Patrick Hunt updated ZOOKEEPER-558:

    Attachment: ZOOKEEPER-558.patch

This patch addresses the sent count issue (counter update was missing), 
additionally I addressed a number
of session establishment/teardown issues that I found while fixing this:

1) better logging of session est/teardown, much of this was based on user 
feedback such as ZOOKEEPER-565 ZOOKEEPER-564
Much of this was cleaning up the log text and using common terminology for the 
messages themselves
1.1) some of 1) required better error handling. A lot of the changed code was 
throwing general IOExceptions, rather than throwing
more specific exceptions allowing higher layer code to properly report the 
2) while closing a client connection the connection close response was not 
always sent to the client
3) in some cases the server would not close the connection (closeconn buffer 
was not being queued) properly. this would happen anyway when the client closed 
the connection, but the server should do the right thing regardless

4) calling close more than once just returns after the first call
5) added testableWaitForShutdown to ZooKeeper to allow tests to validate that 
the client has shutdown all threads
(this is test only)

6) reading a client sockets now reads the entire request in one shot if 
possible. previously we would re-poll the
selector after reading the length header.

> server "sent" stats not being updated
> -------------------------------------
>                 Key: ZOOKEEPER-558
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-558
>             Project: Zookeeper
>          Issue Type: Bug
>          Components: server
>    Affects Versions: 3.2.1
>            Reporter: Patrick Hunt
>            Assignee: Patrick Hunt
>            Priority: Critical
>             Fix For: 3.3.0
>         Attachments: ZOOKEEPER-558.patch
> the server and connection "sent" stat is not being updated. if you run "stat" 
> on the client port the sent packets is much lower than it should be
> seems that sendbuffer is not updating the stats when it shortcircuits the 
> send.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

Reply via email to