On 1/3/2013 10:40 AM, Shawn Heisey wrote:
On 1/3/2013 8:11 AM, Michael Ryan wrote:
We see these EofExceptions in our system occasionally. I believe they
occur when our SolrJ client times out and closes the connection,
before Jetty returns the response.
This would make sense - the load balancer probably drops the healthcheck
connection after its five second timeout. I also have a two second
SoTimeout on the server objects in my app that checks /admin/mbeans (16
times (cores) per server). That app also checks the ping handler on one
core for each server.
I suppose the question is why is it taking so long to make the ping
request? The ping request just makes a query for *:* ... it is a
distributed request, but even so, for that to take more than two (or
even five) seconds seems to indicate there may be a larger problem.
Long stop-the-world GC pauses would explain it, if that's happening. I
guess it's time to turn on GC logging. Jconsole's overall stats don't
suggest that it's a problem, but that is only summary information.
On my 4.1 dev server using Oracle Java7, the only memory tuning options
I have are a max heap of 8GB and -XX:+UseG1GC. On my production 3.5
servers using Oracle Java6, I have the following in addition to a max
heap of 8GB:
-Xmx8192M -XX:NewRatio=1 -XX:+UseParNewGC -XX:+UseConcMarkSweepGC
-XX:+CMSParallelRemarkEnabled
Any pointers would be appreciated!
I have encountered a substantially similar problem after turning on GC
logging. The event did not correspond to anything in the GC log. I
have updated SOLR-4191 with the new information.
https://issues.apache.org/jira/browse/SOLR-4191
Thanks,
Shawn
