⚰️(summary) cleaned tasks failure handling

Code in task failure was assuming that the failure
signal would be called on retry which is not the case.

CHANGELOG.md

@@ -8,9 +8,15 @@ and this project adheres to
 
 ## [Unreleased]
 
+### Added
+
+- ✨(summary) report exception type in failure analytics
+
 ## Fixed
 
 - 🐛(transcription) fix silent bug in speaker assignment
+- 🐛(summary) extend tasks auto retry logic
+- 🐛(summary) properly detect when failure webhook should be sent
 
 ### Changed
 
@@ -18,6 +24,7 @@ and this project adheres to
 - ⬆️(frontend) upgrade i18next from 26.3.1 to 26.3.2
 - ⬆️(frontend) upgrade posthog-js from 1.391.2 to 1.395.0
 - ⬆️(frontend) upgrade @tanstack/react-query from 5.101.0 to 5.101.1
+- ⬆️(frontend) upgrade livekit-client from 2.19.2 to 2.20.0
 
 ## [1.24.0] - 2026-07-21
 

src/frontend/package-lock.json

@@ -29,7 +29,7 @@
         "i18next-browser-languagedetector": "8.2.1",
         "i18next-parser": "9.4.0",
         "i18next-resources-to-backend": "1.2.1",
-        "livekit-client": "2.19.2",
+        "livekit-client": "2.20.0",
         "posthog-js": "1.395.0",
         "react": "18.3.1",
         "react-aria": "3.50.0",
@@ -1028,9 +1028,9 @@
       "license": "Apache-2.0"
     },
     "node_modules/@livekit/protocol": {
-      "version": "1.45.8",
-      "resolved": "https://registry.npmjs.org/@livekit/protocol/-/protocol-1.45.8.tgz",
-      "integrity": "sha512-Q+l57E7w/xxOBFVWzdX5rkAZO7ffyF+rlDzNUYq2SU114+5aTyCq+PK4unaEVDNd4952Af7wteKr3sOgasGuaA==",
+      "version": "1.46.6",
+      "resolved": "https://registry.npmjs.org/@livekit/protocol/-/protocol-1.46.6.tgz",
+      "integrity": "sha512-upzlHP1vi/kZ/QqALZTFskQ0ifqc2f15RKucHYOsIHJsaXvEYanG75mAb7o+Yomfs4XhQ4BaRsdY+TFHXpaqrg==",
       "license": "Apache-2.0",
       "dependencies": {
         "@bufbuild/protobuf": "^1.10.0"
@@ -8125,13 +8125,13 @@
       "license": "MIT"
     },
     "node_modules/livekit-client": {
-      "version": "2.19.2",
-      "resolved": "https://registry.npmjs.org/livekit-client/-/livekit-client-2.19.2.tgz",
-      "integrity": "sha512-Kvk07QYDWRAbmYNLRll04ZIuxMQobW/oLPYnmR1kCy8GGHpU0gqyHf704Rz+29zfy8IJZRjKqeVbzGSKn9sumw==",
+      "version": "2.20.0",
+      "resolved": "https://registry.npmjs.org/livekit-client/-/livekit-client-2.20.0.tgz",
+      "integrity": "sha512-RIJcpvBmOmwz3jTj3rmdY6Dzr55HrhcaJjMgY+HSmoEM+yIRyA40m7r8UKv0hnZWM3z/AYhP1q8C8ciz5UWFKQ==",
       "license": "Apache-2.0",
       "dependencies": {
         "@livekit/mutex": "1.1.1",
-        "@livekit/protocol": "1.45.8",
+        "@livekit/protocol": "1.46.6",
         "events": "^3.3.0",
         "jose": "^6.1.0",
         "loglevel": "^1.9.2",

src/frontend/package.json

@@ -36,7 +36,7 @@
     "i18next-browser-languagedetector": "8.2.1",
     "i18next-parser": "9.4.0",
     "i18next-resources-to-backend": "1.2.1",
-    "livekit-client": "2.19.2",
+    "livekit-client": "2.20.0",
     "posthog-js": "1.395.0",
     "react": "18.3.1",
     "react-aria": "3.50.0",

src/summary/summary/core/analytics.py

@@ -202,13 +202,16 @@ class MetadataManager:
             },
         )
 
-    def capture(self, task_id, event_name):
+    def capture(self, task_id, event_name, extra_properties=None):
         """Capture analytics event with task metadata and clean up."""
         if self._is_disabled or not self.has_task_id(task_id):
             return
 
         metadata = self._get_metadata(task_id)
 
+        if extra_properties:
+            metadata = {**metadata, **extra_properties}
+
         if "start_time" in metadata:
             metadata["execution_time"] = round(time.time() - metadata["start_time"], 2)
             del metadata["start_time"]

src/summary/summary/core/celery_worker.py

@@ -431,7 +431,7 @@ def _should_auto_create_summary(payload: TranscribeTaskJob) -> bool:
 @celery.task(
     max_retries=3,
     queue=settings.call_webhook_queue_v2,
-    autoretry_for=[exceptions.HTTPError],
+    autoretry_for=[exceptions.RequestException],
 )
 def call_webhook_v2_task(
     payload: dict,
@@ -445,7 +445,9 @@ def call_webhook_v2_task(
 
