continuously updating data about metaflow HOT 5 CLOSED

@andaag We do have similar use-cases at Netflix. The recommendation is that you should parameterize your flow to execute on a list of days and some other external system should be able to trigger your workflow with the appropriate list of days you should be processing upon. At Netflix, our production scheduler Meson supports event based triggers which plays elegantly with this pattern. Another, not so elegant, alternative is to embed a light-wieght check before your self.next call to filter out done days.

from metaflow.

How do you then easily join that data? I'm guessing you have one parameterized flow that produces the data, and a separate job that queries all the old versions of this and joins it using the client api?

from metaflow.

Yes. Or if it were internal they leverage the hosting service - see open issues - which makes accessing any specific dateint depending on what we’re deployed (has a versioned REST endpoint) or promoted to be the latest.
The client API will help you - if you want to join all the data for your use case. In the internal ones; they are usually interested in the latest - when running in a cron like fashion.

from metaflow.

@andaag Not sure I understood your question around joining the data appropriately. According to my interpretation, you can use the client API in a separate job (or even in the same flow as a separate step) to look up all the dates that haven't been successfully processed yet (might have to take care of current inflight dates being processed by another instantiation of your workflow appropriately) and trigger your workflow with those dates as parameters (or use those dates as foreach arg if embedding this logic as a step in your flow). Let me know if I ended up answering something else entirely.

from metaflow.

Using the client api + foreach sounds fairly sane 👍. A bit manual compared to the built in support for split/join, but very manageable.

My question was just to doublecheck if there was a more elegant way of doing this.

from metaflow.