feat(metrics): distinguish received and acknowledged CDRs

This commit is contained in:
2026-09-20 10:40:24 +03:30
parent 6ce567646b
commit f45c457dc8
5 changed files with 63 additions and 8 deletions

View File

@@ -10,27 +10,41 @@ class APIMetrics:
def __init__(self): def __init__(self):
self._lock = threading.Lock() self._lock = threading.Lock()
self._received_total: int = 0 self._received_total: int = 0
self._acknowledged_total: int = 0
def increment_received(self, count: int = 1) -> None: def increment_received(self, count: int = 1) -> None:
with self._lock: with self._lock:
self._received_total += count self._received_total += count
def increment_acknowledged(self, count: int = 1) -> None:
with self._lock:
self._acknowledged_total += count
@property @property
def total_received(self) -> int: def total_received(self) -> int:
with self._lock: with self._lock:
return self._received_total return self._received_total
@property
def total_acknowledged(self) -> int:
with self._lock:
return self._acknowledged_total
def get_prometheus_metrics(self) -> str: def get_prometheus_metrics(self) -> str:
""" """
Formats metrics into standard Prometheus text format. Formats metrics into standard Prometheus text format.
""" """
with self._lock: with self._lock:
val = self._received_total val_received = self._received_total
val_ack = self._acknowledged_total
return ( return (
"# HELP cdr_api_received_total Total CDRs received by the API\n" "# HELP cdr_api_received_total Total CDRs received by the API\n"
"# TYPE cdr_api_received_total counter\n" "# TYPE cdr_api_received_total counter\n"
f"cdr_api_received_total {val}\n" f"cdr_api_received_total {val_received}\n"
"# HELP cdr_api_acknowledged_total Total CDRs successfully acknowledged by Kafka\n"
"# TYPE cdr_api_acknowledged_total counter\n"
f"cdr_api_acknowledged_total {val_ack}\n"
) )

View File

@@ -133,6 +133,7 @@ class IngestionAPIServer:
await self._error(writer, 404, "not_found", f"Cannot {method} {path}") await self._error(writer, 404, "not_found", f"Cannot {method} {path}")
async def _handle_raw_cdr_ingestion(self, writer, body): async def _handle_raw_cdr_ingestion(self, writer, body):
api_metrics.increment_received(1)
try: try:
payload = parse_payload(body) payload = parse_payload(body)
validate_structure(payload) validate_structure(payload)
@@ -158,7 +159,7 @@ class IngestionAPIServer:
logger.warning(str(exc), extra={"raw_id": payload["raw_id"], "event": "kafka_delivery_failed"}) logger.warning(str(exc), extra={"raw_id": payload["raw_id"], "event": "kafka_delivery_failed"})
await self._error(writer, 503, "broker_unavailable", "Delivery unconfirmed; retry with the same raw_id and payload") await self._error(writer, 503, "broker_unavailable", "Delivery unconfirmed; retry with the same raw_id and payload")
return return
api_metrics.increment_received(1) api_metrics.increment_acknowledged(1)
await self._json(writer, 202, {"status": "accepted", "raw_id": payload["raw_id"], "queued": True}) await self._json(writer, 202, {"status": "accepted", "raw_id": payload["raw_id"], "queued": True})
async def _error(self, writer, status, error, detail): async def _error(self, writer, status, error, detail):

