Details
-
Task
-
Status: Closed (View Workflow)
-
Minor
-
Resolution: Fixed
-
None
Description
Mini-benchmark has been useful for locally-run benchmarks, recently it has been used to provide benchmarks for MDEV-30628. However, using the script can be tedious since it requires manually modifying the code to change the benchmark configuration, the current implementation also doesn't save standard output of the script (lscpu and other output is lost).
I'm proposing the following CLI options to make the script easier to use:
- --log: logs the entire mini-benchmark standard output to a file (lscpu, etc.)
- --duration: set the duration in seconds
- --name: set the name of the benchmark (for the folder name and log file)
- --workload: the sysbench workload (oltp_read_write, oltp_write_only, etc.)
- --threads: the threads to run the benchmark on
I have already implemented the options locally and they have been useful. For example, with MDEV-30628 I used the CLI options to set the duration to 600 seconds and to only test 64 threads (the issue consistently happened with this duration and thread count). The --name option was useful to denote anything unique about a specific run (some runs were based on different commits, branches).