Usage

By default, the instance will be called lighthouse-collection-$gcloudusername-instance0 but you can replace instance0 by supplying an arg to run.sh.

1. Googlers: set up ssh to gcloud.
2. Spin up machines and start the collection.
   1. (Prerequiste) Install GCloud SDK and authenticate locally, Run brew install google-cloud-sdk && gcloud auth login
   2. (Prerequiste Googlers only) Ensure you're on the corp VPN or you won't be able to SSH into a Google-owned instance (run gcert).
   3. (Optional) Run export TARGET_GIT_REF=<a lighthouse git ref that has been pushed> if wanting to run on anything but main.
   4. (Optional) Run export TARGET_RUNS=<a number> if needing more than 1 run per URL.
   5. (Optional) Run export LIGHTHOUSE_FLAGS=... to configure Lighthouse CLI.
   6. (Optional) Run export LIGHTHOUSE_COLLECTION_GCLOUD_PROJECT=<project name> to set your GCloud project.
   7. Run bash core/scripts/gcp-collection/fleet-create.sh path/to/your/url/list.txt from the repo root.
3. Check status of collection, download data, and kill machines on completion.
   1. Run bash core/scripts/gcp-collection/fleet-status.sh to check on status and download the data.
   2. Run bash core/scripts/gcp-collection/fleet-status.sh --kill to check on the status and terminate finished instances.
4. Analyze the results
   1. Bespoke for now :)
   2. Example analysis

Future Work

• Collect a more comprehensive URL set for urls.txt
• Autodetermine the instance collect instance by n+1 the existing instances
• Add analysis scripts (similar to preprocess.sh and analyze-ab-test.js)