[PATCH] Improve CTDB database recovery

[Martin Schwenke]

On Wed, 23 Mar 2016 15:29:13 +1100, Amitay Isaacs <amitay at gmail.com>
wrote:

> Here are the patches to avoid using a single large buffer for database
> recovery.  This approach can fail for databases with lots of records as
> there is a limit on the size of single talloc buffer (256M).  In the new
> approach, CTDB sends the records in batches with each buffer roughly the
> size specified by RecBufferSizeLimit tunable.
> 
> Also add banning of misbehaving nodes (where pulling records from a node
> fails) during recovery.  This was missing in the parallel database recovery
> code.
> 
> Please review and push.

Reviewed-by: Martin Schwenke <martin at meltin.net>

Pushed, with the extra patch as a fixup to the last patch in this
series.

peace & happiness,
martin