disk option is given and no IgnoreSelected or set to false, only matching disks will be collected. If IgnoreSelected is set to true, all disks are collected except the ones matched.

The dns plugin uses libpcap to capture dns traffic and analyzes it. This option sets the interface that should be used. If this option is not set, or set to "any", the plugin will try to get packets from all interfaces. This may not work on certain platforms, such as Mac OS X.
Ignore packets that originate from this address.
Enabled by default, collects unknown (and thus presented as numeric only) query types.

Sets the socket-file which is to be created.
If running as root change the group of the UNIX-socket after it has been created. Defaults to collectd.
Change the file permissions of the UNIX-socket after it has been created. The permissions must be given as a numeric, octal value as you would pass to chmod(1). Defaults to 0770.
Sets the maximum number of connections that can be handled in parallel. Since this many threads will be started immediately setting this to a very high value will waste valuable resources. Defaults to 5 and will be forced to be at most 16384 to prevent typos and dumb mistakes.

The ethstat plugin collects information about network interface cards (NICs) by talking directly with the underlying kernel driver using ioctl(2).

Synopsis:

 <Plugin "ethstat">
   Interface "eth0"
   Map "rx_csum_offload_errors" "if_rx_errors" "checksum_offload"
   Map "multicast" "if_multicast"
 </Plugin>

Options:

Collect statistical information about interface Name.
By default, the plugin will submit values as type "derive" and type instance set to Name, the name of the metric as reported by the driver. If an appropriate Map option exists, the given Type and, optionally, TypeInstance will be used.
When set to true, only metrics that can be mapped to to a type will be collected, all other metrics will be ignored. Defaults to false.

Please make sure to read collectd-exec(5) before using this plugin. It contains valuable information on when the executable is executed and the output that is expected from it.

Execute the executable Executable as user User. If the user name is followed by a colon and a group name, the effective group is set to that group. The real group and saved-set group will be set to the default group of that user. If no group is given the effective group ID will be the same as the real group ID.

Please note that in order to change the user and/or group the daemon needs superuser privileges. If the daemon is run as an unprivileged user you must specify the same user/group here. If the daemon is run with superuser privileges, you must supply a non-root user here.

The executable may be followed by optional arguments that are passed to the program. Please note that due to the configuration parsing numbers and boolean values may be changed. If you want to be absolutely sure that something is passed as-is please enclose it in quotes.

The Exec and NotificationExec statements change the semantics of the programs executed, i. e. the data passed to them and the response expected from them. This is documented in great detail in collectd-exec(5).

The "filecount" plugin counts the number of files in a certain directory (and its subdirectories) and their combined size. The configuration is very straight forward:

  <Plugin "filecount">
    <Directory "/var/qmail/queue/mess">
      Instance "qmail-message"
    </Directory>
    <Directory "/var/qmail/queue/todo">
      Instance "qmail-todo"
    </Directory>
    <Directory "/var/lib/php5">
      Instance "php5-sessions"
      Name "sess_*"
    </Directory>
  </Plugin>

The example above counts the number of files in QMail's queue directories and the number of PHP5 sessions. Jfiy: The "todo" queue holds the messages that QMail has not yet looked at, the "message" queue holds the messages that were classified into "local" and "remote".

As you can see, the configuration consists of one or more "Directory" blocks, each of which specifies a directory in which to count the files. Within those blocks, the following options are recognized:

Sets the plugin instance to Instance. That instance name must be unique, but it's your responsibility, the plugin doesn't check for that. If not given, the instance is set to the directory name with all slashes replaced by underscores and all leading underscores removed.
Only count files that match Pattern, where Pattern is a shell-like wildcard as understood by fnmatch(3). Only the filename is checked against the pattern, not the entire path. In case this makes it easier for you: This option has been named after the -name parameter to find(1).
Count only files of a specific age: If Age is greater than zero, only files that haven't been touched in the last Age seconds are counted. If Age is a negative number, this is inversed. For example, if -60 is specified, only files that have been modified in the last minute will be counted.

The number can also be followed by a "multiplier" to easily specify a larger timespan. When given in this notation, the argument must in quoted, i. e. must be passed as string. So the -60 could also be written as "-1m" (one minute). Valid multipliers are "s" (second), "m" (minute), "h" (hour), "d" (day), "w" (week), and "y" (year). There is no "month" multiplier. You can also specify fractional numbers, e. g. "0.5d" is identical to "12h".

Count only files of a specific size. When Size is a positive number, only files that are at least this big are counted. If Size is a negative number, this is inversed, i. e. only files smaller than the absolute value of Size are counted.

As with the MTime option, a "multiplier" may be added. For a detailed description see above. Valid multipliers here are "b" (byte), "k" (kilobyte), "m" (megabyte), "g" (gigabyte), "t" (terabyte), and "p" (petabyte). Please note that there are 1000 bytes in a kilobyte, not 1024.

Controls whether or not to recurse into subdirectories. Enabled by default.
Controls whether or not to include "hidden" files and directories in the count. "Hidden" files and directories are those, whose name begins with a dot. Defaults to false, i.e. by default hidden files and directories are ignored.

The GenericJMX plugin is written in Java and therefore documented in collectd-java(5).

The gmond plugin received the multicast traffic sent by gmond, the statistics collection daemon of Ganglia. Mappings for the standard "metrics" are built-in, custom mappings may be added via Metric blocks, see below.

Synopsis:

 <Plugin "gmond">
   MCReceiveFrom "239.2.11.71" "8649"
   <Metric "swap_total">
     Type "swap"
     TypeInstance "total"
     DataSource "value"
   </Metric>
   <Metric "swap_free">
     Type "swap"
     TypeInstance "free"
     DataSource "value"
   </Metric>
 </Plugin>

The following metrics are built-in:

  • load_one, load_five, load_fifteen
  • cpu_user, cpu_system, cpu_idle, cpu_nice, cpu_wio
  • mem_free, mem_shared, mem_buffers, mem_cached, mem_total
  • bytes_in, bytes_out
  • pkts_in, pkts_out

Available configuration options:

Sets sets the multicast group and UDP port to which to subscribe.

Default: 239.2.11.71 / 8649

<Metric Name>
These blocks add a new metric conversion to the internal table. Name, the string argument to the Metric block, is the metric name as used by Ganglia.
Type to map this metric to. Required.
Type-instance to use. Optional.
Data source to map this metric to. If the configured type has exactly one data source, this is optional. Otherwise the option is required.

To get values from hddtemp collectd connects to localhost (127.0.0.1), port 7634/tcp. The Host and Port options can be used to change these default values, see below. "hddtemp" has to be running to work correctly. If "hddtemp" is not running timeouts may appear which may interfere with other statistics..

The hddtemp homepage can be found at <http://www.guzu.net/linux/hddtemp.php>.

Hostname to connect to. Defaults to 127.0.0.1.
TCP-Port to connect to. Defaults to 7634.

Select this interface. By default these interfaces will then be collected. For a more detailed description see IgnoreSelected below.
If no configuration if given, the traffic-plugin will collect data from all interfaces. This may not be practical, especially for loopback- and similar interfaces. Thus, you can use the Interface-option to pick the interfaces you're interested in. Sometimes, however, it's easier/preferred to collect all interfaces except a few ones. This option enables you to do that: By setting IgnoreSelected to true the effect of Interface is inverted: All selected interfaces are ignored and all other interfaces are collected.

Selects sensors to collect or to ignore, depending on IgnoreSelected.
If no configuration if given, the ipmi plugin will collect data from all sensors found of type "temperature", "voltage", "current" and "fanspeed". This option enables you to do that: By setting IgnoreSelected to true the effect of Sensor is inverted: All selected sensors are ignored and all other sensors are collected.
If a sensor appears after initialization time of a minute a notification is sent.
If a sensor disappears a notification is sent.
If you have for example dual power supply and one of them is (un)plugged then a notification is sent.

Select the rules to count. If only Table and Chain are given, this plugin will collect the counters of all rules which have a comment-match. The comment is then used as type-instance.

If Comment or Number is given, only the rule with the matching comment or the nth rule will be collected. Again, the comment (or the number) will be used as the type-instance.

If Name is supplied, it will be used as the type-instance instead of the comment or the number.

Select this irq. By default these irqs will then be collected. For a more detailed description see IgnoreSelected below.
If no configuration if given, the irq-plugin will collect data from all irqs. This may not be practical, especially if no interrupts happen. Thus, you can use the Irq-option to pick the interrupt you're interested in. Sometimes, however, it's easier/preferred to collect all interrupts except a few ones. This option enables you to do that: By setting IgnoreSelected to true the effect of Irq is inverted: All selected interrupts are ignored and all other interrupts are collected.

The Java plugin makes it possible to write extensions for collectd in Java. This section only discusses the syntax and semantic of the configuration options. For more in-depth information on the Java plugin, please read collectd-java(5).

Synopsis:

 <Plugin "java">
   JVMArg "-verbose:jni"
   JVMArg "-Djava.class.path=/opt/collectd/lib/collectd/bindings/java"
   LoadPlugin "org.collectd.java.Foobar"
   <Plugin "org.collectd.java.Foobar">
     # To be parsed by the plugin
   </Plugin>
 </Plugin>

Available configuration options:

Argument that is to be passed to the Java Virtual Machine (JVM). This works exactly the way the arguments to the java binary on the command line work. Execute "java --help" for details.

Please note that all these options must appear before (i. e. above) any other options! When another option is found, the JVM will be started and later options will have to be ignored!

Instantiates a new JavaClass object. The constructor of this object very likely then registers one or more callback methods with the server.

See collectd-java(5) for details.

When the first such option is found, the virtual machine (JVM) is created. This means that all JVMArg options must appear before (i. e. above) all LoadPlugin options!

The entire block is passed to the Java plugin as an org.collectd.api.OConfigItem object.

For this to work, the plugin has to register a configuration callback first, see "config callback" in collectd-java(5). This means, that the Plugin block must appear after the appropriate LoadPlugin block. Also note, that Name depends on the (Java) plugin registering the callback and is completely independent from the JavaClass argument passed to LoadPlugin.

This plugin allows CPU, disk and network load to be collected for virtualized guests on the machine. This means that these characteristics can be collected for guest systems without installing any software on them - collectd only runs on the hosting system. The statistics are collected through libvirt (<http://libvirt.org/>).

Only Connection is required.

Connect to the hypervisor given by uri. For example if using Xen use: 

 Connection "xen:///"

Details which URIs allowed are given at <http://libvirt.org/uri.html>.

Refresh the list of domains and devices every seconds. The default is 60 seconds. Setting this to be the same or smaller than the Interval will cause the list of domains and devices to be refreshed on every iteration.

Refreshing the devices in particular is quite a costly operation, so if your virtualization setup is static you might consider increasing this. If this option is set to 0, refreshing is disabled completely.

Select which domains and devices are collected.

If IgnoreSelected is not given or false then only the listed domains and disk/network devices are collected.

If IgnoreSelected is true then the test is reversed and the listed domains and disk/network devices are ignored, while the rest are collected.

The domain name and device names may use a regular expression, if the name is surrounded by /.../ and collectd was compiled with support for regexps.

The default is to collect statistics for all domains and all their devices.

Example:

 BlockDevice "/:hdb/"
 IgnoreSelected "true"

Ignore all hdb devices on any domain, but other block devices (eg. hda) will be collected.

When the libvirt plugin logs data, it sets the hostname of the collected data according to this setting. The default is to use the guest name as provided by the hypervisor, which is equal to setting name.

uuid means use the guest's UUID. This is useful if you want to track the same guest across migrations.

hostname means to use the global Hostname setting, which is probably not useful on its own because all guests will appear to have the same name.

You can also specify combinations of these fields. For example name uuid means to concatenate the guest name and UUID (with a literal colon character between, thus "foo:1234-1234-1234-1234").

When the libvirt plugin logs interface data, it sets the name of the collected data according to this setting. The default is to use the path as provided by the hypervisor (the "dev" property of the target node), which is equal to setting name.

address means use the interface's mac address. This is useful since the interface path might change between reboots of a guest or across migrations.

Sets the log-level. If, for example, set to notice, then all events with severity notice, warning, or err will be written to the logfile.

Please note that debug is only available if collectd has been compiled with debugging support.

Sets the file to write log messages to. The special strings stdout and stderr can be used to write to the standard output and standard error channels, respectively. This, of course, only makes much sense when collectd is running in foreground- or non-daemon-mode.
Prefix all lines printed by the current time. Defaults to true.
When enabled, all lines are prefixed by the severity of the log message, for example "warning". Defaults to false.

Note: There is no need to notify the daemon after moving or removing the log file (e. g. when rotating the logs). The plugin reopens the file for each line it writes.

The LPAR plugin reads CPU statistics of Logical Partitions, a virtualization technique for IBM POWER processors. It takes into account CPU time stolen from or donated to a partition, in addition to the usual user, system, I/O statistics.

The following configuration options are available:

When enabled, statistics about the processor pool are read, too. The partition needs to have pool authority in order to be able to acquire this information. Defaults to false.
If enabled, the serial of the physical machine the partition is currently running on is reported as hostname and the logical hostname of the machine is reported in the plugin instance. Otherwise, the logical hostname will be used (just like other plugins) and the plugin instance will be empty. Defaults to false.

The "mbmon plugin" uses mbmon to retrieve temperature, voltage, etc.

Be default collectd connects to localhost (127.0.0.1), port 411/tcp. The Host and Port options can be used to change these values, see below. "mbmon" has to be running to work correctly. If "mbmon" is not running timeouts may appear which may interfere with other statistics..

"mbmon" must be run with the -r option ("print TAG and Value format"); Debian's /etc/init.d/mbmon script already does this, other people will need to ensure that this is the case.

Hostname to connect to. Defaults to 127.0.0.1.
TCP-Port to connect to. Defaults to 411.

The "md plugin" collects information from Linux Software-RAID devices (md).

All reported values are of the type "md_disks". Reported type instances are active, failed (present but not operational), spare (hot stand-by) and missing (physically absent) disks.

Select md devices based on device name. The device name is the basename of the device, i.e. the name of the block device without the leading "/dev/". See IgnoreSelected for more details.
Invert device selection: If set to true, all md devices except those listed using Device are collected. If false (the default), only those listed are collected. If no configuration is given, the md plugin will collect data from all md devices.

The "memcachec plugin" connects to a memcached server, queries one or more given pages and parses the returned data according to user specification. The matches used are the same as the matches used in the "curl" and "tail" plugins.

In order to talk to the memcached server, this plugin uses the libmemcached library. Please note that there is another library with a very similar name, libmemcache (notice the missing `d'), which is not applicable.

Synopsis of the configuration:

 <Plugin "memcachec">
   <Page "plugin_instance">
     Server "localhost"
     Key "page_key"
     <Match>
       Regex "(\\d+) bytes sent"
       DSType CounterAdd
       Type "ipt_octets"
       Instance "type_instance"
     </Match>
   </Page>
 </Plugin>

The configuration options are:

<Page Name>
Each Page block defines one page to be queried from the memcached server. The block requires one string argument which is used as plugin instance.
Sets the server address to connect to when querying the page. Must be inside a Page block.
When connected to the memcached server, asks for the page Key.
<Match>
Match blocks define which strings to look for and how matches substrings are interpreted. For a description of match blocks, please see "Plugin tail".

