[jboss-jira] [JBoss JIRA] Commented: (JBVFS-159) Native memory leak due to ZipEntryInputStream
[
https://jira.jboss.org/browse/JBVFS-159?page=com.atlassian.jira.plugin.sy...
]
Marko Strukelj commented on JBVFS-159:
--------------------------------------
I tried your proposed patch on Windows and on Linux, and the results are contrary to each
other.
On Linux I find the memory footprint of the process is noticably improved, but on Windows
it is noticably worsened. The reason seems to be that your patch actually increases heap
usage baseline. But the native zip memory leak on Linux is obviously so big that fixing it
offsets the increased heap usage. On Windows there apparently is no native zip memory leak
and so there's no gain on that front that would offset extra heap usage.
My patch on the other hand doesn't address native zip memory leak at all, but attempts
to reduce heap size used by vfszip metadata. That would maybe allow you to decrease -Xmx
memory size by about 10% which will again translate to decreased process size.
So the combination of both patches should be most beneficial on Linux, while it appears
it's best to avoid using your patch on Windows.
Native memory leak due to ZipEntryInputStream
---------------------------------------------
Key: JBVFS-159
URL: https://jira.jboss.org/browse/JBVFS-159
Project: JBoss VFS
Issue Type: Bug
Security Level: Public(Everyone can see)
Affects Versions: 2.1.2.GA
Environment: Redhat (not sure what version) 2.6.9-78.ELsmp
JBoss 5.1.0.GA
JDK 1.6.0_20
JVM parameter:
-XX:+PrintGCTimeStamps -XX:+PrintGCApplicationStoppedTime -verbose:gc
-Dfile.encoding=iso-8859-1 -server -Djava.net.preferIPv4Stack=true
-Doracle.jdbc.V8Compatible=true -Dsun.rmi.dgc.client.gcInterval=3600000
-Dsun.rmi.dgc.server.gcInterval=3600000 -Dnetworkaddress.cache.ttl=300 -Xss128k -Xmn500m
-Dorg.apache.jasper.runtime.BodyContentImpl.LIMIT_BUFFER=true -XX:+PrintGCDetails
-XX:PermSize=256m -XX:MaxPermSize=256m -Xms1500m -Xmx1500m -XX:+HeapDumpOnOutOfMemoryError
-XX:+UseConcMarkSweepGC
Reporter: Samuel Cai
Assignee: Marko Strukelj
Attachments: change.tar.gz, JBVFS-159.patch, leakdetector.jar, test.zip,
VFSZipMemoryTestCase.java
We used to use JBoss 4.2.1.GA and JDK 1.6.0_11, and trying JBoss 5.1.0.GA and JDK
1.6.0_20 these days.
I found the process size is more larger than before, 2.5G~2.9G compared to 1.9G.
I was thinking this was a bug of JBoss AS, then filed
https://jira.jboss.org/browse/JBAS-8066
After these days investigation, I think this is a memory leak in VFS, maybe only happen
on our specific environment.
I tried a change on class org.jboss.virtual.plugins.context.zip.ZipFileWrapper, method
openStream:
From:
ZipEntryInputStream zis = new ZipEntryInputStream(this, is);
return zis;
To:
//ZipEntryInputStream zis = new ZipEntryInputStream(this, is);
//return zis;
return is;
That is, don't use ZipEntryInputStream, let any class/method invoking openStream to
close zipFile's inputStream immediatelly.
ZipFile will be in open status, but all steams will be closed well.
This makes the process size down to same as JBoss 4's. I tried going through first 3
pages of site, no problems. May need QA team to test more.
Btw, I tried updating VFS to 2.1.3.SP1/2.2.0.M4/3.0.0.CR5, first two have same size
issue, third one couldn't start.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://jira.jboss.org/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira