https://pantsbuild.org/ logo
Join Slack
Powered by
Hi, I'm having some trouble with generate-lockfile...
# general
a
Hi, I'm having some trouble with generate-lockfiles hanging forever and leaving a process using 100% of a core in the background after I cancel the goal. Is there something I could be doing wrong to cause this or should I raise a bug?
e
Can you please provide both 
./pants -V
and the full output of 
./pants -ldebug generate-lockfiles
a
Untitled.txt
Untitled.txt
I've used a few different version 2.12-14 of pants to try to get it to work but currently its 2.14.0.dev5, the debug contained a api key for a private cheeseshop I had to remove
I also get the same issue on two machines one under wsl and one on ubuntu
e
Thanks for the details. I'm a bit confused. The log above ends with lock completed lines. Are you unhappy with the ~85 seconds it took?
a
There are two resolves one finishes the other hangs forever
e
Aha, the 
api
one. Ok.
Can you confirm - you said it above, but to be sure, that the 
api
resolve hang also happens on non-WSL? We went through the exercise yesterday of learning WSL mounts windows drives using 9p which is a network filesystem that does not support filesystem events and perhaps does not support file locks. If it doesn't support file locks then that would foil Pex - which is doing the locking here - and could present as a hang.
a
Yeah I noticed it first on ubuntu and then recreated it on WSL later on
e
And it was Ubuntu on-the-metal? I.E.: The Pants project was on an ext4 or similar filesystem?
a
Well zfs but yeah on the metal
Although ZFS can be a bit weird so maybe that is something
e
ZFS is a local filesystem so it almost certainly supports POSIX locks. Ok. Let me try your two resolves there in the debug output. It's not clear which is 
api
and which is 
alert.
a
I can get you a log of just the failing one if it makes it easier
e
That would be great actually.
a
Untitled.txt
Untitled.txt
e
Ok, great - that's the one I had started. I had to comment out 
proto-bindings==0.1.0
since that's on your private PyPI, but then I get a hang + CPU repro.
Pex should be logging these lines from Pip but is not:
Copy code
$ grep "pip is looking at multiple versions of" /tmp/tmpstuakkbupex-pip-log./pip.log
2022-08-18T07:30:06,421 INFO: pip is looking at multiple versions of click-repl to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:06,421 INFO: pip is looking at multiple versions of click-plugins to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:06,421 INFO: pip is looking at multiple versions of click-didyoumean to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:06,611 INFO: pip is looking at multiple versions of click to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:06,743 INFO: pip is looking at multiple versions of click-repl to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:06,744 INFO: pip is looking at multiple versions of click-plugins to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:06,744 INFO: pip is looking at multiple versions of click-didyoumean to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:07,200 INFO: pip is looking at multiple versions of cheroot to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:07,544 INFO: pip is looking at multiple versions of click to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:12,347 INFO: pip is looking at multiple versions of cheroot to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:30:16,008 INFO: pip is looking at multiple versions of charset-normalizer to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:31:26,353 INFO: pip is looking at multiple versions of charset-normalizer to determine which version is compatible with other requirements. This could take a while.
2022-08-18T07:32:26,952 INFO: pip is looking at multiple versions of certifi to determine which version is compatible with other requirements. This could take a while.
a
Cool glad you get repro without the private package
Yeah from the terminal I run there is no feedback other than the time even when I set the level to trace
e
So this looks like a case of just straight up slow backtracking to find a solution and the backtracking is hitting sdists it has to build along the way.
The Python ecosystem is perverse in that you often have to build an sdist to get metadata to continue the resolve walk,
a
Hmm I guess no easy solution (no pun) then
It does resolve with poetry (These are all based on poetry pyprojects) after a couple of minutes but I guess that is a very unhelpful thing to say haha
e
The solution is Pex upgrading its vendored Pip from 20.3.4 to 22.2.2 (latest). I've been working hard on that and just tried in that branch with a wiped cache:
Copy code
$ rm -rf ~/.pex && time python -mpex.cli lock create --output=lock.json --style=universal --resolver-version pip-2020-resolver --target-system linux --target-system mac --indent=2 --manylinux manylinux2014 --interpreter-constraint "CPython==3.9.*" -r ~/Downloads/requirements.txt

real	1m2.519s
user	0m54.418s
sys	0m3.164s
I'm very close to done, just adding tests and forking out PRs. You can follow along here but I expect Pex 2.1.104 with the Pip upgrade will be released by this weekend and Pants mainline upgraded by Monday: https://github.com/pantsbuild/pex/issues/1805
🎉 2
a
Thanks, that's super helpful.
e
You're welcome. I'll be happy to have Pex using the latest Pip again and going forward. Pip has a good deal of problems, but they also fix the important ones pretty promptly.
a
I'm fairly new to python (More of a Rust guy these days) and I am really surprised by the complexity around having a monorepo setup that packages stuff into docker containers. The fact that you have to create sdists to get metadata out is mindblowing to me. So I am very thankful people like you are working on making it easier.
e
To be fair Rust and Go and these languages supported by big corps with alot of time, money and hindsight have invested in shipping tools with the language up front. Guido & Co. were just trying to ship the interpreter and all else was cobbled as time went by; so 3rdparty handling is a jungle as a result. Ditto Java, although they had less excuse and had big $ available.
a
Oh yeah totally, I'm not criticising I'm just surprised. I've had to setup too many c++ projects, I know how hard these things could be just to get one or two deps working haha
e
Yeah, since Python has supported c extensions from day 1, and thus the variety of builds could be unbounded, the wisdom was roughly, let the user figure it out if they can, don't straight jacket them; so ship a system where you execute a setup.py file. Since that file is Python itself, and not declarative, you really have to execute it to get a dependency list since some of those in the wild actually have conditional logic.
And that was sometime in the early 2000s - it just snowballed from there.
So Pip has a fairly uniquely hard problem to deal with. Cargo / Rust got it right from ~day 1.
b
Reusing this thread to explain my requirement a bit more: I want to generate stubs from proto files in my repo and then package and publish the generated files as a python library. My questions are: 1. What is the goal to generate stubs? 2. Is it necessary to have the grpcio dependencies in 3rdparty/ python folder? I have the dependencies listed in another directory in the same repo where I am going to be importing the generated stubs. I am first building a conda environment which has those requirements fulfilled and then generating the stubs. I am doing all of this in my local machine right now but would want to integrate this entire process in my CI once I get this working.
h
Hey @bored-glass-58755, happy to help, although it would probably be better to have a dedicated thread because those questions don't quite relate to this thread about 
generate-lockfiles
hanging
b
Sorry I used the wrong thread by mistake thinking it was one of my other threads.
👍 1
a
I see the pex PR has gone in last week, is there a timeline for when that change will be picked up in a pants release? Looking at previous releases I imagine another week or two?
e
No timeline currently. You caught most of us in the middle of a sprint on docker command runner support. See the Pex release announcement for more: https://pantsbuild.slack.com/archives/C18RRR4JK/p1661713160748249?thread_ts=1661713160.748249&cid=C18RRR4JK
👍 1
8 Views
Open in Slack
PreviousNext
Powered by