The memcached plugin connects to a memcached server and queries statistics about cache utilization, memory and bandwidth used. <http://www.danga.com/memcached/>

 <Plugin "memcached">
   <Instance "name">
     Host "memcache.example.com"
     Port 11211
   </Instance>
 </Plugin>

The plugin configuration consists of one or more Instance blocks which specify one memcached connection each. Within the Instance blocks, the following options are allowed:

Hostname to connect to. Defaults to 127.0.0.1.
TCP-Port to connect to. Defaults to 11211.
Connect to memcached using the UNIX domain socket at Path. If this setting is given, the Host and Port settings are ignored.

The mic plugin gathers CPU statistics, memory usage and temperatures from Intel's Many Integrated Core (MIC) systems.

Synopsis:

 <Plugin mic>
   ShowCPU true
   ShowCPUCores true
   ShowMemory true
   
   ShowTemperatures true
   Temperature vddg
   Temperature vddq
   IgnoreSelectedTemperature true
   ShowPower true
   Power total0
   Power total1
   IgnoreSelectedPower true   
 </Plugin>

The following options are valid inside the Plugin mic block:

If enabled (the default) a sum of the CPU usage accross all cores is reported.
If enabled (the default) per-core CPU usage is reported.
If enabled (the default) the physical memory usage of the MIC system is reported.
If enabled (the default) various temperatures of the MIC system are reported.
This option controls which temperatures are being reported. Whether matching temperatures are being ignored or only matching temperatures are reported depends on the IgnoreSelectedTemperature setting below. By default all temperatures are reported.
Controls the behavior of the Temperature setting above. If set to false (the default) only temperatures matching a Temperature option are reported or, if no Temperature option is specified, all temperatures are reported. If set to true, matching temperatures are ignored and all other temperatures are reported.

Known temperature names are:

Die of the CPU
Device Memory
Fan In
Fan Out
Voltage ccp
Voltage ddg
Voltage ddq
If enabled (the default) various temperatures of the MIC system are reported.
This option controls which power readings are being reported. Whether matching power readings are being ignored or only matching power readings are reported depends on the IgnoreSelectedPower setting below. By default all power readings are reported.
Controls the behavior of the Power setting above. If set to false (the default) only power readings matching a Power option are reported or, if no Power option is specified, all power readings are reported. If set to true, matching power readings are ignored and all other power readings are reported.

Known power names are:

Total power utilization averaged over Time Window 0 (uWatts).
Total power utilization averaged over Time Window 0 (uWatts).
Instantaneous power (uWatts).
Max instantaneous power (uWatts).
PCI-E connector power (uWatts).
2x3 connector power (uWatts).
2x4 connector power (uWatts).
Core rail (uVolts).
Uncore rail (uVolts).
Memory subsystem rail (uVolts).

The modbus plugin connects to a Modbus "slave" via Modbus/TCP and reads register values. It supports reading single registers (unsigned 16 bit values), large integer values (unsigned 32 bit values) and floating point values (two registers interpreted as IEEE floats in big endian notation).

Synopsis:

 <Data "voltage-input-1">
   RegisterBase 0
   RegisterType float
   Type voltage
   Instance "input-1"
 </Data>
 
 <Data "voltage-input-2">
   RegisterBase 2
   RegisterType float
   Type voltage
   Instance "input-2"
 </Data>
 
 <Host "modbus.example.com">
   Address "192.168.0.42"
   Port    "502"
   Interval 60
   
   <Slave 1>
     Instance "power-supply"
     Collect  "voltage-input-1"
     Collect  "voltage-input-2"
   </Slave>
 </Host>
<Data Name> blocks
Data blocks define a mapping between register numbers and the "types" used by collectd.

Within <Data /> blocks, the following options are allowed:

Configures the base register to read from the device. If the option RegisterType has been set to Uint32 or Float, this and the next register will be read (the register number is increased by one).
Specifies what kind of data is returned by the device. If the type is Int32, Uint32 or Float, two 16 bit registers will be read and the data is combined into one value. Defaults to Uint16.
Specifies the "type" (data set) to use when dispatching the value to collectd. Currently, only data sets with exactly one data source are supported.
Sets the type instance to use when dispatching the value to collectd. If unset, an empty string (no type instance) is used.
<Host Name> blocks
Host blocks are used to specify to which hosts to connect and what data to read from their "slaves". The string argument Name is used as hostname when dispatching the values to collectd.

Within <Host /> blocks, the following options are allowed:

Specifies the node name (the actual network address) used to connect to the host. This may be an IP address or a hostname. Please note that the used libmodbus library only supports IPv4 at the moment.
Specifies the port used to connect to the host. The port can either be given as a number or as a service name. Please note that the Service argument must be a string, even if ports are given in their numerical form. Defaults to "502".
Sets the interval (in seconds) in which the values will be collected from this host. By default the global Interval setting will be used.
<Slave ID>
Over each TCP connection, multiple Modbus devices may be reached. The slave ID is used to specify which device should be addressed. For each device you want to query, one Slave block must be given.

Within <Slave /> blocks, the following options are allowed:

Specify the plugin instance to use when dispatching the values to collectd. By default "slave_ID" is used.
Specifies which data to retrieve from the device. DataName must be the same string as the Name argument passed to a Data block. You can specify this option multiple times to collect more than one value from a slave. At least one Collect option is mandatory.

The "mysql plugin" requires mysqlclient to be installed. It connects to one or more databases when started and keeps the connection up as long as possible. When the connection is interrupted for whatever reason it will try to re-connect. The plugin will complain loudly in case anything goes wrong.

This plugin issues the MySQL "SHOW STATUS" / "SHOW GLOBAL STATUS" command and collects information about MySQL network traffic, executed statements, requests, the query cache and threads by evaluating the "Bytes_{received,sent}", "Com_*", "Handler_*", "Qcache_*" and "Threads_*" return values. Please refer to the MySQL reference manual, 5.1.6. Server Status Variables for an explanation of these values.

Optionally, master and slave statistics may be collected in a MySQL replication setup. In that case, information about the synchronization state of the nodes are collected by evaluating the "Position" return value of the "SHOW MASTER STATUS" command and the "Seconds_Behind_Master", "Read_Master_Log_Pos" and "Exec_Master_Log_Pos" return values of the "SHOW SLAVE STATUS" command. See the MySQL reference manual, 12.5.5.21 SHOW MASTER STATUS Syntax and 12.5.5.31 SHOW SLAVE STATUS Syntax for details.

Synopsis:

  <Plugin mysql>
    <Database foo>
      Host "hostname"
      User "username"
      Password "password"
      Port "3306"
      MasterStats true
    </Database>
    <Database bar>
      Host "localhost"
      Socket "/var/run/mysql/mysqld.sock"
      SlaveStats true
      SlaveNotifications true
    </Database>
  </Plugin>

A Database block defines one connection to a MySQL database. It accepts a single argument which specifies the name of the database. None of the other options are required. MySQL will use default values as documented in the section "mysql_real_connect()" in the MySQL reference manual.

Hostname of the database server. Defaults to localhost.
Username to use when connecting to the database. The user does not have to be granted any privileges (which is synonym to granting the "USAGE" privilege), unless you want to collectd replication statistics (see MasterStats and SlaveStats below). In this case, the user needs the "REPLICATION CLIENT" (or "SUPER") privileges. Else, any existing MySQL user will do.
Password needed to log into the database.
Select this database. Defaults to no database which is a perfectly reasonable option for what this plugin does.
TCP-port to connect to. The port must be specified in its numeric form, but it must be passed as a string nonetheless. For example: 

  Port "3306"

If Host is set to localhost (the default), this setting has no effect. See the documentation for the "mysql_real_connect" function for details.

Specifies the path to the UNIX domain socket of the MySQL server. This option only has any effect, if Host is set to localhost (the default). Otherwise, use the Port option above. See the documentation for the "mysql_real_connect" function for details.
Enable the collection of master / slave statistics in a replication setup. In order to be able to get access to these statistics, the user needs special privileges. See the User documentation above.
If enabled, the plugin sends a notification if the replication slave I/O and / or SQL threads are not running.

The netapp plugin can collect various performance and capacity information from a NetApp filer using the NetApp API.

Please note that NetApp has a wide line of products and a lot of different software versions for each of these products. This plugin was developed for a NetApp FAS3040 running OnTap 7.2.3P8 and tested on FAS2050 7.3.1.1L1, FAS3140 7.2.5.1 and FAS3020 7.2.4P9. It should work for most combinations of model and software version but it is very hard to test this. If you have used this plugin with other models and/or software version, feel free to send us a mail to tell us about the results, even if it's just a short "It works".

To collect these data collectd will log in to the NetApp via HTTP(S) and HTTP basic authentication.

Do not use a regular user for this! Create a special collectd user with just the minimum of capabilities needed. The user only needs the "login-http-admin" capability as well as a few more depending on which data will be collected. Required capabilities are documented below.

Synopsis

 <Plugin "netapp">
   <Host "netapp1.example.com">
    Protocol      "https"
    Address       "10.0.0.1"
    Port          443
    User          "username"
    Password      "aef4Aebe"
    Interval      30
    
    <WAFL>
      Interval 30
      GetNameCache   true
      GetDirCache    true
      GetBufferCache true
      GetInodeCache  true
    </WAFL>
    
    <Disks>
      Interval 30
      GetBusy true
    </Disks>
    
    <VolumePerf>
      Interval 30
      GetIO      "volume0"
      IgnoreSelectedIO      false
      GetOps     "volume0"
      IgnoreSelectedOps     false
      GetLatency "volume0"
      IgnoreSelectedLatency false
    </VolumePerf>
    
    <VolumeUsage>
      Interval 30
      GetCapacity "vol0"
      GetCapacity "vol1"
      IgnoreSelectedCapacity false
      GetSnapshot "vol1"
      GetSnapshot "vol3"
      IgnoreSelectedSnapshot false
    </VolumeUsage>
    
    <Quota>
      Interval 60
    </Quota>
    
    <Snapvault>
      Interval 30
    </Snapvault>
    
    <System>
      Interval 30
      GetCPULoad     true
      GetInterfaces  true
      GetDiskOps     true
      GetDiskIO      true
    </System>
    <VFiler vfilerA>
      Interval 60
      SnapVault true
      # ...
    </VFiler>
   </Host>
 </Plugin>

The netapp plugin accepts the following configuration options:

A host block defines one NetApp filer. It will appear in collectd with the name you specify here which does not have to be its real name nor its hostname (see the Address option below).
A VFiler block may only be used inside a host block. It accepts all the same options as the Host block (except for cascaded VFiler blocks) and will execute all NetApp API commands in the context of the specified VFiler(R). It will appear in collectd with the name you specify here which does not have to be its real name. The VFiler name may be specified using the VFilerName option. If this is not specified, it will default to the name you specify here.

The VFiler block inherits all connection related settings from the surrounding Host block (which appear before the VFiler block) but they may be overwritten inside the VFiler block.

This feature is useful, for example, when using a VFiler as SnapVault target (supported since OnTap 8.1). In that case, the SnapVault statistics are not available in the host filer (vfiler0) but only in the respective VFiler context.

The protocol collectd will use to query this host.

Optional

Type: string

Default: https

Valid options: http, https

The hostname or IP address of the host.

Optional

Type: string

Default: The "host" block's name.

The TCP port to connect to on the host.

Optional

Type: integer

Default: 80 for protocol "http", 443 for protocol "https"

The username and password to use to login to the NetApp.

Mandatory

Type: string

The name of the VFiler in which context to execute API commands. If not specified, the name provided to the VFiler block will be used instead.

Optional

Type: string

Default: name of the VFiler block

Note: This option may only be used inside VFiler blocks.

TODO

The following options decide what kind of data will be collected. You can either use them as a block and fine tune various parameters inside this block, use them as a single statement to just accept all default values, or omit it to not collect any data.

The following options are valid inside all blocks:

Collect the respective statistics every Seconds seconds. Defaults to the host specific setting.

The System block

This will collect various performance data about the whole system.

Note: To get this data the collectd user needs the "api-perf-object-get-instances" capability.

Collect disk statistics every Seconds seconds.
If you set this option to true the current CPU usage will be read. This will be the average usage between all CPUs in your NetApp without any information about individual CPUs.

Note: These are the same values that the NetApp CLI command "sysstat" returns in the "CPU" field.

Optional

Type: boolean

Default: true

Result: Two value lists of type "cpu", and type instances "idle" and "system".

If you set this option to true the current traffic of the network interfaces will be read. This will be the total traffic over all interfaces of your NetApp without any information about individual interfaces.

Note: This is the same values that the NetApp CLI command "sysstat" returns in the "Net kB/s" field.

Or is it?

Optional

Type: boolean

Default: true

Result: One value list of type "if_octects".

If you set this option to true the current IO throughput will be read. This will be the total IO of your NetApp without any information about individual disks, volumes or aggregates.

Note: This is the same values that the NetApp CLI command "sysstat" returns in the "Disk kB/s" field.

Optional

Type: boolean

Default: true

Result: One value list of type "disk_octets".

If you set this option to true the current number of HTTP, NFS, CIFS, FCP, iSCSI, etc. operations will be read. This will be the total number of operations on your NetApp without any information about individual volumes or aggregates.

Note: These are the same values that the NetApp CLI command "sysstat" returns in the "NFS", "CIFS", "HTTP", "FCP" and "iSCSI" fields.

Optional

Type: boolean

Default: true

Result: A variable number of value lists of type "disk_ops_complex". Each type of operation will result in one value list with the name of the operation as type instance.

The WAFL block

This will collect various performance data about the WAFL file system. At the moment this just means cache performance.

Note: To get this data the collectd user needs the "api-perf-object-get-instances" capability.

Note: The interface to get these values is classified as "Diagnostics" by NetApp. This means that it is not guaranteed to be stable even between minor releases.

Collect disk statistics every Seconds seconds.
Optional

Type: boolean

Default: true

Result: One value list of type "cache_ratio" and type instance "name_cache_hit".

Optional

Type: boolean

Default: true

Result: One value list of type "cache_ratio" and type instance "find_dir_hit".

Optional

Type: boolean

Default: true

Result: One value list of type "cache_ratio" and type instance "inode_cache_hit".

Note: This is the same value that the NetApp CLI command "sysstat" returns in the "Cache hit" field.

Optional

Type: boolean

Default: true

Result: One value list of type "cache_ratio" and type instance "buf_hash_hit".

The Disks block

This will collect performance data about the individual disks in the NetApp.

Note: To get this data the collectd user needs the "api-perf-object-get-instances" capability.

Collect disk statistics every Seconds seconds.
If you set this option to true the busy time of all disks will be calculated and the value of the busiest disk in the system will be written.

Note: This is the same values that the NetApp CLI command "sysstat" returns in the "Disk util" field. Probably.

Optional

Type: boolean

Default: true

Result: One value list of type "percent" and type instance "disk_busy".

The VolumePerf block

This will collect various performance data about the individual volumes.

You can select which data to collect about which volume using the following options. They follow the standard ignorelist semantic.

Note: To get this data the collectd user needs the api-perf-object-get-instances capability.

Collect volume performance data every Seconds seconds.
Select the given volume for IO, operations or latency statistics collection. The argument is the name of the volume without the "/vol/" prefix.

