QA (public) » openQA Project (public)

Motivation¶

Some tests generate heavy disk, CPU or memory load and multiple tests of this type running in parallel may cause worker overload that'll lead to timeouts. See #179080 for example. In order to restrict these stress tests to reasonable resource use openQA should offer a feature to restrict scheduling "too many, too heavy" jobs in parallel where "heavy" is not based on qemu parameters but rather test maintainers setting a certain test variable, e.g. WORKER_CLASS=size-heavy

Acceptance Criteria¶

• AC1: Overloading of external resources (e.g. hypervisors) can be prevented by test maintainers
• AC2: By default test schedulers

Suggestions¶

• Should we use special worker class?
• From a discussion between okurz and mkittler we could extend the openQA worker with multiple features for workers ini so that we can go from

[global]
WORKER_CLASS=location-here,my_special

[1]
WORKER_CLASS=foo,location-here,my_special
[2]
WORKER_CLASS=foo,location-here,my_special
[3]
WORKER_CLASS=foo,location-here,my_special

to something like this

[global]
WORKER_CLASS=location-here,my_special

[1-2,5-8]
WORKER_CLASS=foo  # concatenates with the global one to WORKER_CLASS=foo,location-here,my_special

[5] ## additional override for 5 only
WORKER_CLASS=  # delete global entries
WORKER_CLASS=bar,location-elsewhere

The worker class concatenation from global and specific instances would be a breaking feature. As alternative one could consider WORKER_CLASS+= to explicitly concatenate to previously defined values