You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Monitoring system tests: Collision on group.create #2093
See my comment on #2092. This error is slightly surprising. Let's find out more about in exactly what circumstances it occurs, and whether it is deterministic.
@tseaver How hard is it to reproduce these errors? I tried on my local machine and trying on Compute Engine right now and haven't been able to repro. I am doing this because I want to see if there is an equivalent to error['reason'] == 'rateLimitExceeded' in the caught error
@dhermes I saw them multiple times the day I was trying to squash all the flaky issues: if I had to guess, I would say that the issue would be related to load on the back-end.
I don't know why I never tried before, but just running nosetests system_tests/monitoring.py --pdb -vv in two separate bash shells at the same time forced a 503 on the first try.
FWIW @tseaver the 503 occurred in a 404-wrapped delete during tearDown of TestMonitoringGroups.test_group_hierarchy.
Activity
dhermes commentedon Aug 11, 2016
@rimey Can you weigh in here?
[-]Monitoring system tests: Collision on [/-][+]Monitoring system tests: Collision on group.create[/+]tseaver commentedon Aug 11, 2016
Smells like we just need to add a
RetryErrors(ServiceUnavailable)here.tseaver commentedon Aug 11, 2016
I think #2095 addresses this error.
rimey commentedon Aug 11, 2016
@dhermes I'm having trouble assigning this issue to @supriyagarg. Does she need to be a member of GoogleCloudPlatform?
rimey commentedon Aug 11, 2016
I think she needs to be a member of the GoogleCloudPlatform "Contributors" team.
rimey commentedon Aug 12, 2016
See my comment on #2092. This error is slightly surprising. Let's find out more about in exactly what circumstances it occurs, and whether it is deterministic.
dhermes commentedon Aug 18, 2016
Closed by #2095
dhermes commentedon Aug 18, 2016
@tseaver How hard is it to reproduce these errors? I tried on my local machine and trying on Compute Engine right now and haven't been able to repro. I am doing this because I want to see if there is an equivalent to
error['reason'] == 'rateLimitExceeded'in the caught errortseaver commentedon Aug 19, 2016
@dhermes I saw them multiple times the day I was trying to squash all the flaky issues: if I had to guess, I would say that the issue would be related to load on the back-end.
dhermes commentedon Aug 22, 2016
I don't know why I never tried before, but just running
nosetests system_tests/monitoring.py --pdb -vvin two separate bash shells at the same time forced a 503 on the first try.FWIW @tseaver the 503 occurred in a 404-wrapped delete during
tearDownofTestMonitoringGroups.test_group_hierarchy.From the debugger:
Also note that
because the error payload doesn't have the key
payload['error']['errors'].What we really want is a check like
payload['error']['status'] == u'UNAVAILABLE'just like insystem_tests.bigquery._rate_limit_exceeded.feat: support resource_tags for table (#2093)
refactor: reorganize the sqlglot scalar compiler layout - part 2 (#2093)