Windows Python wheels: Bad address src\epoll.cpp:100

[zhanggyarcher]

Hello, recently I set up a new python environment on a new computer as usual, I got automatically installed pyzmq==21.0.2, and I cannot use jupyter notebook, Jupyter lab, vs code at all, the python kernel crashes every time when I open a editor, and after one second, it directly crashes and lose connection to the kernel.

This problem annoyed me for several weeks, and today I saw many people encountered this issue From December 2020 to January 2021. And I have never encountered this issue in the past.

Also, this error message is somewhat not useful and not making sense, finally I realized its pyzmq error.
I degraded it to pyzmq==19.0.2 , it finally resolved this issue.

I hope this could be fixed, or if you could push old versions of pyzmq for users to install as dependencies.
Thank you very much!

[minrk]

Can you test with current pyzmq 22.0.2? As far as I am aware, all of these issues are fixed in the latest release.

pyzmq has no control over error messages from the C runtime thrown by the underlying libzmq.

[zhanggyarcher]

Can you test with current pyzmq 22.0.2? As far as I am aware, all of these issues are fixed in the latest release.

pyzmq has no control over error messages from the C runtime thrown by the underlying libzmq.

Sure, I will have a look and let you know. But could you stop letting them push 21.0.2 as dependencies? Thanks!

[minrk]

But could you stop letting them push 21.0.2 as dependencies?

Who are 'they'? If you pip install pyzmq you should get 22.0.2. That's all I have control over.

[zhanggyarcher]

But could you stop letting them push 21.0.2 as dependencies?

Who are 'they'? If you pip install pyzmq you should get 22.0.2. That's all I have control over.

When I install pyzmq==22.0.2, it still has this crash problem. Then I reverted to 19.0.2, the problem resolved. Everything is working fine.

When I check dependency list, ipykernel is calling to install pyzmq, when user is installing jupyter or jupyter-lab, pyzmq==21.0.2 is automatically installed, so I hope you could let them to install pyzmq==19.0.2, not the later version.
Thank you!

[minrk]

Thank you for testing and confirming this is an issue in the current version!

Can you run:

python3 -m pip install pytest pyzmq==22.0.2
python3 -m pytest -v -m "not wheel and not new_console" --pyargs zmq.tests

and share the output?

Also, can you share some more information about your environment, so I can learn a bit about what makes your environment different from all of our tests?

• Windows Version
• Python version, and how it is installed (from Python.org, conda-forge, etc.)

@bluca any hints as to why this epoll_ctl assert would fail on Windows? This is using the libzmq-v141-x64-4_3_4 artifact on bintray.

so I hope you could let them to install pyzmq==19.0.2, not the later version.

I'm sorry, I really don't know what you mean by this. Anyone is free to install any version, so this is already the case.

[bluca]

• Windows Version
• Python version, and how it is installed (from Python.org, conda-forge, etc.)

@bluca any hints as to why this epoll_ctl assert would fail on Windows? This is using the libzmq-v141-x64-4_3_4 artifact on bintray.

Sorry, don't know - maybe @sigiesec can help

[zhanggyarcher]

@minrk Sorry for the late reply, I will try my best to test these on a new computer as soon as I can(to avoid damaging my environment).
For this issue, I encontered on Windows 10 1909, 20H2, python 3.6.8, 3.7.9, 3.8.6, 3.8.7, so this is not only the issue for python 3.6. So I am not sure the modified title for this issue is suitable or not. And I tried all these versions and I reinstalled windows10 and the python environment more than 20 times, this issue always appear.
I installed python 3.6.8 or other versions via windows x86_64 installer, and then I use pip to install these packages sequentially: tensorflow, tensorflow-gpu, matplotlib, scipy,scikit-learn,torch, jupyterlab. As what I said above, the jupyter need ipykernel, and ipykernel needs pyzmq. When I install jupyterlab or jupyter-notebook, it automatically install ipykernel and automatically install pyzmq=21.0.2, I have no choice. And many users don't know they installed pyzmq, and they never know the version. I manually degraded to pyzmq=19.0.2, it solves the issue.
My friend tried to use anaconda to help me to install the environment, it is ok to open the jupyter to run "hello world", but the tensorflow will fail, it's weird and it's another story.