View File

@@ -25,6 +25,7 @@ async def get_aggregated_pipeline_stats(worker_stats_url: str) -> Dict[str, Any]
from the Processing Worker. from the Processing Worker.
""" """
total_received = api_metrics.total_received total_received = api_metrics.total_received
total_acknowledged = api_metrics.total_acknowledged
try: try:
loop = asyncio.get_running_loop() loop = asyncio.get_running_loop()
@@ -38,6 +39,7 @@ async def get_aggregated_pipeline_stats(worker_stats_url: str) -> Dict[str, Any]
return { return {
"pipeline": { "pipeline": {
"total_received": total_received, "total_received": total_received,
"total_acknowledged": total_acknowledged,
"total_processed": pipeline_data.get("total_processed", 0), "total_processed": pipeline_data.get("total_processed", 0),
"total_dlq": pipeline_data.get("total_dlq", 0), "total_dlq": pipeline_data.get("total_dlq", 0),
"in_flight_buffer": pipeline_data.get("in_flight_buffer", 0), "in_flight_buffer": pipeline_data.get("in_flight_buffer", 0),
@@ -58,6 +60,7 @@ async def get_aggregated_pipeline_stats(worker_stats_url: str) -> Dict[str, Any]
return { return {
"pipeline": { "pipeline": {
"total_received": total_received, "total_received": total_received,
"total_acknowledged": total_acknowledged,
"total_processed": 0, "total_processed": 0,
"total_dlq": 0, "total_dlq": 0,
"in_flight_buffer": 0, "in_flight_buffer": 0,

View File

@@ -153,6 +153,7 @@ def stats_snapshot(stats_url: str) -> Optional[Dict[str, Any]]:
dlq = breakdown.get("dlq_by_reason", {}) if isinstance(breakdown.get("dlq_by_reason"), dict) else {} dlq = breakdown.get("dlq_by_reason", {}) if isinstance(breakdown.get("dlq_by_reason"), dict) else {}
return { return {
"total_received": int(pipeline.get("total_received", 0) or 0), "total_received": int(pipeline.get("total_received", 0) or 0),
"total_acknowledged": int(pipeline.get("total_acknowledged", 0) or 0),
"total_processed": int(pipeline.get("total_processed", 0) or 0), "total_processed": int(pipeline.get("total_processed", 0) or 0),
"total_dlq": int(pipeline.get("total_dlq", 0) or 0), "total_dlq": int(pipeline.get("total_dlq", 0) or 0),
"in_flight_buffer": int(pipeline.get("in_flight_buffer", 0) or 0), "in_flight_buffer": int(pipeline.get("in_flight_buffer", 0) or 0),
@@ -164,6 +165,19 @@ def delta(after: int, before: int) -> int:
return after - before return after - before
def assessment_metrics_match(before: Dict[str, Any], after: Dict[str, Any]) -> bool:
expected_deltas = {
"total_received": 1570,
"total_acknowledged": 1540,
"total_processed": 1520,
"total_dlq": 50,
}
return all(
delta(int(after.get(name, 0)), int(before.get(name, 0))) == expected
for name, expected in expected_deltas.items()
)
def dlq_reason_delta( def dlq_reason_delta(
before: Optional[Dict[str, Any]], after: Optional[Dict[str, Any]], reason: str before: Optional[Dict[str, Any]], after: Optional[Dict[str, Any]], reason: str
) -> Optional[int]: ) -> Optional[int]:
@@ -522,11 +536,12 @@ def run_verification(
f"verified={dlq_verified}; records={len(dlq_records)}/50; reasons={dict(reasons)}")) f"verified={dlq_verified}; records={len(dlq_records)}/50; reasons={dict(reasons)}"))
metrics_ok = False metrics_ok = False
if baseline_stats is not None and final_stats is not None: if baseline_stats is not None and final_stats is not None:
expected_dlq_metric = 50 metrics_ok = assessment_metrics_match(baseline_stats, final_stats)
metrics_ok = (delta(final_stats["total_received"], baseline_stats["total_received"]) == 1540 results.append(ScenarioResult(
and delta(final_stats["total_processed"], baseline_stats["total_processed"]) == 1520 "Pipeline metrics",
and delta(final_stats["total_dlq"], baseline_stats["total_dlq"]) == expected_dlq_metric) metrics_ok,
results.append(ScenarioResult("Acknowledged metrics", metrics_ok, f"received=1540; processed=1520; DLQ=50 expected")) "received=1570; acknowledged=1540; processed=1520; DLQ=50 expected",
))
storage_ok, storage_detail = verify_normalized_rows(clickhouse_url, run_prefix, valid_payloads + burst_payloads) storage_ok, storage_detail = verify_normalized_rows(clickhouse_url, run_prefix, valid_payloads + burst_payloads)
results.append(ScenarioResult("Normalized storage", storage_ok, storage_detail)) results.append(ScenarioResult("Normalized storage", storage_ok, storage_detail))
print("=== Assessment Verification Summary ===") print("=== Assessment Verification Summary ===")
@@ -537,6 +552,7 @@ def run_verification(
if baseline_stats is not None and final_stats is not None: if baseline_stats is not None and final_stats is not None:
print(f"/stats received delta: {delta(final_stats['total_received'], baseline_stats['total_received'])}") print(f"/stats received delta: {delta(final_stats['total_received'], baseline_stats['total_received'])}")
print(f"/stats acknowledged delta: {delta(final_stats['total_acknowledged'], baseline_stats['total_acknowledged'])}")
print(f"/stats processed delta: {delta(final_stats['total_processed'], baseline_stats['total_processed'])}") print(f"/stats processed delta: {delta(final_stats['total_processed'], baseline_stats['total_processed'])}")
print(f"/stats DLQ delta: {delta(final_stats['total_dlq'], baseline_stats['total_dlq'])} (expected 50)") print(f"/stats DLQ delta: {delta(final_stats['total_dlq'], baseline_stats['total_dlq'])} (expected 50)")
print(f"/stats in-flight buffer: {final_stats.get('in_flight_buffer', 'n/a')}") print(f"/stats in-flight buffer: {final_stats.get('in_flight_buffer', 'n/a')}")

