Closed jrbourbeau closed 1 year ago
Digging a bit with the CLI issue..
After installing dask
and distributed
with conda-forge
If I run python -m dask
...
$ python -m dask
shows what we want:
Usage: python -m dask [OPTIONS] COMMAND [ARGS]...
Dask command line interface.
Options:
--version Show the version and exit.
-h, --help Show this message and exit.
Commands:
docs Open Dask documentation (https://docs.dask.org/) in a web browser.
info Information about your dask installation.
scheduler Launch a distributed scheduler.
ssh Launch a distributed cluster over SSH.
worker Launch a distributed worker attached to an existing SCHEDULER.
but
$ dask
shows
Usage: dask [OPTIONS] COMMAND [ARGS]...
Dask command line interface.
Options:
--version Show the version and exit.
-h, --help Show this message and exit.
Commands:
docs Open Dask documentation (https://docs.dask.org/) in a web browser.
info Information about your dask installation.
$ python -m dask
is running dask.__main__:main
while $ dask
is running dask.cli:cli
. (note the pip(setuptools) installation uses dask.__main__:main
). I think the conda-forge entry point should be dask.__main__:main
Opened a PR conda-forge/dask-feedstock#200 fixed in dask-core-feedstock!
Thanks all for helping dig into the CLI issues! 🙏
Given the pain around that, would like to propose hardening the Click requirement in Dask going forward ( https://github.com/dask/dask/pull/9595 ). Ideally would like to include that in this release
I've opened this issue around requiring python<3.11
https://github.com/dask/dask/issues/9605. If folks have thoughts on the topic I'd appreciate feedback on that issue.
We haven't quite gotten https://github.com/dask/distributed/pull/5648 over the finish line yet, so I'm going to push up PRs to put an upper bound of bokeh < 3
for dask
and distributed
given that we know there are compatibility issues with the bokeh=3
release scheduled to come out later today (cc @fjetter @crusaderky)
@gjoseph92 @fjetter are we turning on task queuing on the scheduler by default in this release?
No, it won't be ready in time. See https://github.com/dask/distributed/issues/7213.
bokeh < 3
PRs:
@jrbourbeau not critical, but could be good to get https://github.com/dask/distributed/pull/7224 in if CI passes and you have time to review.
Ah, sorry @gjoseph92 I need to start pushing out the release. Unfortunately I think that one will have to get in in the next release
No worries, queuing is still behind a feature flag so it's not too big of a deal.
2022.10.1 is out on PyPI and conda-forge PRs are open:
Alright, PyPI and conda-forge packages are released.
We ran into a problem building the Python 3.8 dask-notebook
image over in dask-docker
(xref https://github.com/dask/dask-docker/pull/255) due to an issue in the underlying jupyter/base-notebook
notebook (xref https://github.com/jupyter/docker-stacks/issues/1810). I've attempted a couple of possible fixes, but haven't gotten those build to work yet. I've proposed a temporary workaround in https://github.com/dask/dask-docker/pull/256 where we drop Python 3.8 builds. This is a possible workaround, but admittedly feels a bit heavy handed. I'm on PTO next week -- @jacobtomlinson @jakirkham @charlesbluca could I ask you to handle docker images for this release by either finding an actual fix, or by merging the PR that temporarily drops Python 3.8 builds?
FWIW commented on the upstream (Jupyter) issue.
Heads up: https://github.com/dask/dask/issues/9609
Does anyone here have time to investigate what happened here with Bokeh? It appears that we're affecting Airflow negatively.
On Sat, Oct 29, 2022 at 3:53 AM jakirkham @.***> wrote:
FWIW commented on the upstream (Jupyter) issue.
— Reply to this email directly, view it on GitHub https://github.com/dask/community/issues/283#issuecomment-1295775628, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACKZTFYPEGQJAQRJA6DSWLWFTQYXANCNFSM6AAAAAAROE5LGY . You are receiving this because you commented.Message ID: @.***>
Pushing out a hotfix https://github.com/dask/community/issues/284
If there is anything I should know folks should jump in very soon.
There isn't anything I'm aware of that would prevent a hotfix, though dask/distributed#7224 is a non-trivial change. Perhaps @gjoseph92 can comment on whether it's likely to be an issue (I suspect he'd be fine with it)
I think @gjoseph92 is comfortable including that change has he asked for it to be included in the 2022.10.1
release, but it just didn't make it in time (xref https://github.com/dask/community/issues/283#issuecomment-1295438871)
EDIT: queuing is also still behind a feature flag, so I think it's safe to move forward with that change included
I've already chatted with @gjoseph92 and he confirmed.
I've yanked this release from PyPI. Can someone (maybe @jakirkham) help with yanking from conda-forge channels?
could I ask you to handle docker images for this release by either finding an actual fix, or by merging the PR that temporarily drops Python 3.8 builds?
Any opposition to unblocking the 3.8 Docker builds locally with https://github.com/dask/dask-docker/pull/257 until the upstream issue is resolved? This would involve patching the impacted Jupyter dockerfile, which could get hairy in the long run, but it doesn't seem like CI is running very often on dask-docker for this to be an issue.
Would follow these steps for marking packages broken in conda-forge. Unfortunately I don't have bandwidth for this atm
Thanks @jakirkham! Opened up https://github.com/conda-forge/admin-requests/pull/504 to request the packages be marked as broken
As to Jupyter docker-stacks, outlined a path forward in comment ( https://github.com/jupyter/docker-stacks/issues/1810#issuecomment-1297546767 ). If anyone has bandwidth to try, should be a couple line change
Updating this thread, Charles' workaround was merged yesterday. Mamba released 1.0.0 early today. This subsequently fixed Jupyter docker-stacks ( https://github.com/jupyter/docker-stacks/issues/1810#issuecomment-1300850292 ). With that fixed, Charles reverted the workaround ( https://github.com/dask/dask-docker/pull/260 ).
Edit: Also merged Charles PR yesterday to mark conda-forge packages broken.
Sounds like there is nothing more to do here. So going to close this out, but please feel free to reopen if I've missed something.
Thanks all! 🙏
Best effort
Try to close before the release but will not block the release
Blocker
Issues that would cause us to block and postpone the release if not fixed
bokeh=3
compatibility -- see the PRs linked in https://github.com/dask/community/issues/283#issuecomment-1295133553)Comments
cc @ian-r-rose @fjetter @quasiben @jakirkham