fix: harden sub2api codex failover stability

config/platform-infra/sub2api-codex-pool.yaml

@@ -161,7 +161,7 @@ sentinel:
   monitor:
     enabled: true
   actions:
-    enabled: false
+    enabled: true
   schedule: "*/1 * * * *"
   image: python:3.12-alpine
   sdk:

scripts/src/platform-infra-sub2api-codex.ts

@@ -2008,6 +2008,8 @@ function compactGatewayResponsesRecent(block: unknown): unknown {
     slowFinalErrorCount: block.slowFinalErrorCount,
     contextCanceledCount: block.contextCanceledCount,
     ignoredProbeNoiseCount: block.ignoredProbeNoiseCount,
+    failoverBudgetExhaustedCount: Array.isArray(block.failoverBudgetExhausted) ? block.failoverBudgetExhausted.length : undefined,
+    failoverBudgetExhausted: Array.isArray(block.failoverBudgetExhausted) ? block.failoverBudgetExhausted.slice(-3).reverse() : undefined,
     recentFailovers: Array.isArray(block.recentFailovers) ? block.recentFailovers.slice(-4).reverse() : undefined,
     recentForwardFailures: Array.isArray(block.recentForwardFailures) ? block.recentForwardFailures.slice(-4).reverse().map((item) => ({
       ...item,
@@ -5134,6 +5136,51 @@ def ignored_probe_noise(items, section):
             result.append(probe)
     return result
 
+def group_by_request_id(items):
+    grouped = {}
+    for item in items:
+        request_id = item.get("requestId")
+        if not isinstance(request_id, str) or not request_id:
+            continue
+        grouped.setdefault(request_id, []).append(item)
+    return grouped
+
+def failover_budget_exhausted_evidence(failovers, final_errors):
+    final_by_request = {}
+    for item in final_errors:
+        request_id = item.get("requestId")
+        if isinstance(request_id, str) and request_id:
+            final_by_request[request_id] = item
+    exhausted = []
+    for request_id, request_failovers in group_by_request_id(failovers).items():
+        final = final_by_request.get(request_id)
+        if not final:
+            continue
+        last = request_failovers[-1]
+        switch_count = last.get("switchCount")
+        max_switches = last.get("maxSwitches")
+        final_status = final.get("statusCode")
+        if (
+            isinstance(switch_count, int)
+            and isinstance(max_switches, int)
+            and max_switches > 0
+            and switch_count >= max_switches
+            and isinstance(final_status, int)
+            and final_status >= 500
+        ):
+            exhausted.append({
+                "requestId": request_id,
+                "clientRequestId": final.get("clientRequestId") or last.get("clientRequestId"),
+                "path": final.get("path") or last.get("path"),
+                "finalAccountId": final.get("accountId"),
+                "finalStatusCode": final_status,
+                "switchCount": switch_count,
+                "maxSwitches": max_switches,
+                "lastFailoverAccountId": last.get("accountId"),
+                "lastUpstreamStatus": last.get("upstreamStatus"),
+            })
+    return exhausted
+
 def recent_responses_gateway_evidence():
     proc = kubectl(["-n", NAMESPACE, "logs", "deployment/sub2api", "--since=6h", "--tail=2500"])
     stdout = proc.stdout.decode("utf-8", errors="replace")
@@ -5189,6 +5236,7 @@ def recent_responses_gateway_evidence():
     visible_final_errors = filter_ignored_probe_noise(final_errors)
     visible_slow_final_errors = filter_ignored_probe_noise(slow_final_errors)
     visible_context_canceled = filter_ignored_probe_noise(context_canceled)
+    failover_budget_exhausted = failover_budget_exhausted_evidence(visible_failovers, visible_final_errors)
     probe_noise = (
         ignored_probe_noise(failovers, "failovers")
         + ignored_probe_noise(forward_failures, "forwardFailures")
@@ -5207,6 +5255,7 @@ def recent_responses_gateway_evidence():
         "slowFinalErrorCount": len(visible_slow_final_errors),
         "contextCanceledCount": len(visible_context_canceled),
         "ignoredProbeNoiseCount": len(probe_noise),
+        "failoverBudgetExhausted": failover_budget_exhausted[-8:],
         "rawCounts": {
             "failoverCount": len(failovers),
             "forwardFailureCount": len(forward_failures),
@@ -5990,6 +6039,8 @@ def trace_reason(events, final_event):
     failovers = [item for item in events if item.get("type") == "failover"]
     select_failures = [item for item in events if item.get("type") == "select-failed"]
     upstream_errors = [item for item in events if item.get("type") == "upstream-error"]
+    if failover_budget_exhausted(failovers, final_event):
+        return "failover-budget-exhausted"
     if failovers and select_failures:
         return "failover-attempted-no-candidate"
     if failovers:
@@ -6004,6 +6055,47 @@ def trace_reason(events, final_event):
         return "completed"
     return "unknown"
 
+def failover_budget_exhausted(failovers, final_event):
+    if not failovers or not isinstance(final_event, dict):
+        return False
+    last = failovers[-1]
+    switch_count = last.get("switchCount")
+    max_switches = last.get("maxSwitches")
+    final_status = final_event.get("statusCode")
+    return (
+        isinstance(switch_count, int)
+        and isinstance(max_switches, int)
+        and max_switches > 0
+        and switch_count >= max_switches
+        and isinstance(final_status, int)
+        and final_status >= 500
+    )
+
+def trace_untried_schedulable_accounts(failovers, final_event, account_snapshot):
+    if not failover_budget_exhausted(failovers, final_event):
+        return []
+    tried = set()
+    for item in failovers:
+        account_id = item.get("accountId")
+        if isinstance(account_id, int):
+            tried.add(account_id)
+    final_account = final_event.get("accountId") if isinstance(final_event, dict) else None
+    if isinstance(final_account, int):
+        tried.add(final_account)
+    result = []
+    for item in account_snapshot:
+        account_id = item.get("accountId")
+        if not isinstance(account_id, int) or account_id in tried:
+            continue
+        if item.get("schedulable") is True and item.get("status") == "active" and item.get("tempUnschedulableSet") is not True:
+            result.append({
+                "accountId": account_id,
+                "accountName": item.get("accountName"),
+                "priority": item.get("priority"),
+                "concurrency": item.get("concurrency"),
+            })
+    return result
+
 def run_trace():
     payload = json.loads(base64.b64decode(PAYLOAD_B64).decode("utf-8")) if PAYLOAD_B64 else {}
     request_id = payload.get("requestId")
@@ -6063,6 +6155,7 @@ def run_trace():
     final_errors = [item for item in window_events if item.get("type") == "final" and isinstance(item.get("statusCode"), int) and item.get("statusCode") >= 400]
     window_failovers = [item for item in window_events if item.get("type") == "failover"]
     window_select_failures = [item for item in window_events if item.get("type") == "select-failed"]
+    untried_schedulable_accounts = trace_untried_schedulable_accounts(failovers, final_event or {}, account_snapshot)
     reason = trace_reason(events, final_event)
     if not matched:
         outcome = "not-found"
@@ -6112,6 +6205,10 @@ def run_trace():
             "tempUnschedulableCount": len(temp_unsched),
             "adminSchedulableCount": len(admin_sched),
         },
+        "diagnostics": {
+            "failoverBudgetExhausted": failover_budget_exhausted(failovers, final_event or {}),
+            "untriedSchedulableAccounts": untried_schedulable_accounts,
+        },
         "accountSnapshot": account_snapshot,
         "accountSnapshotError": account_snapshot_error,
         "rawLines": [{"line": item.get("_line")} for item in matched[-30:]] if show_lines else [],