[
https://issues.apache.org/jira/browse/IGNITE-848?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14529171#comment-14529171
]
Andrey Gura commented on IGNITE-848:
------------------------------------
1. OOM isn't TTL manager problem. Indeed it is
{{GridCacheMapEntry.onTtlExpired}} bug that leads to
{{GridDhtAtomicCacheEntry}} instance leaks.
2. It is related with example itself: two queries execute sequentially, but
data set have changed when second query started.
3. CPU usage is very high (even streaming stopped) because cleanup worker
infinitely runs but user threads don't help to it. It is bug in
{{DataStreamerImpl.IsolatedUpdater.receive()}} that doesn't force eviction.
> Excessive memory and CPU consumption by TTL manager
> ---------------------------------------------------
>
> Key: IGNITE-848
> URL: https://issues.apache.org/jira/browse/IGNITE-848
> Project: Ignite
> Issue Type: Bug
> Affects Versions: sprint-4
> Reporter: Andrey Gura
> Assignee: Andrey Gura
> Priority: Blocker
> Fix For: sprint-4
>
>
> There is an excessive memory and CPU consumption on data nodes in case of
> expiration policy configured for cache and continuous data streaming performs.
> Some observation:
> 1. Heap fills up very quickly and eventually leads to OOM.
> 2. Inconsistent query results on the {{QueryWords}} example (with out
> expiration policy there are consistent).
> 3. There is suspicious CPU activity when streaming node stopped.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
