feat(grouping): Add model parameter support for v2 similarity grouping rollout

src/sentry/features/temporary.py

@@ -625,6 +625,8 @@ def register_temporary_features(manager: FeatureManager) -> None:
     manager.add("projects:similarity-embeddings", ProjectFeature, FeatureHandlerStrategy.INTERNAL, default=False, api_expose=True)
     manager.add("projects:similarity-indexing", ProjectFeature, FeatureHandlerStrategy.INTERNAL, api_expose=False)
     manager.add("projects:similarity-view", ProjectFeature, FeatureHandlerStrategy.INTERNAL, api_expose=True)
+    # Enable v2 similarity grouping model (part of v2 grouping rollout)
+    manager.add("projects:similarity-grouping-v2-model", ProjectFeature, FeatureHandlerStrategy.FLAGPOLE, api_expose=False)
     # Starfish: extract metrics from the spans
     manager.add("projects:span-metrics-extraction", ProjectFeature, FeatureHandlerStrategy.INTERNAL, api_expose=True)
     manager.add("projects:span-metrics-extraction-addons", ProjectFeature, FeatureHandlerStrategy.INTERNAL, api_expose=False)

src/sentry/grouping/ingest/seer.py

@@ -6,7 +6,7 @@
 from django.conf import settings
 from django.utils import timezone
 
-from sentry import options
+from sentry import features, options
 from sentry import ratelimits as ratelimiter
 from sentry.conf.server import SEER_SIMILARITY_MODEL_VERSION
 from sentry.grouping.grouping_info import get_grouping_info_from_variants_legacy
@@ -18,7 +18,7 @@
 from sentry.models.grouphash import GroupHash
 from sentry.models.project import Project
 from sentry.seer.similarity.similar_issues import get_similarity_data_from_seer
-from sentry.seer.similarity.types import SimilarIssuesEmbeddingsRequest
+from sentry.seer.similarity.types import GroupingVersion, SimilarIssuesEmbeddingsRequest
 from sentry.seer.similarity.utils import (
     SEER_INELIGIBLE_EVENT_PLATFORMS,
     ReferrerOptions,
@@ -272,6 +272,11 @@ def get_seer_similar_issues(
         get_stacktrace_string(get_grouping_info_from_variants_legacy(variants)),
     )
 
+    # Get model configuration from feature flags
+    use_v2_model = features.has("projects:similarity-grouping-v2-model", event.project)
+    model_version = GroupingVersion.V2 if use_v2_model else GroupingVersion.V1
+    training_mode = False  # PR #B will add the smart logic
+
     request_data: SimilarIssuesEmbeddingsRequest = {
         "event_id": event.event_id,
         "hash": event_hash,
@@ -281,10 +286,16 @@ def get_seer_similar_issues(
         "k": options.get("seer.similarity.ingest.num_matches_to_request"),
         "referrer": "ingest",
         "use_reranking": options.get("seer.similarity.ingest.use_reranking"),
+        "model": model_version,
+        "training_mode": training_mode,
     }
     event.data.pop("stacktrace_string", None)
 
-    seer_request_metric_tags = {"platform": event.platform or "unknown"}
+    seer_request_metric_tags: dict[str, str | int | bool] = {
+        "platform": event.platform or "unknown",
+        "model_version": model_version.value,
+        "training_mode": training_mode,
+    }
 
     seer_results = get_similarity_data_from_seer(
         request_data,

src/sentry/issues/endpoints/group_similar_issues_embeddings.py

@@ -6,7 +6,7 @@
 from rest_framework.request import Request
 from rest_framework.response import Response
 
-from sentry import analytics, options
+from sentry import analytics, features, options
 from sentry.api.analytics import GroupSimilarIssuesEmbeddingsCountEvent
 from sentry.api.api_owners import ApiOwner
 from sentry.api.api_publish_status import ApiPublishStatus
@@ -17,7 +17,11 @@
 from sentry.models.group import Group
 from sentry.models.grouphash import GroupHash
 from sentry.seer.similarity.similar_issues import get_similarity_data_from_seer
-from sentry.seer.similarity.types import SeerSimilarIssueData, SimilarIssuesEmbeddingsRequest
+from sentry.seer.similarity.types import (
+    GroupingVersion,
+    SeerSimilarIssueData,
+    SimilarIssuesEmbeddingsRequest,
+)
 from sentry.seer.similarity.utils import (
     ReferrerOptions,
     event_content_has_stacktrace,
@@ -100,6 +104,11 @@ def get(self, request: Request, group: Group) -> Response:
         if not stacktrace_string or not latest_event:
             return Response([])  # No exception, stacktrace or in-app frames, or event
 
