Skip 3 expensive calls for jobs saving in 'waiting' status on UnifiedJob (#15174)

skip update parent logic for 'waiting' on UnifiedJob

by not looking up "status_before" from previous instance
we save 2 to 3 expensive calls (the self lookup of old state, the lookup
of parent, and the update to parent if allow_simultaneous == False or status == 'waiting')
This commit is contained in:
Elijah DeLee 2024-05-13 09:26:03 -05:00 committed by GitHub
parent 946bbe3560
commit 439c3f0c23
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -823,7 +823,7 @@ class UnifiedJob(
update_fields.append(key)
if parent_instance:
if self.status in ('pending', 'waiting', 'running'):
if self.status in ('pending', 'running'):
if parent_instance.current_job != self:
parent_instance_set('current_job', self)
# Update parent with all the 'good' states of it's child
@ -860,7 +860,7 @@ class UnifiedJob(
# If this job already exists in the database, retrieve a copy of
# the job in its prior state.
# If update_fields are given without status, then that indicates no change
if self.pk and ((not update_fields) or ('status' in update_fields)):
if self.status != 'waiting' and self.pk and ((not update_fields) or ('status' in update_fields)):
self_before = self.__class__.objects.get(pk=self.pk)
if self_before.status != self.status:
status_before = self_before.status
@ -902,7 +902,8 @@ class UnifiedJob(
update_fields.append('elapsed')
# Ensure that the job template information is current.
if self.unified_job_template != self._get_parent_instance():
# unless status is 'waiting', because this happens in large batches at end of task manager runs and is blocking
if self.status != 'waiting' and self.unified_job_template != self._get_parent_instance():
self.unified_job_template = self._get_parent_instance()
if 'unified_job_template' not in update_fields:
update_fields.append('unified_job_template')
@ -915,8 +916,9 @@ class UnifiedJob(
# Okay; we're done. Perform the actual save.
result = super(UnifiedJob, self).save(*args, **kwargs)
# If status changed, update the parent instance.
if self.status != status_before:
# If status changed, update the parent instance
# unless status is 'waiting', because this happens in large batches at end of task manager runs and is blocking
if self.status != status_before and self.status != 'waiting':
# Update parent outside of the transaction for Job w/ allow_simultaneous=True
# This dodges lock contention at the expense of the foreign key not being
# completely correct.