fix: coordinate SMS completion and improve operations diagnostics

2026-09-16 18:27:29 +08:00
parent a0209f93bc
commit a350aca883
40 changed files with 2599 additions and 366 deletions
@@ -0,0 +1,26 @@
+type CompletionMetric =
+  'claimed' | 'not_claimed' | 'recovered' | 'fence_rejected' | 'event_committed' | 'retry_wait' | 'needs_review';
+const counts = new Map<CompletionMetric, number>();
+let snapshot: Array<{ state: string; count: number; age: number }> = [];
+const states = ['pending', 'processing', 'retry_wait', 'needs_review'];
+export function countCompletion(event: CompletionMetric) {
+  counts.set(event, (counts.get(event) ?? 0) + 1);
+}
+export function observeCompletion(rows: typeof snapshot) {
+  snapshot = rows;
+}
+export function renderCompletionMetrics() {
+  return [
+    '# TYPE cmpp_completion_events_total counter',
+    ...[...counts].map(([event, value]) => `cmpp_completion_events_total{event="${event}"} ${value}`),
+    '# TYPE cmpp_completion_work gauge',
+    '# TYPE cmpp_completion_oldest_seconds gauge',
+    ...states.flatMap((state) => {
+      const row = snapshot.find((item) => item.state === state);
+      return [
+        `cmpp_completion_work{state="${state}"} ${Number(row?.count ?? 0)}`,
+        `cmpp_completion_oldest_seconds{state="${state}"} ${Math.max(0, Number(row?.age ?? 0))}`,
+      ];
+    }),
+  ];
+}