Provided by: rrdtool_1.7.1-2_amd64 bug


       rrdgraph_libdbi - fetching data for graphing in rrdtool graph via libdbi


       <rrdfile> = sql//<libdbi driver>/<driver-option-name>=<driver-option-value>/...
       [/rrdminstepsize=<stepsize>][/rrdfillmissing=<fill missing n seconds>]//<table>
       /<unixtimestamp column>/<data value column>[/derive]/<where clause 1>/.../<where clause n>


       This pseudo-rrd-filename defines a sql datasource:

                 magic cookie-prefix for a libdbi type datasource

       <libdbi driver>
                 which libdbi driver to use (e.g.: mysql)

                 defines the parameters that are required to connect to the database with the given libdbi driver
                 (These drivers are libdbi dependent - for details please look at the driver documentation of libdbi!)

       /rrdminstepsize=<minimum step size>
                 defines the minimum number of the step-length used for graphing (default: 300 seconds)

       /rrdfillmissing=<fill missing seconds>
                 defines the number of seconds to fill with the last value to avoid NaN boxes due to data-insertation jitter (default: 0 seconds)

                 defines the table from which to fetch the resultset.

                 If there is a need to fetch data from several tables, these tables can be defined by separating the tablenames with a "+"

                 hex-type-encoding via %xx are translated to the actual value, use %% to use %

       <[*]unixtimestamp column>
                 defines the column of <table> which contains the unix-timestamp
                 - if this is a DATETIME field in the database, then prefix with leading '*'

                 hex-type-encoding via %xx are translated to the actual value, use %% to use %

       <data value column>
                 defines the column of <table> which contains the value column, which should be graphed

                 hex-type-encoding via %xx are translated to the actual value, use %% to use %

                 defines that the data value used should be the delta of the 2 consecutive values (to simulate COUNTER or DERIVE type datasources)

       /<where clause(s)>
                 defines one (ore more) where clauses that are joined with AND to filter the entries in the <table>

                 hex-type-encoding via %xx are translated to the actual value, use %% to use %

       the returned value column-names, which can be used as ds-names, are:

       min, avg, max, count and sigma
                 are returned to be used as ds-names in your DS definition.
                 The reason for using this is that if the consolidation function is used for min/avg and max, then the engine is used several times.
                 And this results in the same SQL Statements used several times


       Here an example of a table in a MySQL database:

         DB connect information

         here the table:
           CREATE TABLE RRDValue (
             RRDKeyID      bigint(20) NOT NULL,
             UnixTimeStamp int(11) NOT NULL,
             value         double default NOT NULL,
             PRIMARY KEY  (RRDKeyID,UnixTimeStamp)

       and the RRDKeyID we want to graph for is: 1141942900757789274

       The pseudo rrd-filename to access this is:

       To illustrate this here a command to create a graph that contains the actual values.

         rrdtool graph test.png --imgformat=PNG --start=-1day --end=+3hours --width=1000 --height=600 \
           "DEF:min=$DS_BASE:min:AVERAGE" \
           "LINE1:min#FF0000:value" \
           "DEF:avg=$DS_BASE:avg:AVERAGE" \
           "LINE1:avg#00FF00:average" \
           "DEF:max=$DS_BASE:max:AVERAGE" \
           "LINE1:max#FF0000:max" \
           "DEF:sigma=$DS_BASE:sigma:AVERAGE" \
           "CDEF:upper=avg,4,sigma,*,+" \
           "LINE1:upper#0000FF:+4 sigma" \
           "CDEF:lower=avg,4,sigma,*,-" \
           "LINE1:lower#0000FF:-4 sigma"


       * Naturally you can also use any other kind of driver that libdbi supports - e.g.
       postgres, ...

       * From the way the data source is joined, it should also be possible to do joins over
       different tables
         (separate tables with "," in table and add in the WHERE Clauses the table equal joins.
         This has not been tested!!!)

       * It should also be relatively simple to add to the database using the same data source
         This has not been implemented...

       * The aggregation functions are ignored and several data columns are used instead
         to avoid querying the same SQL several times when minimum, average and maximum are
       needed for graphing...

       * for DB efficiency you should think of having 2 tables, one containing historic values
       and the other containing the latest data.
         This second table should be kept small to allow for the least amount of blocking SQL
         With mysql you can even use myisam table-type for the first and InnoDB for the second.
         This is especially interesting as with tables with +100M rows myisam is much smaller
       then InnoDB.

       * To debug the SQL statements set the environment variable RRDDEBUGSQL and the actual SQL
       statements and the timing is printed to stderr.

Performance issues with MySQL backend

       Previous versions of LibDBI have a big performance issue when retrieving data from a MySQL
       server. Performance impact is exponentially based on the number of values you retrieve
       from the database.  For example, it would take more than 2 seconds to graph 5DS on 150
       hours of data with a precision of 5 minutes (against 100ms when data comes from a RRD
       file). This bug has been fixed in version 0.9.0 of LibDBI.  You can find more information
       on this libdbi-users mailing list thread:


       * at least on Linux please make sure that the libdbi driver is explicitly linked against
         check via ldd /usr/lib/dbd/, that there is a line with
         otherwise at least the perl module RRDs will fail because the dynamic linker cannot find
       some symbols from
         (this only happens when the libdbi driver is actually used the first time!)
         This is KNOWN to be the case with RHEL4 and FC4 and FC5! (But actually this is a bug
       with libdbi make files!)

       * at least version 0.8.1 of libdbi exhibits a bug with BINARY fields
         (shorttext,text,mediumtext,longtext and possibly also BINARY and BLOB fields),
         that can result in coredumps of rrdtool.
         The tool will tell you on stderr if this occurs, so that you know what may be the
         If you are not experiencing these coredumps, then set the environment variable
         and then the message will not get shown.


       Martin Sperl <>