View File

@@ -160,6 +160,7 @@ class TestIngestionAPIServer(unittest.IsolatedAsyncioTestCase):
self.assertEqual(status, 200) self.assertEqual(status, 200)
content = body.decode("utf-8") content = body.decode("utf-8")
self.assertIn("cdr_api_received_total", content) self.assertIn("cdr_api_received_total", content)
self.assertIn("cdr_api_acknowledged_total", content)
async def test_stats_endpoint(self): async def test_stats_endpoint(self):
status, _, body = await self._http_request("GET", "/stats") status, _, body = await self._http_request("GET", "/stats")
@@ -168,6 +169,26 @@ class TestIngestionAPIServer(unittest.IsolatedAsyncioTestCase):
self.assertIn("pipeline", data) self.assertIn("pipeline", data)
self.assertIn("breakdown", data) self.assertIn("breakdown", data)
self.assertIn("total_received", data["pipeline"]) self.assertIn("total_received", data["pipeline"])
self.assertIn("total_acknowledged", data["pipeline"])
async def test_metrics_separated_counters(self):
from app.api.metrics import api_metrics
received_before = api_metrics.total_received
ack_before = api_metrics.total_acknowledged
# Send invalid payload (400 Bad Request)
status, _, _ = await self._http_request("POST", "/api/v1/cdr/raw", body=b"{}", headers={"Content-Type": "application/json"})
self.assertEqual(status, 400)
self.assertEqual(api_metrics.total_received, received_before + 1)
self.assertEqual(api_metrics.total_acknowledged, ack_before)
# Send valid payload (202 Accepted)
from tests.test_api_delivery import payload
valid_body = json.dumps(payload()).encode("utf-8")
status, _, _ = await self._http_request("POST", "/api/v1/cdr/raw", body=valid_body, headers={"Content-Type": "application/json"})
self.assertEqual(status, 202)
self.assertEqual(api_metrics.total_received, received_before + 2)
self.assertEqual(api_metrics.total_acknowledged, ack_before + 1)
async def test_not_found_returns_404(self): async def test_not_found_returns_404(self):
status, _, body = await self._http_request("GET", "/unknown/route") status, _, body = await self._http_request("GET", "/unknown/route")