service-bridge 1.1.0-dev.27 → 1.1.1-dev.29

package/dist/index.js

@@ -591,8 +591,9 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
   const fnAliasMap = new Map;
   const functionChannels = new Map;
   let isOnline = false;
+  let isFlushing = false;
   let stopped = false;
-  let onlineRestoreTimer = null;
+  let isWatchingChannel = false;
   const offlineQueue = [];
   let workerServer = null;
   let workerSessionStream = null;
@@ -900,36 +901,43 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
     }
   }
   _controlReady.then(() => {
-    isOnline = true;
-    flushQueue().catch((err) => {
-      if (isConnectionError(err))
-        scheduleOnlineRestore();
-      else
-        reportSDKError("flush-on-ready", err);
-    });
+    watchChannelConnectivity();
   }).catch(() => {});
-  function scheduleOnlineRestore() {
-    if (stopped || isOnline || onlineRestoreTimer)
+  function watchChannelConnectivity() {
+    isWatchingChannel = false;
+    if (stopped)
       return;
-    onlineRestoreTimer = setTimeout(() => {
-      onlineRestoreTimer = null;
-      if (!stopped) {
+    const channel = stub.getChannel();
+    const state = channel.getConnectivityState(true);
+    if (state === grpc.connectivityState.READY) {
+      if (!isOnline) {
         isOnline = true;
         const queueLen = offlineQueue.length;
         console.info(`[servicebridge] reconnected to runtime${queueLen > 0 ? ` — flushing ${queueLen} queued operation(s)` : ""}`);
-        flushQueue().catch((err) => {
-          if (isConnectionError(err))
-            scheduleOnlineRestore();
-          else
-            reportSDKError("flush-on-restore", err);
+        flushQueue().then(() => {
+          if (isOnline)
+            scheduleNextHeartbeat(100);
+        }).catch((err) => {
+          reportSDKError("flush-on-restore", err);
         });
-        scheduleNextHeartbeat(100);
       }
-    }, 2000);
+    } else if (state === grpc.connectivityState.TRANSIENT_FAILURE) {
+      if (isOnline) {
+        isOnline = false;
+        console.warn("[servicebridge] lost connection to runtime — entering offline mode");
+      }
+    }
+    isWatchingChannel = true;
+    channel.watchConnectivityState(state, Infinity, watchChannelConnectivity);
+  }
+  function ensureChannelWatch() {
+    if (!isWatchingChannel && !stopped) {
+      watchChannelConnectivity();
+    }
   }
   function isConnectionError(e) {
     const code = e?.code;
-    return code === grpc.status.UNAVAILABLE || code === grpc.status.UNKNOWN || code === grpc.status.DEADLINE_EXCEEDED || code === grpc.status.RESOURCE_EXHAUSTED || code === grpc.status.INTERNAL;
+    return code === grpc.status.UNAVAILABLE || code === grpc.status.UNKNOWN;
   }
   function normalizeUnknownErrorMessage(error) {
     return error instanceof Error ? error.message : String(error);
@@ -962,7 +970,7 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
           console.warn("[servicebridge] lost connection to runtime — entering offline mode");
         }
         isOnline = false;
-        scheduleOnlineRestore();
+        ensureChannelWatch();
         enqueueOffline({
           type: "reportCallStart",
           traceId,
@@ -1013,7 +1021,7 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
           console.warn("[servicebridge] lost connection to runtime — entering offline mode");
         }
         isOnline = false;
-        scheduleOnlineRestore();
+        ensureChannelWatch();
         enqueueOffline({
           type: "reportCall",
           traceId,
@@ -1033,66 +1041,71 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
     });
   }
   async function flushQueue() {
-    if (offlineQueue.length === 0)
+    if (isFlushing || offlineQueue.length === 0)
       return;
+    isFlushing = true;
     const snapshot = offlineQueue.slice();
     let flushed = 0;
-    while (offlineQueue.length > 0 && isOnline) {
-      const op = offlineQueue[0];
-      try {
-        if (op.type === "event") {
-          await new Promise((res, rej) => {
-            stub.Publish({
-              topic: op.topic,
-              payload: toJsonBuffer(op.payload),
-              headers: toWireStringMap(op.opts?.headers),
-              trace_id: op.opts?.traceId ?? "",
-              parent_span_id: op.opts?.parentSpanId ?? "",
-              producer_service: service,
-              idempotency_key: op.opts?.idempotencyKey ?? ""
-            }, meta, unaryDeadlineOptions(), (err) => err ? rej(err) : res());
-          });
-        } else if (op.type === "job") {
-          await new Promise((res, rej) => {
-            stub.RegisterJob({
-              cron_expr: op.opts.cron ?? "",
-              timezone: op.opts.timezone ?? "UTC",
-              misfire_policy: op.opts.misfire ?? "fire_now",
-              target_type: op.opts.via ?? "rpc",
-              target_ref: op.target,
-              delay_ms: op.opts.delay ?? 0,
-              service_name: service,
