meshy-node 0.4.1 → 0.4.2

package/main.cjs

@@ -33089,6 +33089,22 @@ function parseNodeSettingsSnapshot(value) {
   }
   return clone(value);
 }
+function parseDevTunnelHealth(value) {
+  if (!isPlainObject(value)) {
+    return void 0;
+  }
+  if (value.status !== "healthy" && value.status !== "failed" || typeof value.checkedAt !== "number" || !Number.isFinite(value.checkedAt)) {
+    return void 0;
+  }
+  return {
+    status: value.status,
+    checkedAt: value.checkedAt,
+    endpoint: typeof value.endpoint === "string" ? value.endpoint : void 0,
+    reason: typeof value.reason === "string" ? value.reason : void 0,
+    statusCode: typeof value.statusCode === "number" && Number.isFinite(value.statusCode) ? value.statusCode : void 0,
+    consecutiveFailures: typeof value.consecutiveFailures === "number" && Number.isFinite(value.consecutiveFailures) ? value.consecutiveFailures : void 0
+  };
+}
 function isClusterInfo(value) {
   return isPlainObject(value) && typeof value.clusterId === "string" && typeof value.createdAt === "number" && Number.isFinite(value.createdAt);
 }
@@ -33112,6 +33128,7 @@ function parseNodeInfo(value) {
     lastHeartbeat: value.lastHeartbeat,
     transportType: typeof value.transportType === "string" ? value.transportType : void 0,
     devtunnelEndpoint: typeof value.devtunnelEndpoint === "string" ? value.devtunnelEndpoint : void 0,
+    devtunnelHealth: parseDevTunnelHealth(value.devtunnelHealth),
     dashboardOrigin: typeof value.dashboardOrigin === "string" ? value.dashboardOrigin : void 0,
     workDir: typeof value.workDir === "string" ? value.workDir : void 0,
     workDirFolders: parseStringArray(value.workDirFolders),
@@ -33515,6 +33532,24 @@ var NodeRegistry = class {
       }
     }
   }
+  updateDevTunnelHealth(id, devtunnelHealth) {
+    const node = this.nodes.get(id);
+    if (node) {
+      if (devtunnelHealth) {
+        node.devtunnelHealth = { ...devtunnelHealth };
+      } else {
+        delete node.devtunnelHealth;
+      }
+      this.store.upsertNode(node);
+    }
+    if (this.selfNode && this.selfNode.id === id) {
+      if (devtunnelHealth) {
+        this.selfNode.devtunnelHealth = { ...devtunnelHealth };
+      } else {
+        delete this.selfNode.devtunnelHealth;
+      }
+    }
+  }
   updateDashboardOrigin(id, dashboardOrigin) {
     const node = this.nodes.get(id);
     if (node) {
@@ -34689,17 +34724,9 @@ var HeartbeatMonitor = class {
   running = false;
   followerMissedMap = /* @__PURE__ */ new Map();
   followerPushReachableMap = /* @__PURE__ */ new Map();
-  /**
-   * Tracks whether a push-heartbeat has EVER succeeded for a follower.
-   * If push has never worked, the follower is on an unreachable network
-   * and should not be marked offline based on push failures alone.
-   */
+  // Tracks whether push-heartbeat has ever succeeded for a follower.
   followerEverReachedMap = /* @__PURE__ */ new Map();
-  /**
-   * Last known leader endpoint — preserved across clearLeader() so
-   * follower keepalives can still attempt to reach the leader even
-   * after the follower timer fires.
