Add integration test for doc mapper update

[rdettai]

Description

Add some high level validation that searches still work after doc mapper update.

It would be interesting to do the same with aggregations.

Part of #5084

How was this PR tested?

NA (these are tests)

[rdettai]

Any juicy scenario suggestion is welcome

[trinity-1686a]

ideas of test i've run manually with various level of success, which represent some edge case i thought of:

• u64 -> string, verify the field is still present after update (initially broken, fixed by cast updated fields #5076)
• with dynamic mode: create index without field1, ingest data with field1, create that field1, ingest some more. Check we can search for both before and after.
• with strict mode: create index without a field1, ingest docs without field1, create field1, ingest data with this field. Check we don't get error from old splits because of searching on unknown field
• index string with raw tokenizer no position; ingest document containing - (dash) in value, such as roller-coaster. update tokenizer to default, with positions. ingest more documents. make a phrase query for that doc (this is expected to work)
• index string with default tokenizer no position; ingest document with - (dash) in value. update to enable positions. ingest more documents. make a phrase query for that doc. this one should work in the future, but is expected to fail currently

[rdettai]

@trinity-1686a I added all your test scnarii

Following your suggestion to test "u64 -> string, verify the field is still present after update", I changed the assert from just the number of hits to the whole hit content. I got 2 surprising results:

• test_update_doc_mapping_json_to_object: I update a json type to an explicit object type, and the fields are dropped in the resulting hit.
• test_update_doc_mapping_strict_to_dynamic: I switch from an explicit (strict) schema to a dynamic one, and the fields are dropped in the resulting hit.

[trinity-1686a]

that's interesting, can you add these test as disabled, and i'll try to make them pass

edit: gh wasn't showing me the whole file, the tests are already there. Still you'll need to disable them before we can merge

[rdettai]

done

[github-actions]

On SSD:

Average search latency is 0.998x that of the reference (lower is better).
Ref run id: 2598, ref commit: 5b5978e
Link

On GCS:

Average search latency is 0.998x that of the reference (lower is better).
Ref run id: 2599, ref commit: 5b5978e
Link