Since the standard ignorelist functionality is used here, you can use a string starting and ending with a slash to specify regular expression matching: To match the volumes "vol0", "vol2" and "vol7", you can use this regular expression:

  GetIO "/^vol[027]$/"

If no regular expression is specified, an exact match is required. Both, regular and exact matching are case sensitive.

If no volume was specified at all for either of the three options, that data will be collected for all available volumes.

When set to true, the volumes selected for IO, operations or latency statistics collection will be ignored and the data will be collected for all other volumes.

When set to false, data will only be collected for the specified volumes and all other volumes will be ignored.

If no volumes have been specified with the above Get* options, all volumes will be collected regardless of the IgnoreSelected* option.

Defaults to false

The VolumeUsage block

This will collect capacity data about the individual volumes.

Note: To get this data the collectd user needs the api-volume-list-info capability.

Collect volume usage statistics every Seconds seconds.
The current capacity of the volume will be collected. This will result in two to four value lists, depending on the configuration of the volume. All data sources are of type "df_complex" with the name of the volume as plugin_instance.

There will be type_instances "used" and "free" for the number of used and available bytes on the volume. If the volume has some space reserved for snapshots, a type_instance "snap_reserved" will be available. If the volume has SIS enabled, a type_instance "sis_saved" will be available. This is the number of bytes saved by the SIS feature.

Note: The current NetApp API has a bug that results in this value being reported as a 32 bit number. This plugin tries to guess the correct number which works most of the time. If you see strange values here, bug NetApp support to fix this.

Repeat this option to specify multiple volumes.

Specify whether to collect only the volumes selected by the GetCapacity option or to ignore those volumes. IgnoreSelectedCapacity defaults to false. However, if no GetCapacity option is specified at all, all capacities will be selected anyway.
Select volumes from which to collect snapshot information.

Usually, the space used for snapshots is included in the space reported as "used". If snapshot information is collected as well, the space used for snapshots is subtracted from the used space.

To make things even more interesting, it is possible to reserve space to be used for snapshots. If the space required for snapshots is less than that reserved space, there is "reserved free" and "reserved used" space in addition to "free" and "used". If the space required for snapshots exceeds the reserved space, that part allocated in the normal space is subtracted from the "used" space again.

Repeat this option to specify multiple volumes.

Specify whether to collect only the volumes selected by the GetSnapshot option or to ignore those volumes. IgnoreSelectedSnapshot defaults to false. However, if no GetSnapshot option is specified at all, all capacities will be selected anyway.

The Quota block

This will collect (tree) quota statistics (used disk space and number of used files). This mechanism is useful to get usage information for single qtrees. In case the quotas are not used for any other purpose, an entry similar to the following in "/etc/quotas" would be sufficient:

  /vol/volA/some_qtree tree - - - - -

After adding the entry, issue "quota on -w volA" on the NetApp filer.

Collect SnapVault(R) statistics every Seconds seconds.

The SnapVault block

This will collect statistics about the time and traffic of SnapVault(R) transfers.

Collect SnapVault(R) statistics every Seconds seconds.

The "netlink" plugin uses a netlink socket to query the Linux kernel about statistics of various interface and routing aspects.

Instruct the plugin to collect interface statistics. This is basically the same as the statistics provided by the "interface" plugin (see above) but potentially much more detailed.

When configuring with Interface only the basic statistics will be collected, namely octets, packets, and errors. These statistics are collected by the "interface" plugin, too, so using both at the same time is no benefit.

When configured with VerboseInterface all counters except the basic ones, so that no data needs to be collected twice if you use the "interface" plugin. This includes dropped packets, received multicast packets, collisions and a whole zoo of differentiated RX and TX errors. You can try the following command to get an idea of what awaits you:

  ip -s -s link list

If Interface is All, all interfaces will be selected.

Collect the octets and packets that pass a certain qdisc, class or filter.

QDiscs and classes are identified by their type and handle (or classid). Filters don't necessarily have a handle, therefore the parent's handle is used. The notation used in collectd differs from that used in tc(1) in that it doesn't skip the major or minor number if it's zero and doesn't print special ids by their name. So, for example, a qdisc may be identified by "pfifo_fast-1:0" even though the minor number of all qdiscs is zero and thus not displayed by tc(1).

If QDisc, Class, or Filter is given without the second argument, i. .e. without an identifier, all qdiscs, classes, or filters that are associated with that interface will be collected.

Since a filter itself doesn't necessarily have a handle, the parent's handle is used. This may lead to problems when more than one filter is attached to a qdisc or class. This isn't nice, but we don't know how this could be done any better. If you have a idea, please don't hesitate to tell us.

As with the Interface option you can specify All as the interface, meaning all interfaces.

Here are some examples to help you understand the above text more easily:

  <Plugin netlink>
    VerboseInterface "All"
    QDisc "eth0" "pfifo_fast-1:0"
    QDisc "ppp0"
    Class "ppp0" "htb-1:10"
    Filter "ppp0" "u32-1:0"
  </Plugin>
The behavior is the same as with all other similar plugins: If nothing is selected at all, everything is collected. If some things are selected using the options described above, only these statistics are collected. If you set IgnoreSelected to true, this behavior is inverted, i. e. the specified statistics will not be collected.

The Network plugin sends data to a remote instance of collectd, receives data from a remote instance, or both at the same time. Data which has been received from the network is usually not transmitted again, but this can be activated, see the Forward option below.

The default IPv6 multicast group is "ff18::efc0:4a42". The default IPv4 multicast group is 239.192.74.66. The default UDP port is 25826.

Both, Server and Listen can be used as single option or as block. When used as block, given options are valid for this socket only. The following example will export the metrics twice: Once to an "internal" server (without encryption and signing) and one to an external server (with cryptographic signature):

 <Plugin "network">
   # Export to an internal server
   # (demonstrates usage without additional options)
   Server "collectd.internal.tld"
   
   # Export to an external server
   # (demonstrates usage with signature options)
   <Server "collectd.external.tld">
     SecurityLevel "sign"
     Username "myhostname"
     Password "ohl0eQue"
   </Server>
 </Plugin>
<Server Host [Port]>
The Server statement/block sets the server to send datagrams to. The statement may occur multiple times to send each datagram to multiple destinations.

The argument Host may be a hostname, an IPv4 address or an IPv6 address. The optional second argument specifies a port number or a service name. If not given, the default, 25826, is used.

The following options are recognized within Server blocks:

Set the security you require for network communication. When the security level has been set to Encrypt, data sent over the network will be encrypted using AES-256. The integrity of encrypted packets is ensured using SHA-1. When set to Sign, transmitted data is signed using the HMAC-SHA-256 message authentication code. When set to None, data is sent without any security.

This feature is only available if the network plugin was linked with libgcrypt.

Sets the username to transmit. This is used by the server to lookup the password. See AuthFile below. All security levels except None require this setting.

This feature is only available if the network plugin was linked with libgcrypt.

Sets a password (shared secret) for this socket. All security levels except None require this setting.

This feature is only available if the network plugin was linked with libgcrypt.

Set the outgoing interface for IP packets. This applies at least to IPv6 packets and if possible to IPv4. If this option is not applicable, undefined or a non-existent interface name is specified, the default behavior is to let the kernel choose the appropriate interface. Be warned that the manual selection of an interface for unicast traffic is only necessary in rare cases.
<Listen Host [Port]>
The Listen statement sets the interfaces to bind to. When multiple statements are found the daemon will bind to multiple interfaces.

The argument Host may be a hostname, an IPv4 address or an IPv6 address. If the argument is a multicast address the daemon will join that multicast group. The optional second argument specifies a port number or a service name. If not given, the default, 25826, is used.

The following options are recognized within "<Listen>" blocks:

Set the security you require for network communication. When the security level has been set to Encrypt, only encrypted data will be accepted. The integrity of encrypted packets is ensured using SHA-1. When set to Sign, only signed and encrypted data is accepted. When set to None, all data will be accepted. If an AuthFile option was given (see below), encrypted data is decrypted if possible.

This feature is only available if the network plugin was linked with libgcrypt.

Sets a file in which usernames are mapped to passwords. These passwords are used to verify signatures and to decrypt encrypted network packets. If SecurityLevel is set to None, this is optional. If given, signed data is verified and encrypted packets are decrypted. Otherwise, signed data is accepted without checking the signature and encrypted data cannot be decrypted. For the other security levels this option is mandatory.

The file format is very simple: Each line consists of a username followed by a colon and any number of spaces followed by the password. To demonstrate, an example file could look like this:

  user0: foo
  user1: bar

Each time a packet is received, the modification time of the file is checked using stat(2). If the file has been changed, the contents is re-read. While the file is being read, it is locked using fcntl(2).

Set the incoming interface for IP packets explicitly. This applies at least to IPv6 packets and if possible to IPv4. If this option is not applicable, undefined or a non-existent interface name is specified, the default behavior is, to let the kernel choose the appropriate interface. Thus incoming traffic gets only accepted, if it arrives on the given interface.
Set the time-to-live of sent packets. This applies to all, unicast and multicast, and IPv4 and IPv6 packets. The default is to not change this value. That means that multicast packets will be sent with a TTL of 1 (one) on most operating systems.
Set the maximum size for datagrams received over the network. Packets larger than this will be truncated. Defaults to 1452 bytes, which is the maximum payload size that can be transmitted in one Ethernet frame using IPv6 / UDP.

On the server side, this limit should be set to the largest value used on any client. Likewise, the value on the client must not be larger than the value on the server, or data will be lost.

Compatibility: Versions prior to version 4.8 used a fixed sized buffer of 1024 bytes. Versions 4.8, 4.9 and 4.10 used a default value of 1024 bytes to avoid problems when sending data to an older server.

If set to true, write packets that were received via the network plugin to the sending sockets. This should only be activated when the Listen- and Server-statements differ. Otherwise packets may be send multiple times to the same multicast group. While this results in more network traffic than necessary it's not a huge problem since the plugin has a duplicate detection, so the values will not loop.
The network plugin cannot only receive and send statistics, it can also create statistics about itself. Collected data included the number of received and sent octets and packets, the length of the receive queue and the number of values handled. When set to true, the Network plugin will make these statistics available. Defaults to false.

This plugin collects the number of connections and requests handled by the "nginx daemon" (speak: engine X), a HTTP and mail server/proxy. It queries the page provided by the "ngx_http_stub_status_module" module, which isn't compiled by default. Please refer to <http://wiki.codemongers.com/NginxStubStatusModule> for more information on how to compile and configure nginx and this module.

The following options are accepted by the "nginx plugin":

Sets the URL of the "ngx_http_stub_status_module" output.
Optional user name needed for authentication.
Optional password needed for authentication.
Enable or disable peer SSL certificate verification. See <http://curl.haxx.se/docs/sslcerts.html> for details. Enabled by default.
Enable or disable peer host name verification. If enabled, the plugin checks if the "Common Name" or a "Subject Alternate Name" field of the SSL certificate matches the host name provided by the URL option. If this identity check fails, the connection is aborted. Obviously, only works when connecting to a SSL enabled server. Enabled by default.
File that holds one or more SSL certificates. If you want to use HTTPS you will possibly need this option. What CA certificates come bundled with "libcurl" and are checked by default depends on the distribution you use.

This plugin sends a desktop notification to a notification daemon, as defined in the Desktop Notification Specification. To actually display the notifications, notification-daemon is required and collectd has to be able to access the X server (i. e., the "DISPLAY" and "XAUTHORITY" environment variables have to be set correctly) and the D-Bus message bus.

The Desktop Notification Specification can be found at <http://www.galago-project.org/specs/notification/>.

Set the timeout, in milliseconds, after which to expire the notification for "OKAY", "WARNING" and "FAILURE" severities respectively. If zero has been specified, the displayed notification will not be closed at all - the user has to do so herself. These options default to 5000. If a negative number has been specified, the default is used as well.

The notify_email plugin uses the ESMTP library to send notifications to a configured email address.

libESMTP is available from <http://www.stafford.uklinux.net/libesmtp/>.

Available configuration options:

Email address from which the emails should appear to come from.

Default: "root@localhost"

Configures the email address(es) to which the notifications should be mailed. May be repeated to send notifications to multiple addresses.

At least one Recipient must be present for the plugin to work correctly.

Hostname of the SMTP server to connect to.

Default: "localhost"

TCP port to connect to.

Default: 25

Username for ASMTP authentication. Optional.
Password for ASMTP authentication. Optional.
Subject-template to use when sending emails. There must be exactly two string-placeholders in the subject, given in the standard printf(3) syntax, i. e. %s. The first will be replaced with the severity, the second with the hostname.

Default: "Collectd notify: %s@%s"

Hostname of the host running ntpd. Defaults to localhost.
UDP-Port to connect to. Defaults to 123.
Sets whether or not to perform reverse lookups on peers. Since the name or IP-address may be used in a filename it is recommended to disable reverse lookups. The default is to do reverse lookups to preserve backwards compatibility, though.
When a peer is a refclock, include the unit ID in the type instance. Defaults to false for backward compatibility.

If two refclock peers use the same driver and this is false, the plugin will try to write simultaneous measurements from both to the same type instance. This will result in error messages in the log and only one set of measurements making it through.

Add a UPS to collect data from. The format is identical to the one accepted by upsc(8).

The olsrd plugin connects to the TCP port opened by the txtinfo plugin of the Optimized Link State Routing daemon and reads information about the current state of the meshed network.

The following configuration options are understood:

Connect to Host. Defaults to "localhost".
Specifies the port to connect to. This must be a string, even if you give the port as a number rather than a service name. Defaults to "2006".
Specifies what information to collect about links, i. e. direct connections of the daemon queried. If set to No, no information is collected. If set to Summary, the number of links and the average of all link quality (LQ) and neighbor link quality (NLQ) values is calculated. If set to Detail LQ and NLQ are collected per link.

Defaults to Detail.

Specifies what information to collect about routes of the daemon queried. If set to No, no information is collected. If set to Summary, the number of routes and the average metric and ETX is calculated. If set to Detail metric and ETX are collected per route.

Defaults to Summary.

Specifies what information to collect about the global topology. If set to No, no information is collected. If set to Summary, the number of links in the entire topology and the average link quality (LQ) is calculated. If set to Detail LQ and NLQ are collected for each link in the entire topology.

Defaults to Summary.

EXPERIMENTAL! See notes below.

The "onewire" plugin uses the owcapi library from the owfs project <http://owfs.org/> to read sensors connected via the onewire bus.

Currently only temperature sensors (sensors with the family code 10, e. g. DS1820, DS18S20, DS1920) can be read. If you have other sensors you would like to have included, please send a sort request to the mailing list.

Hubs (the DS2409 chips) are working, but read the note, why this plugin is experimental, below.

Sets the device to read the values from. This can either be a "real" hardware device, such as a serial port or an USB port, or the address of the owserver(1) socket, usually localhost:4304.

Though the documentation claims to automatically recognize the given address format, with version 2.7p4 we had to specify the type explicitly. So with that version, the following configuration worked for us:

  <Plugin onewire>
    Device "-s localhost:4304"
  </Plugin>

This directive is required and does not have a default value.

