Even after the 6 vcpu gcp bump, we seem to often have corrupted blocks when we try to load in promecieus, attempting recently I found a good 4/5 jobs even successful ones, on both 4.14 and 4.13, appear to have corrupted prom data.

We grab this data with this manner:

https://github.com/openshift/release/blob/master/ci-operator/step-registry/gather/extra/gather-extra-commands.sh#L238-L239

Is this a safe operation or are we doing something silly causing these corrupted archives?

May want to loop in monitoring team for help here.

Could be some correlation to CPU use but unclear, do aws archives tend to work better?