…
|
||
---|---|---|
.. | ||
darwin | ||
linux | ||
.chromium-commit | ||
.gitignore | ||
build.py | ||
build_util.py | ||
init.py | ||
README.md |
Chromium build
We ship our own headless build of Chromium for Linux and Mac OS, using a version of the source that corresponds to the requirements of the Puppeteer node module. The scripts in this folder can be used to accept a commit hash from the Chromium repository, and initialize the build in a workspace.
Why do we do this
Linux: By default, Puppeteer will download a zip file containing the
Chromium browser for any OS. This creates problems on Linux, because Chromium
has a dependency on X11, which is often not installed for a server environment.
We don't want to make a requirement for Linux that you need X11 to run Kibana.
To work around this, we create our own Chromium build, using the
headless_shell
build target. There are no (trustworthy) sources of these builds available
elsewhere.
Mac: We do this on Mac because Elastic signs the Kibanna release artifact with Apple to work with Gatekeeper on Mac OS. Having our own binary of Chromium and bundling it with Kibana is integral to the artifact signing process.
Windows: No custom build is necessary for Windows. We are able to use the full build of Chromium, downloaded from the main Chromium download location, using the revision that corresponds with the Puppeteer dependency.
Build Script Usage
These commands show how to set up an environment to build:
# Allow our scripts to use depot_tools commands
export PATH=$HOME/chromium/depot_tools:$PATH
# Create a dedicated working directory for this directory of Python scripts.
mkdir ~/chromium && cd ~/chromium
# Copy the scripts from the Kibana team's GCS bucket
gsutil cp -r gs://headless_shell_staging/build_chromium .
# Install the OS packages, configure the environment, download the chromium source (25GB)
python ./build_chromium/init.py [arch_name]
# Run the build script with the path to the chromium src directory, the git commit hash
python ./build_chromium/build.py <commit_id> x64
# OR You can build for ARM
python ./build_chromium/build.py <commit_id> arm64
NOTE: The init.py
script updates git config to make it more possible for
the Chromium repo to be cloned successfully. If checking out the Chromium fails
with "early EOF" errors, the instance could be low on memory or disk space.
Getting the Commit Hash
If you need to bump the version of Puppeteer, you need to get a new git commit hash for Chromium that corresponds to the Puppeteer version.
node x-pack/dev-tools/chromium_version.js [PuppeteerVersion]
When bumping the Puppeteer version, make sure you also update the ChromiumArchivePaths.revision
variable in
x-pack/plugins/reporting/server/browsers/chromium/paths.ts
.
Build args
A good how-to on building Chromium from source is here.
We have an linux/args.gn
file that is automatically copied to the build target directory.
To get a list of the build arguments that are enabled, install depot_tools
and run
gn args out/headless --list
. It prints out all of the flags and their
settings, including the defaults. Some build flags are documented
here.
NOTE: Please, make sure you consult @elastic/kibana-security before you change, remove or add any of the build flags.
Directions for Elasticians
If you wish to use a remote VM to build, you'll need access to our GCP account.
NOTE: The builds should be done in Ubuntu on x64 architecture. ARM builds are created in x64 using cross-compiling. CentOS is not supported for building Chromium.
- Login to Google Cloud Console
- Click the "Compute Engine" tab.
- Create a Linux VM:
- 8 CPU
- 30GB memory
- 80GB free space on disk (Try
ncdu /home
to see where space is used.) - git
- python2 (
python
must link topython2
) - lsb_release
- tmux is recommended in case your ssh session is interrupted
- "Cloud API access scopes": must have read / write scope for the Storage API
- Install Google Cloud SDK locally to ssh into the GCP instance
Artifacts
After the build completes, there will be a .zip file and a .md5 file in ~/chromium/chromium/src/out/headless
. These are named like so: chromium-{first_7_of_SHA}-{platform}-{arch}
, for example: chromium-4747cc2-linux-x64
.
The zip files and md5 files are copied to a staging bucket in GCP storage.
Testing
Search the Puppeteer Github repo for known issues that could affect our use case, and make sure to test anywhere that is affected.
Here's the steps on how to test a Puppeteer upgrade, run these tests on Mac, Windows, Linux x64 and Linux arm64:
- Make sure the Reporting plugin is fetching the correct version of the browser
at start-up time, and that it can successfully unzip it and copy the files to
x-pack/plugins/reporting/chromium
- Make sure there are no errors when using the Reporting diagnostic tool
- All functional and API tests that generate PDF and PNG files should pass.
- Use a VM to run Kibana in a low-memory environment and try to generate a PNG of a dashboard that outputs as a 4MB file. Document the minimum requirements in the PR.
Resources
The following links provide helpful context about how the Chromium build works, and its prerequisites:
- Tools for Chromium version information: https://omahaproxy.appspot.com/
- https://www.chromium.org/developers/how-tos/get-the-code/working-with-release-branches
- https://chromium.googlesource.com/chromium/src/+/HEAD/docs/linux/build_instructions.md
- Some build-flag descriptions: https://www.chromium.org/developers/gn-build-configuration
- The serverless Chromium project was indispensable:
b29445aa5a/packages/lambda/builds/chromium/build/build.sh