kav23alex / pg-mark

Postgres benchmarking framework

Postgres Benchmarking Infrastructure

This repository contains scripts and a sample configuration file for benchmarking different PostgreSQL and Postgres Pro versions and flavors.

Notice

Using a dedicated machine is highly recommended at least for minimizing background/parallel activity. Also it's recommended to disable turbo/boost CPU modes to avoid uncontrollable and unpredictable CPU performance fluctuations.

Sample usage

0. Prepare local git repository:

   git clone git://git.postgresql.org/git/postgresql.git postgres.git
   

And install prerequisites: docker, bash, wget, git, tar, 7z, ant, default-jdk
(optional packages for visualization: xsltproc, r-base-core, r-cran-xml, r-cran-ggplot2, r-cran-reshape2)

---

1. Run

   ./prepare-instances.py
   

to get all the Postgres* instances defined in config.xml ready for benchmarking.

You can also specify different configuration file or create only selected instances, e.g.:

./prepare-instances.py -i pg-src-15 pg-src-master

Run

./prepare-instances.py --help

to get more information.

---

2. Run

    ./run-benchmarks.py
   

to perform all the benchmarks Postgres* defined in config.xml for all instances.

You'll get benchmark-results.xml with the normalized benchmarking data and benchmark-results/ directory with a raw benchmarks' output.

You can specify a different configuration file or perform only selected benchmarks for selected instances, e.g.:

./run-benchmarks.py -i pg-src-15 pg-src-master pg-src-15 pg-src-master \
  pg-src-15 pg-src-master -b pgbench_native pgbench_reference ycsb s64da_tpch

Here the instances repeated to get more trustworthy results for comparison of that instances. The benchmark results will be stored as if the instances were named "pg-src-15--1", "pg-src-master--1", "pg-src-15--2", "pg-src-master--2", ...

Run

./run-benchmarks.py --help

to get more information.

---

3. To visualize results, you can use:

   R --no-save < VisualizeResults.R
   

(You'll get Rplots.pdf with graphics presenting some benchmarks results.) Or

xsltproc make-html-tables.xsl benchmark-results.xml >benchmark-results.html

(You'll get benchmark-results.html with tables presenting the benchmarking data.)

---

4. You can also compare benchmark results for several instances in an automated non-visual way. Run

   ./analyze-benchmarks.py -i 'instance-1-pattern' 'instance-2-pattern'
   

For example, to compare results of benchmarking instances shown above, run:

./analyze-benchmarks.py -i 'pg-src-master--.*' 'pg-src-15--.*'

This script can also be used to perform git bisect for finding a commit, that changed some metric. E. g.:

sed "s|\(</pg_instances>\)|<instance id='pg-src-probe' type='src' git_commit='$hash' />\1|g" \
  config.xml > config-probe.xml
time ./prepare-instances.py -c config-probe.xml -i pg-src-probe >prepare.log || exit 125
rm benchmark-results.xml || true
time ./run-benchmarks.py -c config-probe.xml -i pg-src-probe pg-src-probe pg-src-probe \
  pg-src-probe pg-src-probe -b s64da_tpcds
./analyze-benchmarks.py -i 'pg-src-probe--.*' -m s64da_tpcds.query87 -t 2.1 || exit 1

---

Configuration and data structure

A configuration of postgres instances and benchmarks is defined in a single file config.xml (custom configuration files can also be used). It allows to store the complete configuration of a benchmarking session along with the results (benchmark-results.xml).

The structure of config.xml yet to be documented, but it's supposed to be transparent and self-explanatory.

Notice

You can add private repositories and instance definitions with the following extra configuration files:
private_repositories.xml:

<repository id="pgproee"
 url="https://user:[email protected]/pgproee-$PG_VERSION/ubuntu $OS_CODENAME main" />

private_instances.xml:

<private>
    <instance id="pg-proee-apt-15" type="proapt" pgpro_edition="ent"
     repository="pgproee" pg_version="15" />
</private>