JRE/JDK > 1.6.0_17/6u17 consumes all memory before killing geonetwork with java.lang.OutOfMemoryError

[simonp]

Reported by Rik Hendrix:

On a test server, I'm running Geonetwork on JRE version 6u17 (=1.6.0_17) for several months without stability problems.

Last week I upgraded JRE to 6u19. Geonetwork started crashing every few minutes. Did the same upgrade on another server, same result. Downgraded to 6u17, problem solved. This week I upgraded to 6u20, same problem as 6u19.

Apparently there is a memory leak:

# # A fatal error has been detected by the Java Runtime Environment: # # java.lang.OutOfMemoryError?: requested 1073741832 bytes for Chunk::new. Out of swap space? # # Internal Error (allocation.cpp:215), pid=3248, tid=2335914912 # Error: Chunk::new # # JRE version: 6.0_20-b02 # Java VM: Java HotSpot?(TM) Server VM (16.3-b01 mixed mode linux-x86 )

For more details see attachment.

When changing Xmx from 512m to 1024m, Geonetwork stays up and running a few minutes longer, but still crashes after a while.

Did someone notice this behavior already?

[simonp]

Not actually a Geonetwork problem or memory leak in GeoNetwork code - I checked into this this morning as 2.4.3 is about to be released. Pretty much skyrocketing memory usage the moment anything XSLT is done when running under jdk 1.6 20. Looked into the hs error file and found:

Current CompileTask?: C2:880 ! net.sf.saxon.event.ReceivingContentHandler?.startElement(......)

Turning that into google produced a link to something that makes everything settle down nicely:

 http://www.devcomments.com/Saxon-JVM-crash-Out-of-swap-space-OOM-error-withsolution-at89036.htm

I'll commit this for 2.4.3 after a bit more testing etc.

[simonp]

Fixed in svn commit 5997 (actually no problems with the original fix given in the devcomments link - tests fine under 1.5.0_22 and 1.6.0_17 and 1.6.0_20)