+1 Otis about GC tuning.
Also, If you have separate ZK cluster, note that max zk session timeout you
can have is 20 times tickTime setting on ZK cluster. HBase is client to ZK,
and any negotiated session timeout must be 2 and 20 times ZK session tick
time settings (by default).
Hope this helps.
On Sat, Jun 29, 2013 at 1:38 PM, Otis Gospodnetic <
[EMAIL PROTECTED]> wrote:
> Give G1 GC a try if you are on a recent JVM. We didn't have issues
> with ZK, but we gave G1 a try recently and look what happened:
> http://blog.sematext.com/2013/06/24/g1-cms-java-garbage-collector/ -
> maybe that will help you with ZK, too.
> HBase Performance Monitoring -- http://sematext.com/spm
> On Sat, Jun 29, 2013 at 2:03 AM, <[EMAIL PROTECTED]> wrote:
> > Hi All,
> > We hit an issue that HBase server keeps shutting down. From log we can
> see that sometimes we got garbage collection taking too long issue. Also,
> we saw Zookeeper session timeout exception in the log. We tried to increase
> the zookeeper session time out in hbase-site.xml but it didn't help. We
> also tried to increase the heap size in hbase-env.sh but it didn't help
> > Our set up was single node on VM with 4G memory and 1 processor.
> > We're using hbase 0.94.1.
> > Interesting thing is that with the same configuration on a different VM,
> we don't get this issue...
> > Any ideas?
> > Thanks,
> > YuLing