Don't fear the (sub)reaper #1590
Conversation
|
More cowbells! https://youtu.be/Gwb4o2qs7Pc?t=14 |
lib/OpenQA/Worker/Jobs.pm
Outdated
| $worker->{child}->on( | ||
| collected => sub { | ||
| my $self = shift; | ||
| STDERR->printflush("collect status: " . $self->exit_status . "\n"); |
There was a problem hiding this comment.
this changes are not going so much distant from #1579
lib/OpenQA/Worker/Jobs.pm
Outdated
| my $self = shift; | ||
| STDERR->printflush("collect status: " . $self->exit_status . "\n"); | ||
| # TODO: deal when the job was restarted. The process is always killed. | ||
| # This is only problematic if we die, Otherwise this is just to exterminate |
There was a problem hiding this comment.
Now that you are at this, mind changing the wording to process tree?
mudler
force-pushed
the
dontfearthesubreaper
branch
from
a9bca80
to
124f150
Compare
Codecov Report
@@ Coverage Diff @@
## master #1590 +/- ##
==========================================
+ Coverage 82.59% 88.62% +6.03%
==========================================
Files 121 121
Lines 9037 9040 +3
==========================================
+ Hits 7464 8012 +548
+ Misses 1573 1028 -545
Continue to review full report at Codecov.
|
mudler
added
the
acceptance-tests-needed
mudler
force-pushed
the
dontfearthesubreaper
branch
from
acd1aaf
to
0bef6f2
Compare
lib/OpenQA/Worker/Jobs.pm
Outdated
| @@ -914,8 +902,10 @@ sub check_backend { | |||
| log_debug("checking backend state"); | |||
|
|
|||
| return log_debug("backend is running") if $worker->{child}->is_running(); | |||
| return log_debug("backend state not known") unless defined $worker->{child}->exit_status; | |||
There was a problem hiding this comment.
I would move this to log_error
There was a problem hiding this comment.
actually we could remove that timer now, if stop_job would not be so 'blocking' - give me some time to try to re-elaborate this
lib/OpenQA/Worker/Jobs.pm
Outdated
|
|
||
| if ($worker->{child}->is_running()) { | ||
| log_debug("backend is not running anymore"); |
|
with RWP 0.16 the worker is always detecting dead children - as collect_status finds the children. There seems to be an incompatible change in there :( Had to notice after reinstalling the arm worker |
|
So yes, let me be clear a bit #1579 was a premature merge - as collect_status was an internal event and you was required to manually check what you are collecting for - aside from the missing bit shifting to be sure of what is the real status of the process. in >0.16 there is the 'collected' event that it's fired when we have the exit status of the process. This PR requires the latest version (0.19) .
|
|
tweaked MAX_JOB_TIME: (Search for 14311 - will show how without this change we would otherwise loose one qemu process) standard jobs: |
See related Progres issue: https://progress.opensuse.org/issues/32263
tests: we have 2 timers now
mudler
force-pushed
the
dontfearthesubreaper
branch
from
d82d0f0
to
c100b99
Compare
|
|
This is still a wip, and commit message will be replaced :) - needs also new RWP release as sessions were introduced recently - and there was a typo preventing orphaned child resolution during event firing when using subreapers, not in the feature itself (and update cpanfile accordingly once its done).