ignite-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Mashenkov (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (IGNITE-8892) Iterating over large dataset via ScanQuery can fails with OOME.
Date Thu, 28 Jun 2018 14:23:00 GMT

    [ https://issues.apache.org/jira/browse/IGNITE-8892?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16526318#comment-16526318
] 

Andrew Mashenkov edited comment on IGNITE-8892 at 6/28/18 2:22 PM:
-------------------------------------------------------------------

So, the issue here is ScanQuery.keepAll to true by default.
This bug affects all type of queries. Cache.iterator() works fine as it explicitly disabled
keepAll flag.

There is no workaround as keepAll flag is internal feature and can't be set from user side.


was (Author: amashenkov):
So, the issue here is we set ScanQuery.keepAll to true by default.

> Iterating over large dataset via ScanQuery can fails with OOME.
> ---------------------------------------------------------------
>
>                 Key: IGNITE-8892
>                 URL: https://issues.apache.org/jira/browse/IGNITE-8892
>             Project: Ignite
>          Issue Type: Bug
>          Components: cache
>            Reporter: Andrew Mashenkov
>            Priority: Critical
>              Labels: OutOfMemoryError
>             Fix For: 2.7
>
>         Attachments: ScanQueryOOM.java
>
>
> Seems, iterating over query iterator (ScanQuery at least, but may be other affected as
well) on client node cause memory leakage.
> The use case is quite simple.
>  Start server and client. Put much data into cache, then iterate over all entries via
ScanQuery.
>  Looks like JVM crashed due to OOM as GridCacheDistributedQueryManager.futs map contains
to many GridCacheDistributedQueryFuture futures.
> I've put 15kk entries into cache and client failed with OOM after iterating over 10kk
entry.
>  In heapdump I observer 2*10^9 GridCacheDistributedQueryFuture futures. 
> We have to check
>  # if these futures removed from map correctly.
>  # we don't create unnecessary futures.
> PFA repro.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message