Use alluxio as shared file system

Does elasticsearch will support alluxio as the underly shared filesystem. I think it has following advantages:

  1. It will reduce cpu usage since data will be send only to primary replica and build index and the replicas will get the index segment from alluxio. And it is also the same during segment merge process.
  2. Alluxio will cache data on local disk and it has short circuit read. It will not defect performance.
  3. Elastic will become stateless and could be integrated with k8s more naturally.

Each node in Elasticsearch need to have its own copy of shards allocated to it so can not share data through a shared file system.

In the past, Es supports shard file system. Primary replica could write data and follower replica implements a read only engine. They share same data on a shared file system.

If you are referring to shadow replicas, yes this feature has been removed. See discussion at

1 Like

Yes. It is shadow replica. In the past, shadow replica will read data from remote shared file system. Its performance maybe a little worse compared to local file system. But since alluxio could cache data on local file system. It will not defect performance. In the past, many deployment is in on promise environment and users could not find a very reliable shared filesystem. And it is also very difficult to manage both es and the shared file system.
But there are many reliable storage system in public cloud and users do not need to manage them. I think it could reduce cost and simply es's architecture such as replication and recovery.

So the answer is that feature have been removed.
I don't think it will come back again as it was (ie for live indices). But with hot/warm/cold architecture, you can probably store your cold indices on whatever shared FS may be and use the freeze API (

See also:

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.