-              retry_policy_json: op.opts.retryPolicyJson ?? "{}"
-            }, meta, unaryDeadlineOptions(), (err) => err ? rej(err) : res());
-          });
-        } else if (op.type === "workflow") {
-          await new Promise((res, rej) => {
-            stub.RegisterWorkflow({
-              name: op.name,
-              definition: JSON.stringify(op.steps),
-              opts: op.opts ? JSON.stringify(op.opts) : "{}",
-              service_name: service
-            }, meta, unaryDeadlineOptions(), (err) => err ? rej(err) : res());
-          });
-        } else if (op.type === "reportCallStart") {
-          await sendReportCallStart({ ...op });
-        } else if (op.type === "reportCall") {
-          await sendReportCall(op);
-        }
-        offlineQueue.shift();
-        flushed++;
-      } catch (err) {
-        if (isConnectionError(err)) {
-          if (isOnline) {
-            console.warn("[servicebridge] lost connection to runtime — entering offline mode");
+    try {
+      while (offlineQueue.length > 0 && isOnline) {
+        const op = offlineQueue[0];
+        try {
+          if (op.type === "event") {
+            await new Promise((res, rej) => {
+              stub.Publish({
+                topic: op.topic,
+                payload: toJsonBuffer(op.payload),
+                headers: toWireStringMap(op.opts?.headers),
+                trace_id: op.opts?.traceId ?? "",
+                parent_span_id: op.opts?.parentSpanId ?? "",
+                producer_service: service,
+                idempotency_key: op.opts?.idempotencyKey ?? ""
+              }, meta, unaryDeadlineOptions(), (err) => err ? rej(err) : res());
+            });
+          } else if (op.type === "job") {
+            await new Promise((res, rej) => {
+              stub.RegisterJob({
+                cron_expr: op.opts.cron ?? "",
+                timezone: op.opts.timezone ?? "UTC",
+                misfire_policy: op.opts.misfire ?? "fire_now",
+                target_type: op.opts.via ?? "rpc",
+                target_ref: op.target,
+                delay_ms: op.opts.delay ?? 0,
+                service_name: service,
+                retry_policy_json: op.opts.retryPolicyJson ?? "{}"
+              }, meta, unaryDeadlineOptions(), (err) => err ? rej(err) : res());
+            });
+          } else if (op.type === "workflow") {
+            await new Promise((res, rej) => {
+              stub.RegisterWorkflow({
+                name: op.name,
+                definition: JSON.stringify(op.steps),
+                opts: op.opts ? JSON.stringify(op.opts) : "{}",
+                service_name: service
+              }, meta, unaryDeadlineOptions(), (err) => err ? rej(err) : res());
+            });
+          } else if (op.type === "reportCallStart") {
+            await sendReportCallStart({ ...op });
+          } else if (op.type === "reportCall") {
+            await sendReportCall(op);
+          }
+          offlineQueue.shift();
+          flushed++;
+        } catch (err) {
+          if (isConnectionError(err)) {
+            if (isOnline) {
+              console.warn("[servicebridge] lost connection to runtime — entering offline mode");
+            }
+            isOnline = false;
+            ensureChannelWatch();
+            break;
           }
-          isOnline = false;
-          scheduleOnlineRestore();
-          break;
+          reportSDKError("flush-offline-queue", err);
+          offlineQueue.shift();
         }
-        reportSDKError("flush-offline-queue", err);
-        break;
       }
+    } finally {
+      isFlushing = false;
     }
     if (flushed > 0) {
       const remaining = offlineQueue.length;
@@ -1256,7 +1269,7 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
           console.warn("[servicebridge] lost connection to runtime — entering offline mode");
         }
         isOnline = false;
-        scheduleOnlineRestore();
+        ensureChannelWatch();
       } else if (isRegistryResyncRequiredError(err)) {
         await syncRegistrations("heartbeat-resync");
       } else {
@@ -1787,11 +1800,9 @@ function servicebridge(url, serviceKey, serviceOrOpts = {}, maybeGlobalOpts = {}
     stop() {
       stopped = true;
       isOnline = false;
-      if (onlineRestoreTimer)
-        clearTimeout(onlineRestoreTimer);
+      isWatchingChannel = false;
       if (heartbeatTimer)
         clearTimeout(heartbeatTimer);
-      onlineRestoreTimer = null;
       heartbeatTimer = null;
       registeredGroups.clear();
       closeWorkerSession();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "service-bridge",
-  "version": "1.1.0-dev.27",
+  "version": "1.1.1-dev.29",
   "type": "module",
   "description": "ServiceBridge SDK for Node.js — production-ready RPC, durable events, workflows, jobs, and distributed tracing. One Go runtime + PostgreSQL replaces Istio, RabbitMQ, Temporal, and Jaeger.",
   "keywords": [