@nosana/node 1.1.21-rc → 1.1.22-rc

package/dist/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@nosana/node",
-    "version": "1.1.21-rc",
+    "version": "1.1.22-rc",
     "description": "",
     "bin": {
         "nosana-node": "./dist/src/index.js"

package/dist/src/NodeManager/node/task/operations/enforceOperationTimeout.d.ts

@@ -3,13 +3,14 @@ import type { EventEmitter } from 'events';
 /**
  * Enforces a maximum running time for a single operation.
  *
- * The clock starts when the operation emits `start` (i.e. the container is
- * actually running) — the same lifecycle point the rest of
- * `runTaskManagerOperation` keys its timeout enforcement off. If the operation
- * is still running `timeoutSeconds` later, `onTimeout` is invoked; callers use
- * this to abort the op. The timer is cleared as soon as the operation settles
- * (`exit`/`error`) or is torn down (`end`), so a normally completing op never
- * triggers it.
+ * The clock starts on `container:started` — emitted once the container is
+ * actually running, i.e. *after* the image pull — so download time is not
+ * counted against the operation's allotted run time. (The earlier `start`
+ * event, which the provider emits before pulling, is deliberately not used.)
+ * If the operation is still running `timeoutSeconds` later, `onTimeout` is
+ * invoked; callers use this to abort the op. The timer is cleared as soon as
+ * the operation settles (`exit`/`error`) or is torn down (`end`), so a normally
+ * completing op never triggers it.
  *
  * The unit is seconds, matching the on-chain `job.timeout` convention.
  *

package/dist/src/NodeManager/node/task/operations/enforceOperationTimeout.js

@@ -1,13 +1,14 @@
 /**
  * Enforces a maximum running time for a single operation.
  *
- * The clock starts when the operation emits `start` (i.e. the container is
- * actually running) — the same lifecycle point the rest of
- * `runTaskManagerOperation` keys its timeout enforcement off. If the operation
- * is still running `timeoutSeconds` later, `onTimeout` is invoked; callers use
- * this to abort the op. The timer is cleared as soon as the operation settles
- * (`exit`/`error`) or is torn down (`end`), so a normally completing op never
- * triggers it.
+ * The clock starts on `container:started` — emitted once the container is
+ * actually running, i.e. *after* the image pull — so download time is not
+ * counted against the operation's allotted run time. (The earlier `start`
+ * event, which the provider emits before pulling, is deliberately not used.)
+ * If the operation is still running `timeoutSeconds` later, `onTimeout` is
+ * invoked; callers use this to abort the op. The timer is cleared as soon as
+ * the operation settles (`exit`/`error`) or is torn down (`end`), so a normally
+ * completing op never triggers it.
  *
  * The unit is seconds, matching the on-chain `job.timeout` convention.
  *
@@ -23,7 +24,8 @@ export function enforceOperationTimeout(emitter, timeoutSeconds, onTimeout) {
             handle = undefined;
         }
     };
-    emitter.on('start', () => {
+    emitter.on('container:started', () => {
+        // Defensive: clear first so an unexpected duplicate can't stack timers.
         clear();
         handle = setTimeout(() => {
             handle = undefined;

package/dist/src/NodeManager/node/task/operations/runTaskManagerOperation.js

@@ -73,7 +73,8 @@ export async function runTaskManagerOperation(flow, op, dependent) {
     this.abortControllerMap.set(op.id, abort);
     /**
      * Enforce the operation's maximum run time, if one was supplied on its
-     * execution. The clock starts when the op emits `start`; if it's still
+     * execution. The clock starts when the container is running (`container:started`,
+     * i.e. after the image pull), so download time isn't counted. If it's still
      * running when the timeout elapses, we abort it with EXPIRED — the same
      * clean (non-failure) stop used for job-level expiry — and record the reason
      * for diagnostics/auditing.

package/dist/src/NodeManager/provider/Provider.js

@@ -137,6 +137,9 @@ export class Provider {
             if (exist && !exited && container.id) {
                 stateManager = new ContainerStateManager(container, controller, emitter, op.args.restart_policy);
                 await stateManager.startMonitoring();
+                // Reused container is already running (no pull needed) — start the
+                // op-execution timeout clock here too.
+                emitter.emit('container:started');
             }
             else {
                 await this.containerOrchestration.pullImage(op.args.image, op.args.authentication?.docker, controller);
@@ -243,6 +246,10 @@ export class Provider {
                     }
                 }
                 container = await this.containerOrchestration.runFlowContainer(op.args.image ?? flow.jobDefinition.global?.image, createContainerOptions);
+                // The container is now running (image pull is already done). This is the
+                // point the op-execution timeout clock starts from, so pull time isn't
+                // counted against it.
+                emitter.emit('container:started');
                 const info = await container.inspect();
                 emitter.emit('updateOpState', { providerId: container.id });
                 if (!op.args.trusted_execution_env && info.NetworkSettings?.Networks["NOSANA_GATEWAY"].IPAddress) {

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "@nosana/node",
-  "version": "1.1.21-rc",
+  "version": "1.1.22-rc",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@nosana/node",
-      "version": "1.1.21-rc",
+      "version": "1.1.22-rc",
       "license": "ISC",
       "dependencies": {
         "@coral-xyz/anchor": "^0.28.1-beta.1",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nosana/node",
-  "version": "1.1.21-rc",
+  "version": "1.1.22-rc",
   "description": "",
   "bin": {
     "nosana-node": "./dist/src/index.js"