-   */
+  // Preserved across clearLeader() so follower keepalives can still reach the leader.
   lastKnownLeaderEndpoint = null;
   getQueuedMessages;
   messageQueue;
@@ -34922,6 +34949,7 @@ var HeartbeatMonitor = class {
       supportedAgents: self2.supportedAgents,
       devtunnelEnabled: self2.devtunnelEndpoint !== void 0,
       devtunnelEndpoint: self2.devtunnelEndpoint,
+      devtunnelHealth: self2.devtunnelHealth,
       dashboardOrigin: self2.dashboardOrigin,
       workDirFolders: self2.workDir ? listWorkDirFolders(self2.workDir) : void 0,
       settingsSnapshot: this.resolveCurrentSettingsSnapshot()
@@ -34942,6 +34970,7 @@ var HeartbeatMonitor = class {
       workDir,
       devtunnelEnabled,
       devtunnelEndpoint,
+      devtunnelHealth,
       dashboardOrigin,
       workDirFolders,
       settingsSnapshot
@@ -34969,6 +34998,7 @@ var HeartbeatMonitor = class {
         ...transportType ? { transportType } : {},
         ...workDir ? { workDir } : {},
         ...devtunnelEnabled && devtunnelEndpoint ? { devtunnelEndpoint } : {},
+        ...devtunnelHealth ? { devtunnelHealth } : {},
         ...dashboardOrigin ? { dashboardOrigin } : {},
         ...workDirFolders ? { workDirFolders } : {},
         ...settingsSnapshot ? { settingsSnapshot } : {}
@@ -35004,6 +35034,7 @@ var HeartbeatMonitor = class {
     }
     this.nodeRegistry.updateLoad(nodeId, load);
     this.applyReportedDevTunnelState(nodeId, devtunnelEnabled, devtunnelEndpoint);
+    this.nodeRegistry.updateDevTunnelHealth(nodeId, devtunnelEnabled === false ? void 0 : devtunnelHealth);
     this.nodeRegistry.updateDashboardOrigin(nodeId, dashboardOrigin);
     if (workDirFolders) {
       this.nodeRegistry.updateFolders(nodeId, workDirFolders);
@@ -35053,6 +35084,7 @@ var HeartbeatMonitor = class {
       workDir: self2.workDir,
       devtunnelEnabled: self2.devtunnelEndpoint !== void 0,
       devtunnelEndpoint: self2.devtunnelEndpoint,
+      devtunnelHealth: self2.devtunnelHealth,
       dashboardOrigin: self2.dashboardOrigin,
       workDirFolders: self2.workDir ? listWorkDirFolders(self2.workDir) : void 0,
       settingsSnapshot
@@ -37760,6 +37792,9 @@ ${joinErrors.map((e) => `  - ${e}`).join("\n")}`
     if (self2.devtunnelEndpoint) {
       joinBody.devtunnelEndpoint = self2.devtunnelEndpoint;
     }
+    if (self2.devtunnelHealth) {
+      joinBody.devtunnelHealth = self2.devtunnelHealth;
+    }
     if (self2.dashboardOrigin) {
       joinBody.dashboardOrigin = self2.dashboardOrigin;
     }
@@ -38093,6 +38128,7 @@ ${joinErrors.map((e) => `  - ${e}`).join("\n")}`
       workDir: currentSelf.workDir,
       workDirFolders: listWorkDirFolders(this.getWorkDir()),
       ...currentSelf.devtunnelEndpoint ? { devtunnelEndpoint: currentSelf.devtunnelEndpoint } : {},
+      ...currentSelf.devtunnelHealth ? { devtunnelHealth: currentSelf.devtunnelHealth } : {},
       ...currentSelf.dashboardOrigin ? { dashboardOrigin: currentSelf.dashboardOrigin } : {},
       ...currentSelf.settingsSnapshot ? { settingsSnapshot: currentSelf.settingsSnapshot } : {}
     };
@@ -42945,6 +42981,14 @@ var NodeSettingsSnapshotSchema = external_exports.object({
     api: SystemPackageInfoSchema.optional()
   }).optional()
 });
+var DevTunnelHealthSchema = external_exports.object({
+  status: external_exports.enum(["healthy", "failed"]),
+  checkedAt: external_exports.number(),
+  endpoint: external_exports.string().optional(),
+  reason: external_exports.string().optional(),
+  statusCode: external_exports.number().optional(),
+  consecutiveFailures: external_exports.number().optional()
+});
 
 // ../../packages/api/src/schemas/cluster.ts
 var NodeInfoSchema = external_exports.object({
