For auto-materialize max materializations per minu...
# dagster-feedback
c
For auto-materialize max materializations per minute, I'd probably be looking at longer op "cooldowns" on the order of 15 minutes or an hour. I might implement something like that with RetryRequested exceptions myself in the mean time.
o
Hi @Chris Comeau! Definitely a reasonable request -- is it safe to assume you're mostly working with unpartitioned assets? The key friction when designing the rate limiting behavior was that the biggest thing we wanted to avoid was a small change unexpectedly causing hundreds of partitioned runs to be launched within a single minute, so this was the case we optimized the API for (it's easy to say "you can launch up to 5 materializations per minute"). However, in the unpartitioned case, you'd rather set something in terms of a period, rather than a frequency (i.e. "1 materialization every 15 minutes"). I wonder if it might make sense to allow you to set this value to a float (at least for unpartitioned assets), i.e.
max_materializations_per_minute=1/15
. Do you have thoughts along those lines?
c
Right, it's for limiting load from coarse unpartitioned table refreshes. Allowing a float there would be fine