Closed ShyamsundarR closed 2 months ago
The expected use case is to refresh the cache from another process, so actual job should never need to refresh the cache. But I think we can avoid the race by using unique temporary file so 2 concurrent fetches will not break each other.
e.g The following run failed stating .tmp file was not found (occurred twice in about 6-7 runs). Looking at the code, as both runners use the same tmp file location it is possible that on a race one of the runners deleted the tmp file and later the other runner attempted to rename the same tmp file and failed:
The env YAML used was: