What were some of the pain points you face(d) - looking back at your Metaflow adoption?
Disclaimer: I work in Netflix ML Platform that helped open-source Metaflow originally.
We've run into some issues with getting AWS Batch to play nicely, though I wouldn't say it is Metaflow specific. Initially we did quite a bit of troubleshooting the Stuck in RUNNABLE errors. We sometimes have issues with batch jobs that can't be satisfied by our compute environment causing other jobs in a queue to be blocked.
There are other small issues, but overall our ML engineers are very happy with it as a tool.
There are other small issues, but overall our ML engineers are very happy with it as a tool.