First, make sure you’re using lhci autorun or invoking lhci healthcheck --fatal in your script. Check the output logs of the healthcheck for anything that’s failing and try resolving those problems first using the guide below before filing an issue.
You have a couple of options to teach Lighthouse CI how to login to your page.
Checkout the issues in Lighthouse to see if someone else has run into the same problem before (Hint: closed issues often have solutions that you might find useful too!). Here are some of the most common problems we see…
puppeteer version first, or follow the examples in the getting started guide to add Chrome to your CI provider.--collect.settings.chromeFlags="--no-sandbox". If running in a docker container, ensure the container has SYS_ADMIN privileges.Most CI systems prevent secrets from leaking into the environment of untrusted pull requests from forks. This is a helpful security measure to prevent privileged information from falling into malicious hands.
If your LHCI server is already publicly accessible and you don’t mind allowing contributors to potentially spam data into it, then you can safely make the LHCI_TOKEN public as well. It is an additive-write-only credential that cannot destroy historical data.
While it’s not possible retrieve a lost GitHub App token, you can uninstall and reinstall the app to get a fresh token.
Some CI systems download a shallow copy of the repository to improve cloning time, but this prevents contextual information like the shared base ancestor commit from being picked up by Lighthouse CI. Ensure you configure your provider’s git depth to a value large enough to cover most of your feature branches.
.travis.yml
git:
#### Example: clone with the past 20 commits ######
depth: 20
#### Example: disable git depth entirely ######
depth: false
This is a precautionary measure to prevent the effective deletion of historical data from the server. As workaround, you can create empty commits to rerun builds git commit --allow-empty -m 'rerun CI' && git push.
If the URLs that you audit contain random components (ports, UUIDs, hashes, etc), then you’ll need to tell LHCI how to normalize your data when you upload it with the --url-replacement-patterns option.
When using this option, you’ll lose the default :PORT and UUID replacements, so be sure to copy those into your configuration if necessary. See the example in the lighthuose-ci repo itself for how to configure these patterns.
Webpages are fickle beasts and variance is a common problem when measuring properties of pages in a real browser as Lighthouse does. The Lighthouse team has several good documents on the subject, but the tl;dr is…
--collect.numberOfRuns=X option to increase this).setTimeout calls, A/B tests, etc).Different environments and ways of running Lighthouse will tend to see different results for the same underlying reasons described in the above FAQ on variance (see the several good documents on the subject). Before resorting to variance isolation though, first double check that the different environments are actually being run with the same Lighthouse settings. The most common differences we see that folks might forget to consider…
The team does not have the bandwidth to assist with debugging this problem. Please do not file an issue about it if you’re unable to get scores from different environments to match.
If you are using actions/checkout@v3 to checkout your repository and based on the size and traffic on your repo this error might happen due to how the checkout action is configured and is not a LHCI issue.
Checkout action by default doesn’t clone the entire repo and the number of commits to fetch is set to 1 for performance reasons. When LHCI runs the health check the ancestor hash might be missing because the branch / hash is not there in the local history.
In order to fix it change / add the following:
name: Run Lighthouse tests
jobs:
lhci:
name: Lighthouse
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
with:
fetch-depth: 20
- name: Fetch base_ref HEAD to use it as Ancestor hash in LHCI
run: git fetch --depth=1 origin +refs/heads/$:refs/remotes/origin/$
The additions are fetch-depth: 20 added to actions/checkout@v3 and a new step to fetch base_ref HEAD to use it as ancestor hash in LHCI.
The fetch depth set to 20 is a good default, but might not work in all cases, you can adjust it based on your needs.