⚰️(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
 

src/backend/core/utils.py

@@ -423,31 +423,6 @@ def detect_mimetype(file_buffer: bytes, filename: str | None = None) -> str:
     return mimetype_from_content or "application/octet-stream"
 
 
-def _get_s3_client(*, override_domain: bool = True):
-    """Return an S3 client, honoring the AWS_S3_DOMAIN_REPLACE endpoint override.
-
-    AWS_S3_DOMAIN_REPLACE is used when the backend and frontend reach object
-    storage under different domains (this is the case in the docker compose stack
-    used in development: the frontend connects to the object storage on localhost
-    while the backend uses the object storage service name declared in the stack).
-    The domain name is used to compute the signature, so it can't be changed
-    dynamically by the frontend; we build a dedicated boto3 client pointed at that
-    endpoint. Otherwise we reuse the default storage client.
-    """
-    if settings.AWS_S3_DOMAIN_REPLACE and override_domain:
-        return boto3.client(
-            "s3",
-            aws_access_key_id=settings.AWS_S3_ACCESS_KEY_ID,
-            aws_secret_access_key=settings.AWS_S3_SECRET_ACCESS_KEY,
-            endpoint_url=settings.AWS_S3_DOMAIN_REPLACE,
-            config=botocore.client.Config(
-                region_name=settings.AWS_S3_REGION_NAME,
-                signature_version=settings.AWS_S3_SIGNATURE_VERSION,
-            ),
-        )
-    return default_storage.connection.meta.client
-
-
 def generate_upload_policy(file):
     """
     Generate a S3 upload policy for a given file.
@@ -458,7 +433,26 @@ def generate_upload_policy(file):
 
     key = file.temporary_file_key
 
-    s3_client = _get_s3_client()
+    # This settings should be used if the backend application and the frontend application
+    # can't connect to the object storage with the same domain. This is the case in the
+    # docker compose stack used in development. The frontend application will use localhost
+    # to connect to the object storage while the backend application will use the object storage
+    # service name declared in the docker compose stack.
+    # This is needed because the domain name is used to compute the signature. So it can't be
+    # changed dynamically by the frontend application.
+    if settings.AWS_S3_DOMAIN_REPLACE:
+        s3_client = boto3.client(
+            "s3",
+            aws_access_key_id=settings.AWS_S3_ACCESS_KEY_ID,
+            aws_secret_access_key=settings.AWS_S3_SECRET_ACCESS_KEY,
+            endpoint_url=settings.AWS_S3_DOMAIN_REPLACE,
+            config=botocore.client.Config(
+                region_name=settings.AWS_S3_REGION_NAME,
+                signature_version=settings.AWS_S3_SIGNATURE_VERSION,
+            ),
+        )
+    else:
+        s3_client = default_storage.connection.meta.client
 
     # Generate the policy
     policy = s3_client.generate_presigned_url(
@@ -479,7 +473,26 @@ def generate_download_s3_url(
     if not key:
         raise ValueError("key cannot be empty")
 
-    s3_client = _get_s3_client(override_domain=override_domain)
+    # This setting should be used if the backend application and the frontend application
+    # can't connect to the object storage with the same domain. This is the case in the
+    # docker compose stack used in development. The frontend application will use localhost
+    # to connect to the object storage while the backend application will use the object storage
+    # service name declared in the docker compose stack.
+    # This is needed because the domain name is used to compute the signature. So it can't be
+    # changed dynamically by the frontend application.
+    if settings.AWS_S3_DOMAIN_REPLACE and override_domain:
+        s3_client = boto3.client(
+            "s3",
+            aws_access_key_id=settings.AWS_S3_ACCESS_KEY_ID,
+            aws_secret_access_key=settings.AWS_S3_SECRET_ACCESS_KEY,
+            endpoint_url=settings.AWS_S3_DOMAIN_REPLACE,
+            config=botocore.client.Config(
+                region_name=settings.AWS_S3_REGION_NAME,
+                signature_version=settings.AWS_S3_SIGNATURE_VERSION,
+            ),
+        )
+    else:
+        s3_client = default_storage.connection.meta.client
 
     return s3_client.generate_presigned_url(
         ClientMethod="get_object",

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"],
+        ]
+    )