backupccl: pause backup schedule in TestFullClusterBackup #71907
Conversation
|
This change is |
Currently the backup schedule in TestFullClusterBackup can be processed by the job scheduler in the original DB after the backup has been taken, thus causing a difference between the original and restored clusters. Prevent this processing by setting the first run in the future and pausing the schedule. Fixes cockroachdb#71435 Release note: None
rhu713
force-pushed
the
backup-test-fix
branch
from
f4f35bb to
5de7501
Compare
There was a problem hiding this comment.
Reviewable status:
complete! 0 of 0 LGTMs obtained (waiting on @dt and @rhu713)
pkg/ccl/backupccl/full_cluster_backup_restore_test.go, line 151 at r1 (raw file):
// Populate system.scheduled_jobs table with a first run in the future to prevent immediate adoption. firstRun := timeutil.Now().Add(time.Hour).Format(timeutil.TimestampWithoutTZFormat) sqlDB.Exec(t, `CREATE SCHEDULE FOR BACKUP data.bank INTO $1 RECURRING '@hourly' FULL BACKUP ALWAYS WITH SCHEDULE OPTIONS first_run = $2`, LocalFoo, firstRun)
Ah nice. I think when we were looking at the test failure we erroneously misread the output as being about the sql stats compaction job. I wonder though, is it possible that the same sort of issue happens with the stats compaction job?
There was a problem hiding this comment.
Reviewable status:
pkg/ccl/backupccl/full_cluster_backup_restore_test.go, line 151 at r1 (raw file):
Previously, stevendanna (Steven Danna) wrote…
Ah nice. I think when we were looking at the test failure we erroneously misread the output as being about the sql stats compaction job. I wonder though, is it possible that the same sort of issue happens with the stats compaction job?
I think it's unlikely for the stats compaction job since there's a 2-5 minute delay for the first scan of the job scheduler, and the stats jobs are paused right away in the test. But I think in theory it could happen:
https://github.com/cockroachdb/cockroach/blob/master/pkg/jobs/job_scheduler.go#L411
|
bors r+ |
|
Build failed (retrying...): |
|
Build succeeded: |
Currently the backup schedule in TestFullClusterBackup can be processed by the
job scheduler in the original DB after the backup has been taken, thus causing
a difference between the original and restored clusters. Prevent this
processing by setting the first run in the future and pausing the schedule.
Fixes #71435
Release note: None