+        # Get model configuration from feature flags
+        use_v2_model = features.has("projects:similarity-grouping-v2-model", group.project)
+        model_version = GroupingVersion.V2 if use_v2_model else GroupingVersion.V1
+        training_mode = False  # TODO: currently hardcoded, follow up PR will add the logic
+
         similar_issues_params: SimilarIssuesEmbeddingsRequest = {
             "event_id": latest_event.event_id,
             "hash": latest_event.get_primary_hash(),
@@ -109,6 +118,8 @@ def get(self, request: Request, group: Group) -> Response:
             "read_only": True,
             "referrer": "similar_issues",
             "use_reranking": options.get("seer.similarity.similar_issues.use_reranking"),
+            "model": model_version,
+            "training_mode": training_mode,
         }
         # Add optional parameters
         if request.GET.get("k"):

src/sentry/seer/similarity/grouping_records.py

@@ -15,7 +15,7 @@
 )
 from sentry.net.http import connection_from_url
 from sentry.seer.signed_seer_api import make_signed_seer_api_request
-from sentry.seer.similarity.types import RawSeerSimilarIssueData
+from sentry.seer.similarity.types import GroupingVersion, RawSeerSimilarIssueData
 from sentry.utils import json, metrics
 
 logger = logging.getLogger(__name__)
@@ -36,6 +36,8 @@ class CreateGroupingRecordsRequest(TypedDict):
     data: list[CreateGroupingRecordData]
     stacktrace_list: list[str]
     use_reranking: bool | None
+    model: NotRequired[GroupingVersion]  # Model version, defaults to V1 for backward compatibility
+    training_mode: NotRequired[bool]
 
 
 class BulkCreateGroupingRecordsResponse(TypedDict):
@@ -62,6 +64,8 @@ def post_bulk_grouping_records(
             [len(stacktrace) for stacktrace in grouping_records_request["stacktrace_list"]]
         ),
         "use_reranking": grouping_records_request.get("use_reranking"),
+        "model": grouping_records_request.get("model"),
+        "training_mode": grouping_records_request.get("training_mode"),
     }
 
     try:

src/sentry/seer/similarity/similar_issues.py

@@ -53,7 +53,8 @@ def get_similarity_data_from_seer(
     logger_extra = {
         k: v
         for k, v in similar_issues_request.items()
-        if k in {"event_id", "project_id", "hash", "referrer", "use_reranking"}
+        if k
+        in {"event_id", "project_id", "hash", "referrer", "use_reranking", "model", "training_mode"}
     }
     logger.info(
         "get_seer_similar_issues.request",

src/sentry/seer/similarity/types.py

@@ -1,13 +1,21 @@
 import logging
 from collections.abc import Mapping
 from dataclasses import dataclass
+from enum import StrEnum
 from typing import Any, ClassVar, NotRequired, Self, TypedDict
 
 from sentry.models.grouphash import GroupHash
 
 logger = logging.getLogger(__name__)
 
 
+class GroupingVersion(StrEnum):
+    """Model version for similarity grouping."""
+
+    V1 = "v1"
+    V2 = "v2"
+
+
 class IncompleteSeerDataError(Exception):
     pass
 
@@ -31,6 +39,8 @@ class SimilarIssuesEmbeddingsRequest(TypedDict):
     event_id: NotRequired[str]
     referrer: NotRequired[str]
     use_reranking: NotRequired[bool]
+    model: NotRequired[GroupingVersion]  # Model version, defaults to V1 for backward compatibility
+    training_mode: NotRequired[bool]  # whether to just insert embedding without querying
 
 
 class RawSeerSimilarIssueData(TypedDict):

src/sentry/tasks/embeddings_grouping/utils.py

@@ -11,7 +11,7 @@
 from google.api_core.exceptions import DeadlineExceeded, ServiceUnavailable
 from snuba_sdk import Column, Condition, Entity, Limit, Op, Query, Request
 
