[
https://issues.apache.org/jira/browse/IGNITE-3209?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15331344#comment-15331344
]
Semen Boikov commented on IGNITE-3209:
--------------------------------------
Reviewed,
I have one concern: it seems that synchronization could be broken by this
change: GridTaskWorker.onResponse process response while
GridJobResultImpl.setOccupied is 'true', but now occupied flag can be reset
(line 901) and then response can be processed asynchronously by affinity future
listener.
Also I added one more test to check affinityCall from client node where
affinity cache is not configured.
> Need to wait for affinity assignment change during affinityCall failover
> ------------------------------------------------------------------------
>
> Key: IGNITE-3209
> URL: https://issues.apache.org/jira/browse/IGNITE-3209
> Project: Ignite
> Issue Type: Bug
> Components: compute
> Affects Versions: 1.6
> Reporter: Valentin Kulichenko
> Assignee: Andrey Gura
> Fix For: 1.7
>
>
> {{AlwaysFailoverSpi.failover()}} method makes several attempts (5 by default)
> to get new primary node for the affinity key. Affinity assignment takes time,
> so there is a big chance to make all these attempts before new node is
> returned.
> We need to add discovery event that initiated failover to {{FailoverContext}}
> and wait for affinity is assigned.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
