Stale issue / PR sprint

[jrbourbeau]

As discussed in the maintainers meeting today, it's a good time for another stale issue / PR sprint. I'll propose next Thursday, October 14, from 9 AM - 1 PM CT in the Dask whereby channel. This should hopefully provide a large enough windows for folks (both US and Europe based) to come and go as they please / their schedule allows.

cc @jcrist @ncclementi @jsignell @quasiben @jacobtomlinson @douglasdavis who expressed an interest in the sprint. Does this time window work for you?

[GenevieveBuckley]

BTW, I'd also be happy to chip in towards this effort.

My timezone is UTC+11, so that works for a UK/Euro morning, or US folks who might want to have an evening round at it. I get it if that's not convenient for anyone else.

Also, if there's anything asynchronous I can do to help, let me know. Maybe sorting through old issues/PRs and tagging them, so other people can spend more effort on reviewing? I don't currently have permissions to do that now, but I'm sure we could get that worked out if there was a specific task in mind.

[jsignell]

Oooh let's add @GenevieveBuckley to the triage team

[jrbourbeau]

+1! @GenevieveBuckley I just added you to the @dask/triage team. You should be able to re-open, close, and label issues/PRs in dask/dask and dask/distributed now

[GenevieveBuckley]

Thanks!
Ok, if I go through old issues/PRs, what do you want me to tag? I could go through and label everything older than a certain date as "stale", but that may or may not be the most helpful thing to do...

[fjetter]

FYI there is an open PR to enable a GH actions to close stale PRs dask/distributed#5402 I think we should open that discussion again, see also #60

[jsignell]

if I go through old issues/PRs, what do you want me to tag?

I think needs-info, good-*-issue and the regular tags would be useful. Sometimes when you stare at the issue tracker for long enough you can also spot duplicates.

[GenevieveBuckley]

Ok, I've been working my way backwards from the oldest open issues. I'm up to the start of August 2019.

So far I've:

• Closed 15 issues
• Found 17 more that might be issues we can close. I'll share this list a little later on. Right now I've pinged some of the original authors, so things might change a bit.
• Added good first/second issue to 8 issues, and "needs info" to a few others (not sure how many)

There several places that might be good for easy wins in our sprint:

1. Change default quantiles implementation to use tdigest dask#6566 (corresponding issue dask.dataframe quantile fails spectacularly in some edge cases dask#731)
2. Add item method to Dask Array dask#3630 (corresponding issue Implementing item method for Dask Arrays dask#2959)
3. Remove sep from fastparquet calls dask#8206 (corresponding issue Correct fastparquet call dask#8201)
4. Warn user when tokenize function is slow dask#5631 (corresponding issue Better educate users when hashing/tokenizing large numpy arrays dask#4275)
5. RTD config file in the repo dask#2568 (this needs a maintainer to do it, someone with permissions to the dask read the docs instance)
6. (Documentation update) Helm dask notes seem to have a typo in setting the environment variable dask#4253
7. (Documentation update) map_partitions tries to partition a pd.DataFrame given as argument to a mapped function dask#2807 (comment)
8. List optional cytoolz dependency dask#2812

[GenevieveBuckley]

Closed another 13 issues today.

Here are some more places to look for potential easy wins:
9. dask/dask#8221 (corresponding issue dask/dask#5865)
10. dask/dask#6456 (corresponding issue dask/dask#5695)
11. dask/dask#6344 (corresponding issue dask/dask#6315)
12. dask/dask#6276 (corresponding issue dask/dask#6275)
13. dask/dask#6627 (corresponding issue dask/dask#6161)

I'm up to November 2020 in dask's open issues, working towards newer issues (haven't looked at open PRs or dask/distributed just yet).

[GenevieveBuckley]

Closed another 7 issues today (that's 35 in total 🎉).

I've finished looking through all the open dask issues. I have not looked at the distributed issues, or the dask or distributed PR backlogs.

Again, I found a bunch more places the maintainer team can look at for easy wins in this stale issue/PR sprint. They are:
14. dask/dask#6768
15. dask/dask#8259 (corresponding issue dask/dask#6808)
16. dask/dask#7167
17. dask/dask#7688 (corresponding issue dask/dask#7647)
18. dask/dask#8138 (it doesn't close the related issue, this is more of an in the meantime, raise and exception situation)

I also have a list of issues that maybe could be closed, but need some kind of discussion or interaction before that happens. Happy to share that too, but I think it makes sense to look at that list of possible easy wins first.

[GenevieveBuckley]

Update: closed 24 issues in the dask/distributed repository today (up to October 2017, working forwards through time).

I don't have a list of potential easy wins for you this time, most distributed stuff seems fairly complex. I did label a few "good second issue", but I'm just guessing. I have not been thoroughly labeling everything like I did in the dask repo (but I have asked Jacob if he can make a "discussion" label).

I'm running out of steam a little bit, so I might take a break from looking through old issues, depending on how I feel.

[GenevieveBuckley]

Update: closed another 25 issues in the dask/distributed repository today (up to 10th November 2018, working forwards through time). I find the distributed issues harder to skim than issues in dask/dask, so it's much slower going and I might easily be missing stuff.

Potential easy wins for the sprint in dask/distributed:

• added timestamp to log lines by default distributed#3987 (corresponding issue Logging formatting distributed#1650)
• ENH: Close Adaptive during Cluster shutdown distributed#5093 (corresponding issue Option to shutdown adaptive cluster distributed#1898)
• Better error for missing modules during deserialization distributed#5419 (corresponding issue Cryptic error when package not installed on worker distributed#2230)
• pytest collection errors after following setup guidelines distributed#5212

[GenevieveBuckley]

Update: closed 19 issues today (up to 25th June 2019, working forwards through time in the dask/distributed open issues).

Potential easy wins:

• Remove --nprocs keyword from dask-worker command? distributed#2471

[GenevieveBuckley]

Update: closed 12 issues over the last two days.

Found another possible easy-ish win:

• Use moving average to reset rectangles in scheduler dashboard distributed#3190 (corresponding issue Scheduler dashboard status showing one big square instead of zoomed out view distributed#3130)

I'm up to Jan/Feb 2020 in the dask/distributed open issues, so everything newer than that, I haven't looked at. I'm going to stop here, I think we're now at the point of diminishing returns. It makes more sense to focus maintenance efforts on the PR backlog now.

I might be able to pair with someone to go through PRs. The napari project is doing something similar next week.

[GenevieveBuckley]

Here's another one for the list of potential easy wins:

• Add dask.ml module dask#6384