fix(telemetry): restore worker agent delivery and truthful system status
This commit is contained in:
@@ -38,6 +38,8 @@ export function useWorkerTelemetry(
|
||||
setError(null);
|
||||
return;
|
||||
}
|
||||
setTelemetry(null);
|
||||
setError(null);
|
||||
const controller = new AbortController();
|
||||
const stopPolling = startSequentialPolling(async () => {
|
||||
setLoading(true);
|
||||
@@ -48,6 +50,7 @@ export function useWorkerTelemetry(
|
||||
setError(null);
|
||||
} catch (reason: unknown) {
|
||||
if (controller.signal.aborted) return;
|
||||
setTelemetry(null);
|
||||
setError(
|
||||
reason instanceof Error
|
||||
? reason.message
|
||||
|
||||
@@ -114,6 +114,9 @@ export interface WorkerTelemetry {
|
||||
power_watts?: number;
|
||||
temperature_celsius?: number;
|
||||
memory_used_percent?: number;
|
||||
sm_clock_mhz?: number | null;
|
||||
memory_clock_mhz?: number | null;
|
||||
driver_version?: string | null;
|
||||
} | null;
|
||||
triton: {
|
||||
ready: boolean;
|
||||
@@ -131,6 +134,13 @@ export interface WorkerTelemetry {
|
||||
completed_runs: number | null;
|
||||
failed_runs: number | null;
|
||||
model_load_seconds: number | null;
|
||||
profile_name?: string | null;
|
||||
input_state?: string | null;
|
||||
wait_reason?: string | null;
|
||||
live_children?: number | null;
|
||||
input_pauses?: number | null;
|
||||
pending_bundles?: number | null;
|
||||
buffer_bytes?: number | null;
|
||||
stages: WorkerPipelineStage[];
|
||||
};
|
||||
network: {
|
||||
@@ -189,7 +199,9 @@ export async function fetchWorkerTelemetry(
|
||||
`/api/v1/system/contours/${encodeURIComponent(contourId)}/telemetry?history=90`,
|
||||
{
|
||||
method: "GET",
|
||||
signal,
|
||||
signal: signal
|
||||
? AbortSignal.any([signal, AbortSignal.timeout(5_000)])
|
||||
: AbortSignal.timeout(5_000),
|
||||
},
|
||||
),
|
||||
"missioncore.worker-telemetry/v1",
|
||||
|
||||
Reference in New Issue
Block a user