[rabbitmq-discuss] Upgrading RabbitMQ cluster

[Simon MacMullen]

On 04/09/2013 7:15PM, Michael Klishin wrote:
> Viraj Gupte:
>
>> {"init terminating in do_boot",{function_clause,[{lists,usplit_1,['rabbit at crxv10','rabbit at crxv11',{['rabbit at crxv10','rabbit at crxv11'],['rabbit at crxv10']},[],[]]},{rabbit_upgrade,maybe_upgrade_mnesia,0},{rabbit,prepare,0},{init,eval_script,8},{init,do_boot,3}]}}
>
> I suspect that the upgrade step in the release you are trying assumes a different
> [internal] database schema from what you actually have after upgrading to 3.1.

Nearly. The issue actually is that the format of the known cluster 
members file changed. But as with Mnesia, we support automatic upgrades 
for this file but not downgrades.

> Without writing code to perform a reverse migration, broker won't start. Unfortunately,
> some migrations simply cannot be reversed (e.g. if a column was removed
> from a table).

Yeah. I don't think in practice we have made such upgrades, but that's 
certainly an issue.

The bigger issue of course is that RabbitMQ 2.8.7 is going to have a 
difficult time knowing what changes have occurred after it was released. 
There could be an external tool written for downgrades I suppose. Not a 
high priority for us though.

Cheers, Simon

-- 
Simon MacMullen
RabbitMQ, Pivotal