feat(seer): Use Seer model_used response field for seer_model on GroupHashMetadata

[yuvmen]

Use Seer's model_used response field to record what model Seer actually used, separately from what we requested.

When Sentry calls Seer for similarity grouping, GroupHashMetadata.seer_model was previously set from what Sentry requested. If Seer internally falls back from v2 to v1, we had no visibility into that. Seer now returns a model_used field in the response — this PR consumes it.

• seer_model ← what Seer actually used (model_used from response, falling back to the requested version)
• seer_latest_training_model ← what we sent (always the requested version)

Changes:

• Add model_used: NotRequired[str] to SimilarIssuesEmbeddingsResponse
• get_similarity_data_from_seer now returns tuple[list[SeerSimilarIssueData], str | None]
• get_seer_similar_issues now returns a 3-tuple threading model_used through
• maybe_check_seer_for_matching_grouphash sets seer_model from the response value

Backward compatible: model_used is parsed with .get() defaulting to None, and seer_model_used or model_version.value means behavior is identical to today if the field is absent. No migration needed.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: model_used discarded when Seer returns empty results
  • Changed return value from ([], None) to ([], model_used) at line 188 to preserve the model version when Seer returns no matches.

Or push these changes by commenting:

@cursor push 47bcd7300d

Preview (47bcd7300d)

diff --git a/src/sentry/seer/similarity/similar_issues.py b/src/sentry/seer/similarity/similar_issues.py
--- a/src/sentry/seer/similarity/similar_issues.py
+++ b/src/sentry/seer/similarity/similar_issues.py
@@ -185,7 +185,7 @@
             sample_rate=options.get("seer.similarity.metrics_sample_rate"),
             tags={**metric_tags, "outcome": "no_similar_groups"},
         )
-        return ([], None)
+        return ([], model_used)
 
     # This may get overwritten as we process the results, but by this point we know that Seer at
     # least found *something*

_{This Bugbot Autofix run was free. To enable autofix for future PRs, go to the Cursor dashboard.}

[github-actions]

Backend Test Failures

Failures on 303bd12 in this run:

tests/sentry/deletions/test_grouphash.py::DeleteGroupHashTest::test_deleting_grouphash_matched_by_seer — log

tests/sentry/deletions/test_grouphash.py:60: in test_deleting_grouphash_matched_by_seer
    assert new_event.group_id == existing_group_id
E   assert 22 == 21
E    +  where 22 = <sentry.services.eventstore.models.Event at 0x7f473eb6e780: event_id=5d1b267a4c2c40839db1c2cdcdf9f316>.group_id

tests/sentry/deletions/test_grouphash.py::DeleteGroupHashTest::test_deleting_grouphash_matched_by_seer_after_unmerge — log

tests/sentry/deletions/test_grouphash.py:104: in test_deleting_grouphash_matched_by_seer_after_unmerge
    assert new_event.group_id == existing_event.group.id
E   assert 30 == 29
E    +  where 30 = <sentry.services.eventstore.models.Event at 0x7fda3ffc8b50: event_id=da9678ec6b18489c8066c19d0bd5c866>.group_id
E    +  and   29 = <Group at 0x7fda45175090: id=29, project_id=4557816015618048>.id
E    +    where <Group at 0x7fda45175090: id=29, project_id=4557816015618048> = <sentry.services.eventstore.models.Event at 0x7fda3ffc8aa0: event_id=d18456c8c2c04611822494c587dc1a4b>.group

tests/sentry/deletions/test_group.py::DeleteGroupTest::test_delete_grouphashes_and_metadata — log

tests/sentry/deletions/test_group.py:322: in test_delete_grouphashes_and_metadata
    assert event_a.group_id == event_b.group_id
E   assert 19 == 20
E    +  where 19 = <sentry.services.eventstore.models.Event at 0x7f51c08e68e0: event_id=b618552c579c4350b3a5bafddb7e5ea5>.group_id
E    +  and   20 = <sentry.services.eventstore.models.Event at 0x7f51c0608680: event_id=db791883cfb741ff9a2da791afd09d37>.group_id