-Connection closed exceptions with slow fsync and CancelledKeyExceptions
mattgordon 2012-11-08, 01:48
We have been trying to understand why our ZooKeeper cluster will occasionally
have a wave of connection closed exceptions. We have switched to
-XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode for garbage collection with
no noticeable improvements.
The symptoms are:
(1) All nodes show messages like "fsync-ing the write ahead log in
SyncThread:0 took 6309ms which will adversely effect operation latency. See
the ZooKeeper troubleshooting guide" with times typically around 5 seconds.
At least once, this fsync appeared in the leaders log immediately before a
ERROR [CommitProcessor:0:NIOServerCnxn@445] - Unexpected Exception:
Our clients received ZookeeperConnectionClosed exceptions at this time and
all traffic on the ZooKeeper cluster essentially went to zero for a moment
before resuming normal operation with new connections.
(2) Probably unrelated since I haven't correlated it temporally with the
client errors, but running "sudo strace -r -T -f -p 9574 -e
trace=fsync,fdatasync -o trace.txt" turns up some messages like "10581
0.000246 — SIGSEGV (Segmentation fault) @ 0 (0) —"
ZK Version: 3.3.4
Cluster has 5 nodes running in EC2
Here is a screenshot showing ZooKeeper network traffic going to zero at the
time of the connection closed exceptions: http://i.imgur.com/dfNh0.png
Anyone have ideas on what the cause of these "waves" of
CancelledKeyExceptions could be from?
View this message in context: http://zookeeper-user.578899.n2.nabble.com/Connection-closed-exceptions-with-slow-fsync-and-CancelledKeyExceptions-tp7578166.html
Sent from the zookeeper-user mailing list archive at Nabble.com.
Vitalii Tymchyshyn 2012-11-12, 00:12