action #120735
coordination #68794: [qe-core][functional][epic] rework postfail hooks
[qe-core] PoC: Decrease typing on `post_fail_hooks` log collection
0%
Description
Right now we have several routines that are doing a lot of typing, and a lot of it could be executed in a single script, (if what has to be typed is longer than 80 characters, os-autoinst will upload one script instead, and execute that), however we could benefit from running scripts directly to collect all the information in a single go, per log level.
While we're at it, we can extract the export_logs_* to Utils::Logging::
or a better namespace, so we can shed some lines from opensusebasetest
- lib/opensusebasetest.pm:471:sub export_logs_basic {
- lib/opensusebasetest.pm:504:sub export_logs {
- lib/opensusebasetest.pm:539:sub export_logs_locale {
- lib/opensusebasetest.pm:596:sub export_logs_desktop {
Once this is done, we should shave some minutes of typing (see parent ticket) which can be up to 30m
Suggestions¶
- Take note on how long do we expend on collecting logs for a single test that fails across different architectures
- Prepare a proposal with the changes and discuss them with the team
Acceptance Criteria¶
- AC1: Proposal has been created (doesn't need to be merged)
- AC2: export_logs* has been extracted to its own module
Related issues
History
#1
Updated by szarate 2 months ago
- Copied from action #120720: [qe-core] Use Utils::Logging::save_ulog to redefine upload_logs in opensusebasetest added
#3
Updated by szarate 2 months ago
- Copied to action #120882: [qe-core] ensure `os-autoinst` reports properly the execution time of `post_fail_hooks` added
#5
Updated by szarate about 2 months ago
- Sprint changed from QE-Core: November Sprint (Nov 09 - Dec 07) to QE-Core: December Sprint (Dec 07 - Jan 04)
#8
Updated by dimstar about 1 month ago
pdostal wrote:
https://github.com/os-autoinst/os-autoinst-distri-opensuse/pull/16083
That has intorduced an interesting regression, as seen for example on https://openqa.opensuse.org/tests/2968643#step/kate/4
historically, 'kate' has been a stable test, no change in the snapshot to make believe it should fail today. Firefox, that fails before it, is 'known' (in the KDE live setup)
It seems that after firefox, we go 'serial' for the post_fail_hook (as introduced here) - then reload LASTGOOD snapshot, at which state we lose the information on how the serial state is. The next few tests all run on X11 console (but never explicitly switch to it - is the system even aware it is on X11 now or is VNC interaction already an accident?); kate then does ensure_installed, which tries to interact with the serial console, at which point things explode