elasticsearch icon indicating copy to clipboard operation
elasticsearch copied to clipboard
verified publisher icon elastic

Allow ack unpromotable refresh request before shard is created

Open ywangd opened this issue 1 year ago • 2 comments

If a search node lags behind in cluster state, it is possible that a refresh request arrives before the shard is even ready. This in turn fails the shard and leads to unnecessary failure and recovery. This PR allow search node to ack the refresh request before the shard is created to prevent that from happening. The subsequent recovery process of the search shard should ensure it using the latest commit data.

Depends on: serverless 2277

Resolves: serverless 2258

ywangd avatar Jun 27 '24 07:06 ywangd

Pinging @elastic/es-distributed (Team:Distributed)

elasticsearchmachine avatar Jun 27 '24 07:06 elasticsearchmachine

As suggested, updated this PR to just respond OK to the refresh request (depends on serverless 2195). Thanks!

ywangd avatar Jun 28 '24 07:06 ywangd

@elasticmachine update branch

ywangd avatar Jul 02 '24 02:07 ywangd