Snapshot/Restore: Add ability to restore partial snapshots

[geekpete]

Is there a way to copy a single index from a snapshot marked as "incomplete" by the restore function to the data dir of a single node elastic cluster in order to restore that index data by reopening it as a closed index?

eg, to restore a single index from a multi-index snapshot (an index marked as incompleted).

from whatever source location with a dir structure like:

snapshotname/indices/myindexname/*

to:

/var/lib/elasticsearch/mycluster/nodes/0/indices/myindexname/*

will elasticsearch see this as a closed index and provide the ability to online it again?

Failing that, is there any other possible way to salvage data from a snapshotted index that is marked as incomplete, especially if the size of the index appears to be correct?

[imotov]

You should be able to restore all indices that were successfully snapshotted except the incomplete index by excluding the incomplete index from the restore operation:

$ curl -XPOST "localhost:9200/_snapshot/my_backup/snapshot_1/_restore" -d '{
    "indices": "-corrupted_index",
    "include_global_state": false
}'

There is no way simple to force incomplete index to be restored at the moment. What type of failures do you see when you run the following command:

$ curl -XGET localhost:9200/_snapshot/repository_name/snapshot_name

[geekpete]

I have it working. :)

You can adjust the metadata files in the snapshot to allow the restore to work.

After restoring the incomplete index, I get an incomplete/uninitialisable shard that must be replaced with a blank one from a temporary index to get the cluster back to green (as I was able to prove the data for that shard was simply missing from the snapshot.)

But an override to the usual behaviour would be nice to do the following:
-allow restoration of incomplete indexes from snapshots
-provide the ability to zero out the data in bad/incomplete shards if the data is missing to ensure green state when restored
-provide detailed feedback on the shards that have been reset to blank and how much data and how many documents were lost

I do realise that once again my case is probably an edge case so not that high priority compared to other features/bugs. :)

Thanks.