defer to @catherinewu for final accept to make sure docs are aligned
Jan 11 2021
woof those times are insane
add test case to show re-use of run_id fails with good error
I think this is in a slightly confusing state, I would guess due to how it started vs the latest approach. Sending to you for one more spin for some mix of more tests or renaming / rewording
Jan 9 2021
more clean up
pylint hax engage
Jan 8 2021
I have a good feeling that this will feel better all pulled in to one spot
comments inline - I think this will be much clearer if all of this complexity is brought in to one spot with some comments
i take it the new defaults will be in a followup diff?
based on our earlier conversation it seems we may punt on this for release due to the complexity around:
- type based required resource keys
- interactions with default value loading
to your q
rebase on fixes
some fixes, still looking in to lint
ooof this sucks
What if we added a map method to InvokedSolidOutputHandle that raises a NotImplementedError?
Jan 7 2021
can we add some magic to help with the lint warnings?
forEach -> map
If we have map, do we also need forEach?
can you clarify what you want feedback on at this stage? if there isn't any RFCish stuff left prep this for final review
im still worried about keeping these config fields up to date - but ill let you make the final call
Jan 6 2021
rebase, adding step_handle to DagsterEvent shown in D5755
Jan 5 2021
the full custom alternative path is less of a concern since that is unlikely - but I do worry that the config setup does not actually work if attempted to be loaded given the provided test plan
is there an escape hatch here for providing a custom one?
take a note in some form to test this once we publish a prerelease
to your queue post discussion, summary was roughly trying to simplify things a bit by not having overlapping responsibilities between input and object managers, and moving towards a RootInputManager set up where it is only responsible for root loads
ill defer to @max to make the final call
clearing queue for 0.10.0 burn down - request review post release if there is more discussion this can generate
to your queue
Do we need this in for 0.10.0? If yes we should figure out a path forward, likely punting on solving the name issue. If not lets revisit post release
Jan 4 2021
alright thats an acceptable enough base line i think - and chances are it will improve a little before launch
We should think about what kind of affordances we might want to put into dagit that prompt people to go deploy the daemon though.
from what i can tell there are no differences to this approach besides the additional typing information on __annotations__
Dec 22 2020
some more test call sites to patch up but this gives a gist of what this could look like
does the our existing mypy check in buildkite exercise this? should we add a step?
I am pretty sure this is legit given how a lot of the dagstermill impl predates so much of the shifting that has happened around the core execution pathway but I will defer to @max to be sure