Bill Farner created AURORA-99:
---------------------------------
Summary: Do not queue task for local gc if it started after the
AdjustRetainedTask message was sent
Key: AURORA-99
URL: https://issues.apache.org/jira/browse/AURORA-99
Project: Aurora
Issue Type: Task
Components: Executor
Reporter: Bill Farner
Priority: Minor
Quoting [~wickman]:
Hmm, reading the logic in reconcile_states:
{noformat}
Local vs Scheduler => Action
===================================
ACTIVE ACTIVE => no-op
ACTIVE STARTING => no-op
ACTIVE TERMINAL => maybe kill task*
ACTIVE !EXISTS => maybe kill task*
TERMINAL ACTIVE => send actual status**
TERMINAL STARTING => send actual status**
TERMINAL TERMINAL => no-op
TERMINAL !EXISTS => gc locally
!EXISTS ACTIVE => send LOST**
!EXISTS STARTING => no-op
!EXISTS TERMINAL => gc remotely
{noformat}
What happened (I believe) is that when the AdjustRetainedTasks was sent, the
task did not exist, but by the time that the GC was run, the task had run to
completion locally.
We should probably not queue anything for local deletion if its start time
occurred after the AdjustRetainedTasks was sent.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
