Write Consistency Level for index/delete/delete_by_query/bulk with one/quorum/all. Defaults to quorum.

[kimchy]

When performing a "write" operation, allow to control if it can be performed only when one shard is active, a quorum of shards are active (within a replication group), or all.

This can be controlled per API call (REST parameter is consistency and can be set to one, quorum, or all. It defaults to a node level setting of action.write_consistency which in turn defaults to quorum. This basically means that the default is quorum.

What does this means? Basically, in a 1 shard with 2 replicas, there will have to be at least 2 shards active (quorum) within the cluster for the operation to be performed.

In a 1 shard with 1 replica, at least 1 shard will need to be active (in this case quorum and one is the same).

[kimchy]

Write Consistency Level for index/delete/delete_by_query/bulk with one/quorum/all. Defaults to quorum, closed by 5d1d927.

[clintongormley]

What happens if a quorum isn't available - will the write hang or throw an error?

[kimchy]

It will wait for the provided timeout parameter for the expected consistency level to be met, and if its not, it will bail with an exception.

[ppearcy]

Hey,
This should never effect consistency of search results, correct? Fundamentally, what does this setting get you? Maybe, a extra durability in case of a getting killed before snapshotting?

Not sure if this is completely on topic for this, but when a doc is indexed, it can never fail on one replica and succeed on another, correct? I ask, because I have been and still am seeing minor consistency issues when I test bringing nodes up and down on latest master.

Thanks!

[ppearcy]

Err... nevermind about the second part... Everything has stayed consistent on 0.13-SNAPSHOT.

[kimchy]

No, its just for writes, it basically allows to control for a write to succeed only if a specific number of shards to be available.