Windows raw kernel launch failure fix (#11603)

Author: IanMatthewHuff

news/2 Fixes/11574.md

@@ -0,0 +1 @@
+Fix intermittent launch failure with raw kernels on windows.

package-lock.json

@@ -2929,6 +2929,7 @@
         "@koa/cors": "^3.0.0",
         "@loadable/component": "^5.12.0",
         "@nteract/messaging": "^7.0.0",
+        "@types/tcp-port-used": "^1.0.0",
         "ansi-regex": "^4.1.0",
         "arch": "^2.1.0",
         "azure-storage": "^2.10.3",
@@ -2972,6 +2973,7 @@
         "strip-ansi": "^5.2.0",
         "sudo-prompt": "^8.2.0",
         "svg-to-pdfkit": "^0.1.8",
+        "tcp-port-used": "^1.0.1",
         "tmp": "^0.0.29",
         "tree-kill": "^1.2.2",
         "typescript-char": "^0.0.0",

package.json

@@ -313,7 +313,8 @@ export enum Telemetry {
     RawKernelSessionStartSuccess = 'DS_INTERNAL.RAWKERNEL_SESSION_START_SUCCESS',
     RawKernelSessionStartUserCancel = 'DS_INTERNAL.RAWKERNEL_SESSION_START_USER_CANCEL',
     RawKernelSessionStartTimeout = 'DS_INTERNAL.RAWKERNEL_SESSION_START_TIMEOUT',
-    RawKernelSessionStartException = 'DS_INTERNAL.RAWKERNEL_SESSION_START_EXCEPTION'
+    RawKernelSessionStartException = 'DS_INTERNAL.RAWKERNEL_SESSION_START_EXCEPTION',
+    RawKernelProcessLaunch = 'DS_INTERNAL.RAWKERNEL_PROCESS_LAUNCH'
 }
 
 export enum NativeKeyboardCommandTelemetry {

src/client/datascience/constants.ts

@@ -3,6 +3,7 @@
 'use strict';
 
 import { ChildProcess } from 'child_process';
+import * as tcpPortUsed from 'tcp-port-used';
 import { Event, EventEmitter } from 'vscode';
 import { PYTHON_LANGUAGE } from '../../common/constants';
 import { traceError, traceInfo, traceWarning } from '../../common/logger';
@@ -11,6 +12,8 @@ import { IProcessServiceFactory, ObservableExecutionResult } from '../../common/
 import { Resource } from '../../common/types';
 import { noop, swallowExceptions } from '../../common/utils/misc';
 import { PythonInterpreter } from '../../interpreter/contracts';
+import { captureTelemetry } from '../../telemetry';
+import { Telemetry } from '../constants';
 import { IJupyterKernelSpec } from '../types';
 import { findIndexOfConnectionFile } from './kernelFinder';
 import { PythonKernelLauncherDaemon } from './kernelLauncherDaemon';
@@ -61,6 +64,8 @@ export class KernelProcess implements IKernelProcess {
             await this.kernelDaemon?.interrupt();
         }
     }
+
+    @captureTelemetry(Telemetry.RawKernelProcessLaunch, undefined, true)
     public async launch(): Promise<void> {
         if (this.launchedOnce) {
             throw new Error('Kernel has already been launched.');
@@ -103,6 +108,9 @@ export class KernelProcess implements IKernelProcess {
                 }
             }
         );
+
+        // Don't return until our heartbeat channel is open for connections
+        return this.waitForHeartbeat();
     }
 
     public async dispose(): Promise<void> {
@@ -119,6 +127,20 @@ export class KernelProcess implements IKernelProcess {
         swallowExceptions(() => this.connectionFile?.dispose());
     }
 
+    // Make sure that the heartbeat channel is open for connections
+    private async waitForHeartbeat() {
+        try {
+            // Wait until the port is open for connection
+            // First parameter is wait between retries, second parameter is total wait before error
+            await tcpPortUsed.waitUntilUsed(this.connection.hb_port, 200, 30_000);
+        } catch (error) {
+            // Make sure to dispose if we never get a heartbeat
+            this.dispose().ignoreErrors();
+            traceError('Timed out waiting to get a heartbeat from kernel process.');
+            throw new Error('Timed out waiting to get a heartbeat from kernel process.');
+        }
+    }
+
     private async createAndUpdateConnectionFile() {
         this.connectionFile = await this.file.createTemporaryFile('.json');
         await this.file.writeFile(this.connectionFile.filePath, JSON.stringify(this._connection), {

src/client/datascience/kernel-launcher/kernelProcess.ts

@@ -2098,6 +2098,7 @@ export interface IEventNamePropertyMapping {
     // Raw kernel timing events
     [Telemetry.RawKernelSessionConnect]: never | undefined;
     [Telemetry.RawKernelStartRawSession]: never | undefined;
+    [Telemetry.RawKernelProcessLaunch]: never | undefined;
 
     // Raw kernel single events
     [Telemetry.RawKernelSessionStartSuccess]: never | undefined;