@ddtcorex/dsh-maestro-supervisor 0.7.7 → 0.7.8

package/lib/health-poller.js

@@ -74,8 +74,21 @@ export async function pollHealth(opts = {}) {
     // (but otherwise fine) response was indistinguishable from a real crash, and
     // combined with a low down-threshold this caused a self-sustaining restart
     // loop (see Supervisor.downThreshold). 12s gives boot room without masking
-    // a genuinely dead process for long.
-    const fetchFn = opts.fetch ?? defaultFetch(opts.url ?? 'http://127.0.0.1:3080/', opts.timeoutMs ?? 12000);
+    // a genuinely dead process for long. 20s (Option A 2026-09-03) tolerates
+    // high load (loadavg 12) stalls; configurable via domains.supervisor.pollTimeoutMs.
+    // If opts.timeoutMs is not injected, read from supervisor config (maestro settings).
+    let effectiveTimeout = opts.timeoutMs;
+    if (effectiveTimeout === undefined) {
+        try {
+            const { readSupervisorConfig } = await import('./config.js');
+            const cfg = await readSupervisorConfig();
+            if (typeof cfg.pollTimeoutMs === 'number' && cfg.pollTimeoutMs > 0)
+                effectiveTimeout = cfg.pollTimeoutMs;
+        }
+        catch { }
+        effectiveTimeout ??= 20000;
+    }
+    const fetchFn = opts.fetch ?? defaultFetch(opts.url ?? 'http://127.0.0.1:3080/', effectiveTimeout);
     const psAliveFn = opts.psAlive ?? defaultPsAlive;
     const logTailFn = opts.logTail ?? defaultLogTail;
     let httpCode;

package/lib/supervisor.d.ts

@@ -84,6 +84,8 @@ export declare class Supervisor {
     private getResumeWithinMs;
     private getEffectiveIntervalMs;
     private getEffectiveDownThreshold;
+    private getEffectiveDegradedThreshold;
+    private getEffectivePollTimeoutMs;
     private findInterruptedRecent;
     private collectGitDiff;
     private attemptAutoResume;

package/lib/supervisor.js

@@ -211,7 +211,7 @@ export class Supervisor {
                 return cfg.intervalMs;
         }
         catch { }
-        return 3000;
+        return 5000;
     }
     async getEffectiveDownThreshold() {
         if (this.deps.downThreshold !== undefined)
@@ -222,8 +222,26 @@ export class Supervisor {
                 return cfg.downThreshold;
         }
         catch { }
+        return 6;
+    }
+    async getEffectiveDegradedThreshold() {
+        try {
+            const cfg = await readSupervisorConfig();
+            if (typeof cfg.degradedThreshold === 'number' && cfg.degradedThreshold > 0)
+                return cfg.degradedThreshold;
+        }
+        catch { }
         return 5;
     }
+    async getEffectivePollTimeoutMs() {
+        try {
+            const cfg = await readSupervisorConfig();
+            if (typeof cfg.pollTimeoutMs === 'number' && cfg.pollTimeoutMs > 0)
+                return cfg.pollTimeoutMs;
+        }
+        catch { }
+        return 20000;
+    }
     async findInterruptedRecent(withinMs) {
         const ms = withinMs ?? this.getResumeWithinMs();
         // Prefer injected mock for testability
@@ -383,8 +401,8 @@ export class Supervisor {
                 }
             }
             const now = this.deps.getTime ? this.deps.getTime() : Date.now();
-            // degraded needs 3 consecutive (not downThreshold which tests set to 1) to avoid flapping
-            const degradedThreshold = 3;
+            // degraded needs consecutive hits to avoid flapping — Option A 2026-09-03: 3→5 (via config)
+            const degradedThreshold = await this.getEffectiveDegradedThreshold();
             this.consecutiveDegraded++;
             if (this.consecutiveDegraded < degradedThreshold) {
                 if (now - this.lastDegradedNotify < 60000)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ddtcorex/dsh-maestro-supervisor",
-  "version": "0.7.7",
+  "version": "0.7.8",
   "description": "Supervisor daemon for DSH Web resilience — auto-detect crashes, rollback to LKG, report",
   "type": "module",
   "main": "./lib/index.js",