You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@poi.apache.org by "Dominik Stadler (JIRA)" <ji...@apache.org> on 2019/03/03 09:11:00 UTC
[jira] [Resolved] (XMLBEANS-525) Reduce locking/contention in
Path.getCompiledPath()
[ https://issues.apache.org/jira/browse/XMLBEANS-525?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Dominik Stadler resolved XMLBEANS-525.
--------------------------------------
Resolution: Won't Fix
Assignee: Dominik Stadler
Fix Version/s: unspecified
Contention did not show any more in subsequent tests so it seems the initial setup was maybe too limit in memory/hardware to be a valid test-setup.
> Reduce locking/contention in Path.getCompiledPath()
> ---------------------------------------------------
>
> Key: XMLBEANS-525
> URL: https://issues.apache.org/jira/browse/XMLBEANS-525
> Project: XMLBeans
> Issue Type: Improvement
> Affects Versions: Version 3.0.2
> Reporter: Dominik Stadler
> Assignee: Dominik Stadler
> Priority: Critical
> Labels: 303-candidate
> Fix For: unspecified
>
>
> When running the large regression test-suite for Apache POI with multiple threads (each working on a single file), I saw that the lock in Path.getCompiledPath() is causing big contention, keeping the test from utilizing multiple CPU cores fully.
> I tested with manually disabling this locking completely, then CPU is utilized much better, however there are potential race-conditions for the static maps.
> It seems in this case where we process a huge amount of different files, write-locks do occur constantly, i.e. there are always new expressions being put into the cache and thus write-locks do occur and keep other threads from accessing the cache.
> Possible options:
> * test with using using more fine-grained locking on each map separately
> * maybe we can use concurrent maps and Java 8 Map.computeIfAbsent() directly instead of the lock-get-put-unlock combination.
> * Add an option to make the caches per thread via thread-local so no locking is needed any more
> * ?
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@poi.apache.org
For additional commands, e-mail: dev-help@poi.apache.org