Rebase

Nits

blk

py27

Fixup py27

Up

sounds like we are going to not try to paper over the differences in priority handling between celery, multiprocess[, k8s, ...] -- so let's revert the tagging change in the celery tests and reinstitute the sorting step in dagster-celery

if we do this, I'd like to do a deprecation process where in 0.7.x it emits warnings before going away in 0.8.0

Fixup

py23

Blk

Up

Up

Black

Fixup job spec

Test spec

ok

Rebase

i can split them into a separate diff but would like to release both

Use package_data

Up

Up

Up

Up

Up

Up

super, see comment

Migration guide?

ah great. I vote for Noneable

Omittable?

There is some potential for confusion with both Noneable and Nullable if these are meant to imply that a value can be omitted (vs. explicitly set to Python None or to YAML/JSON null

Fixup

Set Dockerhub creds

Up

Up

blk

Up

This makes me want to add mypy annotations to the public API. I know it's cumbersome.

yes

but @prha can still repro

Rebase

Rebase

can't repro @prha's backfill issue -- including before the WAL change on the event logs

i played with the isolation level a little bit but couldn't identify an impact

Explicitly test metadata

Tests

Tst

Up

Up

Rebase

blk

resolves the issue afaict

hmm. these are (runMetadata.steps[step.key] || EMPTY_STEP_METADATA).expectationResults etc., which get provided by the RunMetadataProvider. It sure seems to me like extractMetadataFromLogs should execute before the component renders, and this should never be undefined, but I can reliably reproduce.

Up

Fixup

Up

Up