Not sure if we have any experts on the topic here but I thought I’d give
it a shot.
We are running Rails 2.3.5 with a warbler war in Tomcat 5 on RHEL 5.4.
I’m noticing some file handle leakage on our production server. Using
“lsof -u tomcat| wc -l” I can see the number steadily increase over the
entire day until tomcat is restarted early in the morning.
I have tried to duplicate this on our staging server which has the same
version of all the software. I cannot duplicate on the staging
environment.
Looking at the content of the lsof output the most noticeable difference
is that in production I see a stead buildup of lines like this:
java 20893 tomcat 129u 0000 0,11 0
1665311 eventpoll
java 20893 tomcat 100u FIFO 0,6
1641223 pipe
I see none of these for the tomcat user on the staging server.
There are usually about twice as many pipe entries as eventpoll entries.
Anyone have any ideas why these might be building up?
It is possible that there are some OS config differences between staging
and production as the prod servers were setup by a hosting company.
Any help is greatly appreciated.