We've identified the problem and rolled out a fix.

Runners have been intermittently failing on network issue or I/O errors. We're currently looking into the issue.