Riak Search and Yokozuna Backup Strategy
matthewv at basho.com
Mon Jan 27 15:19:04 EST 2014
On Jan 27, 2014, at 3:06 PM, Elias Levy <fearsome.lucidity at gmail.com> wrote:
> On Mon, Jan 27, 2014 at 11:57 AM, Matthew Von-Maszewski <matthewv at basho.com> wrote:
> Google's designed leveldb to always assume it was not cleanly shutdown. If the startup can read the most recent MANIFEST file, leveldb cleans up the rest of the mess.
> However, using the backup strategy previously discussed (backup sst_? directories numerically … rsync or tar) creates a scenario where the MANIFEST file is likely incorrect in a manner that leveldb will not detect (hmm … maybe it could detect it but no one thought to add a test). That is why we always recommend (require!) that you manually run repair after a restore.
> So, yes, but its not sufficient. Got it. Thanks.
> And I assume that currently requires attaching to the Riak console and executing something like:
> eleveldb:repair("/var/db/riak/leveldb/442446784738847563128068650529343492278651453440", ).
> for every vnode LevelDB in any eleveldb backend?
So, yes. But the underlying theory is that restore via backups is the last ditch effort when all the N=3 / active anti-entropy (and multi-datacenter replication) niceties have failed (like when your datacenter is flooded).
That said, Basho's product management is working aggressively to define a better, more holistic backup/restore feature to add after the 2.0 release.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the riak-users