Selects sensors to collect or to ignore, depending on IgnoreSelected, see below. Sensors are specified without the family byte at the beginning, to you'd use "F10FCA000800", and not include the leading 10. family byte and point.
If no configuration if given, the onewire plugin will collect data from all sensors found. This may not be practical, especially if sensors are added and removed regularly. Sometimes, however, it's easier/preferred to collect only specific sensors or all sensors except a few specified ones. This option enables you to do that: By setting IgnoreSelected to true the effect of Sensor is inverted: All selected interfaces are ignored and all other interfaces are collected.
Sets the interval in which all sensors should be read. If not specified, the global Interval setting is used.

EXPERIMENTAL! The "onewire" plugin is experimental, because it doesn't yet work with big setups. It works with one sensor being attached to one controller, but as soon as you throw in a couple more senors and maybe a hub or two, reading all values will take more than ten seconds (the default interval). We will probably add some separate thread for reading the sensors and some cache or something like that, but it's not done yet. We will try to maintain backwards compatibility in the future, but we can't promise. So in short: If it works for you: Great! But keep in mind that the config might change, though this is unlikely. Oh, and if you want to help improving this plugin, just send a short notice to the mailing list. Thanks :)

The OpenVPN plugin reads a status file maintained by OpenVPN and gathers traffic statistics about connected clients.

To set up OpenVPN to write to the status file periodically, use the --status option of OpenVPN. Since OpenVPN can write two different formats, you need to set the required format, too. This is done by setting --status-version to 2.

So, in a nutshell you need:

  openvpn $OTHER_OPTIONS \
    --status "/var/run/openvpn-status" 10 \
    --status-version 2

Available options:

Specifies the location of the status file.
When enabled, the filename of the status file will be used as plugin instance and the client's "common name" will be used as type instance. This is required when reading multiple status files. Enabling this option is recommended, but to maintain backwards compatibility this option is disabled by default.
Sets whether or not statistics about the compression used by OpenVPN should be collected. This information is only available in single mode. Enabled by default.
Sets whether or not traffic information is collected for each connected client individually. If set to false, currently no traffic data is collected at all because aggregating this data in a save manner is tricky. Defaults to true.
When enabled, the number of currently connected clients or users is collected. This is especially interesting when CollectIndividualUsers is disabled, but can be configured independently from that option. Defaults to false.

The "oracle" plugin uses the OracleX Call Interface (OCI) to connect to an OracleX Database and lets you execute SQL statements there. It is very similar to the "dbi" plugin, because it was written around the same time. See the "dbi" plugin's documentation above for details.

  <Plugin oracle>
    <Query "out_of_stock">
      Statement "SELECT category, COUNT(*) AS value FROM products WHERE in_stock = 0 GROUP BY category"
      <Result>
        Type "gauge"
        # InstancePrefix "foo"
        InstancesFrom "category"
        ValuesFrom "value"
      </Result>
    </Query>
    <Database "product_information">
      ConnectID "db01"
      Username "oracle"
      Password "secret"
      Query "out_of_stock"
    </Database>
  </Plugin>

Query blocks

The Query blocks are handled identically to the Query blocks of the "dbi" plugin. Please see its documentation above for details on how to specify queries.

Database blocks

Database blocks define a connection to a database and which queries should be sent to that database. Each database needs a "name" as string argument in the starting tag of the block. This name will be used as "PluginInstance" in the values submitted to the daemon. Other than that, that name is not used.

Defines the "database alias" or "service name" to connect to. Usually, these names are defined in the file named "$ORACLE_HOME/network/admin/tnsnames.ora".
Hostname to use when dispatching values for this database. Defaults to using the global hostname of the collectd instance.
Username used for authentication.
Password used for authentication.
Associates the query named QueryName with this database connection. The query needs to be defined before this statement, i. e. all query blocks you want to refer to must be placed above the database block you want to refer to them from.

This plugin embeds a Perl-interpreter into collectd and provides an interface to collectd's plugin system. See collectd-perl(5) for its documentation.

The Pinba plugin receives profiling information from Pinba, an extension for the PHP interpreter. At the end of executing a script, i.e. after a PHP-based webpage has been delivered, the extension will send a UDP packet containing timing information, peak memory usage and so on. The plugin will wait for such packets, parse them and account the provided information, which is then dispatched to the daemon once per interval.

Synopsis:

 <Plugin pinba>
   Address "::0"
   Port "30002"
   # Overall statistics for the website.
   <View "www-total">
     Server "www.example.com"
   </View>
   # Statistics for www-a only
   <View "www-a">
     Host "www-a.example.com"
     Server "www.example.com"
   </View>
   # Statistics for www-b only
   <View "www-b">
     Host "www-b.example.com"
     Server "www.example.com"
   </View>
 </Plugin>

The plugin provides the following configuration options:

Configures the address used to open a listening socket. By default, plugin will bind to the any address "::0".
Configures the port (service) to bind to. By default the default Pinba port "30002" will be used. The option accepts service names in addition to port numbers and thus requires a string argument.
<View Name> block
The packets sent by the Pinba extension include the hostname of the server, the server name (the name of the virtual host) and the script that was executed. Using View blocks it is possible to separate the data into multiple groups to get more meaningful statistics. Each packet is added to all matching groups, so that a packet may be accounted for more than once.
Matches the hostname of the system the webserver / script is running on. This will contain the result of the gethostname(2) system call. If not configured, all hostnames will be accepted.
Matches the name of the virtual host, i.e. the contents of the $_SERVER["SERVER_NAME"] variable when within PHP. If not configured, all server names will be accepted.
Matches the name of the script name, i.e. the contents of the $_SERVER["SCRIPT_NAME"] variable when within PHP. If not configured, all script names will be accepted.

The Ping plugin starts a new thread which sends ICMP "ping" packets to the configured hosts periodically and measures the network latency. Whenever the "read" function of the plugin is called, it submits the average latency, the standard deviation and the drop rate for each host.

Available configuration options:

Host to ping periodically. This option may be repeated several times to ping multiple hosts.
Sets the interval in which to send ICMP echo packets to the configured hosts. This is not the interval in which statistics are queries from the plugin but the interval in which the hosts are "pinged". Therefore, the setting here should be smaller than or equal to the global Interval setting. Fractional times, such as "1.24" are allowed.

Default: 1.0

Time to wait for a response from the host to which an ICMP packet had been sent. If a reply was not received after Seconds seconds, the host is assumed to be down or the packet to be dropped. This setting must be smaller than the Interval setting above for the plugin to work correctly. Fractional arguments are accepted.

Default: 0.9

Sets the Time-To-Live of generated ICMP packets.
Sets the source address to use. host may either be a numerical network address or a network hostname.
Sets the outgoing network device to be used. name has to specify an interface name (e. g. "eth0"). This might not be supported by all operating systems.
Trigger a DNS resolve after the host has not replied to Packets packets. This enables the use of dynamic DNS services (like dyndns.org) with the ping plugin.

Default: -1 (disabled)

The "postgresql" plugin queries statistics from PostgreSQL databases. It keeps a persistent connection to all configured databases and tries to reconnect if the connection has been interrupted. A database is configured by specifying a Database block as described below. The default statistics are collected from PostgreSQL's statistics collector which thus has to be enabled for this plugin to work correctly. This should usually be the case by default. See the section "The Statistics Collector" of the PostgreSQL Documentation for details.

By specifying custom database queries using a Query block as described below, you may collect any data that is available from some PostgreSQL database. This way, you are able to access statistics of external daemons which are available in a PostgreSQL database or use future or special statistics provided by PostgreSQL without the need to upgrade your collectd installation.

Starting with version 5.2, the "postgresql" plugin supports writing data to PostgreSQL databases as well. This has been implemented in a generic way. You need to specify an SQL statement which will then be executed by collectd in order to write the data (see below for details). The benefit of that approach is that there is no fixed database layout. Rather, the layout may be optimized for the current setup.

The PostgreSQL Documentation manual can be found at <http://www.postgresql.org/docs/manuals/>.

  <Plugin postgresql>
    <Query magic>
      Statement "SELECT magic FROM wizard WHERE host = $1;"
      Param hostname
      <Result>
        Type gauge
        InstancePrefix "magic"
        ValuesFrom magic
      </Result>
    </Query>
    <Query rt36_tickets>
      Statement "SELECT COUNT(type) AS count, type \
                        FROM (SELECT CASE \
                                     WHEN resolved = 'epoch' THEN 'open' \
                                     ELSE 'resolved' END AS type \
                                     FROM tickets) type \
                        GROUP BY type;"
      <Result>
        Type counter
        InstancePrefix "rt36_tickets"
        InstancesFrom "type"
        ValuesFrom "count"
      </Result>
    </Query>
    <Writer sqlstore>
      Statement "SELECT collectd_insert($1, $2, $3, $4, $5, $6, $7, $8, $9);"
      StoreRates true
    </Writer>
    <Database foo>
      Host "hostname"
      Port "5432"
      User "username"
      Password "secret"
      SSLMode "prefer"
      KRBSrvName "kerberos_service_name"
      Query magic
    </Database>
    <Database bar>
      Interval 300
      Service "service_name"
      Query backend # predefined
      Query rt36_tickets
    </Database>
    <Database qux>
      # ...
      Writer sqlstore
      CommitInterval 10
    </Database>
  </Plugin>

The Query block defines one database query which may later be used by a database definition. It accepts a single mandatory argument which specifies the name of the query. The names of all queries have to be unique (see the MinVersion and MaxVersion options below for an exception to this rule). The following configuration options are available to define the query:

In each Query block, there is one or more Result blocks. Result blocks define how to handle the values returned from the query. They define which column holds which value and how to dispatch that value to the daemon. Multiple Result blocks may be used to extract multiple values from a single query.

Specify the sql query statement which the plugin should execute. The string may contain the tokens $1, $2, etc. which are used to reference the first, second, etc. parameter. The value of the parameters is specified by the Param configuration option - see below for details. To include a literal $ character followed by a number, surround it with single quotes (').

Any SQL command which may return data (such as "SELECT" or "SHOW") is allowed. Note, however, that only a single command may be used. Semicolons are allowed as long as a single non-empty command has been specified only.

The returned lines will be handled separately one after another.

Specify the parameters which should be passed to the SQL query. The parameters are referred to in the SQL query as $1, $2, etc. in the same order as they appear in the configuration file. The value of the parameter is determined depending on the value of the Param option as follows:
The configured hostname of the database connection. If a UNIX domain socket is used, the parameter expands to "localhost".
The name of the database of the current connection.
The name of the database plugin instance. See the Instance option of the database specification below for details.
The username used to connect to the database.
The interval with which this database is queried (as specified by the database specific or global Interval options).

Please note that parameters are only supported by PostgreSQL's protocol version 3 and above which was introduced in version 7.4 of PostgreSQL.

The type name to be used when dispatching the values. The type describes how to handle the data and where to store it. See types.db(5) for more details on types and their configuration. The number and type of values (as selected by the ValuesFrom option) has to match the type of the given name.

This option is required inside a Result block.

Specify how to create the "TypeInstance" for each data set (i. e. line). InstancePrefix defines a static prefix that will be prepended to all type instances. InstancesFrom defines the column names whose values will be used to create the type instance. Multiple values will be joined together using the hyphen ("-") as separation character.

The plugin itself does not check whether or not all built instances are different. It is your responsibility to assure that each is unique.

Both options are optional. If none is specified, the type instance will be empty.

Names the columns whose content is used as the actual data for the data sets that are dispatched to the daemon. How many such columns you need is determined by the Type setting as explained above. If you specify too many or not enough columns, the plugin will complain about that and no data will be submitted to the daemon.

The actual data type, as seen by PostgreSQL, is not that important as long as it represents numbers. The plugin will automatically cast the values to the right type if it know how to do that. For that, it uses the strtoll(3) and strtod(3) functions, so anything supported by those functions is supported by the plugin as well.

This option is required inside a Result block and may be specified multiple times. If multiple ValuesFrom options are specified, the columns are read in the given order.

Specify the minimum or maximum version of PostgreSQL that this query should be used with. Some statistics might only be available with certain versions of PostgreSQL. This allows you to specify multiple queries with the same name but which apply to different versions, thus allowing you to use the same configuration in a heterogeneous environment.

The version has to be specified as the concatenation of the major, minor and patch-level versions, each represented as two-decimal-digit numbers. For example, version 8.2.3 will become 80203.

The following predefined queries are available (the definitions can be found in the postgresql_default.conf file which, by default, is available at "prefix/share/collectd/"):

This query collects the number of backends, i. e. the number of connected clients.
This query collects the numbers of committed and rolled-back transactions of the user tables.
This query collects the numbers of various table modifications (i. e. insertions, updates, deletions) of the user tables.
This query collects the numbers of various table scans and returned tuples of the user tables.
This query collects the numbers of live and dead rows in the user tables.
This query collects disk block access counts for user tables.
This query collects the on-disk size of the database in bytes.

In addition, the following detailed queries are available by default. Please note that each of those queries collects information by table, thus, potentially producing a lot of data. For details see the description of the non-by_table queries above.

The Writer block defines a PostgreSQL writer backend. It accepts a single mandatory argument specifying the name of the writer. This will then be used in the Database specification in order to activate the writer instance. The names of all writers have to be unique. The following options may be specified:

This mandatory option specifies the SQL statement that will be executed for each submitted value. A single SQL statement is allowed only. Anything after the first semicolon will be ignored.

Nine parameters will be passed to the statement and should be specified as tokens $1, $2, through $9 in the statement string. The following values are made available through those parameters:

$1
The timestamp of the queried value as a floating point number.
$2
The hostname of the queried value.
$3
The plugin name of the queried value.
$4
The plugin instance of the queried value. This value may be NULL if there is no plugin instance.
$5
The type of the queried value (cf. types.db(5)).
$6
The type instance of the queried value. This value may be NULL if there is no type instance.
$7
An array of names for the submitted values (i. e., the name of the data sources of the submitted value-list).
$8
An array of types for the submitted values (i. e., the type of the data sources of the submitted value-list; "counter", "gauge", ...). Note, that if StoreRates is enabled (which is the default, see below), all types will be "gauge".
$9
An array of the submitted values. The dimensions of the value name and value arrays match.

In general, it is advisable to create and call a custom function in the PostgreSQL database for this purpose. Any procedural language supported by PostgreSQL will do (see chapter "Server Programming" in the PostgreSQL manual for details).

If set to true (the default), convert counter values to rates. If set to false counter values are stored as is, i. e. as an increasing integer number.

The Database block defines one PostgreSQL database for which to collect statistics. It accepts a single mandatory argument which specifies the database name. None of the other options are required. PostgreSQL will use default values as documented in the section "CONNECTING TO A DATABASE" in the psql(1) manpage. However, be aware that those defaults may be influenced by the user collectd is run as and special environment variables. See the manpage for details.

Specify the interval with which the database should be queried. The default is to use the global Interval setting.
This option may be used for database connections which have "writers" assigned (see above). If specified, it causes a writer to put several updates into a single transaction. This transaction will last for the specified amount of time. By default, each update will be executed in a separate transaction. Each transaction generates a fair amount of overhead which can, thus, be reduced by activating this option. The draw-back is, that data covering the specified amount of time will be lost, for example, if a single statement within the transaction fails or if the database server crashes.
Specify the hostname or IP of the PostgreSQL server to connect to. If the value begins with a slash, it is interpreted as the directory name in which to look for the UNIX domain socket.