@@ -42959,6 +43003,7 @@ var NodeInfoSchema = external_exports.object({
   lastHeartbeat: external_exports.number(),
   transportType: external_exports.string().optional(),
   devtunnelEndpoint: external_exports.string().optional(),
+  devtunnelHealth: DevTunnelHealthSchema.optional(),
   dashboardOrigin: external_exports.string().optional(),
   workDir: external_exports.string().optional(),
   workDirFolders: external_exports.array(external_exports.string()).optional(),
@@ -42995,6 +43040,7 @@ var JoinClusterBody = external_exports.object({
   workDir: external_exports.string().optional(),
   workDirFolders: external_exports.array(external_exports.string()).optional(),
   devtunnelEndpoint: external_exports.string().url().optional(),
+  devtunnelHealth: DevTunnelHealthSchema.optional(),
   dashboardOrigin: external_exports.string().url().optional(),
   settingsSnapshot: NodeSettingsSnapshotSchema.optional(),
   tasks: external_exports.array(JoinTaskSchema).default([])
@@ -43039,6 +43085,7 @@ var NodeInfoSchema2 = external_exports.object({
   lastHeartbeat: external_exports.number(),
   transportType: external_exports.string().optional(),
   devtunnelEndpoint: external_exports.string().optional(),
+  devtunnelHealth: DevTunnelHealthSchema.optional(),
   dashboardOrigin: external_exports.string().optional(),
   workDir: external_exports.string().optional(),
   workDirFolders: external_exports.array(external_exports.string()).optional(),
@@ -50729,6 +50776,82 @@ function buildRuntimeMetadata(storagePath) {
   };
 }
 
+// src/bootstrap/tunnel-health.ts
+var DEFAULT_TUNNEL_REACHABILITY_INTERVAL_MS = 6e4;
+var DEFAULT_TUNNEL_REACHABILITY_TIMEOUT_MS = 5e3;
+var DEFAULT_TUNNEL_RESTART_FAILURE_THRESHOLD = 2;
+function createTunnelReachabilityMonitor(options = {}) {
+  const failureThreshold = options.failureThreshold ?? DEFAULT_TUNNEL_RESTART_FAILURE_THRESHOLD;
+  const fetchImpl = options.fetchImpl ?? fetch;
+  const healthPath = options.healthPath ?? "/api/system/health";
+  const intervalMs = options.intervalMs ?? DEFAULT_TUNNEL_REACHABILITY_INTERVAL_MS;
+  const now = options.now ?? (() => Date.now());
+  const timeoutMs = options.timeoutMs ?? DEFAULT_TUNNEL_REACHABILITY_TIMEOUT_MS;
+  let consecutiveFailures = 0;
+  let lastCheckedAt = null;
+  function recordFailure(reason, statusCode) {
+    consecutiveFailures += 1;
+    return {
+      available: false,
+      checked: true,
+      consecutiveFailures,
+      reason,
+      shouldRestart: consecutiveFailures >= failureThreshold,
+      ...statusCode !== void 0 ? { statusCode } : {}
+    };
+  }
+  return {
+    async check(endpoint) {
+      const currentTime = now();
+      if (lastCheckedAt !== null && currentTime - lastCheckedAt < intervalMs) {
+        return {
+          available: true,
+          checked: false,
+          consecutiveFailures,
+          shouldRestart: false
+        };
+      }
+      lastCheckedAt = currentTime;
+      if (!endpoint) {
+        return recordFailure("missing endpoint");
+      }
+      try {
+        const response = await fetchWithTimeout2(fetchImpl, buildHealthUrl(endpoint, healthPath), timeoutMs);
+        if (!response.ok) {
+          return recordFailure(`health check returned ${response.status}`, response.status);
+        }
+        consecutiveFailures = 0;
+        return {
+          available: true,
+          checked: true,
+          consecutiveFailures,
+          shouldRestart: false,
+          statusCode: response.status
+        };
+      } catch (err) {
+        return recordFailure(err instanceof Error ? err.message : String(err));
+      }
+    },
+    reset() {
+      consecutiveFailures = 0;
+      lastCheckedAt = null;
+    }
+  };
+}
