refactor(enterprise): Remove custom Prometheus metrics app (#12253)

Co-authored-by: openhands <openhands@all-hands.dev>

enterprise/poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 2.1.3 and should not be changed by hand.
+# This file is automatically @generated by Poetry 2.2.1 and should not be changed by hand.
 
 [[package]]
 name = "aiofiles"
@@ -5860,7 +5860,7 @@ wsproto = ">=1.2.0"
 
 [[package]]
 name = "openhands-ai"
-version = "0.0.0-post.5803+a8098505c"
+version = "1.1.0"
 description = "OpenHands: Code Less, Make More"
 optional = false
 python-versions = "^3.12,<3.14"
@@ -6858,22 +6858,6 @@ files = [
 [package.extras]
 twisted = ["twisted"]
 
-[[package]]
-name = "prometheus-fastapi-instrumentator"
-version = "7.1.0"
-description = "Instrument your FastAPI app with Prometheus metrics"
-optional = false
-python-versions = ">=3.8"
-groups = ["main"]
-files = [
-    {file = "prometheus_fastapi_instrumentator-7.1.0-py3-none-any.whl", hash = "sha256:978130f3c0bb7b8ebcc90d35516a6fe13e02d2eb358c8f83887cdef7020c31e9"},
-    {file = "prometheus_fastapi_instrumentator-7.1.0.tar.gz", hash = "sha256:be7cd61eeea4e5912aeccb4261c6631b3f227d8924542d79eaf5af3f439cbe5e"},
-]
-
-[package.dependencies]
-prometheus-client = ">=0.8.0,<1.0.0"
-starlette = ">=0.30.0,<1.0.0"
-
 [[package]]
 name = "prompt-toolkit"
 version = "3.0.52"
@@ -14508,4 +14492,4 @@ cffi = ["cffi (>=1.17) ; python_version >= \"3.13\" and platform_python_implemen
 [metadata]
 lock-version = "2.1"
 python-versions = "^3.12,<3.14"
-content-hash = "fac67a8991a3e2c840a23702dc90f99e98d381f3537ad50b4c4739cdbde941ca"
+content-hash = "ab703edc73639f22f498894d16bf7170fe3ab9c2697761cdd494587caee77973"

enterprise/pyproject.toml

@@ -29,7 +29,6 @@ cloud-sql-python-connector = "^1.16.0"
 psycopg2-binary = "^2.9.10"
 pg8000 = "^1.31.2"
 stripe = "^11.5.0"
-prometheus-fastapi-instrumentator = "^7.0.2"
 python-json-logger = "^3.2.1"
 python-keycloak = "^5.3.1"
 asyncpg = "^0.30.0"

enterprise/saas_server.py

@@ -18,7 +18,6 @@ from server.auth.constants import (  # noqa: E402
 )
 from server.constants import PERMITTED_CORS_ORIGINS  # noqa: E402
 from server.logger import logger  # noqa: E402
-from server.metrics import metrics_app  # noqa: E402
 from server.middleware import SetAuthCookieMiddleware  # noqa: E402
 from server.rate_limit import setup_rate_limit_handler  # noqa: E402
 from server.routes.api_keys import api_router as api_keys_router  # noqa: E402
@@ -61,9 +60,6 @@ def is_saas():
     return {'saas': True}
 
 
-# This requires a trailing slash to access, like /api/metrics/
-base_app.mount('/internal/metrics', metrics_app())
-
 base_app.include_router(readiness_router)  # Add routes for readiness checks
 base_app.include_router(api_router)  # Add additional route for github auth
 base_app.include_router(oauth_router)  # Add additional route for oauth callback

enterprise/server/metrics.py

@@ -1,43 +0,0 @@
-from typing import Callable
-
-from prometheus_client import Gauge, make_asgi_app
-from server.clustered_conversation_manager import ClusteredConversationManager
-
-from openhands.server.shared import (
-    conversation_manager,
-)
-
-RUNNING_AGENT_LOOPS_GAUGE = Gauge(
-    'saas_running_agent_loops',
-    'Count of running agent loops, aggregate by session_id to dedupe',
-    ['session_id'],
-)
-
-
-async def _update_metrics():
-    """Update any prometheus metrics that are not updated during normal operation."""
-    if isinstance(conversation_manager, ClusteredConversationManager):
-        running_agent_loops = (
-            await conversation_manager.get_running_agent_loops_locally()
-        )
-        # Clear so we don't keep counting old sessions.
-        # This is theoretically a race condition but this is scraped on a regular interval.
-        RUNNING_AGENT_LOOPS_GAUGE.clear()
-        # running_agent_loops shouldn't be None, but can be.
-        if running_agent_loops is not None:
-            for sid in running_agent_loops:
-                RUNNING_AGENT_LOOPS_GAUGE.labels(session_id=sid).set(1)
-
-
-def metrics_app() -> Callable:
-    metrics_callable = make_asgi_app()
-
-    async def wrapped_handler(scope, receive, send):
-        """
-        Call _update_metrics before serving Prometheus metrics endpoint.
-        Not wrapped in a `try`, failing would make metrics endpoint unavailable.
-        """
-        await _update_metrics()
-        await metrics_callable(scope, receive, send)
-
-    return wrapped_handler

enterprise/server/saas_monitoring_listener.py

@@ -1,4 +1,3 @@
-from prometheus_client import Counter, Histogram
 from server.logger import logger
 
 from openhands.core.config.openhands_config import OpenHandsConfig
@@ -9,45 +8,27 @@ from openhands.events.observation import (
 )
 from openhands.server.monitoring import MonitoringListener
 
-AGENT_STATUS_ERROR_COUNT = Counter(
-    'saas_agent_status_errors', 'Agent Status change events to status error'
-)
-CREATE_CONVERSATION_COUNT = Counter(
-    'saas_create_conversation', 'Create conversation attempts'
-)
-AGENT_SESSION_START_HISTOGRAM = Histogram(
-    'saas_agent_session_start',
-    'AgentSession starts with success and duration',
-    labelnames=['success'],
-)
-
 
 class SaaSMonitoringListener(MonitoringListener):
-    """
-    Forward app signals to Prometheus.
-    """
+    """Forward app signals to structured logging for GCP native monitoring."""
 
     def on_session_event(self, event: Event) -> None:
-        """
-        Track metrics about events being added to a Session's EventStream.
-        """
+        """Track metrics about events being added to a Session's EventStream."""
         if (
             isinstance(event, AgentStateChangedObservation)
             and event.agent_state == AgentState.ERROR
         ):
-            AGENT_STATUS_ERROR_COUNT.inc()
             logger.info(
                 'Tracking agent status error',
                 extra={'signal': 'saas_agent_status_errors'},
             )
 
     def on_agent_session_start(self, success: bool, duration: float) -> None:
-        """
-        Track an agent session start.
+        """Track an agent session start.
+
         Success is true if startup completed without error.
         Duration is start time in seconds observed by AgentSession.
         """
-        AGENT_SESSION_START_HISTOGRAM.labels(success=success).observe(duration)
         logger.info(
             'Tracking agent session start',
             extra={
@@ -58,11 +39,10 @@ class SaaSMonitoringListener(MonitoringListener):
         )
 
     def on_create_conversation(self) -> None:
-        """
-        Track the beginning of conversation creation.
+        """Track the beginning of conversation creation.
+
         Does not currently capture whether it succeed.
         """
-        CREATE_CONVERSATION_COUNT.inc()
         logger.info(
             'Tracking create conversation', extra={'signal': 'saas_create_conversation'}
         )