Thank you!

All the package versions are listed below.
`
PS C:\Windows\system32> pip list
Package Version

---

absl-py 0.11.0
anyio 2.1.0
argon2-cffi 20.1.0
astor 0.8.1
astunparse 1.6.3
async-generator 1.10
attrs 20.3.0
backcall 0.2.0
bleach 3.3.0
cached-property 1.5.2
cachetools 4.2.1
certifi 2020.12.5
cffi 1.14.4
chardet 4.0.0
colorama 0.4.4
contextvars 2.4
cryptography 3.2.1
cycler 0.10.0
dataclasses 0.8
decorator 4.4.2
defusedxml 0.6.0
entrypoints 0.3
flatbuffers 1.12
future 0.18.2
gast 0.3.3
google-auth 1.25.0
google-auth-oauthlib 0.4.2
google-pasta 0.2.0
grpcio 1.32.0
h5py 2.10.0
idna 2.10
immutables 0.15
importlib-metadata 3.4.0
ipykernel 5.4.3
ipython 7.16.1
ipython-genutils 0.2.0
jedi 0.18.0
Jinja2 2.11.3
json5 0.9.5
jsonschema 3.2.0
jupyter-client 6.1.11
jupyter-core 4.7.1
jupyter-kite 1.2.0
jupyter-server 1.3.0
jupyterlab 2.2.9
jupyterlab-pygments 0.1.2
jupyterlab-server 1.2.0
Keras-Applications 1.0.8
Keras-Preprocessing 1.1.2
kiwisolver 1.3.1
Markdown 3.3.3
MarkupSafe 1.1.1
matplotlib 3.3.4
mistune 0.8.4
mysqlclient 2.0.3
nbclient 0.5.2
nbconvert 6.0.7
nbformat 5.1.2
nest-asyncio 1.5.1
nodejs 0.1.1
notebook 6.2.0
npm 0.1.1
numpy 1.19.5
oauthlib 3.1.0
opt-einsum 3.3.0
optional-django 0.1.0
packaging 20.9
pandas 1.1.5
pandocfilters 1.4.3
parso 0.8.1
pickleshare 0.7.5
Pillow 8.1.0
pip 18.1
pipdeptree 2.0.0
prometheus-client 0.9.0
prompt-toolkit 3.0.14
protobuf 3.14.0
pyasn1 0.4.8
pyasn1-modules 0.2.8
pycparser 2.20
Pygments 2.7.4
pyparsing 2.4.7
pyrsistent 0.17.3
python-dateutil 2.8.1
pytz 2021.1
pywin32 300
pywinpty 0.5.7
pyzmq 19.0.2
requests 2.25.1
requests-oauthlib 1.3.0
rsa 4.7
Send2Trash 1.5.0
setuptools 53.0.0
six 1.15.0
sniffio 1.2.0
tb-nightly 2.5.0a20210209
tensorboard 2.4.1
tensorboard-plugin-wit 1.8.0
tensorflow 2.4.1
tensorflow-estimator 2.4.0
tensorflow-gpu 2.4.1
termcolor 1.1.0
terminado 0.9.2
testpath 0.4.4
tf-estimator-nightly 2.5.0.dev2021020901
torch 1.7.0+cpu
torchaudio 0.7.0
torchvision 0.8.1+cpu
tornado 6.1
traitlets 4.3.3
typing-extensions 3.7.4.3
urllib3 1.24.3
wcwidth 0.2.5
webencodings 0.5.1
Werkzeug 1.0.1
wheel 0.36.2
wrapt 1.12.1
zipp 3.4.0
You are using pip version 18.1, however version 21.0.1 is available.
You should consider upgrading via the 'python -m pip install --upgrade pip' command.
`

[minrk]

Thanks for the info! I'll see if I can figure out how to debug it. Since it does not occur on my Windows VMs, I don't really feel like I have any actions available at the moment.

