[
https://issues.apache.org/jira/browse/AURORA-1332?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14619519#comment-14619519
]
Bill Farner commented on AURORA-1332:
-------------------------------------
We already have an error response message when an update is a no-op, which you
could argue conflicts with the behavior requested here. For example, if the
client scopes the update to 0-3, but all of 0-3 are already updated, should we
return an error? I would buy the argument that we should honor this ticket,
and *not* return an error when an update is a no-op, but send back a signal
that the update required no work. This behavior would play nicely with the
pattern of idempotent requests.
> Updater does not skip already updated instances with /INSTANCES option.
> -----------------------------------------------------------------------
>
> Key: AURORA-1332
> URL: https://issues.apache.org/jira/browse/AURORA-1332
> Project: Aurora
> Issue Type: Bug
> Components: Scheduler
> Reporter: Maxim Khutornenko
> Assignee: Bill Farner
>
> The new updater is too restrictive when processing /INSTANCES option. E.g. if
> instance 0 is already updated but the client sends /0-3 range, the entire
> request is rejected with: "updateOnlyTheseInstances contains instances
> irrelevant to the update: [0]".
> https://github.com/apache/aurora/blob/2dc1d59f1e772e220b3bfb26480c3b90c688800f/src/main/java/org/apache/aurora/scheduler/thrift/SchedulerThriftInterface.java#L1175-L1181
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
