Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Zookeeper, mail # user - Standalone session expiration


Copy link to this message
-
Re: Standalone session expiration
Jared Cantwell 2011-07-02, 17:00
Forgot to reply to this. It looks like the socket linger issue was  
causing this. I didn't track it completely since disabling linger  
solved the issue. My guess is that the sockets were hung on close for  
2 seconds each, which somehow caused the server to kill other  
connections (maybe a lock is held?).

Jared

On Jul 2, 2011, at 10:55 AM, Ted Dunning <[EMAIL PROTECTED]> wrote:

> Has anybody understood this scenario yet?
>
> On Thu, Jun 30, 2011 at 8:15 AM, Jared Cantwell <[EMAIL PROTECTED]
> >wrote:
>
>> Hello again,
>>
>> I am seeing a strange issue that I'm hoping someone can give me  
>> insight
>> into.  For simple testing, I have a standalone server setup.  I  
>> connected
>> to
>> this server from 2 nodes, one of which is the node hosting the  
>> standalone
>> server.  After opening a small number of connections from each node  
>> (3 or 4
>> clients/node), I powered off the node not hosting the standalone  
>> server.
>> As
>> expected, the logs show the server expiring all sessions for  
>> connections to
>> that node.  The problem comes 10 seconds later when the server  
>> decides to
>> also expire all local connections too.  As a result, the clients on  
>> the
>> node
>> that is still alive (and hosting the standalone server) all try
>> reconnecting, but their connections are denied for having expired--  
>> over
>> and
>> over again.
>>
>> I am working on getting some consolidated logs, so I'll reply to  
>> this when
>> I
>> have them.  I was wondering if anyone knows of an issue or has any  
>> initial
>> thoughts?
>>
>> Some things I am going to try:
>> 1. Start a 3 node quorum and connect clients from a 4th node.  Then  
>> kill
>> the
>> 4th node and see if other connections are killed too.  If this  
>> works OK
>> then
>> it would point to an issue with the standalone server mode.
>> 2. Connect 3 nodes to my standalone server.  Power off one node and  
>> see if
>> connections to the other node is killed.  This will determine if its
>> killing
>> all other connections, or just local connections for some strange  
>> reasons.
>>
>> ~Jared
>>