Were you able to run pyzmq tests with the commands I gave above? That might help narrow down the issue.

When I install jupyterlab or jupyter-notebook, it automatically install ipykernel and automatically install pyzmq=21.0.2, I have no choice

You do have a choice. You can always specify pinning of any package in the install command. For example:

pip install jupyterlab 'pyzmq<20'

ipykernel does not request 'pyzmq==22.0.2', it requests 'pyzmq' which gets the latest version by default. If you want to specify a different version, you can do that.

My friend tried to use anaconda to help me to install the environment

If you are using conda, it's a good idea to use conda to install packages like tensorflow and pyzmq that have DLL dependencies instead of mixing conda and pip. Tensorflow wheels in particular are known to have serious compatibility issues. For example:

conda create -n tensorflow tensorflow-gpu jupyterlab python=3.8

to create an environment called 'tensorflow' with tensorflow-gpu and jupyterlab with Python 3.8.

[zhanggyarcher]

@minrk the result is:

[minrk]

@zhanggyarcher could you download libzmq-v141-x64-4_3_4.zip from https://dl.bintray.com/zeromq/generic/ and run test_reqrep_tcp.exe and see if you get a similar error? If you do, could you also download the v140 version and test that?

[zhanggyarcher]

@minrk nothing has displayed, no error message

[zoloc]

Hello, recently I set up a new python environment on a new computer as usual, I got automatically installed pyzmq==21.0.2, and I cannot use jupyter notebook, Jupyter lab, vs code at all, the python kernel crashes every time when I open a editor, and after one second, it directly crashes and lose connection to the kernel.

This problem annoyed me for several weeks, and today I saw many people encountered this issue From December 2020 to January 2021. And I have never encountered this issue in the past.

Also, this error message is somewhat not useful and not making sense, finally I realized its pyzmq error.
I degraded it to pyzmq==19.0.2 , it finally resolved this issue.

I hope this could be fixed, or if you could push old versions of pyzmq for users to install as dependencies.
Thank you very much!

Thanks for the solution, I've fixed it

[anupsaha1]

I am having error while installing pyzmq 19.0.2
(base) C:\Users\Cipher_X>This can take several minutes. Press CTRL-C to abort.
'This' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>failed
'failed' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>UnsatisfiableError: The following specifications were found
'UnsatisfiableError:' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>to be incompatible with the existing python installation in your environment:
'to' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>Specifications:
'Specifications:' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X> - anaconda -> python[version='2.7.13|2.7.14|2.7.14|2.7.15|2.7.15|2.7.16|3.6.10|3.6.10|3.7.10|3.7.11|3.9.7|3.8.12|3.8.8|3.9.4|3.7.9|3.8.5|3.7.7|3.8.3|3.8.1|3.7.6|3.6.9|3.7.4|3.6.8|3.7.3|3.7.3|3.6.7|3.7.1|3.7.0|3.6.6|3.6.5|3.5.5|3.6.4|3.5.4|3.6.3|3.5.4|3.6.2|3.5.4|>=3.9,<3.10.0a0|>=3.8,<3.9.0a0|>=2.7,<2.8.0a0|>=3.7,<3.8.0a0|>=3.6,<3.7.0a0|>=3.5,<3.6.0a0',build='hd3c4935_11|h6679aeb_11|hb034564_12|h8c3f1cb_23|h6538335_1|h0c2934d_0|hea74fb7_0|hea74fb7_0|hcb6e200_5|h9f7ef89_2|h8c8aaf0_1|h9f7ef89_7|h5263a28_0|h5500b2f_0|h9f7ef89_0|h81c818b_4|h9f7ef89_2|h5fd99cc_1|h6244533_0|h6244533_0|hdbf39b2_5|h6244533_0|h6244533_0|h6244533_1|h60c2a47_0|he1778fa_2|h5fd99cc_8_cpython|h60c2a47_2|hcb6e200_0|h8c8aaf0_0|h8c8aaf0_6|he216670_0|h0c2934d_2|h1357f44_23|h9e2ca53_1|hdec4e59_20|h59f5a59_20']
| was unexpected at this time.
(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>Your python: python=3.9
'Your' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>If python is on the left-most side of the chain, that's the version you've asked for.
is was unexpected at this time.
(base) C:\Users\Cipher_X>When python appears to the right, that indicates that the thing on the left is somehow
'When' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>not available for the python version you are constrained to. Note that conda will not
'not' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>change your python version to a different minor version unless you explicitly specify
Invalid parameter(s)
CHANGE { LOGON | PORT | USER }

