[ https://issues.apache.org/jira/browse/HBASE-10506?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14236599#comment-14236599 ]
hongyu bi commented on HBASE-10506: ----------------------------------- fail-fast doesn't decrease the callQueueSize which may lead to "Call queue is full" > Fail-fast if client connection is lost before the real call be executed in > RPC layer > ------------------------------------------------------------------------------------ > > Key: HBASE-10506 > URL: https://issues.apache.org/jira/browse/HBASE-10506 > Project: HBase > Issue Type: Bug > Components: IPC/RPC > Affects Versions: 0.94.3 > Reporter: Liang Xie > Assignee: Liang Xie > Fix For: 0.98.0, 0.96.2, 0.99.0, 0.94.17 > > Attachments: HBASE-10506-0.94.txt, HBASE-10506-trunk.txt > > > In current HBase rpc impletement, there is no any connection double-checking > just before the "call" be invoked, considing there's a gc or other OS > scheduling or the call queue is full enough(e.g. the server side is slow/hang > due to some issues), and if the client side has a small rpc timeout value, it > could be possible when this request be taken from call queue, the client > connection is lost in that moment. we'd better has some fail-fast code before > the reall "call" be invoked, it just waste the server side resource. > Here is a strace trace from our production env: > 2014-02-11,18:16:19,525 WARN org.apache.hadoop.ipc.HBaseServer: IPC Server > Responder, call get([B@3eae6c77, > {"timeRange":[0,9223372036854775807],"totalColumns":1,"cacheBlocks":true,"families":{"X":["T"]},"maxVersions":1,"row":"074103000000001-m8997060"}), > rpc version=1, client version=29, methodsFingerPrint=-241105381 from > 10.101.10.181:43252: output error > 2014-02-11,18:16:19,526 WARN org.apache.hadoop.ipc.HBaseServer: IPC Server > handler 151 on 12600 caught a ClosedChannelException, this means that the > server was processing a request but the client went away. The error message > was: null > 2014-02-11,18:16:19,797 ERROR > org.apache.hadoop.hbase.regionserver.HRegionServer: > org.apache.hadoop.hbase.ipc.CallerDisconnectedException: Aborting call > get([B@3f10ffd2, > {"timeRange":[0,9223372036854775807],"totalColumns":1,"cacheBlocks":true,"families":{"X":["T"]},"maxVersions":1,"row":"4245978-m7281526"}), > rpc version=1, client version=29, methodsFingerPrint=-241105381 from > 10.101.10.181:43259 after 0 ms, since caller disconnected > at > org.apache.hadoop.hbase.ipc.HBaseServer$Call.throwExceptionIfCallerDisconnected(HBaseServer.java:450) > at > org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.nextInternal(HRegion.java:3633) > at > org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.next(HRegion.java:3590) > at > org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.next(HRegion.java:3615) > at org.apache.hadoop.hbase.regionserver.HRegion.get(HRegion.java:4414) > at org.apache.hadoop.hbase.regionserver.HRegion.get(HRegion.java:4387) > at > org.apache.hadoop.hbase.regionserver.HRegionServer.get(HRegionServer.java:2075) > at sun.reflect.GeneratedMethodAccessor29.invoke(Unknown Source) > at > sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25) > at java.lang.reflect.Method.invoke(Method.java:597) > at > org.apache.hadoop.hbase.ipc.SecureRpcEngine$Server.call(SecureRpcEngine.java:460) > at org.apache.hadoop.hbase.ipc.HBaseServer$Handler.run(HBaseServer.java:1457) > 2014-02-11,18:16:19,802 WARN org.apache.hadoop.ipc.HBaseServer: IPC Server > Responder, call get([B@3f10ffd2, > {"timeRange":[0,9223372036854775807],"totalColumns":1,"cacheBlocks":true,"families":{"X":["T"]},"maxVersions":1,"row":"4245978-m7281526"}), > rpc version=1, client version=29, methodsFingerPrint=-241105381 from > 10.101.10.181:43259: output error > 2014-02-11,18:16:19,802 WARN org.apache.hadoop.ipc.HBaseServer: IPC Server > handler 46 on 12600 caught a ClosedChannelException, this means that the > server was processing a request but the client went away. The error message > was: null > With this fix, we can reduce this hit probability at least:) the upstream > hadoop has this checking already, see: > https://github.com/apache/hadoop-common/blob/trunk/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/ipc/Server.java#L2034-L2036 -- This message was sent by Atlassian JIRA (v6.3.4#6332)