In our old approach, a workflow file contains a job that uploads the PR number as an artifact. While the PR is still open, the workflow_run triggered by it will download the artifact and use the information to clean up all except the last used cache associated with that original workflow. When a PR is merged or closed, there will be a post-pr workflow that uploads the PR number as an artifact and triggers a workflow_run that clean up all caches associated with the PR. The reason we did it this way was in the cache cleanup workflows, we did not find an easy way to get the number of the PR triggering them. This is not convenient because we have to add jobs uploading artifacts to workflow files.
After some experiments, we have found a reliable way to find the PR number without using artifacts. The workflow_run's payload always contains the head SHA of the commit that triggers it, whether the PR comes from a fork or not. We can then use gh pr list to search for that head and obtain the PR number.
In our old approach, a workflow file contains a job that uploads the PR number as an artifact. While the PR is still open, the workflow_run triggered by it will download the artifact and use the information to clean up all except the last used cache associated with that original workflow. When a PR is merged or closed, there will be a post-pr workflow that uploads the PR number as an artifact and triggers a workflow_run that clean up all caches associated with the PR. The reason we did it this way was in the cache cleanup workflows, we did not find an easy way to get the number of the PR triggering them. This is not convenient because we have to add jobs uploading artifacts to workflow files.
After some experiments, we have found a reliable way to find the PR number without using artifacts. The workflow_run's payload always contains the head SHA of the commit that triggers it, whether the PR comes from a fork or not. We can then use
gh pr listto search for that head and obtain the PR number.