Fix serverless test

[carlosdelest]

This test fails on serverless, as the wait for index to be green happened after the second insertion into the synonyms index. It should happen after the first to ensure the index has been created and search shards are available in order to perform the second insert.

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[carlosdelest]

@benwtrent @mayya-sharipova may I get your review in this test change?

Thanks!

[benwtrent]

I think this is OK. But I would expect users to want the ability to ensure that the synonyms are fully reloaded (e.g. the API waits until some timeout for it to occur).

Could we open an issue to add a flag to the API that allows it to spin wait until the synonyms are fully reloaded and available? Of course, there is always the case where it remains unavailable due to bad state, etc. So, there will have to be some timeout associated with it.

[carlosdelest]

But I would expect users to want the ability to ensure that the synonyms are fully reloaded (e.g. the API waits until some timeout for it to occur).

The API does not return until all analyzers have been reloaded. But that doesn't mean that it succeeds on the operation, which is what happened on the test.

We could add a final check on the API requests that updates the synonym(s) to perform a request for them afterwards, and deal there with the potential 503 returned in serverless - waiting for a timeout if needed. I'd put that as part of the default behaviour as the operation is expected to return when analyzers have been reloaded and we're all set. I've opened #121441 to deal with that.

I'm not sure how other system indices deal with this situation, if they do it at all. 🤔