@stage5/lumine 0.2.56 → 0.2.57

package/lib/sponsor.js

@@ -312,6 +312,9 @@ async function runSponsorDuty({
   process.on("SIGINT", signalHandler);
   process.on("SIGTERM", signalHandler);
 
+  const dutyHeartbeatEveryMs =
+    Math.max(10, Number(started.heartbeatEverySeconds || 30)) * 1_000;
+  let lastDutyHeartbeatSuccessAt = Date.now();
   const heartbeatDuty = async () => {
     if (dutyHeartbeatRunning) return;
     dutyHeartbeatRunning = true;
@@ -324,6 +327,7 @@ async function runSponsorDuty({
         body: { leaseToken },
       });
       dutyLeaseError = null;
+      lastDutyHeartbeatSuccessAt = Date.now();
       if (canonicalDuty.state === "stopped") stopRequested = true;
     } catch (error) {
       const status = await sponsorRequest({
@@ -340,15 +344,24 @@ async function runSponsorDuty({
         stopRequested = true;
         return;
       }
-      dutyLeaseError = error;
-      stopRequested = true;
+      // Same lease policy as the per-job heartbeat: end the duty only when the
+      // server rejects the lease or the lease window has fully elapsed. A
+      // transient network failure retries on the next tick.
+      const terminalDutyError = !isRetryableSponsorDutyRequestError(error);
+      const leaseWindowElapsed =
+        Date.now() - lastDutyHeartbeatSuccessAt >=
+        Math.max(120_000, dutyHeartbeatEveryMs * 3);
+      if (terminalDutyError || leaseWindowElapsed) {
+        dutyLeaseError = error;
+        stopRequested = true;
+      }
     } finally {
       dutyHeartbeatRunning = false;
     }
   };
   const dutyHeartbeat = setInterval(
     () => void heartbeatDuty(),
-    Math.max(10, Number(started.heartbeatEverySeconds || 30)) * 1_000,
+    dutyHeartbeatEveryMs,
   );
 
   console.log(`Shared Zero/Ciel sponsor duty started with ${provider}.`);
@@ -368,13 +381,28 @@ async function runSponsorDuty({
         Number(canonicalDuty.capacity?.maxConcurrentTasks || 1),
       );
       while (!stopRequested && activeJobs.size < concurrency) {
-        const claim = await sponsorRequest({
-          options,
-          auth,
-          method: "POST",
-          path: "/jobs/claim",
-          body: { dutySessionId: dutyId, leaseToken },
-        });
+        let claim;
+        try {
+          claim = await sponsorRequest({
+            options,
+            auth,
+            method: "POST",
+            path: "/jobs/claim",
+            body: { dutySessionId: dutyId, leaseToken },
+          });
+        } catch (error) {
+          if (!isRetryableSponsorDutyRequestError(error)) {
+            dutyLeaseError = error;
+            stopRequested = true;
+            break;
+          }
+          // A failed claim must not kill the duty process; the heartbeat owns
+          // lease validity, and the outer loop polls again after a sleep.
+          console.error(
+            `Workshop claim failed; retrying next poll: ${error?.message || error}`,
+          );
+          break;
+        }
         if (!claim.job) break;
         const work = processSponsorJob({
           options,
@@ -1133,6 +1161,12 @@ async function retrySponsorTransport(operation) {
   throw lastError;
 }
 
+function isRetryableSponsorDutyRequestError(error) {
+  const status = Number(error?.status || 0);
+  if (!status) return true;
+  return status === 408 || status === 425 || status === 429 || status >= 500;
+}
+
 async function sponsorRequest({
   options,
   auth,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stage5/lumine",
-  "version": "0.2.56",
+  "version": "0.2.57",
   "description": "Command line tools for launching Lumine builds on Twinkle.",
   "type": "module",
   "bin": {