Running benchmark¶

Introduction¶

This document describes how to run benchmarking tool.

Zaqar Contributors can use this tool to test how the particular code change affects Zaqar’s performance.

Usage¶

1. First install and run zaqar-server.

   For example, you can setup Zaqar in development environment.

   See Setting up a development environment.

2. In your terminal cd into your local Zaqar repo and install additional requirements:

   $ pip install -r bench-requirements.txt
   

3. Copy the configuration file to ~/.zaqar:

   $ cp etc/zaqar-benchmark.conf.sample ~/.zaqar/zaqar-benchmark.conf
   

4. In this configuration file specify where zaqar-server can be found:

   server_url = http://localhost:8888
   

5. The benchmarking tool needs a set of messages to work with. Specify the path to the file with messages in the configuration file. Alternatively, put it in the directory with the configuration file and name it zaqar-benchmark-messages.json. As a starting point, you can use the sample file from the etc directory:

   $ cp etc/zaqar-benchmark-messages.json ~/.zaqar/
   

   If the file is not found or no file is specified, a single hard-coded message is used for all requests.

6. Run the benchmarking tool using the following command:

   $ zaqar-bench
   

   By default, the command will run a performance test for 5 seconds, using one producer process with 10 greenlet workers, and one observer process with 5 workers. The consumer role is disabled by default.

   You can override these defaults in the config file or on the command line using a variety of options. For example, the following command runs a performance test for 30 seconds using 4 producer processes with 20 workers each, plus 4 consumer processes with 20 workers each.

   Note that the observer role is also disabled in this example by setting its number of workers to zero:

   $ zaqar-bench -pp 4 -pw 10 -cp 4 -cw 20 -ow 0 -t 30
   

   By default, the results are in human-readable format. For JSON output add the --noverbose flag. The non-verbose output looks similar to the following:

   $ zaqar-bench --noverbose
   Using 'envvars' credentials
   Using 'keystone' authentication method
   Benchmarking Zaqar API v2...
   {"params": {"consumer": {"processes": 1, "workers": 0}, "observer": {"processes": 1, "workers": 5}, "producer": {"processes": 1, "workers": 10}}, "consumer": {"claim_total_requests": 0, "ms_per_claim": 0, "total_reqs": 0, "reqs_per_sec": 0, "successful_reqs": 0, "duration_sec": 0, "ms_per_delete": 0, "messages_processed": 0}, "producer": {"duration_sec": 8.569170951843262, "ms_per_req": 201.715140507139, "total_reqs": 29, "successful_reqs": 29, "reqs_per_sec": 3.384224700729303}, "observer": {"duration_sec": 8.481178045272827, "ms_per_req": 407.40778711107043, "total_reqs": 18, "successful_reqs": 18, "reqs_per_sec": 2.122346672115049}}
   

   By default, zaqar-bench is benchmarking Zaqar API version 2. To run benchmark against other API versions use -api parameter. For example:

   $ zaqar-bench -api 1.1