[Cosmos] Hybrid Search query pipeline #38275
Conversation
|
API change check API changes are not detected in this pull request. |
| @@ -173,7 +173,7 @@ def __init__(self, execution_context, aggregate_operators): | |||
| for operator in aggregate_operators: | |||
| if operator == "Average": | |||
| self._local_aggregators.append(_AverageAggregator()) | |||
| elif operator == "Count": | |||
| elif operator in ("Count", "CountIf"): | |||
There was a problem hiding this comment.
There are no changes to _CountAggregator in this PR - how is this supposed to start supporting CountIf?
| @@ -187,7 +187,7 @@ def __init__(self, execution_context, aggregate_operators): | |||
| for operator in aggregate_operators: | |||
| if operator == "Average": | |||
| self._local_aggregators.append(_AverageAggregator()) | |||
| elif operator == "Count": | |||
| elif operator in ("Count", "CountIf"): | |||
There was a problem hiding this comment.
No changes to CountAggregator - how woudl this support new Aggregate?
| @@ -136,6 +145,22 @@ def _create_pipelined_execution_context(self, query_execution_info): | |||
| self._query, | |||
| self._options, | |||
| query_execution_info) | |||
| elif query_execution_info.has_hybrid_search_query_info(): | |||
There was a problem hiding this comment.
can thge exception factory be refactroed to share across sync and async?
| self.test_db = self.client.create_database(str(uuid.uuid4())) | ||
| self.test_container = self.test_db.create_container( | ||
| id="FTS" + self.TEST_CONTAINER_ID, | ||
| partition_key=PartitionKey(path="/id"), |
There was a problem hiding this comment.
Same comment I left for Java
think teh structure of the docs is not very helpful - You would need test cases to target individual physical partitions (both via PK in query request options or just via query plan by where ocndition) as well as real cross partition queries - technically you could still do it with id - but the addiitonal fulltext / hybrid-search becomes use-less on a single doc. So, I would argue there should be few logical partitions with 100 docs (coudl be same ids form the file) each -a nd then have tests searching across partiitons as wlel as scoped to individual partition - ideally also with HPK. You kind of have to add all these test cases to the matrix because we can't rely on it magically working given the extra step getting global statistics.
I think the test matrix will need to get extended quite a bit - maybe let's jump on a call Monday morning to discuss this - and also what tests are required for merge - vs. ok to add later.
Adds support for performing full text search queries through the introduction of the hybrid search query pipeline. This consists of the newly added
hybrid_search_aggregator, which performs the necessary query steps to obtain the needed results.With these changes, the SDK can now interpret queries utilizing key functions like
FullTextContains(),FullTextContainsAll(),FullTextContainsAny(),Order By Rank <FullTextFunction>(), andOrder By Rank RRF().The design doc for the implementation can be found here: Hybrid Search Doc.
The new README in this PR also has additional information.
Still missing in this PR at the moment: