New HA not keeping statistics aligned?

[Pavel Moravec <pm...@redhat.com>]

Hi all,
I noticed new HA does not align (queue, exchange,..) statistics cluster-wide. See following scenario in qpid 0.18:

1) 2node cluster with node1 active and node2 ready
2) publishing and consuming some messages to 'qpid.replicate':all queue
3) restart node2
4) restart node1 & promote node2 as active

Now node2 has zeroed statistics about #consumers, #consumed messages, #flowStopped, .. etc. It only has statistics from "create the queue and populate outstanding messages".

Is this by purpose or is it a bug / limitation?

(I would expect the second as I don't care which broker is primary and as the cluster was running and providing service continually "just" with the bounce when failing over active node, statistics should cover whole scenario)


Kind regards,
Pavel



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
For additional commands, e-mail: users-help@qpid.apache.org

Re: New HA not keeping statistics aligned?

[Alan Conway <ac...@redhat.com>]

On Fri, 2013-01-04 at 08:20 -0500, Pavel Moravec wrote:
> Hi all,
> I noticed new HA does not align (queue, exchange,..) statistics cluster-wide. See following scenario in qpid 0.18:
> 
> 1) 2node cluster with node1 active and node2 ready
> 2) publishing and consuming some messages to 'qpid.replicate':all queue
> 3) restart node2
> 4) restart node1 & promote node2 as active
> 
> Now node2 has zeroed statistics about #consumers, #consumed messages, #flowStopped, .. etc. It only has statistics from "create the queue and populate outstanding messages".
> 
> Is this by purpose or is it a bug / limitation?
> 
> (I would expect the second as I don't care which broker is primary and as the cluster was running and providing service continually "just" with the bounce when failing over active node, statistics should cover whole scenario)

It is intentional. The new HA does not try to maintain identical state
in all brokers like the old cluster did. The management stats reflect
the actual stat of each broker.

A READY broker enough state to be able to take over on a failure. The
primary delays completion of messages till they are completed by the
backup, so in a failure any messages not on the backup will be replayed
when the sender fails over and so not lost.

Cheers,
Alan.


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@qpid.apache.org
For additional commands, e-mail: users-help@qpid.apache.org