I've only looked at the code a little, and likely missed something, but does this if block decrement the call queue, if the client already closed the connection?
https://github.com/apache/hbase/blob/07a771866f18e8ec532c14f624fa908815bd88c7/hbase-server/src/main/java/org/apache/hadoop/hbase/ipc/CallRunner.java#L74 On Thu, Aug 7, 2014 at 11:32 PM, Walter King <[email protected]> wrote: > Yes, sorry, CallQueueTooBigException. but that value never returns to > zero, even when number of requests goes to zero. The call queue too big > happens if any regionserver is up for a long enough period of time, so I > have to periodically restart them. Also at that 15:30 time I wasn't > seeing that exception, but it seems like that is one time in which a call > didnt properly decrement the callqueuesize because it was at zero before > and has never hit zero again - today the minimum is even higher. > > > On Thu, Aug 7, 2014 at 9:14 PM, Qiang Tian <[email protected]> wrote: > >> bq. "Eventually we ran into ipc queue size full messages being returned to >> clients trying large batch puts, as it approaches a gigabyte." >> >> Do you mean CallQueueTooBigException? it looks not the queue size, but the >> data size that client sends..configured by >> "hbase.ipc.server.max.callqueue.size". >> >> I guess when you client got the exception, it closed the exception and >> causing other shared connection RPC failed. >> >> >> 2014-08-06 22:27:57,253 WARN [RpcServer.reader=9,port=60020] >> ipc.RpcServer >> (RpcServer.java:doRead(794)) - RpcServer.listener,port=60020: count of >> bytes read: 0 >> java.io.IOException: Connection reset by peer >> at sun.nio.ch.FileDispatcherImpl.read0(Native Method) >> at sun.nio.ch.SocketDispatcher.read(SocketDispatcher.java:39) >> at sun.nio.ch.IOUtil.readIntoNativeBuffer(IOUtil.java:223) >> at sun.nio.ch.IOUtil.read(IOUtil.java:197) >> at sun.nio.ch.SocketChannelImpl.read(SocketChannelImpl.java:379) >> at org.apache.hadoop.hbase.ipc.RpcServer.channelRead(RpcServer.java:2229) >> at >> >> org.apache.hadoop.hbase.ipc.RpcServer$Connection.readAndProcess(RpcServer.java:1415) >> at >> org.apache.hadoop.hbase.ipc.RpcServer$Listener.doRead(RpcServer.java:790) >> at >> >> org.apache.hadoop.hbase.ipc.RpcServer$Listener$Reader.doRunLoop(RpcServer.java:581) >> at >> >> org.apache.hadoop.hbase.ipc.RpcServer$Listener$Reader.run(RpcServer.java:556) >> at >> >> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145) >> at >> >> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615) >> at java.lang.Thread.run(Thread.java:744) >> 2014-08-06 22:27:57,257 WARN [RpcServer.handler=18,port=60020] >> ipc.RpcServer (RpcServer.java:processResponse(1041)) - >> RpcServer.respondercallId: 84968 service: ClientService methodName: Multi >> size: 17.7 K connection: 10.248.130.152:49780: output error >> 2014-08-06 22:27:57,258 WARN [RpcServer.handler=18,port=60020] >> ipc.RpcServer (CallRunner.java:run(135)) - >> RpcServer.handler=18,port=60020: >> caught a ClosedChannelException, this means that the server was processing >> a request but the client went away. The error message was: null >> 2014-08-06 22:27:57,260 WARN [RpcServer.handler=61,port=60020] >> ipc.RpcServer (RpcServer.java:processResponse(1041)) - >> RpcServer.respondercallId: 83907 service: ClientService methodName: Multi >> size: 17.1 K connection: 10.248.1.56:53615: output error >> 2014-08-06 22:27:57,263 WARN [RpcServer.handler=61,port=60020] >> ipc.RpcServer (CallRunner.java:run(135)) - >> RpcServer.handler=61,port=60020: >> caught a ClosedChannelException, this means that the server was processing >> a request but the client went away. The error message was: null >> >> >> >> On Fri, Aug 8, 2014 at 2:57 AM, Walter King <[email protected]> wrote: >> >> > >> https://gist.github.com/walterking/4c5c6f5e5e4a4946a656#file-gistfile1-txt >> > >> > http://adroll-test-sandbox.s3.amazonaws.com/regionserver.stdout.log.gz >> > >> > These are logs from that particular server, and the debug dump from >> now(no >> > restart in between). The times in the graph are pacific, so it should >> be >> > around 2014-08-06 22:25:00. I do see some exceptions around there. >> > >> > >
