Recover after failure (lost shards)

Heya,

I just ran into a major node HDD failure and lost some shards on as index,
so this is bad, but not that bad since we are in the process or reindexing
everything. However, I would like to replace the lost shards with "dummy"
empty shards just to retrieve the green cluster state.
I've read somewhere (
http://elasticsearch-users.115913.n3.nabble.com/Recovering-after-shard-failure-td4018776.html)
that creating a similar index on a local cluster and moving the empty
missing shards to some of the cluster nodes and they will be recovered.

However, when I tried it, I got this error;

I was wondering, is this method still valid with 0.20.x versions ? And if
not, is there a way to do it ? I would be nice to have this... I've seen
some post about it, but the only solution I have found is the one I
mentionned.

Many thanks,

Jerome

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Oh, forgot to mention; we don't have replicas on this index. (We are
currently adding them, while reindexing)

Don't do that by the way (no replicas), its a massive pain. :slight_smile:

On Tuesday, May 7, 2013 4:33:57 PM UTC-4, Jérôme Gagnon wrote:

Heya,

I just ran into a major node HDD failure and lost some shards on as index,
so this is bad, but not that bad since we are in the process or reindexing
everything. However, I would like to replace the lost shards with "dummy"
empty shards just to retrieve the green cluster state.
I've read somewhere (
http://elasticsearch-users.115913.n3.nabble.com/Recovering-after-shard-failure-td4018776.html)
that creating a similar index on a local cluster and moving the empty
missing shards to some of the cluster nodes and they will be recovered.

However, when I tried it, I got this error;
https://gist.github.com/jgagnon1/a4ebc27b17e8451af59e

I was wondering, is this method still valid with 0.20.x versions ? And if
not, is there a way to do it ? I would be nice to have this... I've seen
some post about it, but the only solution I have found is the one I
mentionned.

Many thanks,

Jerome

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Thanks to Shay, I finally got it working; Basically the solution was the
good one; but I made two mistakes;

  1. Shard folder permissions; the shard folder must have the same
    owner/permission as the other (user that run ES process)
  2. Created a "dummy index" from another cluster which was running
    ElasticSearch 0.90.x which was creating Lucene 4.x shard instead of 3.6

Shay told me that in 0.90.x the error message when ES is not able to load
some Lucene shard are much clearer, good to know!

But then again, we are running without replicas on one index due to several
constraint (IO, etc...), and this is not the way to go, we've been butt
hurt (and still are) many time due to that, and we are moving forward with
a better architecture.

On the end, I got the cluster green and learned more. and finally went to
sleep, everything's good :slight_smile:

Jerome

On Tuesday, May 7, 2013 4:44:00 PM UTC-4, Jérôme Gagnon wrote:

Oh, forgot to mention; we don't have replicas on this index. (We are
currently adding them, while reindexing)

Don't do that by the way (no replicas), its a massive pain. :slight_smile:

On Tuesday, May 7, 2013 4:33:57 PM UTC-4, Jérôme Gagnon wrote:

Heya,

I just ran into a major node HDD failure and lost some shards on as
index, so this is bad, but not that bad since we are in the process or
reindexing everything. However, I would like to replace the lost shards
with "dummy" empty shards just to retrieve the green cluster state.
I've read somewhere (
http://elasticsearch-users.115913.n3.nabble.com/Recovering-after-shard-failure-td4018776.html)
that creating a similar index on a local cluster and moving the empty
missing shards to some of the cluster nodes and they will be recovered.

However, when I tried it, I got this error;
https://gist.github.com/jgagnon1/a4ebc27b17e8451af59e

I was wondering, is this method still valid with 0.20.x versions ? And if
not, is there a way to do it ? I would be nice to have this... I've seen
some post about it, but the only solution I have found is the one I
mentionned.

Many thanks,

Jerome

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.