Server crash - in function memcpy
OS- Solaris 9 and 10 (sparc)
WS- SJSWS 6.1 SP5-8
JAVA- j2sdk 1.4.2_13
Frequent server crashes with the following in the logs-
errors:[18/Jul/2007:14:16:04] catastrophe ( 5938): CORE3260: Server crash detected (signal SIGBUS)
errors:[18/Jul/2007:14:16:04] info ( 5938): CORE3261: Crash occurred in NSAPI SAF send-file
errors:[18/Jul/2007:14:16:04] info ( 5938): CORE3262: Crash occurred in function memcpy from module /platform/sun4v/lib/libc_psr.so.
Also seems to happen when xml files (served to clients frequently) are updated in the docroot, which is nfs mounted and shared to multiple webservers.
Any thoughts?
Thanks,
Jess
[694 byte] By [
jesspdxa] at [2007-11-27 10:59:07]
# 1
pl. upgrade to latest service pack 8 - as it addresses important security vulnerabilites
http://www.sun.com/download/products.xml?id=4694392a
with respect to core dump, if you have dbx (comes with Studio 12 - you can get it for free from (http://developers.sun.com/sunstudio/index.jsp) then you can provide us with the stack trace.
also, since this issue seems to be resulting a crash in libc module, instaling the latest solaris patch cluster might be the first step
http://sunsolve.sun.com/pub-cgi/show.pl?target=patches/patch-access
# 3
This occurs in SP5 - SP8, also Solaris patches are up to date.
We have had a case open with Sun for months, but sometimes one gets better advice right here.
That said, I think we found the issue last night. The docroot resides on nfs and an xml file that gets served frequently gets updated by a separate java process. If the update occurs while the web server is serving a request for this file it crashes (nfs stale file handle issue). Looks like it may get listed as a bug, but in the meantime we have turned on nsfc file caching and it seems to alleviate the issue.
Is using nfs for approot/docroot not a common practice? It makes deployment much more simple.
Thanks,
Jess