+function buildHealthUrl(endpoint, healthPath) {
+  return new URL(healthPath, endpoint).toString();
+}
+async function fetchWithTimeout2(fetchImpl, url, timeoutMs) {
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(), timeoutMs);
+  timer.unref?.();
+  try {
+    return await fetchImpl(url, { method: "GET", signal: controller.signal });
+  } finally {
+    clearTimeout(timer);
+  }
+}
+
 // src/bootstrap/start-node.ts
 function createSettingsSnapshotProvider(options) {
   let cachedSnapshot = null;
@@ -50818,6 +50941,9 @@ function createTunnelManager(options) {
   let shareTransport = null;
   let dashboardHealthTimer = null;
   let nodeTunnelHealthTimer = null;
+  let dashboardHealthCheckRunning = false;
+  let nodeTunnelHealthCheckRunning = false;
+  const nodeTunnelReachability = createTunnelReachabilityMonitor();
   function createDashboardTransportConfig() {
     return {
       type: "devtunnel",
@@ -50857,6 +50983,28 @@ function createTunnelManager(options) {
   function shouldPublishDashboardTunnel() {
     return authMetadata.enabled && (meshyNode.getTransportType() === "devtunnel" || meshyNode.isDevTunnelEnabled());
   }
+  function getPublishedNodeTunnelEndpoint() {
+    const self2 = meshyNode.getNodeRegistry().getSelf();
+    return meshyNode.getTransportType() === "devtunnel" ? self2.endpoint : self2.devtunnelEndpoint;
+  }
+  function setPublishedNodeTunnelHealth(health) {
+    const selfId = meshyNode.getNodeRegistry().getSelf().id;
+    meshyNode.getNodeRegistry().updateDevTunnelHealth(selfId, health);
+  }
+  function markPublishedNodeTunnelHealthy(endpoint) {
+    setPublishedNodeTunnelHealth({
+      status: "healthy",
+      checkedAt: Date.now(),
+      ...endpoint ? { endpoint } : {}
+    });
+  }
+  function markPublishedNodeTunnelFailed(details) {
+    setPublishedNodeTunnelHealth({
+      status: "failed",
+      checkedAt: Date.now(),
+      ...details
+    });
+  }
   async function stopDashboardTransport(reason) {
     if (!dashboardTransport) {
       if (dashboardOrigin) {
@@ -50960,31 +51108,100 @@ function createTunnelManager(options) {
     if (!nodeTunnelHealthTimer) {
       nodeTunnelHealthTimer = setInterval(() => {
         void (async () => {
-          if (!meshyNode.hasPublishedNodeTunnel()) {
+          if (nodeTunnelHealthCheckRunning) {
             return;
           }
-          const healthy = await meshyNode.isPublishedNodeTunnelHealthy().catch(() => false);
-          if (healthy) {
-            return;
-          }
-          meshyNode.getLogger().warn("published node tunnel became unhealthy; restarting with stable id", {
-            mainTransportType: meshyNode.getTransportType(),
-            sidecarEnabled: meshyNode.isDevTunnelEnabled()
-          });
+          nodeTunnelHealthCheckRunning = true;
           try {
-            const endpoint = await meshyNode.restartPublishedNodeTunnel();
-            meshyNode.getLogger().info("restarted published node tunnel", {
-              endpoint,
-              stableUrl: true,
-              mainTransportType: meshyNode.getTransportType(),
-              sidecarEnabled: meshyNode.isDevTunnelEnabled()
+            if (!meshyNode.hasPublishedNodeTunnel()) {
+              nodeTunnelReachability.reset();
+              setPublishedNodeTunnelHealth(void 0);
+              return;
+            }
+            const publishedEndpoint = getPublishedNodeTunnelEndpoint();
+            const healthy = await meshyNode.isPublishedNodeTunnelHealthy().catch(() => false);
+            if (!healthy) {
+              markPublishedNodeTunnelFailed({
+                endpoint: publishedEndpoint,
+                reason: "devtunnel process is not running"
+              });