 @celery.task(
     bind=True,
-    autoretry_for=[exceptions.HTTPError],
+    autoretry_for=[
+        exceptions.RequestException,
+    ],
     max_retries=settings.celery_max_retries,
     queue=settings.transcribe_queue_v2,
 )
@@ -575,12 +577,6 @@ def task_retry_handler_transcript(request=None, reason=None, einfo=None, **kwarg
     metadata_manager.retry(request.id)
 
 
-@signals.task_failure.connect(sender=process_audio_transcribe_v2_task)
-def task_failure_handler_transcript(task_id, exception=None, **kwargs):
-    """Signal handler called when task execution fails permanently."""
-    metadata_manager.capture(task_id, settings.posthog_transcript_failure)
-
-
 @signals.task_failure.connect(sender=process_audio_transcribe_v2_task)
 def handle_transcribe_v2_failed(
     sender,
@@ -594,30 +590,26 @@ def handle_transcribe_v2_failed(
 ):
     """Handle the failure of transcribe_v2_task.
 
-    This function is triggered when the transcribe_v2_task fails.
-    It sends a webhook failure payload to notify the client of the failure.
+    Tracks the failure event in analytics and sends a failure webhook to the client.
     """
-    n_retries_left = sender.max_retries - sender.request.retries - 1
-    if n_retries_left > 0:
-        logger.info(
-            "Transcribe task %s failed, %s retries left.",
-            task_id,
-            n_retries_left,
-        )
-    else:
-        logger.warn(
-            "Transcribe task %s failed, no more retries left, sending failure webhook.",
-            task_id,
-        )
-        call_webhook_v2_task.apply_async(
-            args=[
-                TranscribeWebhookFailurePayload(
-                    job_id=task_id,
-                    error_code="unknown_error",
-                ).model_dump(),
-                args[0]["tenant_id"],
-            ]
-        )
+    logger.error(
+        "Transcribe task %s failed, no more retries left, sending failure webhook.",
+        task_id,
+    )
+    metadata_manager.capture(
+        task_id,
+        settings.posthog_transcript_failure,
+        {"exception_type": type(exception).__name__},
+    )
+    call_webhook_v2_task.apply_async(
+        args=[
+            TranscribeWebhookFailurePayload(
+                job_id=task_id,
+                error_code="unknown_error",
+            ).model_dump(),
+            args[0]["tenant_id"],
+        ]
+    )
 
 
 @celery.task(
@@ -681,12 +673,6 @@ def task_retry_handler_summary(request=None, reason=None, einfo=None, **kwargs):
     metadata_manager.retry(request.id)
 
 
-@signals.task_failure.connect(sender=summarize_v2_task)
-def task_failure_handler_summary(task_id, exception=None, **kwargs):
-    """Signal handler called when task execution fails permanently."""
-    metadata_manager.capture(task_id, settings.posthog_summary_failure)
-
-
 @signals.task_failure.connect(sender=summarize_v2_task)
 def handle_summarize_v2_failed(
     sender,
@@ -700,27 +686,23 @@ def handle_summarize_v2_failed(
 ):
     """Handle the failure of summarize_v2_task.
 
-    This function is triggered when the summarize_v2_task fails.
-    It sends a webhook failure payload to notify the client of the failure.
+    Tracks the failure event in analytics and sends a failure webhook to the client.
     """
-    n_retries_left = sender.max_retries - sender.request.retries - 1
-    if n_retries_left > 0:
-        logger.info(
-            "Summary task %s failed, %s retries left.",
-            task_id,
-            n_retries_left,
-        )
-    else:
-        logger.warn(
-            "Summary task %s failed, no more retries left, sending failure webhook.",
-            task_id,
-        )
-        call_webhook_v2_task.apply_async(
-            args=[
-                SummarizeWebhookFailurePayload(
-                    job_id=task_id,
-                    error_code="unknown_error",
-                ).model_dump(),
-                args[0]["tenant_id"],
-            ]
-        )
+    logger.warn(
+        "Summary task %s failed, no more retries left, sending failure webhook.",
+        task_id,
+    )
+    metadata_manager.capture(
+        task_id,
+        settings.posthog_summary_failure,
+        {"exception_type": type(exception).__name__},
+    )
+    call_webhook_v2_task.apply_async(
+        args=[
+            SummarizeWebhookFailurePayload(
+                job_id=task_id,
+                error_code="unknown_error",
+            ).model_dump(),
+            args[0]["tenant_id"],
+        ]
+    )

src/summary/tests/unit/test_user_assign.py

@@ -1,7 +1,6 @@
 """Tests for the speaker-to-user assignment service."""
 
 import math
-from dataclasses import dataclass, field
 from datetime import datetime
 
 from summary.core import user_assign
@@ -17,11 +16,9 @@ from summary.core.user_assign import (
 )
 