This option is also used to determine the hostname that is associated with a collected data set. If it has been omitted or either begins with with a slash or equals localhost it will be replaced with the global hostname definition of collectd. Any other value will be passed literally to collectd when dispatching values. Also see the global Hostname and FQDNLookup options.

Specify the TCP port or the local UNIX domain socket file extension of the server.
Specify the username to be used when connecting to the server.
Specify the password to be used when connecting to the server.
Specify whether to use an SSL connection when contacting the server. The following modes are supported:
Specify the plugin instance name that should be used instead of the database name (which is the default, if this option has not been specified). This allows to query multiple databases of the same name on the same host (e.g. when running multiple database server versions in parallel).
Do not use SSL at all.
First, try to connect without using SSL. If that fails, try using SSL.
First, try to connect using SSL. If that fails, try without using SSL.
Use SSL only.
Specify the Kerberos service name to use when authenticating with Kerberos 5 or GSSAPI. See the sections "Kerberos authentication" and "GSSAPI" of the PostgreSQL Documentation for details.
Specify the PostgreSQL service name to use for additional parameters. That service has to be defined in pg_service.conf and holds additional connection parameters. See the section "The Connection Service File" in the PostgreSQL Documentation for details.
Specifies a query which should be executed in the context of the database connection. This may be any of the predefined or user-defined queries. If no such option is given, it defaults to "backends", "transactions", "queries", "query_plans", "table_states", "disk_io" and "disk_usage" (unless a Writer has been specified). Else, the specified queries are used only.
Assigns the specified writer backend to the database connection. This causes all collected data to be send to the database using the settings defined in the writer configuration (see the section "FILTER CONFIGURATION" below for details on how to selectively send data to certain plugins).

Each writer will register a flush callback which may be used when having long transactions enabled (see the CommitInterval option above). When issuing the FLUSH command (see collectd-unixsock(5) for details) the current transaction will be committed right away. Two different kinds of flush callbacks are available with the "postgresql" plugin:

Flush all writer backends.
Flush all writers of the specified database only.

The "powerdns" plugin queries statistics from an authoritative PowerDNS nameserver and/or a PowerDNS recursor. Since both offer a wide variety of values, many of which are probably meaningless to most users, but may be useful for some. So you may chose which values to collect, but if you don't, some reasonable defaults will be collected.

  <Plugin "powerdns">
    <Server "server_name">
      Collect "latency"
      Collect "udp-answers" "udp-queries"
      Socket "/var/run/pdns.controlsocket"
    </Server>
    <Recursor "recursor_name">
      Collect "questions"
      Collect "cache-hits" "cache-misses"
      Socket "/var/run/pdns_recursor.controlsocket"
    </Recursor>
    LocalSocket "/opt/collectd/var/run/collectd-powerdns"
  </Plugin>
The Server block defines one authoritative server to query, the Recursor does the same for an recursing server. The possible options in both blocks are the same, though. The argument defines a name for the server / recursor and is required.
Using the Collect statement you can select which values to collect. Here, you specify the name of the values as used by the PowerDNS servers, e. g. "dlg-only-drops", "answers10-100".

The method of getting the values differs for Server and Recursor blocks: When querying the server a "SHOW *" command is issued in any case, because that's the only way of getting multiple values out of the server at once. collectd then picks out the values you have selected. When querying the recursor, a command is generated to query exactly these values. So if you specify invalid fields when querying the recursor, a syntax error may be returned by the daemon and collectd may not collect any values at all.

If no Collect statement is given, the following Server values will be collected:

The following Recursor values will be collected by default:

Please note that up to that point collectd doesn't know what values are available on the server and values that are added do not need a change of the mechanism so far. However, the values must be mapped to collectd's naming scheme, which is done using a lookup table that lists all known values. If values are added in the future and collectd does not know about them, you will get an error much like this:

  powerdns plugin: submit: Not found in lookup table: foobar = 42

In this case please file a bug report with the collectd team.

Configures the path to the UNIX domain socket to be used when connecting to the daemon. By default "${localstatedir}/run/pdns.controlsocket" will be used for an authoritative server and "${localstatedir}/run/pdns_recursor.controlsocket" will be used for the recursor.
Querying the recursor is done using UDP. When using UDP over UNIX domain sockets, the client socket needs a name in the file system, too. You can set this local name to Path using the LocalSocket option. The default is "prefix/var/run/collectd-powerdns".

Select more detailed statistics of processes matching this name. The statistics collected for these selected processes are size of the resident segment size (RSS), user- and system-time used, number of processes and number of threads, io data (where available) and minor and major pagefaults.
Similar to the Process option this allows to select more detailed statistics of processes matching the specified regex (see regex(7) for details). The statistics of all matching processes are summed up and dispatched to the daemon using the specified name as an identifier. This allows to "group" several processes together. name must not contain slashes.

Collects a lot of information about various network protocols, such as IP, TCP, UDP, etc.

Available configuration options:

Selects whether or not to select a specific value. The string being matched is of the form "Protocol:ValueName", where Protocol will be used as the plugin instance and ValueName will be used as type instance. An example of the string being used would be "Tcp:RetransSegs".

You can use regular expressions to match a large number of values with just one configuration option. To select all "extended" TCP values, you could use the following statement:

  Value "/^TcpExt:/"

Whether only matched values are selected or all matched values are ignored depends on the IgnoreSelected. By default, only matched values are selected. If no value is configured at all, all values will be selected.

If set to true, inverts the selection made by Value, i. e. all matching values will be ignored.

This plugin embeds a Python-interpreter into collectd and provides an interface to collectd's plugin system. See collectd-python(5) for its documentation.

The "routeros" plugin connects to a device running RouterOS, the Linux-based operating system for routers by MikroTik. The plugin uses librouteros to connect and reads information about the interfaces and wireless connections of the device. The configuration supports querying multiple routers:

  <Plugin "routeros">
    <Router>
      Host "router0.example.com"
      User "collectd"
      Password "secr3t"
      CollectInterface true
      CollectCPULoad true
      CollectMemory true
    </Router>
    <Router>
      Host "router1.example.com"
      User "collectd"
      Password "5ecret"
      CollectInterface true
      CollectRegistrationTable true
      CollectDF true
      CollectDisk true
    </Router>
  </Plugin>

As you can see above, the configuration of the routeros plugin consists of one or more <Router> blocks. Within each block, the following options are understood:

Hostname or IP-address of the router to connect to.
Port name or port number used when connecting. If left unspecified, the default will be chosen by librouteros, currently "8728". This option expects a string argument, even when a numeric port number is given.
Use the user name User to authenticate. Defaults to "admin".
Set the password used to authenticate.
When set to true, interface statistics will be collected for all interfaces present on the device. Defaults to false.
When set to true, information about wireless LAN connections will be collected. Defaults to false.
When set to true, information about the CPU usage will be collected. The number is a dimensionless value where zero indicates no CPU usage at all. Defaults to false.
When enabled, the amount of used and free memory will be collected. How used memory is calculated is unknown, for example whether or not caches are counted as used space. Defaults to false.
When enabled, the amount of used and free disk space will be collected. Defaults to false.
When enabled, the number of sectors written and bad blocks will be collected. Defaults to false.

The Redis plugin connects to one or more Redis servers and gathers information about each server's state. For each server there is a Node block which configures the connection parameters for this node.

  <Plugin redis>
    <Node "example">
        Host "localhost"
        Port "6379"
        Timeout 2000
    </Node>
  </Plugin>

The information shown in the synopsis above is the default configuration which is used by the plugin if no configuration is present.

The Node block identifies a new Redis node, that is a new Redis instance running in an specified host and port. The name for node is a canonical identifier which is used as plugin instance. It is limited to 64 characters in length.
The Host option is the hostname or IP-address where the Redis instance is running on.
The Port option is the TCP port on which the Redis instance accepts connections. Either a service name of a port number may be given. Please note that numerical port numbers must be given as a string, too.
Use Password to authenticate when connecting to Redis.
The Timeout option set the socket timeout for node response. Since the Redis read function is blocking, you should keep this value as low as possible. Keep in mind that the sum of all Timeout values for all Nodes should be lower than Interval defined globally.

The "rrdcached" plugin uses the RRDtool accelerator daemon, rrdcached(1), to store values to RRD files in an efficient manner. The combination of the "rrdcached" plugin and the "rrdcached" daemon is very similar to the way the "rrdtool" plugin works (see below). The added abstraction layer provides a number of benefits, though: Because the cache is not within "collectd" anymore, it does not need to be flushed when "collectd" is to be restarted. This results in much shorter (if any) gaps in graphs, especially under heavy load. Also, the "rrdtool" command line utility is aware of the daemon so that it can flush values to disk automatically when needed. This allows to integrate automated flushing of values into graphing solutions much more easily.

There are disadvantages, though: The daemon may reside on a different host, so it may not be possible for "collectd" to create the appropriate RRD files anymore. And even if "rrdcached" runs on the same host, it may run in a different base directory, so relative paths may do weird stuff if you're not careful.

So the recommended configuration is to let "collectd" and "rrdcached" run on the same host, communicating via a UNIX domain socket. The DataDir setting should be set to an absolute path, so that a changed base directory does not result in RRD files being created / expected in the wrong place.

Address of the daemon as understood by the "rrdc_connect" function of the RRD library. See rrdcached(1) for details. Example: 

  <Plugin "rrdcached">
    DaemonAddress "unix:/var/run/rrdcached.sock"
  </Plugin>
Set the base directory in which the RRD files reside. If this is a relative path, it is relative to the working base directory of the "rrdcached" daemon! Use of an absolute path is recommended.
Enables or disables the creation of RRD files. If the daemon is not running locally, or DataDir is set to a relative path, this will not work as expected. Default is true.
When enabled, new RRD files are enabled asynchronously, using a separate thread that runs in the background. This prevents writes to block, which is a problem especially when many hundreds of files need to be created at once. However, since the purpose of creating the files asynchronously is not to block until the file is available, values before the file is available will be discarded. When disabled (the default) files are created synchronously, blocking for a short while, while the file is being written.
Force the stepsize of newly created RRD-files. Ideally (and per default) this setting is unset and the stepsize is set to the interval in which the data is collected. Do not use this option unless you absolutely have to for some reason. Setting this option may cause problems with the "snmp plugin", the "exec plugin" or when the daemon is set up to receive data from other hosts.
Force the heartbeat of newly created RRD-files. This setting should be unset in which case the heartbeat is set to twice the StepSize which should equal the interval in which data is collected. Do not set this option unless you have a very good reason to do so.
The "rrdtool plugin" calculates the number of PDPs per CDP based on the StepSize, this setting and a timespan. This plugin creates RRD-files with three times five RRAs, i. e. five RRAs with the CFs MIN, AVERAGE, and MAX. The five RRAs are optimized for graphs covering one hour, one day, one week, one month, and one year.

So for each timespan, it calculates how many PDPs need to be consolidated into one CDP by calculating:
number of PDPs = timespan / (stepsize * rrarows)

Bottom line is, set this no smaller than the width of you graphs in pixels. The default is 1200.

Adds an RRA-timespan, given in seconds. Use this option multiple times to have more then one RRA. If this option is never used, the built-in default of (3600, 86400, 604800, 2678400, 31622400) is used.

For more information on how RRA-sizes are calculated see RRARows above.

Set the "XFiles Factor". The default is 0.1. If unsure, don't set this option. Factor must be in the range "[0.0-1.0)", i.e. between zero (inclusive) and one (exclusive).

You can use the settings StepSize, HeartBeat, RRARows, and XFF to fine-tune your RRD-files. Please read rrdcreate(1) if you encounter problems using these settings. If you don't want to dive into the depths of RRDtool, you can safely ignore these settings.

Set the directory to store RRD files under. By default RRD files are generated beneath the daemon's working directory, i.e. the BaseDir.
When enabled, new RRD files are enabled asynchronously, using a separate thread that runs in the background. This prevents writes to block, which is a problem especially when many hundreds of files need to be created at once. However, since the purpose of creating the files asynchronously is not to block until the file is available, values before the file is available will be discarded. When disabled (the default) files are created synchronously, blocking for a short while, while the file is being written.
Force the stepsize of newly created RRD-files. Ideally (and per default) this setting is unset and the stepsize is set to the interval in which the data is collected. Do not use this option unless you absolutely have to for some reason. Setting this option may cause problems with the "snmp plugin", the "exec plugin" or when the daemon is set up to receive data from other hosts.
Force the heartbeat of newly created RRD-files. This setting should be unset in which case the heartbeat is set to twice the StepSize which should equal the interval in which data is collected. Do not set this option unless you have a very good reason to do so.
The "rrdtool plugin" calculates the number of PDPs per CDP based on the StepSize, this setting and a timespan. This plugin creates RRD-files with three times five RRAs, i.e. five RRAs with the CFs MIN, AVERAGE, and MAX. The five RRAs are optimized for graphs covering one hour, one day, one week, one month, and one year.

So for each timespan, it calculates how many PDPs need to be consolidated into one CDP by calculating:
number of PDPs = timespan / (stepsize * rrarows)

Bottom line is, set this no smaller than the width of you graphs in pixels. The default is 1200.

Adds an RRA-timespan, given in seconds. Use this option multiple times to have more then one RRA. If this option is never used, the built-in default of (3600, 86400, 604800, 2678400, 31622400) is used.

For more information on how RRA-sizes are calculated see RRARows above.

Set the "XFiles Factor". The default is 0.1. If unsure, don't set this option. Factor must be in the range "[0.0-1.0)", i.e. between zero (inclusive) and one (exclusive).
When the "rrdtool" plugin uses a cache (by setting CacheTimeout, see below) it writes all values for a certain RRD-file if the oldest value is older than (or equal to) the number of seconds specified. If some RRD-file is not updated anymore for some reason (the computer was shut down, the network is broken, etc.) some values may still be in the cache. If CacheFlush is set, then the entire cache is searched for entries older than CacheTimeout seconds and written to disk every Seconds seconds. Since this is kind of expensive and does nothing under normal circumstances, this value should not be too small. 900 seconds might be a good value, though setting this to 7200 seconds doesn't normally do much harm either.
If this option is set to a value greater than zero, the "rrdtool plugin" will save values in a cache, as described above. Writing multiple values at once reduces IO-operations and thus lessens the load produced by updating the files. The trade off is that the graphs kind of "drag behind" and that more memory is used.
When collecting many statistics with collectd and the "rrdtool" plugin, you will run serious performance problems. The CacheFlush setting and the internal update queue assert that collectd continues to work just fine even under heavy load, but the system may become very unresponsive and slow. This is a problem especially if you create graphs from the RRD files on the same machine, for example using the "graph.cgi" script included in the "contrib/collection3/" directory.

This setting is designed for very large setups. Setting this option to a value between 25 and 80 updates per second, depending on your hardware, will leave the server responsive enough to draw graphs even while all the cached values are written to disk. Flushed values, i. e. values that are forced to disk by the FLUSH command, are not effected by this limit. They are still written as fast as possible, so that web frontends have up to date data when generating graphs.