+              meshyNode.getLogger().warn("published node tunnel process became unhealthy; restarting with stable id", {
+                mainTransportType: meshyNode.getTransportType(),
+                sidecarEnabled: meshyNode.isDevTunnelEnabled()
+              });
+              try {
+                const endpoint = await meshyNode.restartPublishedNodeTunnel();
+                nodeTunnelReachability.reset();
+                markPublishedNodeTunnelHealthy(endpoint);
+                meshyNode.getLogger().info("restarted published node tunnel", {
+                  endpoint,
+                  stableUrl: true,
+                  mainTransportType: meshyNode.getTransportType(),
+                  sidecarEnabled: meshyNode.isDevTunnelEnabled()
+                });
+              } catch (err) {
+                markPublishedNodeTunnelFailed({
+                  endpoint: publishedEndpoint,
+                  reason: err instanceof Error ? err.message : String(err)
+                });
+                meshyNode.getLogger().warn("failed to restart published node tunnel", {
+                  error: err instanceof Error ? err.message : String(err),
+                  mainTransportType: meshyNode.getTransportType(),
+                  sidecarEnabled: meshyNode.isDevTunnelEnabled()
+                });
+              }
+              return;
+            }
+            const reachability = await nodeTunnelReachability.check(publishedEndpoint);
+            if (!reachability.checked) {
+              return;
+            }
+            if (reachability.available) {
+              markPublishedNodeTunnelHealthy(publishedEndpoint);
+              return;
+            }
+            markPublishedNodeTunnelFailed({
+              consecutiveFailures: reachability.consecutiveFailures,
+              endpoint: publishedEndpoint,
+              reason: reachability.reason,
+              statusCode: reachability.statusCode
             });
-          } catch (err) {
-            meshyNode.getLogger().warn("failed to restart published node tunnel", {
-              error: err instanceof Error ? err.message : String(err),
+            meshyNode.getLogger().warn("published node tunnel endpoint health check failed", {
+              consecutiveFailures: reachability.consecutiveFailures,
+              failureThresholdReached: reachability.shouldRestart,
               mainTransportType: meshyNode.getTransportType(),
-              sidecarEnabled: meshyNode.isDevTunnelEnabled()
+              reason: reachability.reason,
+              sidecarEnabled: meshyNode.isDevTunnelEnabled(),
+              statusCode: reachability.statusCode
             });
+            if (!reachability.shouldRestart) {
+              return;
+            }
+            try {
+              const endpoint = await meshyNode.restartPublishedNodeTunnel();
+              nodeTunnelReachability.reset();
+              markPublishedNodeTunnelHealthy(endpoint);
+              meshyNode.getLogger().info("restarted published node tunnel", {
+                endpoint,
+                stableUrl: true,
+                mainTransportType: meshyNode.getTransportType(),
+                sidecarEnabled: meshyNode.isDevTunnelEnabled()
+              });
+            } catch (err) {
+              markPublishedNodeTunnelFailed({
+                consecutiveFailures: reachability.consecutiveFailures,
+                endpoint: publishedEndpoint,
+                reason: err instanceof Error ? err.message : String(err),
+                statusCode: reachability.statusCode
+              });
+              meshyNode.getLogger().warn("failed to restart published node tunnel", {
+                error: err instanceof Error ? err.message : String(err),
