fix(esql): refresh keep_alive TTL on async query GET requests

[emreyalvac]

Summary

Fixes a bug in the ESQL async query API where updating the keep_alive value via a GET /_query/async/{id}?keep_alive=... request did not properly extend the lifetime of the async result, causing unexpected 404s after the original TTL expired.

Details

• The initial keep_alive duration specified in the POST request was honored, but follow-up GET requests with a new keep_alive value failed to refresh the TTL.
• This PR ensures that if a new keep_alive value is passed in the follow-up GET request, it will properly update the expiration time of the async result.

How to Reproduce (Before Fix)

1. Submit an async ESQL query with a short keep_alive, e.g. 15s.
2. Within the TTL window, send a GET /_query/async/{id}?keep_alive=60s
3. Wait >15s
4. Expect 404 (bug)

After Fix

• Async results will persist according to the most recent keep_alive value provided in the GET call.
• Prevents premature cleanup.

Related

Closes: #130619

[emreyalvac]

@kertal can you check?

[kertal]

thx @emreyalvac not sure who's the best to review this, since I just reported it , and it's something we need in Kibana. dear @elasticsearch-esql , can you help me (or know who can take care of it, thx) . If this is the solution I assume it needs at least one test to cover it

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[astefan]

buildkite test this

[astefan]

@emreyalvac thank you for providing a fix in this PR. Would you mind looking into the test failures and, also, add a test for this fix?

[astefan]

CC @ChrisHegarty if you have a quicker turnaround on this one 🙏 .

[ChrisHegarty]

@elasticmachine update branch

[ChrisHegarty]

I locally hacked EsqlAsyncActionIT in order to minimally reproduce the issue, and verify the fix. I managed to reproduce the observed problem, but the fix does not work. It immediately returns org.elasticsearch.ResourceNotFoundException.

@emreyalvac did you reproduce and verity the change?

[ChrisHegarty]

buildkite test this

[ChrisHegarty]

One can set a keep-alive time through the initial query, and this works. I'm not sure that the keep-alive update over async get was ever implemented. It's been more than a year since I wrote this code, and it has changed quite a bit. I think that there is some more work to be done if we wanna support this - more than is in this PR.

[emreyalvac]

@ChrisHegarty

With this PR, I can extend the keep_alive parameter for GET requests.

[ChrisHegarty]

With this PR, I can extend the keep_alive parameter for GET requests.

When I locally mocked up an incomplete test for this, I found that it did not work - at least not in my environment, tho it is possible that I did something wrong. That said, org.elasticsearch.xpack.esql.action.AsyncEsqlQueryActionIT.testBasicAsyncExecution fails with this change, because of org.elasticsearch.ResourceNotFoundException: ..., so something is is clearly not right with this change as it stands.

If you believe that it works, then can you please add a test that demonstrates that, as well as fixing whatever has been broken in AsyncEsqlQueryActionIT.testBasicAsyncExecution.

Finally, the test scenario coverage should include updating the keep_alive TTL both 1) when the query is still executing, as well as 2) after it has completed (but it still retained).

[kertal]

I also tested this PR in a Kibana PR + Elasticsearch PR a-la-carte deployment, and while the change seems to be in the right area of the code, it doesn't work, when testing it manually