For example: If you have 100,000 RRD files and set WritesPerSecond to 30 updates per second, writing all values to disk will take approximately 56 minutes. Together with the flushing ability that's integrated into "collection3" you'll end up with a responsive and fast system, up to date graphs and basically a "backup" of your values every hour.

When set, the actual timeout for each value is chosen randomly between CacheTimeout-RandomTimeout and CacheTimeout+RandomTimeout. The intention is to avoid high load situations that appear when many values timeout at the same time. This is especially a problem shortly after the daemon starts, because all values were added to the internal cache at roughly the same time.

The Sensors plugin uses lm_sensors to retrieve sensor-values. This means that all the needed modules have to be loaded and lm_sensors has to be configured (most likely by editing /etc/sensors.conf. Read sensors.conf(5) for details.

The lm_sensors homepage can be found at <http://secure.netroedge.com/~lm78/>.

Read the lm_sensors configuration from File. When unset (recommended), the library's default will be used.
Selects the name of the sensor which you want to collect or ignore, depending on the IgnoreSelected below. For example, the option "Sensor it8712-isa-0290/voltage-in1" will cause collectd to gather data for the voltage sensor in1 of the it8712 on the isa bus at the address 0290.
If no configuration if given, the sensors-plugin will collect data from all sensors. This may not be practical, especially for uninteresting sensors. Thus, you can use the Sensor-option to pick the sensors you're interested in. Sometimes, however, it's easier/preferred to collect all sensors except a few ones. This option enables you to do that: By setting IgnoreSelected to true the effect of Sensor is inverted: All selected sensors are ignored and all other sensors are collected.

The sigrok plugin uses libsigrok to retrieve measurements from any device supported by the sigrok <http://sigrok.org/> project.

Synopsis

 <Plugin sigrok>
   LogLevel 3
   <Device "AC Voltage">
      Driver "fluke-dmm"
      MinimumInterval 10
      Conn "/dev/ttyUSB2"
   </Device>
   <Device "Sound Level">
      Driver "cem-dt-885x"
      Conn "/dev/ttyUSB1"
   </Device>
 </Plugin>
The sigrok logging level to pass on to the collectd log, as a number between 0 and 5 (inclusive). These levels correspond to "None", "Errors", "Warnings", "Informational", "Debug "and "Spew", respectively. The default is 2 ("Warnings"). The sigrok log messages, regardless of their level, are always submitted to collectd at its INFO log level.
<Device Name>
A sigrok-supported device, uniquely identified by this section's options. The Name is passed to collectd as the plugin instance.
The sigrok driver to use for this device.
If the device cannot be auto-discovered, or more than one might be discovered by the driver, ConnectionSpec specifies the connection string to the device. It can be of the form of a device path (e.g. "/dev/ttyUSB2"), or, in case of a non-serial USB-connected device, the USB VendorID.ProductID separated by a period (e.g. 0403.6001). A USB device can also be specified as Bus.Address (e.g. 1.41).
For serial devices with non-standard port settings, this option can be used to specify them in a form understood by sigrok, e.g. "9600/8n1". This should not be necessary; drivers know how to communicate with devices they support.
Specifies the minimum time between measurement dispatches to collectd, in seconds. Since some sigrok supported devices can acquire measurements many times per second, it may be necessary to throttle these. For example, the RRD plugin cannot process writes more than once per second.

The default MinimumInterval is 0, meaning measurements received from the device are always dispatched to collectd. When throttled, unused measurements are discarded.

Since the configuration of the "snmp plugin" is a little more complicated than other plugins, its documentation has been moved to an own manpage, collectd-snmp(5). Please see there for details.

The statsd plugin listens to a UDP socket, reads "events" in the statsd protocol and dispatches rates or other aggregates of these numbers periodically.

The plugin implements the Counter, Timer, Gauge and Set types which are dispatched as the collectd types "derive", "latency", "gauge" and "objects" respectively.

The following configuration options are valid:

Bind to the hostname / address Host. By default, the plugin will bind to the "any" address, i.e. accept packets sent to any of the hosts addresses.
UDP port to listen to. This can be either a service name or a port number. Defaults to 8125.
These options control what happens if metrics are not updated in an interval. If set to False, the default, metrics are dispatched unchanged, i.e. the rate of counters and size of sets will be zero, timers report "NaN" and gauges are unchanged. If set to True, the such metrics are not dispatched and removed from the internal cache.
Calculate and dispatch the configured percentile, i.e. compute the latency, so that Percent of all reported timers are smaller than or equal to the computed latency. This is useful for cutting off the long tail latency, as it's often done in Service Level Agreements (SLAs).

If not specified, no percentile is calculated / dispatched.

The Swap plugin collects information about used and available swap space. On Linux and Solaris, the following options are available:

Configures how to report physical swap devices. If set to false (the default), the summary over all swap devices is reported only, i.e. the globally used and available space over all devices. If true is configured, the used and available space of each device will be reported separately.

This option is only available if the Swap plugin can read "/proc/swaps" (under Linux) or use the swapctl(2) mechanism (under Solaris).

When enabled, the swap I/O is reported in bytes. When disabled, the default, swap I/O is reported in pages. This option is available under Linux only.

Sets the log-level. If, for example, set to notice, then all events with severity notice, warning, or err will be submitted to the syslog-daemon.

Please note that debug is only available if collectd has been compiled with debugging support.

Controls which notifications should be sent to syslog. The default behaviour is not to send any. Less severe notifications always imply logging more severe notifications: Setting this to OKAY means all notifications will be sent to syslog, setting this to WARNING will send WARNING and FAILURE notifications but will dismiss OKAY notifications. Setting this option to FAILURE will only send failures to syslog.

The "table plugin" provides generic means to parse tabular data and dispatch user specified values. Values are selected based on column numbers. For example, this plugin may be used to get values from the Linux proc(5) filesystem or CSV (comma separated values) files.

  <Plugin table>
    <Table "/proc/slabinfo">
      Instance "slabinfo"
      Separator " "
      <Result>
        Type gauge
        InstancePrefix "active_objs"
        InstancesFrom 0
        ValuesFrom 1
      </Result>
      <Result>
        Type gauge
        InstancePrefix "objperslab"
        InstancesFrom 0
        ValuesFrom 4
      </Result>
    </Table>
  </Plugin>

The configuration consists of one or more Table blocks, each of which configures one file to parse. Within each Table block, there are one or more Result blocks, which configure which data to select and how to interpret it.

The following options are available inside a Table block:

If specified, instance is used as the plugin instance. So, in the above example, the plugin name "table-slabinfo" would be used. If omitted, the filename of the table is used instead, with all special characters replaced with an underscore ("_").
Any character of string is interpreted as a delimiter between the different columns of the table. A sequence of two or more contiguous delimiters in the table is considered to be a single delimiter, i. e. there cannot be any empty columns. The plugin uses the strtok_r(3) function to parse the lines of a table - see its documentation for more details. This option is mandatory.

A horizontal tab, newline and carriage return may be specified by "\\t", "\\n" and "\\r" respectively. Please note that the double backslashes are required because of collectd's config parsing.

The following options are available inside a Result block:

Sets the type used to dispatch the values to the daemon. Detailed information about types and their configuration can be found in types.db(5). This option is mandatory.
If specified, prepend prefix to the type instance. If omitted, only the InstancesFrom option is considered for the type instance.
If specified, the content of the given columns (identified by the column number starting at zero) will be used to create the type instance for each row. Multiple values (and the instance prefix) will be joined together with dashes (-) as separation character. If omitted, only the InstancePrefix option is considered for the type instance.

The plugin itself does not check whether or not all built instances are different. ItXs your responsibility to assure that each is unique. This is especially true, if you do not specify InstancesFrom: You have to make sure that the table only contains one row.

If neither InstancePrefix nor InstancesFrom is given, the type instance will be empty.

Specifies the columns (identified by the column numbers starting at zero) whose content is used as the actual data for the data sets that are dispatched to the daemon. How many such columns you need is determined by the Type setting above. If you specify too many or not enough columns, the plugin will complain about that and no data will be submitted to the daemon. The plugin uses strtoll(3) and strtod(3) to parse counter and gauge values respectively, so anything supported by those functions is supported by the plugin as well. This option is mandatory.

The "tail plugin" follows logfiles, just like tail(1) does, parses each line and dispatches found values. What is matched can be configured by the user using (extended) regular expressions, as described in regex(7).

  <Plugin "tail">
    <File "/var/log/exim4/mainlog">
      Instance "exim"
      <Match>
        Regex "S=([1-9][0-9]*)"
        DSType "CounterAdd"
        Type "ipt_bytes"
        Instance "total"
      </Match>
      <Match>
        Regex "\\<R=local_user\\>"
        ExcludeRegex "\\<R=local_user\\>.*mail_spool defer"
        DSType "CounterInc"
        Type "counter"
        Instance "local_user"
      </Match>
    </File>
  </Plugin>

The config consists of one or more File blocks, each of which configures one logfile to parse. Within each File block, there are one or more Match blocks, which configure a regular expression to search for.

The Instance option in the File block may be used to set the plugin instance. So in the above example the plugin name "tail-foo" would be used. This plugin instance is for all Match blocks that follow it, until the next Instance option. This way you can extract several plugin instances from one logfile, handy when parsing syslog and the like.

Each Match block has the following options to describe how the match should be performed:

Sets the regular expression to use for matching against a line. The first subexpression has to match something that can be turned into a number by strtoll(3) or strtod(3), depending on the value of "CounterAdd", see below. Because extended regular expressions are used, you do not need to use backslashes for subexpressions! If in doubt, please consult regex(7). Due to collectd's config parsing you need to escape backslashes, though. So if you want to match literal parentheses you need to do the following: 

  Regex "SPAM \\(Score: (-?[0-9]+\\.[0-9]+)\\)"
Sets an optional regular expression to use for excluding lines from the match. An example which excludes all connections from localhost from the match: 

  ExcludeRegex "127\\.0\\.0\\.1"
Sets how the values are cumulated. Type is one of:
Calculate the average.
Use the smallest number only.
Use the greatest number only.
Use the last number found.
The matched number is a counter. Simply sets the internal counter to this value. Variants exist for "COUNTER", "DERIVE", and "ABSOLUTE" data sources.
Add the matched value to the internal counter. In case of DeriveAdd, the matched number may be negative, which will effectively subtract from the internal counter.
Increase the internal counter by one. These DSType are the only ones that do not use the matched subexpression, but simply count the number of matched lines. Thus, you may use a regular expression without submatch in this case.

As you'd expect the Gauge* types interpret the submatch as a floating point number, using strtod(3). The Counter* and AbsoluteSet types interpret the submatch as an unsigned integer using strtoull(3). The Derive* types interpret the submatch as a signed integer using strtoll(3). CounterInc and DeriveInc do not use the submatch at all and it may be omitted in this case.

Sets the type used to dispatch this value. Detailed information about types and their configuration can be found in types.db(5).
This optional setting sets the type instance to use.

The tail_csv plugin reads files in the CSV format, e.g. the statistics file written by Snort.

Synopsis:

 <Plugin "tail_csv">
   <Metric "snort-dropped">
       Type "percent"
       Instance "dropped"
       Index 1
   </Metric>
   <File "/var/log/snort/snort.stats">
       Instance "snort-eth0"
       Interval 600
       Collect "snort-dropped"
   </File>
 </Plugin>

The configuration consists of one or more Metric blocks that define an index into the line of the CSV file and how this value is mapped to collectd's internal representation. These are followed by one or more Instance blocks which configure which file to read, in which interval and which metrics to extract.

<Metric Name>
The Metric block configures a new metric to be extracted from the statistics file and how it is mapped on collectd's data model. The string Name is only used inside the Instance blocks to refer to this block, so you can use one Metric block for multiple CSV files.
Configures which Type to use when dispatching this metric. Types are defined in the types.db(5) file, see the appropriate manual page for more information on specifying types. Only types with a single data source are supported by the tail_csv plugin. The information whether the value is an absolute value (i.e. a "GAUGE") or a rate (i.e. a "DERIVE") is taken from the Type's definition.
If set, TypeInstance is used to populate the type instance field of the created value lists. Otherwise, no type instance is used.
Configure to read the value from the field with the zero-based index Index. If the value is parsed as signed integer, unsigned integer or double depends on the Type setting, see above.
<File Path>
Each File block represents one CSV file to read. There must be at least one File block but there can be multiple if you have multiple CSV files.
Sets the plugin instance used when dispatching the values.
Specifies which Metric to collect. This option must be specified at least once, and you can use this option multiple times to specify more than one metric to be extracted from this statistic file.
Configures the interval in which to read values from this instance / file. Defaults to the plugin's default interval.
Rather than using the local time when dispatching a value, read the timestamp from the field with the zero-based index Index. The value is interpreted as seconds since epoch. The value is parsed as a double and may be factional.

The "teamspeak2 plugin" connects to the query port of a teamspeak2 server and polls interesting global and virtual server data. The plugin can query only one physical server but unlimited virtual servers. You can use the following options to configure it:

The hostname or ip which identifies the physical server. Default: 127.0.0.1
The query port of the physical server. This needs to be a string. Default: "51234"
This option has to be added once for every virtual server the plugin should query. If you want to query the virtual server on port 8767 this is what the option would look like: 

  Server "8767"

This option, although numeric, needs to be a string, i. e. you must use quotes around it! If no such statement is given only global information will be collected.

The TED plugin connects to a device of "The Energy Detective", a device to measure power consumption. These devices are usually connected to a serial (RS232) or USB port. The plugin opens a configured device and tries to read the current energy readings. For more information on TED, visit <http://www.theenergydetective.com/>.

Available configuration options:

Path to the device on which TED is connected. collectd will need read and write permissions on that file.

Default: /dev/ttyUSB0

Apparently reading from TED is not that reliable. You can therefore configure a number of retries here. You only configure the retries here, to if you specify zero, one reading will be performed (but no retries if that fails); if you specify three, a maximum of four readings are performed. Negative values are illegal.

Default: 0

The "tcpconns plugin" counts the number of currently established TCP connections based on the local port and/or the remote port. Since there may be a lot of connections the default if to count all connections with a local port, for which a listening socket is opened. You can use the following options to fine-tune the ports you are interested in:

If this option is set to true, statistics for all local ports for which a listening socket exists are collected. The default depends on LocalPort and RemotePort (see below): If no port at all is specifically selected, the default is to collect listening ports. If specific ports (no matter if local or remote ports) are selected, this option defaults to false, i. e. only the selected ports will be collected unless this option is set to true specifically.
Count the connections to a specific local port. This can be used to see how many connections are handled by a specific daemon, e. g. the mailserver. You have to specify the port in numeric form, so for the mailserver example you'd need to set 25.
Count the connections to a specific remote port. This is useful to see how much a remote service is used. This is most useful if you want to know how many connections a local service has opened to remote services, e. g. how many connections a mail server or news server has to other mail or news servers, or how many connections a web proxy holds to web servers. You have to give the port in numeric form.

By default, the Thermal plugin tries to read the statistics from the Linux "sysfs" interface. If that is not available, the plugin falls back to the "procfs" interface. By setting this option to true, you can force the plugin to use the latter. This option defaults to false.
Selects the name of the thermal device that you want to collect or ignore, depending on the value of the IgnoreSelected option. This option may be used multiple times to specify a list of devices.
Invert the selection: If set to true, all devices except the ones that match the device names specified by the Device option are collected. By default only selected devices are collected if a selection is made. If no selection is configured at all, all devices are selected.