 
-@dataclass
-class FakeTranscription:
-    """Mimics the OpenAI Transcription pydantic model for testing."""
-
-    segments: list = field(default_factory=list)
+def make_transcription(segments: list | None = None) -> dict:
+    """Build a WhisperX transcription dict, matching WhisperXResponse.model_dump()."""
+    return {"segments": segments or []}
 
 
 RECORDING_START = datetime.fromisoformat("2026-03-17T15:30:33.000001")
@@ -68,7 +65,7 @@ METADATA_SINGLE_USER = {
     ],
 }
 
-DIARIZATION_SINGLE_SPEAKER = FakeTranscription(
+DIARIZATION_SINGLE_SPEAKER = make_transcription(
     segments=[
         {
             "start": 1.363,
@@ -182,7 +179,7 @@ class TestBuildSpeakerTimelines:
 
     def test_segment_without_words_falls_back_to_segment_bounds(self):
         """Segments missing a `words` key use the segment start/end as one interval."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[{"start": 1.5, "end": 3.5, "speaker": "SPEAKER_00"}],
         )
         result = _build_speaker_timelines(transcription)
@@ -190,7 +187,7 @@ class TestBuildSpeakerTimelines:
 
     def test_segment_with_only_none_word_timestamps_falls_back(self):
         """If every word has None start/end, fall back to segment bounds."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 1.0,
@@ -208,7 +205,7 @@ class TestBuildSpeakerTimelines:
 
     def test_short_words_only_uses_segment_start_and_last_word_end(self):
         """With no overly long words, the interval runs segment start to end."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 1.0,
@@ -231,7 +228,7 @@ class TestBuildSpeakerTimelines:
         max_word_duration = (
             user_assign.settings.resolve_speaker_identities_max_word_duration
         )
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 0.0,
@@ -252,7 +249,7 @@ class TestBuildSpeakerTimelines:
 
     def test_long_word_in_middle_splits_segment(self):
         """Short words around a long word produce two intervals (before-cap + after)."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 0.0,
@@ -275,7 +272,7 @@ class TestBuildSpeakerTimelines:
 
     def test_tail_word_is_capped_at_max_duration(self):
         """The trailing word's end is capped at word.start + max_word_duration."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 0.0,
@@ -302,7 +299,7 @@ class TestBuildSpeakerTimelines:
                 update={"resolve_speaker_identities_enable_split_on_words": False},
             ),
         )
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 0.0,
@@ -322,7 +319,7 @@ class TestBuildSpeakerTimelines:
 
     def test_multiple_speakers_keep_separate_timelines(self):
         """Segments from different speakers populate independent timeline entries."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {
                     "start": 0.0,
@@ -402,7 +399,7 @@ class TestResolveSpeakerIdentities:
             ],
             "participants": [{"participantId": "user-a", "name": "Shared Mic"}],
         }
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 1.0, "end": 3.0, "speaker": "SPEAKER_00"},
                 {"start": 5.0, "end": 7.0, "speaker": "SPEAKER_01"},
@@ -445,7 +442,7 @@ class TestResolveSpeakerIdentities:
                 {"participantId": "user-b", "name": "Bob"},
             ],
         }
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 1.5, "end": 3.5, "speaker": "SPEAKER_00"},
                 {"start": 5.5, "end": 7.5, "speaker": "SPEAKER_01"},
@@ -493,7 +490,7 @@ class TestResolveSpeakerIdentities:
                 {"participantId": "user-b", "name": "Bob"},
             ],
         }
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 1.5, "end": 5.5, "speaker": "SPEAKER_00"},
                 {"start": 4.0, "end": 7.5, "speaker": "SPEAKER_01"},
@@ -529,7 +526,7 @@ class TestResolveSpeakerIdentities:
             ],
             "participants": [{"participantId": "user-a", "name": "Brief User"}],
         }
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 1.0, "end": 10.0, "speaker": "SPEAKER_00"},
             ],
@@ -561,7 +558,7 @@ class TestResolveSpeakerIdentities:
             ],
             "participants": [{"participantId": "user-a", "name": "Early User"}],
         }
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 0.0, "end": 3.0, "speaker": "SPEAKER_00"},
             ],
@@ -576,7 +573,7 @@ class TestResolveSpeakerIdentities:
         """No segments produces empty result."""
         result = resolve_speaker_identities(
             METADATA_SINGLE_USER,
-            FakeTranscription(segments=[]),
+            make_transcription(segments=[]),
             RECORDING_START,
             RECORDING_END,
         )
@@ -584,7 +581,7 @@ class TestResolveSpeakerIdentities:
 
     def test_segment_without_speaker_ignored(self):
         """Segments missing speaker key are skipped."""
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 1.0, "end": 3.0, "text": "no speaker"},
             ],
@@ -608,7 +605,7 @@ class TestResolveSpeakerIdentities:
             ],
             "participants": [{"participantId": "user-a", "name": "Still Talking"}],
         }
-        transcription = FakeTranscription(
+        transcription = make_transcription(
             segments=[
                 {"start": 2.0, "end": 9.0, "speaker": "SPEAKER_00"},
             ],