scheduler: Handle cycles in directly chained dependencies #3490
Conversation
Otherwise the unhandled exception prevents the scheduler from processing *any* further scheduled jobs and jobs are stuck in the scheduled state. See https://progress.opensuse.org/issues/32545#note-37 for a graph with an example of such a problematic directly chained cluster.
Codecov Report
@@ Coverage Diff @@
## master #3490 +/- ##
==========================================
+ Coverage 95.52% 95.57% +0.05%
==========================================
Files 366 366
Lines 31632 31660 +28
==========================================
+ Hits 30215 30259 +44
+ Misses 1417 1401 -16
Continue to review full report at Codecov.
|
so they prevent other jobs from being assigned
| @@ -141,6 +143,33 @@ subtest 'assign multiple jobs to worker' => sub { | |||
| ok($job_token, 'job token present'); | |||
| }; | |||
|
|
|||
| # prevent writing to a log file to enable use of combined_like in the following tests | |||
| my $usual_log = $t->app->log; | |||
| $t->app->log(Mojo::Log->new(level => 'debug')); | |||
There was a problem hiding this comment.
Maybe this should just be enabled at the start of the test? 🤔
There was a problem hiding this comment.
No, it shouldn't. I also intentionally disable it immediately after the tests using it again. That's because other logs messages would otherwise interfere with the test output (unless explicitly handled).
There was a problem hiding this comment.
can you do that within the subtest as well as the restoring?
|
Ah, nice that you've already created a ticket for the failure. Unfortunately I'm currently not sure where it comes from. And the OBS failures should hopefully go away with #3491. |
Otherwise the unhandled exception prevents the scheduler from processing
any further scheduled jobs and jobs are stuck in the scheduled state.
See https://progress.opensuse.org/issues/32545#note-37 for a graph with
an example of such a problematic directly chained cluster.