Hi,
We're using 6.2.3, and I noticed some errors in our application logs yesterday that indicated a few shards (eg, 1-4 out of total of 90) failed while processing a scroll request. Unfortunately I don't have access to the Elasticsearch logs, only the application logs, and they just show number of failed shards from the _shards.failed field in the response, but no reason.
Since they're scrolls, I thought it may be a timeout, but to affect only a few shards seems odd.
What situations could cause this to happen?