[
https://issues.apache.org/jira/browse/LOG4J2-327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13723836#comment-13723836
]
Sebastian Oerding commented on LOG4J2-327:
------------------------------------------
Hi,
1. the excessive memory usage was there with not using the AsyncLoggers.
However with using the AsyncLoggers still a huge amount of memory was consumed
but not as mach as with using the RollingFile.
I built log4j 2 and tried it:
2. Yes the IllegalArgumentException has been vanished.
3. The memory usage: Seems to be better, test scenario was not completed due to
time. However as far as it goes we had memory usage without logging approx. 250
MB, with old log4j 550 - 600 MB (and crash), with current log4j at peak 440 MB,
in average 400 MB.
While these numbers omit all settings and and thus not count as a benchmark in
any way, they give a relative impression. We have not checked for corrupted
strings, can't say whether they still occur.
Test code / The big string:
Sorry I failed writing a small test case. And I can't put 15,000 files on some
external system for FTP access. I think that similar to
https://issues.apache.org/jira/browse/LOG4J2-290 this has to do with
multi-threading and only happens at very specific conditions that are difficult
to reproduce. I'll provide more information once I have some.
> Log4j 2 drastical performance loss
> ----------------------------------
>
> Key: LOG4J2-327
> URL: https://issues.apache.org/jira/browse/LOG4J2-327
> Project: Log4j 2
> Issue Type: Bug
> Components: Appenders, Core, Layouts
> Affects Versions: 2.0-beta7, 2.0-beta8
> Environment: Windows 7 64 bit, JDk 1.6.0_43
> Reporter: Sebastian Oerding
> Priority: Blocker
> Labels: performance
>
> Hello,
> we encountered a severe performance loss with log4j 2. Unfortunately we have
> no exact error description such as a stacktrace but I'll give it may best
> shot.
> We are developing an application working with mass data. During the
> development we are running tests such as importing 15,000 files via FTP or
> creating 500,000 business objects and workiong with them.
> Switching from log4j 1.2.17 to log4j 2 we ran into really severe performance
> issues.
> With the beta7 we even get an OutOfMemoryError. Using a profiler we see that
> there was a huge amount of string objects (that were log messages according
> to a control sample).
> Switching to the beta8 the profiler shows there are only 10 % of the string
> objects left that we had before but it still was a big number. Especially the
> strings live for a really long time (> 100 generations).
> Using the RollingFileAppender and ConsoleAppender, (not a FlumeAppender for
> which a memory has been fixed in beta8) and default garbage collection we had
> the following behaviour with beta8:
> Starting the application (each time exactly the same test scenario) the
> machine fastly reaches the memory limit given for the JVM. Afterwards
> periodically garbage collection is invoked, causing a long "Stop the World
> Phase"
> (8 minutes in another scenario), freeing memory down a "base level" and the
> application continues until the memory is full again. The base level remains
> the same for next invocations of the garbage collection.
> This may relate to tickets: https://issues.apache.org/jira/browse/LOG4J2-323,
> https://issues.apache.org/jira/browse/LOG4J2-322,
> https://issues.apache.org/jira/browse/LOG4J2-304.
> Furthermore when looking into one string object being not GCed (372 KB) I
> noticed that it contained a lot of log messages being concatenated. The
> application contains no log statement that logs such a big message.
> Looking at the string towards the end I found even complete log events
> including the log level and so on not only the message. This feels being
> related to https://issues.apache.org/jira/browse/LOG4J2-290, somehow log
> events /
> strings get concatenated / corrupted.
> Maybe it is even a problem with the ParameterizedMessage which we use a lot
> of.
> Switching to the AsyncAppender (FastRollingFile) asynhronously the
> application performs better than with the RollingFile but still bad. Running
> the same scenario with the FastRollingFile "immediateFlush" and "append"
> parameters ommitted in the configuration the FastRollingFile crashed due to a
> IllegalArgumentExeption in class ByteBuffer (stacktrace is lost).
> The blocker priority may be at bit extremen but currently we are thinking of
> going back to log4j 1.2.17, throwing away several days of work and requiring
> a few days for adjusting our code. Hence this is a blocker for us as we could
> not release our software in the current state.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
