vscode

Table of Contents

Table of Contents
CPU profiling unresponsive window via tracing
Using samply on macOS
Using samply on windows
Using samply on linux
Using etw on windows

CPU profiling unresponsive window via tracing

When a window (renderer process) is unresponsive the devtools might be unable to collect the required cpu profile. A reliable way to profile in such cases is with the tracing infra using the steps below,

Restart VS Code with the following flags --trace --trace-category-filter="renderer,blink,v8,disabled-by-default-v8.cpu_profiler"
Once the unresponsive dialog pop ups, click on the Reopen button
In the new window, execute "F1 > Developer: Stop Tracing"

This will generate a trace json file which can be loaded into any of the following clients that supports the Google's Trace Event format

chrome://tracing in any chromium browser
ui.perfetto.dev
Firefox profiler

Using samply on macOS

If you have an unsigned build of VSCode, then samply can be used to capture both JIT and native traces

Download the latest release of https://github.com/mstange/samply/releases
Start from the command line with following samply.exe record -s -o profile.json.gz --browsers -- <path>/Visual\ Studio\ Code.app/Contents/MacOS/Electron --js-flags="--perf-basic-prof --perf-prof-unwinding-info --interpreted-frames-native-stack"
Perform the steps for unresponsiveness or slow startups etc
Exit the application and the profile will be saved to profile.json.gz

You can inspect the profile with samply.exe load --breakpad-symbol-dir <path> profile.json.gz

Using samply on windows

If you want to capture only application related cpu profile with both JIT and native traces

Download the latest release of https://github.com/mstange/samply/releases
Start from the command line with following samply.exe record -s -o profile.json.gz --browsers -- <path>/Microsoft\ VS\ Code/Code.exe --js-flags="--enable-etw-stack-walking --interpreted-frames-native-stack"
Perform the steps for unresponsiveness or slow startups etc
Exit the application and the profile will be saved to profile.json.gz

You can inspect the profile with samply.exe load --breakpad-symbol-dir <path> profile.json.gz

Using samply on linux

Download the latest release of https://github.com/mstange/samply/releases
Start from the command line with following samply.exe record -s -o profile.json.gz --browsers -- <path>/code --js-flags="--perf-basic-prof --perf-prof-unwinding-info --interpreted-frames-native-stack"
Perform the steps for unresponsiveness or slow startups etc
Exit the application and the profile will be saved to profile.json.gz

You can inspect the profile with samply.exe load --breakpad-symbol-dir <path> profile.json.gz

Using etw on windows

Open Windows Performance Recorder from System start menu, it is not installed by default. You can install it via Windows Assessment and Deployment Kit (ADK) and choose Windows Performance Toolkit.
Select the following options under Resource Analysis
1. CPU Usage
2. File I/O activity
3. Registry I/O activity
4. Pool usage
5. VirtualAlloc usage
6. GPU activity
Set Performance Scenario to General, Detail level to verbose and Logging mode to file
Start the recorder
Start the VSCode application from the terminal with the following additional flags --js-flags="--enable-etw-stack-walking --interpreted-frames-native-stack"
Perform the actions that trigger the issue (slow startups, out of memory, unresponsive threads etc)
Stop the recorder and save the file

The .etl file should not be attached to public issues since they contain PII and only shared via trusted channels.

Project Management

Contributing

Documentation

Want to contribute to this Wiki?

Fork it and send a pull request.