"no shard follow tasks" when pausing following on follower index

Elastic Stack Elasticsearch
1

:wave: everyone!

We are using CCR (elasticsearch v 6.8.x) for quite a time now to run in code where we automatically delete old indices.
The documentation states: The follower index must be paused and closed before invoking the unfollow API. (see: unfollow API documentation)

So, our worker kicks in and starts to invoke the pause API for a follower index, unsuccessfully.

method [POST], host [http://10.176.x.139:9200], URI [/ccr-visit.../_ccr/pause_follow], status line [HTTP/1.1 400 Bad Request]
{
    "error": {
        "root_cause": [
            {
                "type": "illegal_argument_exception",
                "reason": "no shard follow tasks for [ccr-visit...]"
            }
        ],
        "type": "illegal_argument_exception",
        "reason": "no shard follow tasks for [ccr-visit...]"
    },
    "status": 400
}

No logs on elastic side though.
Maybe it's worth noting, that the leader index got created - therefore also the follower - but never got data written to.

Is there a way to prevent this message, or is this expected for empty follower indices?

Thanks in advance!

2

Hi @kley,

Did you delete a follower index before pausing? If yes, you can safely ignore the error. We can improve the error handling for this situation. With https://github.com/elastic/elasticsearch/pull/44702, CCR should automatically unfollow if a follower index gets deleted. Ideally, we should pause, close, and unfollow before deleting a follower index.

I hope this helps.

3

I opened https://github.com/elastic/elasticsearch/pull/48915 to return a more appropriate error message.

4

Hey Nhat!
No I did not delete the index at all. It's still there.
So, I invoked _cat to show all indices and that this particular index is still there and then called the pause API which returns this error:

image
image.png1758×487 38.7 KB

5

Hi @kley,

Was there any master failover in the follower cluster? Can you share the cluster state of the follower cluster (to my email)? Thank you.