The human-centric platform for production ML & AI

Outerbounds

Hi guys, What happens if metaflow ‘metadata service’ is down while running a workflow? When it gets back up again. Will it be able to read the metadata of the previous steps?

there a built-in retry policy that takes care of downtime lasting &lt;30s or so. Hence it is likely that just rebooting the service won't have any visible impact, or a temporary network glitch

if you add `@retry` to your step (or run/deploy `--with retry`), it'll be even more resilient, handling minutes of downtime

beyond that, `resume` will continue from the last successful step, so yes, it'll be able to read the metadata of all steps that succeeded prior to the downtime