Re: [PATCH libevl 2/4] evl-test: run latmus in the end of the test suite
Philippe Gerum <[email protected]> Mon, 06 Jul 2026 11:00:45 +0200
| Newsgroups | dev.linux.lists.xenomai |
|---|---|
| Message-ID | <[email protected]> |
Tobias Schaffner <[email protected]> writes: > Run latmus at the end of the session, before tearing down, to allow measuring > the worst-case latency under load if a stress-command is given. > > Add --latmus-args to pass extra arguments to latmus (e.g. -A threshold or -T to > override the runtime). > > latmus runs quietly for 10 seconds by default; a runtime passed through > --latmus-args takes precedence. Its own output is dropped by default and > can be redirected with --latmus-out. > > Examples: > # default: tests + a 10s latmus run, latmus output discarded > evl test > > # fail if worst-case latency exceeds 150us > evl test --latmus-args "-A 150" > > # measure latency under load for 60s, save the report > evl test --stress-cmd "stress-ng --cpu 4" \ > --latmus-args "-T 60 -A 200" --latmus-out /tmp/latmus.log > The purpose is fine by me, but the integration logic looks too complicated. Would it be possible to simply have a 'latency' test (as a shell script) available from $libexec/evl/tests, that would launch the latmus executable under the hood with the provided arguments? Spawning it would amount to using the common syntax: $ evl test latency [-- latmus-args ...] With this, we could achieve the same goal with no addition required to evl-test whatsoever: $ evl test -k $ evl test latency -- -A 150 -T 10s > /tmp/latmus.log -- Philippe.