-from sentry import nodestore, options
+from sentry import features, nodestore, options
 from sentry.conf.server import SEER_SIMILARITY_MODEL_VERSION
 from sentry.grouping.grouping_info import get_grouping_info_from_variants_legacy
 from sentry.grouping.grouptype import ErrorGroupType
@@ -25,6 +25,7 @@
     post_bulk_grouping_records,
 )
 from sentry.seer.similarity.types import (
+    GroupingVersion,
     IncompleteSeerDataError,
     SeerSimilarIssueData,
     SimilarHashMissingGroupError,
@@ -491,12 +492,20 @@ def send_group_and_stacktrace_to_seer(
         f"{BACKFILL_NAME}.send_group_and_stacktrace_to_seer",
         sample_rate=options.get("seer.similarity.metrics_sample_rate"),
     ):
+        # Get model configuration from feature flags
+        project = Project.objects.get_from_cache(id=project_id)
+        use_v2_model = features.has("projects:similarity-grouping-v2-model", project)
+        model_version = GroupingVersion.V2 if use_v2_model else GroupingVersion.V1
+        training_mode = False  # TODO: currently hardcoded, follow up PR will add the logic
+
         return _make_seer_call(
             CreateGroupingRecordsRequest(
                 group_id_list=groups_to_backfill_with_no_embedding_has_snuba_row_and_nodestore_row,
                 data=nodestore_results["data"],
                 stacktrace_list=nodestore_results["stacktrace_list"],
                 use_reranking=options.get("similarity.backfill_use_reranking"),
+                model=model_version,
+                training_mode=training_mode,
             ),
             project_id,
         )
@@ -508,13 +517,21 @@ def send_group_and_stacktrace_to_seer_multithreaded(
     nodestore_results,
     project_id,
 ):
+    # Get model configuration from feature flags
+    project = Project.objects.get_from_cache(id=project_id)
+    use_v2_model = features.has("projects:similarity-grouping-v2-model", project)
+    model_version = GroupingVersion.V2 if use_v2_model else GroupingVersion.V1
+    training_mode = False  # TODO: currently hardcoded, follow up PR will add the logic
+
     def process_chunk(chunk_data, chunk_stacktrace):
         return _make_seer_call(
             CreateGroupingRecordsRequest(
                 group_id_list=chunk_data["group_ids"],
                 data=chunk_data["data"],
                 stacktrace_list=chunk_stacktrace,
                 use_reranking=options.get("similarity.backfill_use_reranking"),
+                model=model_version,
+                training_mode=training_mode,
             ),
             project_id,
         )

tests/sentry/grouping/seer_similarity/test_get_seer_similar_issues.py

@@ -10,7 +10,7 @@
 from sentry.models.grouphash import GroupHash
 from sentry.models.grouphashmetadata import GroupHashMetadata
 from sentry.models.project import Project
-from sentry.seer.similarity.types import SeerSimilarIssueData
+from sentry.seer.similarity.types import GroupingVersion, SeerSimilarIssueData
 from sentry.seer.similarity.utils import get_stacktrace_string
 from sentry.services.eventstore.models import Event
 from sentry.testutils.cases import TestCase
@@ -108,8 +108,15 @@ def test_sends_expected_data_to_seer(self, mock_get_similarity_data: MagicMock)
                 "k": options.get("seer.similarity.ingest.num_matches_to_request"),
                 "referrer": "ingest",
                 "use_reranking": True,
+                "model": GroupingVersion.V1,
+                "training_mode": False,
+            },
+            {
+                "platform": "python",
+                "model_version": "v1",
+                "training_mode": False,
+                "hybrid_fingerprint": False,
             },
-            {"platform": "python", "hybrid_fingerprint": False},
         )
 
