Quorum queues and partition strategy

[Arnaud Morin]

Hey team,

Is there any recommendation on using a specific partition strategy when
using quorum queues?


Our strategy so far is pause_minority with a cluster of 3 nodes.

We had an outage this morning that lead to a split for the 3 nodes (all
nodes were alone - I am not yet sure why, but this is what I an see from
logs).

As a result, each node started a shutdown operation, resulting in a
massive dumping of error messages like this:
reason: {{badmatch,{error,noproc}},

I dont remember having such issues when using classic mirrored queues.


Should we disable pause_minority?
Are quorum queues able to manage a split brain better than mirrored
queues without any strategy?

Cheers,

[Karl Nilsson]

Quorum queues don't behave any different whichever strategy you use. The RabbitMQ cluster does however.

Partition handling strategies will disappear with RabbitMQ 4.0 and the new meta data store.

I haven't done extensive testing but I would probably choose autoheal or ignore.

--
You received this message because you are subscribed to the Google Groups "rabbitmq-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rabbitmq-user...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/rabbitmq-users/ZK/nSUwVGTCfhpnU%40sync2.

--

Karl Nilsson

[Arnaud Morin]

Hey,
Thanks for your answer!
I am wondering what would happen with ignore?
If the three nodes are splitted, the queues are no longer having
"quorum", are they stopped?
What happen when the cluster come back alive then? Are the queues able
to re-elect a master and continue working? Or are they stucked forever
and need force-delete?
Cheers,

> *Karl Nilsson*

>
> --
> You received this message because you are subscribed to the Google Groups "rabbitmq-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to rabbitmq-user...@googlegroups.com.

> To view this discussion on the web, visit https://groups.google.com/d/msgid/rabbitmq-users/CAHC35TBiWPG44dTKFp5YtmWozuHA12qi5mkYN_p_nAe_Grn2kA%40mail.gmail.com.

[Karl Nilsson]

You read more here: https://www.rabbitmq.com/partitions.html#automatic-handling

Autoheal will choose a "winner" and restart the other nodes. You may lose any entities (queues, exchange, users etc) that were declared close to the partition event as the recovery isn't meta data safe but unless you have high entity churn autoheal is probably the way to go.

To view this discussion on the web, visit https://groups.google.com/d/msgid/rabbitmq-users/ZK/sHmde1Dyia19J%40sync2.

--

Karl Nilsson