[
https://issues.apache.org/jira/browse/ZOOKEEPER-832?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14136066#comment-14136066
]
Andrew Purtell commented on ZOOKEEPER-832:
------------------------------------------
If the zxid is not recognized by the server, the client will attempt reconnects
indefinitely and never succeed. This can contribute gigabytes to service
logging and bring down the service with the embedded ZK client *and others*.
The client should error out somehow permanently. Can this be fixed here?
Perhaps other issues can be addressed in follow up? Unquestionably the
continuous and perpetual client retries is a bug, they will never succeed, and
a fleet of clients in such state can impact network services.
If we otherwise recreate application znode state in the reinitialized cluster,
why should applications be prevented from riding over a loss of transient ZK
state, if the use case supports it? For example if the ZK client would error
out with a session expired status under these conditions, HBase would create a
new ZooKeeper client object, attempt a reconnect using it, and might succeed.
(I think, haven't tested it.) Understood this is not a normal case but it comes
up frequently in federated test environments where the ZK cluster is under your
control but not necessarily the applications.
> Invalid session id causes infinite loop during automatic reconnect
> ------------------------------------------------------------------
>
> Key: ZOOKEEPER-832
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-832
> Project: ZooKeeper
> Issue Type: Bug
> Components: server
> Affects Versions: 3.3.1, 3.4.5, 3.5.0
> Environment: All
> Reporter: Ryan Holmes
> Assignee: Germán Blanco
> Fix For: 3.4.7, 3.5.1
>
> Attachments: ZOOKEEPER-832.patch, ZOOKEEPER-832.patch,
> ZOOKEEPER-832.patch, ZOOKEEPER-832.patch, ZOOKEEPER-832.patch
>
>
> Steps to reproduce:
> 1.) Connect to a standalone server using the Java client.
> 2.) Stop the server.
> 3.) Delete the contents of the data directory (i.e. the persisted session
> data).
> 4.) Start the server.
> The client now automatically tries to reconnect but the server refuses the
> connection because the session id is invalid. The client and server are now
> in an infinite loop of attempted and rejected connections. While this
> situation represents a catastrophic failure and the current behavior is not
> incorrect, it appears that there is no way to detect this situation on the
> client and therefore no way to recover.
> The suggested improvement is to send an event to the default watcher
> indicating that the current state is "session invalid", similar to how the
> "session expired" state is handled.
> Server log output (repeats indefinitely):
> 2010-08-05 11:48:08,283 - INFO
> [NIOServerCxn.Factory:0.0.0.0/0.0.0.0:2181:NIOServerCnxn$Factory@250] -
> Accepted socket connection from /127.0.0.1:63292
> 2010-08-05 11:48:08,284 - INFO
> [NIOServerCxn.Factory:0.0.0.0/0.0.0.0:2181:NIOServerCnxn@751] - Refusing
> session request for client /127.0.0.1:63292 as it has seen zxid 0x44 our last
> zxid is 0x0 client must try another server
> 2010-08-05 11:48:08,284 - INFO
> [NIOServerCxn.Factory:0.0.0.0/0.0.0.0:2181:NIOServerCnxn@1434] - Closed
> socket connection for client /127.0.0.1:63292 (no session established for
> client)
> Client log output (repeats indefinitely):
> 11:47:17 org.apache.zookeeper.ClientCnxn startConnect INFO line 1000 -
> Opening socket connection to server localhost/127.0.0.1:2181
> 11:47:17 org.apache.zookeeper.ClientCnxn run WARN line 1120 - Session
> 0x12a3ae4e893000a for server null, unexpected error, closing socket
> connection and attempting reconnect
> java.net.ConnectException: Connection refused
> at sun.nio.ch.SocketChannelImpl.checkConnect(Native Method)
> at
> sun.nio.ch.SocketChannelImpl.finishConnect(SocketChannelImpl.java:574)
> at org.apache.zookeeper.ClientCnxn$SendThread.run(ClientCnxn.java:1078)
> 11:47:17 org.apache.zookeeper.ClientCnxn cleanup DEBUG line 1167 - Ignoring
> exception during shutdown input
> java.nio.channels.ClosedChannelException
> at
> sun.nio.ch.SocketChannelImpl.shutdownInput(SocketChannelImpl.java:638)
> at sun.nio.ch.SocketAdaptor.shutdownInput(SocketAdaptor.java:360)
> at
> org.apache.zookeeper.ClientCnxn$SendThread.cleanup(ClientCnxn.java:1164)
> at org.apache.zookeeper.ClientCnxn$SendThread.run(ClientCnxn.java:1129)
> 11:47:17 org.apache.zookeeper.ClientCnxn cleanup DEBUG line 1174 - Ignoring
> exception during shutdown output
> java.nio.channels.ClosedChannelException
> at
> sun.nio.ch.SocketChannelImpl.shutdownOutput(SocketChannelImpl.java:649)
> at sun.nio.ch.SocketAdaptor.shutdownOutput(SocketAdaptor.java:368)
> at
> org.apache.zookeeper.ClientCnxn$SendThread.cleanup(ClientCnxn.java:1171)
> at org.apache.zookeeper.ClientCnxn$SendThread.run(ClientCnxn.java:1129)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