     @patch("sentry.grouping.ingest.seer.metrics.incr")
@@ -152,6 +159,8 @@ def test_sends_second_seer_request_when_seer_matches_are_unusable(
                 "project_id": self.project.id,
                 "stacktrace": new_stacktrace_string,
                 "exception_type": "FailedToFetchError",
+                "model": GroupingVersion.V1,
+                "training_mode": False,
             }
 
             assert mock_get_similarity_data.call_count == 2
@@ -164,7 +173,12 @@ def test_sends_second_seer_request_when_seer_matches_are_unusable(
                         "referrer": "ingest",
                         "use_reranking": True,
                     },
-                    {"platform": "python", "hybrid_fingerprint": False},
+                    {
+                        "platform": "python",
+                        "model_version": "v1",
+                        "training_mode": False,
+                        "hybrid_fingerprint": False,
+                    },
                 ),
                 # Second call to store the event's data since the match that came back from Seer
                 # wasn't usable
@@ -175,7 +189,11 @@ def test_sends_second_seer_request_when_seer_matches_are_unusable(
                         "referrer": "ingest_follow_up",
                         "use_reranking": False,
                     },
-                    {"platform": "python"},
+                    {
+                        "platform": "python",
+                        "model_version": "v1",
+                        "training_mode": False,
+                    },
                 ),
             ]
 

tests/sentry/grouping/seer_similarity/test_seer.py

@@ -4,6 +4,7 @@
 from sentry import options
 from sentry.grouping.ingest.seer import maybe_check_seer_for_matching_grouphash
 from sentry.models.grouphash import GroupHash
+from sentry.seer.similarity.types import GroupingVersion
 from sentry.seer.similarity.utils import MAX_FRAME_COUNT
 from sentry.services.eventstore.models import Event
 from sentry.testutils.cases import TestCase
@@ -60,8 +61,15 @@ def test_simple(self, mock_get_similarity_data: MagicMock) -> None:
                 "k": 1,
                 "referrer": "ingest",
                 "use_reranking": True,
+                "model": GroupingVersion.V1,
+                "training_mode": False,
+            },
+            {
+                "platform": "python",
+                "model_version": "v1",
+                "training_mode": False,
+                "hybrid_fingerprint": False,
             },
-            {"platform": "python", "hybrid_fingerprint": False},
         )
 
     @patch("sentry.grouping.ingest.seer.record_did_call_seer_metric")
@@ -182,6 +190,13 @@ def test_too_many_frames_bypassed_platform(self, mock_get_similarity_data: Magic
                 "k": 1,
                 "referrer": "ingest",
                 "use_reranking": True,
+                "model": GroupingVersion.V1,
+                "training_mode": False,
+            },
+            {
+                "platform": "python",
+                "model_version": "v1",
+                "training_mode": False,
+                "hybrid_fingerprint": False,
             },
-            {"platform": "python", "hybrid_fingerprint": False},
         )

tests/sentry/issues/endpoints/test_group_similar_issues_embeddings.py

@@ -179,6 +179,8 @@ def test_simple(
             "read_only": True,
             "referrer": "similar_issues",
             "use_reranking": True,
+            "model": "v1",
+            "training_mode": False,
             "k": 1,
         }
 
@@ -350,6 +352,8 @@ def test_incomplete_return_data(
                     "read_only": True,
                     "referrer": "similar_issues",
                     "use_reranking": True,
+                    "model": "v1",
+                    "training_mode": False,
                 },
                 "raw_similar_issue_data": {
                     "should_group": True,
@@ -632,6 +636,8 @@ def test_no_optional_params(self, mock_seer_request: mock.MagicMock) -> None:
                     "read_only": True,
                     "referrer": "similar_issues",
                     "use_reranking": True,
+                    "model": "v1",
+                    "training_mode": False,
                 },
             ),
             headers={"content-type": "application/json;charset=utf-8"},
@@ -660,6 +666,8 @@ def test_no_optional_params(self, mock_seer_request: mock.MagicMock) -> None:
                     "read_only": True,
                     "referrer": "similar_issues",
                     "use_reranking": True,
+                    "model": "v1",
+                    "training_mode": False,
                     "k": 1,
                 },
             ),
@@ -691,6 +699,8 @@ def test_no_optional_params(self, mock_seer_request: mock.MagicMock) -> None:
                     "read_only": True,
                     "referrer": "similar_issues",
                     "use_reranking": True,
+                    "model": "v1",
+                    "training_mode": False,
                 },
             ),
             headers={"content-type": "application/json;charset=utf-8"},