Squash merge 1.12.1-otel-ee into release/e-1.12.1

api/tasks/enterprise_telemetry_task.py

@@ -0,0 +1,52 @@
+"""Celery worker for enterprise metric/log telemetry events.
+
+This module defines the Celery task that processes telemetry envelopes
+from the enterprise_telemetry queue. It deserializes envelopes and
+dispatches them to the EnterpriseMetricHandler.
+"""
+
+import json
+import logging
+
+from celery import shared_task
+
+from enterprise.telemetry.contracts import TelemetryEnvelope
+from enterprise.telemetry.metric_handler import EnterpriseMetricHandler
+
+logger = logging.getLogger(__name__)
+
+
+@shared_task(queue="enterprise_telemetry")
+def process_enterprise_telemetry(envelope_json: str) -> None:
+    """Process enterprise metric/log telemetry envelope.
+
+    This task is enqueued by the TelemetryGateway for metric/log-only
+    events. It deserializes the envelope and dispatches to the handler.
+
+    Best-effort processing: logs errors but never raises, to avoid
+    failing user requests due to telemetry issues.
+
+    Args:
+        envelope_json: JSON-serialized TelemetryEnvelope.
+    """
+    try:
+        # Deserialize envelope
+        envelope_dict = json.loads(envelope_json)
+        envelope = TelemetryEnvelope.model_validate(envelope_dict)
+
+        # Process through handler
+        handler = EnterpriseMetricHandler()
+        handler.handle(envelope)
+
+        logger.debug(
+            "Successfully processed telemetry envelope: tenant_id=%s, event_id=%s, case=%s",
+            envelope.tenant_id,
+            envelope.event_id,
+            envelope.case,
+        )
+    except Exception:
+        # Best-effort: log and drop on error, never fail user request
+        logger.warning(
+            "Failed to process enterprise telemetry envelope, dropping event",
+            exc_info=True,
+        )

api/tasks/ops_trace_task.py

@@ -39,12 +39,24 @@ def process_trace_tasks(file_info):
         trace_info["documents"] = [Document.model_validate(doc) for doc in trace_info["documents"]]
 
     try:
+        trace_type = trace_info_info_map.get(trace_info_type)
+        if trace_type:
+            trace_info = trace_type(**trace_info)
+
+        from extensions.ext_enterprise_telemetry import is_enabled as is_ee_telemetry_enabled
+
+        if is_ee_telemetry_enabled():
+            from enterprise.telemetry.enterprise_trace import EnterpriseOtelTrace
+
+            try:
+                EnterpriseOtelTrace().trace(trace_info)
+            except Exception:
+                logger.exception("Enterprise trace failed for app_id: %s", app_id)
+
         if trace_instance:
             with current_app.app_context():
-                trace_type = trace_info_info_map.get(trace_info_type)
-                if trace_type:
-                    trace_info = trace_type(**trace_info)
                 trace_instance.trace(trace_info)
+
         logger.info("Processing trace tasks success, app_id: %s", app_id)
     except Exception as e:
         logger.info("error:\n\n\n%s\n\n\n\n", e)
@@ -52,4 +64,12 @@ def process_trace_tasks(file_info):
         redis_client.incr(failed_key)
         logger.info("Processing trace tasks failed, app_id: %s", app_id)
     finally:
-        storage.delete(file_path)
+        try:
+            storage.delete(file_path)
+        except Exception as e:
+            logger.warning(
+                "Failed to delete trace file %s for app_id %s: %s",
+                file_path,
+                app_id,
+                e,
+            )