Hey everyone 🙂
I need help getting two fixes for 1.14 out which affect the kf PyTorch plugin and which were unfortunately introduced while implementing the “distributed error aggregation” RFC
https://github.com/flyteorg/flyte/pull/5598, see also
https://github.com/flyteorg/flyte/pull/5741. I apologise for only discovering this now, we tested this locally with k3d and these issues turned out to only appear in combination with google cloud storage:
• “List objects” google cloud storage permission is required for flytepropeller in order to list error files. I add this to the gcp terraform module in union-oss/deploy-flyte in
this PR and propose to either add this as a required change to the “Breaking” section in the 1.14 release notes or make it an opt-in feature like
this.
• The
List
function of
StowStorage
needs a
fix for google cloud storage. Would it be possible to make a patch release for this?
Again, sorry that I discover this only now.