Open dirkmc opened 2 years ago
Started looking into this a bit more as it's been popping up pretty frequently. The interesting bit here is that the commp mismatches are showing expected and actual commp from the other deals, so it's not an incorrect computation. I expect that the wires are getting crossed in the test somewhere and we're seeing some collision in data prep.
https://app.circleci.com/pipelines/github/filecoin-project/boost/2470/workflows/fc0e1c49-b324-4c8e-ac3d-0bb18f692b49/jobs/21387/parallel-runs/0/steps/0-114