diff --git a/src/service_api/main.py b/src/service_api/main.py
index 78bc407bdb82b9e581ca56b16bafe0b9c763f634..65099caa674bc698e3575cb314c11263cc918d07 100644
--- a/src/service_api/main.py
+++ b/src/service_api/main.py
@@ -15,6 +15,7 @@ from src.structures import TrafficStats
logging.config.dictConfig(LOGGING_CONFIG)
logger = logging.getLogger(__name__)
+logger.setLevel(logging.DEBUG)
T = TypeVar("T")
@@ -41,24 +42,25 @@ async def _query_instant(
return None, None
except (KeyError, ValueError) as parse_error:
- logger.error("Error parsing response data:", exc_info=True)
+ logger.error("Error parsing response data:", exc_info=parse_error)
raise HTTPException(
status_code=status.HTTP_502_BAD_GATEWAY, detail=f"Error parsing Prometheus data: {parse_error}"
- ) from parse_error
+ ) from None
except httpx.RequestError as request_error:
- logger.error("HTTP request error:", exc_info=True)
+ logger.error("HTTP request error:", exc_info=request_error)
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail=f"HTTP request error: {request_error}"
- ) from request_error
+ ) from None
except httpx.HTTPStatusError as status_error:
logger.error(
- f"HTTP status error: {status_error.response.status_code} - {status_error.response.text}", exc_info=True
+ f"HTTP status error: {status_error.response.status_code} - {status_error.response.text}",
+ exc_info=status_error,
)
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail=f"HTTP status error: {status_error}"
- ) from status_error
+ ) from None
async def _query_series(client: httpx.AsyncClient, cfg: PromConfig, query: str) -> list[dict[str, Any]]:
@@ -73,24 +75,25 @@ async def _query_series(client: httpx.AsyncClient, cfg: PromConfig, query: str)
return []
except (AttributeError, KeyError, ValueError) as parse_error:
- logger.error("Error parsing series response data:", exc_info=True)
+ logger.error("Error parsing series response data:", exc_info=parse_error)
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, detail="Error parsing Prometheus series data."
- ) from parse_error
+ ) from None
except httpx.RequestError as request_error:
- logger.error(f"HTTP request error: {request_error}", exc_info=True)
+ logger.error(f"HTTP request error: {request_error}", exc_info=request_error)
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail=f"HTTP request error: {request_error}"
- ) from request_error
+ ) from None
except httpx.HTTPStatusError as status_error:
logger.error(
- f"HTTP status error: {status_error.response.status_code} - {status_error.response.text}", exc_info=True
+ f"HTTP status error: {status_error.response.status_code} - {status_error.response.text}",
+ exc_info=status_error,
)
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail=f"HTTP status error: {status_error}"
- ) from status_error
+ ) from None
async def _get_pod_inbound_traffic_bps(
@@ -149,9 +152,9 @@ async def _get_pod_traffic_per_link_bytes_per_second(
return traffic_bytes
- except Exception as default_error:
- logger.error("Error parsing response data:", exc_info=True)
- raise HTTPException(status_code=500, detail="Error parsing data.") from default_error
+ except Exception as ex:
+ logger.error("Error parsing response data:", exc_info=ex)
+ raise HTTPException(status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, detail="Error parsing data.") from None
async def _get_pod_inbound_traffic_rate(
@@ -198,12 +201,12 @@ async def _get_outbound_traffic_rate_by_status_code(
return traffic_res_rate_by_code
- except (AttributeError, KeyError, ValueError) as parse_error:
- logger.error(f"Error parsing series response data: {parse_error}")
+ except (AttributeError, KeyError, ValueError) as ex:
+ logger.error("Error parsing series response data:", exc_info=ex)
raise HTTPException(
status_code=status.HTTP_503_SERVICE_UNAVAILABLE,
- detail=f"Error parsing Prometheus series data: {parse_error}",
- ) from parse_error
+ detail="Error parsing Prometheus series data.",
+ ) from None
async def _get_pod_outbound_traffic_latency(
@@ -228,10 +231,10 @@ async def _get_pod_outbound_traffic_latency(
return results
except Exception as ex:
- logger.error(f"Request error while fetching pod traffic response stats:\n{ex}")
+ logger.error("Request error while fetching pod traffic response stats.", exc_info=ex)
raise HTTPException(
- status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail=f"Request error while fetching data: {ex}"
- ) from ex
+ status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail="Request error while fetching data"
+ ) from None
app = FastAPI()
@@ -290,14 +293,14 @@ async def get_workloads_list():
return workloads
-async def get_pod_by_name_or_uid(client: httpx.AsyncClient, pod_name_or_uid: str, cfg: PromConfig) -> dict:
+async def get_pod_by_name_or_uid(client: httpx.AsyncClient, pod_name_or_uid: str, cfg: PromConfig) -> dict[str, Any]:
try:
uid = UUID(pod_name_or_uid)
query = f'kube_pod_labels{{uid="{uid}"}}'
pod, _ = await _query_instant(client, cfg, query)
if pod is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="Unable to obtain pod info.")