The Threshold plugin checks values collected or received by collectd against a configurable threshold and issues notifications if values are out of bounds.

Documentation for this plugin is available in the collectd-threshold(5) manual page.

The TokyoTyrant plugin connects to a TokyoTyrant server and collects a couple metrics: number of records, and database size on disk.

The hostname or ip which identifies the server. Default: 127.0.0.1
The query port of the server. This needs to be a string, even if the port is given in its numeric form. Default: 1978

Sets the socket-file which is to be created.
If running as root change the group of the UNIX-socket after it has been created. Defaults to collectd.
Change the file permissions of the UNIX-socket after it has been created. The permissions must be given as a numeric, octal value as you would pass to chmod(1). Defaults to 0770.
If set to true, delete the socket file before calling bind(2), if a file with the given name already exists. If collectd crashes a socket file may be left over, preventing the daemon from opening a new socket when restarted. Since this is potentially dangerous, this defaults to false.

This plugin, if loaded, causes the Hostname to be taken from the machine's UUID. The UUID is a universally unique designation for the machine, usually taken from the machine's BIOS. This is most useful if the machine is running in a virtual environment such as Xen, in which case the UUID is preserved across shutdowns and migration.

The following methods are used to find the machine's UUID, in order:

  • Check /etc/uuid (or UUIDFile).
  • Check for UUID from HAL (<http://www.freedesktop.org/wiki/Software/hal>) if present.
  • Check for UUID from "dmidecode" / SMBIOS.
  • Check for UUID from Xen hypervisor.

If no UUID can be found then the hostname is not modified.

Take the UUID from the given file (default /etc/uuid).

The varnish plugin collects information about Varnish, an HTTP accelerator.

Synopsis:

 <Plugin "varnish">
   <Instance "example">
     CollectCache       true
     CollectConnections true
     CollectBackend     true
     CollectSHM         true
     CollectESI         false
     CollectFetch       false
     CollectHCB         false
     CollectSMA         false
     CollectSMS         false
     CollectSM          false
     CollectTotals      false
     CollectWorkers     false
   </Instance>
 </Plugin>

The configuration consists of one or more <Instance Name> blocks. Name is the parameter passed to "varnishd -n". If left empty, it will collectd statistics from the default "varnishd" instance (this should work fine in most cases).

Inside each <Instance> blocks, the following options are recognized:

Cache hits and misses. True by default.
Number of client connections received, accepted and dropped. True by default.
Back-end connection statistics, such as successful, reused, and closed connections. True by default.
Statistics about the shared memory log, a memory region to store log messages which is flushed to disk when full. True by default.
Statistics about ban operations, such as number of bans added, retired, and number of objects tested against ban operations. Only available with Varnish 3.x. False by default.
DNS director lookup cache statistics. Only available with Varnish 3.x. False by default.
Edge Side Includes (ESI) parse statistics. False by default.
Statistics about fetches (HTTP requests sent to the backend). False by default.
Inserts and look-ups in the crit bit tree based hash. Look-ups are divided into locked and unlocked look-ups. False by default.
Statistics on cached objects: number of objects expired, nuked (prematurely expired), saved, moved, etc. False by default.
Statistics about purge operations, such as number of purges added, retired, and number of objects tested against purge operations. Only available with Varnish 2.x. False by default.
Client session statistics. Number of past and current sessions, session herd and linger counters, etc. False by default.
malloc or umem (umem_alloc(3MALLOC) based) storage statistics. The umem storage component is Solaris specific. Only available with Varnish 2.x. False by default.
synth (synthetic content) storage statistics. This storage component is used internally only. False by default.
file (memory mapped file) storage statistics. Only available with Varnish 2.x. False by default.
Current varnish internal state statistics. Number of current sessions, objects in cache store, open connections to backends (with Varnish 2.x), etc. False by default.
Collects overview counters, such as the number of sessions created, the number of requests and bytes transferred. False by default.
Varnish uptime. False by default.
Number of total (available + discarded) VCL (config files). False by default.
Collect statistics about worker threads. False by default.

The "vmem" plugin collects information about the usage of virtual memory. Since the statistics provided by the Linux kernel are very detailed, they are collected very detailed. However, to get all the details, you have to switch them on manually. Most people just want an overview over, such as the number of pages read from swap space.

Enables verbose collection of information. This will start collecting page "actions", e. g. page allocations, (de)activations, steals and so on. Part of these statistics are collected on a "per zone" basis.

This plugin doesn't have any options. VServer support is only available for Linux. It cannot yet be found in a vanilla kernel, though. To make use of this plugin you need a kernel that has VServer support built in, i. e. you need to apply the patches and compile your own kernel, which will then provide the /proc/virtual filesystem that is required by this plugin.

The VServer homepage can be found at <http://linux-vserver.org/>.

Note: The traffic collected by this plugin accounts for the amount of traffic passing a socket which might be a lot less than the actual on-wire traffic (e. g. due to headers and retransmission). If you want to collect on-wire traffic you could, for example, use the logging facilities of iptables to feed data for the guest IPs into the iptables plugin.

The "write_graphite" plugin writes data to Graphite, an open-source metrics storage and graphing project. The plugin connects to Carbon, the data layer of Graphite, via TCP or UDP and sends data via the "line based" protocol (per default using port 2003). The data will be sent in blocks of at most 1428 bytes to minimize the number of network packets.

Synopsis:

 <Plugin write_graphite>
   <Node "example">
     Host "localhost"
     Port "2003"
     Protocol "udp"
     LogSendErrors true
     Prefix "collectd"
   </Node>
 </Plugin>

The configuration consists of one or more <Node Name> blocks. Inside the Node blocks, the following options are recognized:

Hostname or address to connect to. Defaults to "localhost".
Service name or port number to connect to. Defaults to 2003.
Protocol to use when connecting to Graphite. Defaults to "tcp".
If set to true (the default), logs errors when sending data to Graphite. If set to false, it will not log the errors. This is especially useful when using Protocol UDP since many times we want to use the "fire-and-forget" approach and logging errors fills syslog with unneeded messages.
When set, String is added in front of the host name. Dots and whitespace are not escaped in this string (see EscapeCharacter below).
When set, String is appended to the host name. Dots and whitespace are not escaped in this string (see EscapeCharacter below).
Carbon uses the dot (".") as escape character and doesn't allow whitespace in the identifier. The EscapeCharacter option determines which character dots, whitespace and control characters are replaced with. Defaults to underscore ("_").
If set to true (the default), convert counter values to rates. If set to false counter values are stored as is, i. e. as an increasing integer number.
If set to true, the plugin instance and type instance will be in their own path component, for example "host.cpu.0.cpu.idle". If set to false (the default), the plugin and plugin instance (and likewise the type and type instance) are put into one component, for example "host.cpu-0.cpu-idle".
If set the true, append the name of the Data Source (DS) to the "metric" identifier. If set to false (the default), this is only done when there is more than one DS.

The write_mongodb plugin will send values to MongoDB, a schema-less NoSQL database.

Synopsis:

 <Plugin "write_mongodb">
   <Node "default">
     Host "localhost"
     Port "27017"
     Timeout 1000
     StoreRates true
   </Node>
 </Plugin>

The plugin can send values to multiple instances of MongoDB by specifying one Node block for each instance. Within the Node blocks, the following options are available:

Hostname or address to connect to. Defaults to "localhost".
Service name or port number to connect to. Defaults to 27017.
Set the timeout for each operation on MongoDB to Timeout milliseconds. Setting this option to zero means no timeout, which is the default.
If set to true (the default), convert counter values to rates. If set to false counter values are stored as is, i.e. as an increasing integer number.
Sets the information used when authenticating to a MongoDB database. The fields are optional (in which case no authentication is attempted), but if you want to use authentication all three fields must be set.

This output plugin submits values to an http server by POST them using the PUTVAL plain-text protocol. Each destination you want to post data to needs to have one URL block, within which the destination can be configured further, for example by specifying authentication data.

Synopsis:

 <Plugin "write_http">
   <URL "http://example.com/post-collectd">
     User "collectd"
     Password "weCh3ik0"
   </URL>
 </Plugin>

URL blocks need one string argument which is used as the URL to which data is posted. The following options are understood within URL blocks.

Optional user name needed for authentication.
Optional password needed for authentication.
Enable or disable peer SSL certificate verification. See <http://curl.haxx.se/docs/sslcerts.html> for details. Enabled by default.
Enable or disable peer host name verification. If enabled, the plugin checks if the "Common Name" or a "Subject Alternate Name" field of the SSL certificate matches the host name provided by the URL option. If this identity check fails, the connection is aborted. Obviously, only works when connecting to a SSL enabled server. Enabled by default.
File that holds one or more SSL certificates. If you want to use HTTPS you will possibly need this option. What CA certificates come bundled with "libcurl" and are checked by default depends on the distribution you use.
Format of the output to generate. If set to Command, will create output that is understood by the Exec and UnixSock plugins. When set to JSON, will create output in the JavaScript Object Notation (JSON).

Defaults to Command.

If set to true, convert counter values to rates. If set to false (the default) counter values are stored as is, i. e. as an increasing integer number.

The write_riemann plugin will send values to Riemann, a powerfull stream aggregation and monitoring system. The plugin sends Protobuf encoded data to Riemann using UDP packets.

Synopsis:

 <Plugin "write_riemann">
   <Node "example">
     Host "localhost"
     Port "5555"
     Protocol UDP
     StoreRates true
     AlwaysAppendDS false
     TTLFactor 2.0
   </Node>
   Tag "foobar"
 </Plugin>

The following options are understood by the write_riemann plugin:

<Node Name>
The plugin's configuration consists of one or more Node blocks. Each block is given a unique Name and specifies one connection to an instance of Riemann. Indise the Node block, the following per-connection options are understood:
Hostname or address to connect to. Defaults to "localhost".
Service name or port number to connect to. Defaults to 5555.
Specify the protocol to use when communicating with Riemann. Defaults to UDP.
If set to true (the default), convert counter values to rates. If set to false counter values are stored as is, i.e. as an increasing integer number.

This will be reflected in the "ds_type" tag: If StoreRates is enabled, converted values will have "rate" appended to the data source type, e.g. "ds_type:derive:rate".

If set the true, append the name of the Data Source (DS) to the "service", i.e. the field that, together with the "host" field, uniquely identifies a metric in Riemann. If set to false (the default), this is only done when there is more than one DS.
Riemann events have a Time to Live (TTL) which specifies how long each event is considered active. collectd populates this field based on the metrics interval setting. This setting controls the factor with which the interval is multiplied to set the TTL. The default value is 2.0. Unless you know exactly what you're doing, you should only increase this setting from its default value.
Add the given string as an additional tag to the metric being sent to Riemann.

Starting with version 4.3.0 collectd has support for monitoring. By that we mean that the values are not only stored or sent somewhere, but that they are judged and, if a problem is recognized, acted upon. The only action collectd takes itself is to generate and dispatch a "notification". Plugins can register to receive notifications and perform appropriate further actions.

Since systems and what you expect them to do differ a lot, you can configure thresholds for your values freely. This gives you a lot of flexibility but also a lot of responsibility.

Every time a value is out of range a notification is dispatched. This means that the idle percentage of your CPU needs to be less then the configured threshold only once for a notification to be generated. There's no such thing as a moving average or similar - at least not now.

Also, all values that match a threshold are considered to be relevant or "interesting". As a consequence collectd will issue a notification if they are not received for Timeout iterations. The Timeout configuration option is explained in section "GLOBAL OPTIONS". If, for example, Timeout is set to "2" (the default) and some hosts sends it's CPU statistics to the server every 60 seconds, a notification will be dispatched after about 120 seconds. It may take a little longer because the timeout is checked only once each Interval on the server.

When a value comes within range again or is received after it was missing, an "OKAY-notification" is dispatched.

Here is a configuration example to get you started. Read below for more information.

 <Threshold>
   <Type "foo">
     WarningMin    0.00
     WarningMax 1000.00
     FailureMin    0.00
     FailureMax 1200.00
     Invert false
     Instance "bar"
   </Type>
   <Plugin "interface">
     Instance "eth0"
     <Type "if_octets">
       FailureMax 10000000
       DataSource "rx"
     </Type>
   </Plugin>
   <Host "hostname">
     <Type "cpu">
       Instance "idle"
       FailureMin 10
     </Type>
     <Plugin "memory">
       <Type "memory">
         Instance "cached"
         WarningMin 100000000
       </Type>
     </Plugin>
   </Host>
 </Threshold>

There are basically two types of configuration statements: The "Host", "Plugin", and "Type" blocks select the value for which a threshold should be configured. The "Plugin" and "Type" blocks may be specified further using the "Instance" option. You can combine the block by nesting the blocks, though they must be nested in the above order, i. e. "Host" may contain either "Plugin" and "Type" blocks, "Plugin" may only contain "Type" blocks and "Type" may not contain other blocks. If multiple blocks apply to the same value the most specific block is used.

The other statements specify the threshold to configure. They must be included in a "Type" block. Currently the following statements are recognized:

Sets the upper bound of acceptable values. If unset defaults to positive infinity. If a value is greater than FailureMax a FAILURE notification will be created. If the value is greater than WarningMax but less than (or equal to) FailureMax a WARNING notification will be created.
Sets the lower bound of acceptable values. If unset defaults to negative infinity. If a value is less than FailureMin a FAILURE notification will be created. If the value is less than WarningMin but greater than (or equal to) FailureMin a WARNING notification will be created.
Some data sets have more than one "data source". Interesting examples are the "if_octets" data set, which has received ("rx") and sent ("tx") bytes and the "disk_ops" data set, which holds "read" and "write" operations. The system load data set, "load", even has three data sources: "shortterm", "midterm", and "longterm".

Normally, all data sources are checked against a configured threshold. If this is undesirable, or if you want to specify different limits for each data source, you can use the DataSource option to have a threshold apply only to one data source.

If set to true the range of acceptable values is inverted, i. e. values between FailureMin and FailureMax (WarningMin and WarningMax) are not okay. Defaults to false.
Sets how often notifications are generated. If set to true one notification will be generated for each value that is out of the acceptable range. If set to false (the default) then a notification is only generated if a value is out of range but the previous value was okay.

This applies to missing values, too: If set to true a notification about a missing value is generated once every Interval seconds. If set to false only one such notification is generated until the value appears again.

If set to true, the minimum and maximum values given are interpreted as percentage value, relative to the other data sources. This is helpful for example for the "df" type, where you may want to issue a warning when less than 5 % of the total space is available. Defaults to false.
Delay creating the notification until the threshold has been passed Number times. When a notification has been generated, or when a subsequent value is inside the threshold, the counter is reset. If, for example, a value is collected once every 10 seconds and Hits is set to 3, a notification will be dispatched at most once every 30 seconds.

This is useful when short bursts are not a problem. If, for example, 100% CPU usage for up to a minute is normal (and data is collected every 10 seconds), you could set Hits to 6 to account for this.

When set to non-zero, a hysteresis value is applied when checking minimum and maximum bounds. This is useful for values that increase slowly and fluctuate a bit while doing so. When these values come close to the threshold, they may "flap", i.e. switch between failure / warning case and okay case repeatedly.

