[Gluster-users] Why oh why does gluster delay?

[Atin Mukherjee]

On 12/02/2015 02:08 AM, Andrus, Brian Contractor wrote:
> Atin,
> 
> In debug, the glusterd log file is pretty sizeable. I do see regular:
> [2015-12-01 20:29:14.407105] E [MSGID: 106167] [glusterd-handshake.c:2073:__glusterd_peer_dump_version_cbk] 0-management: Error through RPC layer, retry again later
> 
> And ones like:
> [2015-12-01 20:29:45.057998] C [rpc-clnt-ping.c:165:rpc_clnt_ping_timer_expired] 0-management: server 172.18.56.55:24007 has not responded in the last 30 seconds, disconnecting.
> 
> Hopefully there is something in there that can give a clue as to what is going on.
This seems like the nodes are in a flaky network.
> 
> 
> Incidentally, to enable debug using the RPMs, the code in /etc/sysconfig/glusterd is inaccurate. The init file looks for LOG_FILE and LOG_LEVEL rather than the GLUSTERD_LOGLEVEL and GLUSTERD_LOGFILE that are in there as examples.
> 
> Brian Andrus
> ITACS/Research Computing
> Naval Postgraduate School
> Monterey, California
> 
> 
> -----Original Message-----
> From: Atin Mukherjee [mailto:amukherj at redhat.com] 
> Sent: Monday, November 30, 2015 8:09 PM
> To: Andrus, Brian Contractor <bdandrus at nps.edu>; gluster-users at gluster.org
> Subject: Re: [Gluster-users] Why oh why does gluster delay?
> 
> 
> 
> On 12/01/2015 04:15 AM, Andrus, Brian Contractor wrote:
>> All,
>>
>>  
>>
>> I am seeing it VERY consistently that when I do a 'gluster peer status'
>> or 'gluster pool list', the system 'hangs' for up to 1 minute before 
>> spitting back results.
> That's interesting, 'gluster peer status' or 'pool list' is a local node operation and doesn't incur RPC calls. Until and unless it takes time to communicate back to CLI because of a poor N/W, I don't see any reason why would it take such a long time. Can you run glusterd with debug log enabled and share the *glusterd.log file with us?
> 
> ~Atin
>>
>>  
>>
>> I have 10 nodes all on the same network and currently ZERO volumes or 
>> bricks configured. Just trying to get good performance for the cluster 
>> to be talking to itself....
>>
>>  
>>
>>  
>>
>> What is gluster doing that takes so long to respond? Seems there may 
>> be a more efficient way of doing it, whatever it is...
>>
>>  
>>
>>  
>>
>>  
>>
>> Brian Andrus
>>
>> ITACS/Research Computing
>>
>> Naval Postgraduate School
>>
>> Monterey, California
>>
>> voice: 831-656-6238
>>
>>  
>>
>>  
>>
>>
>>
>> _______________________________________________
>> Gluster-users mailing list
>> Gluster-users at gluster.org
>> http://www.gluster.org/mailman/listinfo/gluster-users
>>