MONGO-PERF:

Mongo-perf (not to be confused with mongoperf) is a micro benchmarking tool for the MongoDB server. It measures throughput of commands with regards to the number of threads.

OVERVIEW:

This repo contains scripts to run benchmark tests for MongoDB.

DEPENDENCIES:

General Benchmarking Dependencies

• Python >= 2.7.X < 3.0
• mongo shell >= 2.7.7-pre- (at revision 881b3a97fb5080b4e5d5ce11ad016da73ea23931 or newer)

Installing Python Dependencies pip install -r requirements.txt

Python Benchmarking Dependencies

• argparse

Python Reporting Dependencies

• bottle
• pymongo

HOW TO RUN:

To run a micro benchmarking test or tests:
python benchrun.py -f <list of testfiles> -t <list of thread configs> [-m <number of dbs>] [-s <shell path>]

For example, to run the simple_insert.js test case on 1, 2, and 4 threads, no multi-db and using the basic mongo shell:
python benchrun.py -f testcases/simple_insert.js -t 1 2 4

To run the single test case 'Queries.Emptyon 1, 2, and 4 thread:python benchrun.py -f testcases/* --includeFilter Queries.Empty -t 1 2 4`

To run all insert and update test cases on 1, 2, and 4 threads, for 10 seconds each using the basic mongo shell:
python benchrun.py -f testcases/* -t 1 2 4 --includeFilter insert update --trialTime 10

To run all insert and update test cases that are also have the tag core on 1, 2, and 4 threads, for 10 seconds each using the basic mongo shell:
python benchrun.py -f testcases/* -t 1 2 4 --includeFilter insert update --includeFilter core --trialTime 10

For a complete list of options :
python benchrun.py --help

RESULT CHANGES

Mongo-perf is built upon the Mongo shell benchrun command. The results format of benchrun changed in Mongo 3.1.5 and 3.0.5. Because of the result changes, mongo-perf results from before 3.1.5 or 3.0.5 may not be directly comparable to results after 3.1.5 or 3.0.5.

As of Mongo 3.1.5 and 3.0.5 the benchrun command measures op performance on the client side, instead of on the server side. Any and only those ops passed into the op array of the benchrun command are counted as ops for the purpose of reporting throughput. In some cases this may cause the reported throughput to be higher than previous version of mongo-perf (for instance, if the "let" operation is used), or lower than before (for instance, if the shell issues getMore commands in addition to a query in order to complete an op).