If, for example, the threshold is configures as

  WarningMax 100.0
  Hysteresis 1.0

then a Warning notification is created when the value exceeds 101 and the corresponding Okay notification is only created once the value falls below 99, thus avoiding the "flapping".

Starting with collectd 4.6 there is a powerful filtering infrastructure implemented in the daemon. The concept has mostly been copied from ip_tables, the packet filter infrastructure for Linux. We'll use a similar terminology, so that users that are familiar with iptables feel right at home.

The following are the terms used in the remainder of the filter configuration documentation. For an ASCII-art schema of the mechanism, see "General structure" below.

A match is a criteria to select specific values. Examples are, of course, the name of the value or it's current value.

Matches are implemented in plugins which you have to load prior to using the match. The name of such plugins starts with the "match_" prefix.

A target is some action that is to be performed with data. Such actions could, for example, be to change part of the value's identifier or to ignore the value completely.

Some of these targets are built into the daemon, see "Built-in targets" below. Other targets are implemented in plugins which you have to load prior to using the target. The name of such plugins starts with the "target_" prefix.

The combination of any number of matches and at least one target is called a rule. The target actions will be performed for all values for which all matches apply. If the rule does not have any matches associated with it, the target action will be performed for all values.
A chain is a list of rules and possibly default targets. The rules are tried in order and if one matches, the associated target will be called. If a value is handled by a rule, it depends on the target whether or not any subsequent rules are considered or if traversal of the chain is aborted, see "Flow control" below. After all rules have been checked, the default targets will be executed.

The following shows the resulting structure:

 +---------+
 ! Chain   !
 +---------+
      !
      V
 +---------+  +---------+  +---------+  +---------+
 ! Rule    !->! Match   !->! Match   !->! Target  !
 +---------+  +---------+  +---------+  +---------+
      !
      V
 +---------+  +---------+  +---------+
 ! Rule    !->! Target  !->! Target  !
 +---------+  +---------+  +---------+
      !
      V
      :
      :
      !
      V
 +---------+  +---------+  +---------+
 ! Rule    !->! Match   !->! Target  !
 +---------+  +---------+  +---------+
      !
      V
 +---------+
 ! Default !
 ! Target  !
 +---------+

There are four ways to control which way a value takes through the filter mechanism:

The built-in jump target can be used to "call" another chain, i. e. process the value with another chain. When the called chain finishes, usually the next target or rule after the jump is executed.
The stop condition, signaled for example by the built-in target stop, causes all processing of the value to be stopped immediately.
Causes processing in the current chain to be aborted, but processing of the value generally will continue. This means that if the chain was called via Jump, the next target or rule after the jump will be executed. If the chain was not called by another chain, control will be returned to the daemon and it may pass the value to another chain.
Most targets will signal the continue condition, meaning that processing should continue normally. There is no special built-in target for this condition.

The configuration reflects this structure directly:

 PostCacheChain "PostCache"
 <Chain "PostCache">
   <Rule "ignore_mysql_show">
     <Match "regex">
       Plugin "^mysql$"
       Type "^mysql_command$"
       TypeInstance "^show_"
     </Match>
     <Target "stop">
     </Target>
   </Rule>
   <Target "write">
     Plugin "rrdtool"
   </Target>
 </Chain>

The above configuration example will ignore all values where the plugin field is "mysql", the type is "mysql_command" and the type instance begins with "show_". All other values will be sent to the "rrdtool" write plugin via the default target of the chain. Since this chain is run after the value has been added to the cache, the MySQL "show_*" command statistics will be available via the "unixsock" plugin.

Configure the name of the "pre-cache chain" and the "post-cache chain". The argument is the name of a chain that should be executed before and/or after the values have been added to the cache.

To understand the implications, it's important you know what is going on inside collectd. The following diagram shows how values are passed from the read-plugins to the write-plugins:

   +---------------+
   !  Read-Plugin  !
   +-------+-------+
           !
 + - - - - V - - - - +
 : +---------------+ :
 : !   Pre-Cache   ! :
 : !     Chain     ! :
 : +-------+-------+ :
 :         !         :
 :         V         :
 : +-------+-------+ :  +---------------+
 : !     Cache     !--->!  Value Cache  !
 : !     insert    ! :  +---+---+-------+
 : +-------+-------+ :      !   !
 :         !   ,------------'   !
 :         V   V     :          V
 : +-------+---+---+ :  +-------+-------+
 : !  Post-Cache   +--->! Write-Plugins !
 : !     Chain     ! :  +---------------+
 : +---------------+ :
 :                   :
 :  dispatch values  :
 + - - - - - - - - - +

After the values are passed from the "read" plugins to the dispatch functions, the pre-cache chain is run first. The values are added to the internal cache afterwards. The post-cache chain is run after the values have been added to the cache. So why is it such a huge deal if chains are run before or after the values have been added to this cache?

Targets that change the identifier of a value list should be executed before the values are added to the cache, so that the name in the cache matches the name that is used in the "write" plugins. The "unixsock" plugin, too, uses this cache to receive a list of all available values. If you change the identifier after the value list has been added to the cache, this may easily lead to confusion, but it's not forbidden of course.

The cache is also used to convert counter values to rates. These rates are, for example, used by the "value" match (see below). If you use the rate stored in the cache before the new value is added, you will use the old, previous rate. Write plugins may use this rate, too, see the "csv" plugin, for example. The "unixsock" plugin uses these rates too, to implement the "GETVAL" command.

Last but not last, the stop target makes a difference: If the pre-cache chain returns the stop condition, the value will not be added to the cache and the post-cache chain will not be run.

Adds a new chain with a certain name. This name can be used to refer to a specific chain, for example to jump to it.

Within the Chain block, there can be Rule blocks and Target blocks.

Adds a new rule to the current chain. The name of the rule is optional and currently has no meaning for the daemon.

Within the Rule block, there may be any number of Match blocks and there must be at least one Target block.

Adds a match to a Rule block. The name specifies what kind of match should be performed. Available matches depend on the plugins that have been loaded.

The arguments inside the Match block are passed to the plugin implementing the match, so which arguments are valid here depends on the plugin being used. If you do not need any to pass any arguments to a match, you can use the shorter syntax:

 Match "foobar"

Which is equivalent to:

 <Match "foobar">
 </Match>
Add a target to a rule or a default target to a chain. The name specifies what kind of target is to be added. Which targets are available depends on the plugins being loaded.

The arguments inside the Target block are passed to the plugin implementing the target, so which arguments are valid here depends on the plugin being used. If you do not need any to pass any arguments to a target, you can use the shorter syntax:

 Target "stop"

This is the same as writing:

 <Target "stop">
 </Target>

The following targets are built into the core daemon and therefore need no plugins to be loaded:

Signals the "return" condition, see the "Flow control" section above. This causes the current chain to stop processing the value and returns control to the calling chain. The calling chain will continue processing targets and rules just after the jump target (see below). This is very similar to the RETURN target of iptables, see iptables(8).

This target does not have any options.

Example:

 Target "return"
Signals the "stop" condition, see the "Flow control" section above. This causes processing of the value to be aborted immediately. This is similar to the DROP target of iptables, see iptables(8).

This target does not have any options.

Example:

 Target "stop"
Sends the value to "write" plugins.

Available options:

Name of the write plugin to which the data should be sent. This option may be given multiple times to send the data to more than one write plugin.

If no plugin is explicitly specified, the values will be sent to all available write plugins.

Example:

 <Target "write">
   Plugin "rrdtool"
 </Target>
Starts processing the rules of another chain, see "Flow control" above. If the end of that chain is reached, or a stop condition is encountered, processing will continue right after the jump target, i. e. with the next target or the next rule. This is similar to the -j command line option of iptables, see iptables(8).

Available options:

Jumps to the chain Name. This argument is required and may appear only once.

Example:

 <Target "jump">
   Chain "foobar"
 </Target>

Matches a value using regular expressions.

Available options:

Match values where the given regular expressions match the various fields of the identifier of a value. If multiple regular expressions are given, all regexen must match for a value to match.
When set to true, the result of the match is inverted, i.e. all value lists where all regular expressions apply are not matched, all other value lists are matched. Defaults to false.

Example:

 <Match "regex">
   Host "customer[0-9]+"
   Plugin "^foobar$"
 </Match>
Matches values that have a time which differs from the time on the server.

This match is mainly intended for servers that receive values over the "network" plugin and write them to disk using the "rrdtool" plugin. RRDtool is very sensitive to the timestamp used when updating the RRD files. In particular, the time must be ever increasing. If a misbehaving client sends one packet with a timestamp far in the future, all further packets with a correct time will be ignored because of that one packet. What's worse, such corrupted RRD files are hard to fix.

This match lets one match all values outside a specified time range (relative to the server's time), so you can use the stop target (see below) to ignore the value, for example.

Available options:

Matches all values that are ahead of the server's time by Seconds or more seconds. Set to zero for no limit. Either Future or Past must be non-zero.
Matches all values that are behind of the server's time by Seconds or more seconds. Set to zero for no limit. Either Future or Past must be non-zero.

Example:

 <Match "timediff">
   Future  300
   Past   3600
 </Match>

This example matches all values that are five minutes or more ahead of the server or one hour (or more) lagging behind.

Matches the actual value of data sources against given minimum / maximum values. If a data-set consists of more than one data-source, all data-sources must match the specified ranges for a positive match.

Available options:

Sets the smallest value which still results in a match. If unset, behaves like negative infinity.
Sets the largest value which still results in a match. If unset, behaves like positive infinity.
Inverts the selection. If the Min and Max settings result in a match, no-match is returned and vice versa. Please note that the Invert setting only effects how Min and Max are applied to a specific value. Especially the DataSource and Satisfy settings (see below) are not inverted.
Select one or more of the data sources. If no data source is configured, all data sources will be checked. If the type handled by the match does not have a data source of the specified name(s), this will always result in no match (independent of the Invert setting).
Specifies how checking with several data sources is performed. If set to Any, the match succeeds if one of the data sources is in the configured range. If set to All the match only succeeds if all data sources are within the configured range. Default is All.

Usually All is used for positive matches, Any is used for negative matches. This means that with All you usually check that all values are in a "good" range, while with Any you check if any value is within a "bad" range (or outside the "good" range).

Either Min or Max, but not both, may be unset.

Example:

 # Match all values smaller than or equal to 100. Matches only if all data
 # sources are below 100.
 <Match "value">
   Max 100
   Satisfy "All"
 </Match>
 
 # Match if the value of any data source is outside the range of 0 - 100.
 <Match "value">
   Min   0
   Max 100
   Invert true
   Satisfy "Any"
 </Match>
Matches all values with one or more data sources of type COUNTER and where all counter values are zero. These counters usually never increased since they started existing (and are therefore uninteresting), or got reset recently or overflowed and you had really, really bad luck.

Please keep in mind that ignoring such counters can result in confusing behavior: Counters which hardly ever increase will be zero for long periods of time. If the counter is reset for some reason (machine or service restarted, usually), the graph will be empty (NAN) for a long time. People may not understand why.

Calculates a hash value of the host name and matches values according to that hash value. This makes it possible to divide all hosts into groups and match only values that are in a specific group. The intended use is in load balancing, where you want to handle only part of all data and leave the rest for other servers.

The hashing function used tries to distribute the hosts evenly. First, it calculates a 32 bit hash value using the characters of the hostname:

  hash_value = 0;
  for (i = 0; host[i] != 0; i++)
    hash_value = (hash_value * 251) + host[i];

The constant 251 is a prime number which is supposed to make this hash value more random. The code then checks the group for this host according to the Total and Match arguments:

  if ((hash_value % Total) == Match)
    matches;
  else
    does not match;

Please note that when you set Total to two (i. e. you have only two groups), then the least significant bit of the hash value will be the XOR of all least significant bits in the host name. One consequence is that when you have two hosts, "server0.example.com" and "server1.example.com", where the host name differs in one digit only and the digits differ by one, those hosts will never end up in the same group.

Available options:

Divide the data into Total groups and match all hosts in group Match as described above. The groups are numbered from zero, i. e. Match must be smaller than Total. Total must be at least one, although only values greater than one really do make any sense.

You can repeat this option to match multiple groups, for example:

  Match 3 7
  Match 5 7

The above config will divide the data into seven groups and match groups three and five. One use would be to keep every value on two hosts so that if one fails the missing data can later be reconstructed from the second host.

Example:

 # Operate on the pre-cache chain, so that ignored values are not even in the
 # global cache.
 <Chain "PreCache">
   <Rule>
     <Match "hashed">
       # Divide all received hosts in seven groups and accept all hosts in
       # group three.
       Match 3 7
     </Match>
     # If matched: Return and continue.
     Target "return"
   </Rule>
   # If not matched: Return and stop.
   Target "stop"
 </Chain>

Creates and dispatches a notification.

Available options:

This required option sets the message of the notification. The following placeholders will be replaced by an appropriate value:
%{host}
%{plugin}
%{plugin_instance}
%{type}
%{type_instance}
These placeholders are replaced by the identifier field of the same name.
%{ds:name}
These placeholders are replaced by a (hopefully) human readable representation of the current rate of this data source. If you changed the instance name (using the set or replace targets, see below), it may not be possible to convert counter values to rates.

Please note that these placeholders are case sensitive!

Sets the severity of the message. If omitted, the severity "WARNING" is used.

Example:

  <Target "notification">
    Message "Oops, the %{type_instance} temperature is currently %{ds:value}!"
    Severity "WARNING"
  </Target>
Replaces parts of the identifier using regular expressions.

Available options:

Match the appropriate field with the given regular expression Regex. If the regular expression matches, that part that matches is replaced with Replacement. If multiple places of the input buffer match a given regular expression, only the first occurrence will be replaced.

You can specify each option multiple times to use multiple regular expressions one after another.

Example:

 <Target "replace">
   # Replace "example.net" with "example.com"
   Host "\\<example.net\\>" "example.com"
 
   # Strip "www." from hostnames
   Host "\\<www\\." ""
 </Target>
Sets part of the identifier of a value to a given string.

Available options:

Set the appropriate field to the given string. The strings for plugin instance and type instance may be empty, the strings for host and plugin may not be empty. It's currently not possible to set the type of a value this way.

Example:

 <Target "set">
   PluginInstance "coretemp"
   TypeInstance "core3"
 </Target>

If you use collectd with an old configuration, i. e. one without a Chain block, it will behave as it used to. This is equivalent to the following configuration:

 <Chain "PostCache">
   Target "write"
 </Chain>

If you specify a PostCacheChain, the write target will not be added anywhere and you will have to make sure that it is called where appropriate. We suggest to add the above snippet as default target to your "PostCache" chain.

Ignore all values, where the hostname does not contain a dot, i. e. can't be an FQDN.

 <Chain "PreCache">
   <Rule "no_fqdn">
     <Match "regex">
       Host "^[^\.]*$"
     </Match>
     Target "stop"
   </Rule>
   Target "write"
 </Chain>

collectd(1), collectd-exec(5), collectd-perl(5), collectd-unixsock(5), types.db(5), hddtemp(8), iptables(8), kstat(3KSTAT), mbmon(1), psql(1), regex(7), rrdtool(1), sensors(1)

Florian Forster <octo@verplant.org>