+                mainTransportType: meshyNode.getTransportType(),
+                sidecarEnabled: meshyNode.isDevTunnelEnabled()
+              });
+            }
+          } finally {
+            nodeTunnelHealthCheckRunning = false;
           }
         })();
       }, config.cluster.heartbeatInterval);
@@ -50992,17 +51209,24 @@ function createTunnelManager(options) {
     if (!dashboardHealthTimer) {
       dashboardHealthTimer = setInterval(() => {
         void (async () => {
-          if (!dashboardTransport) {
+          if (dashboardHealthCheckRunning) {
             return;
           }
-          const healthy = await dashboardTransport.isHealthy().catch(() => false);
-          if (healthy) {
-            return;
+          dashboardHealthCheckRunning = true;
+          try {
+            if (!dashboardTransport) {
+              return;
+            }
+            const healthy = await dashboardTransport.isHealthy().catch(() => false);
+            if (!healthy) {
+              meshyNode.getLogger().warn("dashboard tunnel process became unhealthy; restarting with stable id", {
+                port: config.node.port
+              });
+              await restartDashboardTransport("healthcheck");
+            }
+          } finally {
+            dashboardHealthCheckRunning = false;
           }
-          meshyNode.getLogger().warn("dashboard tunnel became unhealthy; restarting with stable id", {
-            port: config.node.port
-          });
-          await restartDashboardTransport("healthcheck");
         })();
       }, config.cluster.heartbeatInterval);
     }
@@ -51016,6 +51240,7 @@ function createTunnelManager(options) {
       clearInterval(nodeTunnelHealthTimer);
       nodeTunnelHealthTimer = null;
     }
+    nodeTunnelReachability.reset();
   }
   async function stop() {
     stopHealthChecks();
@@ -51168,6 +51393,9 @@ async function startNode(args) {
     switchTransport: async (type) => {
       const endpoint = await meshyNode.switchTransport(type);
       persistNodeStartupTransport(config.storage.path, type);
+      if (type !== "devtunnel" && !meshyNode.isDevTunnelEnabled()) {
+        meshyNode.getNodeRegistry().updateDevTunnelHealth(meshyNode.getNodeRegistry().getSelf().id, void 0);
+      }
       await tunnelManager.syncDashboardTransport(`switchTransport:${type}`);
       return endpoint;
     },
@@ -51180,6 +51408,9 @@ async function startNode(args) {
     },
     disableDevTunnel: async () => {
       await meshyNode.disableDevTunnel();
+      if (meshyNode.getTransportType() !== "devtunnel") {
+        meshyNode.getNodeRegistry().updateDevTunnelHealth(meshyNode.getNodeRegistry().getSelf().id, void 0);
+      }
       persistNodeStartupTransport(
         config.storage.path,
         meshyNode.getTransportType()

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "meshy-node",
-  "version": "0.4.1",
+  "version": "0.4.2",
   "private": false,
   "description": "Standalone Meshy node package with bundled runtime and dashboard assets.",
   "type": "commonjs",

package/runtime-metadata.json

@@ -1,14 +1,14 @@
 {
   "packageName": "meshy-node",
-  "packageVersion": "0.4.1",
+  "packageVersion": "0.4.2",
   "packages": {
     "workspace": {
       "name": "meshy",
-      "version": "0.4.1"
+      "version": "0.4.2"
     },
     "node": {
       "name": "meshy-node",
-      "version": "0.4.1"
+      "version": "0.4.2"
     },
     "core": {
       "name": "@meshy/core",
@@ -25,7 +25,7 @@
   },
   "repository": {
     "url": "https://github.com/ai-microsoft/meshy",
-    "branch": "release0_4_1",
-    "commit": "554b574"
+    "branch": "TunnelRestart",
+    "commit": "c895cea"
   }
 }