Bill Farner created AURORA-1149:
-----------------------------------
Summary: Decide the appropriate way to handle TASK_ERROR
Key: AURORA-1149
URL: https://issues.apache.org/jira/browse/AURORA-1149
Project: Aurora
Issue Type: Story
Components: Scheduler, Usability
Reporter: Bill Farner
This may mean the state machine needs to accommodate this state. The outcome
should depend on whether we believe TASK_ERROR is truly permanently fatal, or
whether it could represent a transient configuration error in some component
(e.g. scheduler, master). If we believe TASK_ERROR is indeed always fatal,
this should be a terminal and non-reschedulable state regardless of whether the
task is a service. However, if it can be transient, we may still want to try
to revive the task to avoid temporary configuration errors from tearing down
services.
Personally, i'm leaning towards the retry case.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