return pod
@@ -310,7 +313,7 @@ async def get_pod_by_name_or_uid(client: httpx.AsyncClient, pod_name_or_uid: str
pod, _ = await _query_instant(client, cfg, query)
if pod is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="Unable to obtain pod info.")
return pod
@@ -332,37 +335,42 @@ async def get_pod_details(pod_name_or_uid: str) -> Optional[PodInfoExtended]:
_, value = await _query_instant(client, cfg, query)
return int(value)
- async def get_num_replicas(client, pod):
+ async def get_num_replicas(client, pod) -> Optional[int]:
# Step 1: Get the ReplicaSet owner of the pod
query = f'kube_pod_owner{{namespace="{pod["namespace"]}", pod="{pod["pod"]}"}}'
metric, _ = await _query_instant(client, cfg, query)
if metric is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ # raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="Unable to obtain pod owner")
+ return None
replicaset_name = metric.get("owner_name")
if replicaset_name is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ # raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="Unable to obtain pod owner")
+ return None
# Step 2: Get the Deployment owner of the ReplicaSet
query = f'kube_replicaset_owner{{namespace="{pod["namespace"]}", replicaset="{replicaset_name}"}}'
metric, _ = await _query_instant(client, cfg, query)
if metric is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ # raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="Unable to obtain replica owner")
+ return None
deployment_name = metric.get("owner_name")
if deployment_name is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ # raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="Unable to obtain replica owner")
+ return None
# Step 3: Figure out how many deployments are there
query = f'kube_deployment_spec_replicas{{namespace="{pod["namespace"]}", deployment="{deployment_name}"}}'
_, value = await _query_instant(client, cfg, query)
if value is None:
- raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ # raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY)
+ return None
return int(value)
@@ -370,8 +378,21 @@ async def get_pod_details(pod_name_or_uid: str) -> Optional[PodInfoExtended]:
pod = await get_pod_by_name_or_uid(client, pod_name_or_uid, cfg)
uid = pod.get("uid")
+
+ if uid is None:
+ raise HTTPException(
+ status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+ detail=f"Invalid UID '{uid}'",
+ )
+
pod_name = pod.get("pod")
+ if pod_name is None:
+ raise HTTPException(
+ status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+ detail=f"Could not obtain pod name '{uid}' -> '{pod_name}'",
+ )
+
# obtain last valid state
pod["current_state"] = get_current_state(client, uid)
@@ -407,4 +428,7 @@ async def get_traffic_stats(pod_name_or_uid: str):
pod = await get_pod_by_name_or_uid(client, pod_name_or_uid, cfg)
results = await _get_pod_traffic_per_link_bytes_per_second(client, pod["pod"], timestamp, cfg)
+ if not results:
+ raise HTTPException(status_code=status.HTTP_503_SERVICE_UNAVAILABLE, detail="Traffic data not available.")
+
return results