Restoring a single DB in a cluster

[Garth Gutenberg <ga...@gmail.com>]

Hello developers.  I can't find any documentation on this, so if it exists,
please point me at it.  Otherwise, I need some input.

What is the official, sanctioned way to restore a DB in a cluster
configuration?

Scenario:
3 servers in a cluster.  A client deletes a bunch of docs from their DB, or
some other DB-specific catastrophe happens.  What is the correct approach
for restoring this DB from a backup?  Let's assume that I have disk
snapshots of every node in the cluster taken at *approximately* the same
time from the day before.

In the old days, pre-cluster, it was as simple as just copying the
<db>.couch file... but with shards and clusters, this just doesn't seem
feasible.  Any input would be greatly appreciated.

Re: Restoring a single DB in a cluster

[Paul Davis <pa...@gmail.com>]

Actually thats basically it. The only catch is that you have to make sure
your dbs and nodes dbs are right. Assuming your nodes db hasn't changed and
you have a snapshot of the dbs db, i'd just take the dbs db and name it
something else and pull the db doc off it amd re store it in the live dbs
db. You could also hand craft ilthat doc in the worst case. Or if you're
lucky and the dbs db hasn't been compacted you may be able to just open the
old revision.

But basically restoring the individual shard files will work just fine.
It'll heal itself for any discrepancy from the snapshot as long as you
restore the same shards to the same nodes. The only hiccup that we don't
cover very well is the dbs doc and the neccesarry consistency with the live
nodes. And by handle i just mean there's not very good tooling for it.

I'm rattling this off while having a smoke outside the bar so if I'm not
being clear feel free to ask more questions and I'll get back to you
durring normal thinking hours.


On Sun, Mar 26, 2017 at 4:50 PM Garth Gutenberg <ga...@gmail.com>
wrote:

> Hello developers.  I can't find any documentation on this, so if it exists,
> please point me at it.  Otherwise, I need some input.
>
> What is the official, sanctioned way to restore a DB in a cluster
> configuration?
>
> Scenario:
> 3 servers in a cluster.  A client deletes a bunch of docs from their DB, or
> some other DB-specific catastrophe happens.  What is the correct approach
> for restoring this DB from a backup?  Let's assume that I have disk
> snapshots of every node in the cluster taken at *approximately* the same
> time from the day before.
>
> In the old days, pre-cluster, it was as simple as just copying the
> <db>.couch file... but with shards and clusters, this just doesn't seem
> feasible.  Any input would be greatly appreciated.
>