Load testing

A basic tool for simulating PuppetDB loads is included with the standard PuppetDB distribution. This tool is currently experimental and is likely to change in future releases.

The load testing tool can submit catalogs, facts, and/or reports at a defined interval, for a specified number of nodes. It is also able to vary the submitted catalogs over time to simulate catalog changes (which can cause a higher PuppetDB load). The intent of the tool is to get a rough idea about how the system (and software) will handle load under realistic conditions. The tool will use catalogs/reports/facts specified by the user, including those exported from a production/running system, to simulate real-world conditions. While originally designed as a PuppetDB developer tool, we expect that many users may find it useful.

Getting data for the tool

The load testing tool does not yet have the ability to generate its own data. To run simulations, you will need a collection of catalogs, facts and/or reports. The easiest source of this information is the export tool included with PuppetDB (more information here). The export will produce a tar.gz file containing facts, catalogs, and reports.

Unzip the exported data. When unzipped/untared, find the puppetdb-bak/catalogs, puppetdb-bak/facts and a puppetdb-back/reports directories. One or all of these directories can be used as input to the load testing tool.

Running the load testing tool (benchmark)

Before running the load testing tool, make sure you have the full path to your example data. You'll also need a config file (such as config.ini) with the host and port information for the PuppetDB instance you wish to load test. The config file format is the same as the one PuppetDB uses, but you only need two entries:

  host=<host name here>
  port=<port here>

There is currently no script for running the tool, so you'll need a command like the one below:

$ java -cp /opt/puppetlabs/server/apps/puppetdb/puppetdb.jar clojure.main \
    -m puppetlabs.puppetdb.cli.benchmark \
    --config myconfig.ini \
    --catalogs /tmp/puppetdb-bak/catalogs \
    --runinterval 30 --numhosts 1000 --rand-perc 10

Note that if you run it from the source tree via leiningen, you should make sure to use trampoline, i.e. lein trampoline run benchmark ... so that the tool can shut down and clean up normally.

Running on an agent other than the primary server

Running the PuppetDB benchmark tool on the primary server can cause resource constraints which may significantly skew performance on the primary server. If you would like to run the benchmark tool on an agent this can be achieved following the instructions below.

  • On the primary server, modify /etc/puppetlabs/puppetdb/conf.d/jetty.ini. In the [jetty] section, set either:

    • host= # http access from all agents

    • host=<agent ip address> # access from specific agent

  • Install java on the agent

  • On the agent, in the config.ini file set the port to the puppetdb port for http traffic (defaults to 8080)

After these steps have been completed you should be able to run the benchmark tool on the agent using the java -cp ... command described above.

Arguments accepted by the benchmark command

  • --config / -c: path to the INI file that has the host/port configuration for the PuppetDB instance to be tested.

  • --catalogs / -C: directory containing catalogs to use for testing (probably from a previous PuppetDB export).

  • --reports / -R: directory containing reports to use for testing (probably from a previous PuppetDB export).

  • --facts / -F: directory containing facts to use for testing (probably from a previous PuppetDB export).

  • --archive / -A: tarball archive obtained via a PuppetDB export. This option is incompatible with the preceding four.

  • --runinterval / -i: integer indicating the amount of time in minutes between puppet runs for each simulated node. Typical values are 30 or 60. Mutually exclusive with --nummsgs. This option requires some temporary filesystem space, which will be allocated in TMPDIR (if set in the environment), java.io.tmpdir (if that JVM property is set), or the default JVM location.

  • --nummsgs / -N: integer indicating the number of commands and/or reports to send for each host, after which benchmark will exit. Mutually exclusive with --runinterval.

  • --numhosts / -n: number of separate hosts that the tool should simulate.

  • --rand-perc / -r: what percentage of catalogs submissions should be changed (this simulates typical catalog changes, such as adding a resource, edge, or something similar). More changes to catalogs will cause a higher load on PuppetDB. A typical change percentage is 10.

  • --threads / -t: number of threads to use for command submission; defaults to four times the number of available processors.

Note: If --facts, --catalogs, --reports, and --archive are unspecified, the PuppetDB sample data will be used. This data includes catalogs, facts, and reports.