(base) C:\Users\Cipher_X>that.
'that.' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>The following specifications were found to be incompatible with each other:
'The' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>Output in format: Requested package -> Available versions
'Output' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>
(base) C:\Users\Cipher_X>Package html5lib conflicts for:
'Package' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>anaconda -> bleach==2.1.4=py27_0 -> html5lib[version='>=0.99999999|>=0.99999999pre,!=1.0b1,!=1.0b2,!=1.0b3,!=1.0b4,!=1.0b5,!=1.0b6,!=1.0b7,!=1.0b8']
'!' is not recognized as an internal or external command,
operable program or batch file.

(base) C:\Users\Cipher_X>anaconda -> html5lib[version='0.999999999|0.999999999|0.999999999|1.0.1|1.0.1|1.0.1|1.0.1|1.0.1|1.0.1|1.0.1|1.1',build='py27h47766b3_0|py36ha09b1f3_0|py27h5a33001_0|py27_0|py37_0|py_0|pyhd3eb1b0_0|py36_0|py_0|py36h047fa9f_0|py35h047fa9f_0|py35h79d4e7f_0']
'0.999999999' is not recognized as an internal or external command,
operable program or batch file.

[anupsaha1]

@kevin-bates
please have a look

[anupsaha1]

this is the message appearing when I want to run jupyter notebook(it's obvious it won't run but for understanding the error I think this will help you )
Traceback (most recent call last):
File "C:\Users\Cipher_X\anaconda3\Scripts\jupyter-notebook-script.py", line 6, in
from notebook.notebookapp import main
File "C:\Users\Cipher_X\anaconda3\lib\site-packages\notebook\notebookapp.py", line 78, in
from .services.kernels.kernelmanager import MappingKernelManager, AsyncMappingKernelManager
File "C:\Users\Cipher_X\anaconda3\lib\site-packages\notebook\services\kernels\kernelmanager.py", line 18, in
from jupyter_client.session import Session
File "C:\Users\Cipher_X\anaconda3\lib\site-packages\jupyter_client_init_.py", line 4, in
from .connect import *
File "C:\Users\Cipher_X\anaconda3\lib\site-packages\jupyter_client\connect.py", line 21, in
import zmq
ModuleNotFoundError: No module named 'zmq'

[minrk]

@anupsaha1 I'm not sure if this is related to the issue, but all I can see is pasting lots of non-CMD text into a CMD prompt.

ModuleNotFoundError: No module named 'zmq'

This suggests you don't have pyzmq. You can try installing it:

conda instal pyzmq

[anupsaha1]

pyzmq==19.0.2

thanks mate it's worked
it was giving me an error for the first time installing pyzmq==19.0.2 but this time it's worked
finally, after a few days of searching, I got the solution wlah!

[malsesto]

Late to the party but I encountered the same issue today with version 23.0.2
System: Win 11
Python: 3.10.5

[knidarkness]

@malsesto had this same issue today, for me the solution was to rollback the Python version to 3.8.13 and then pyzmq to 19.0.2 via the Anaconda Navigator > Environments > (current environment).

[minrk]

@Godsix since you've been able to reproduce this in #1754 by building libzmq from source, can you build libzmq once each from commits zeromq/libzmq@b2921d7 and zeromq/libzmq@b77543a (both commits from https://github.com/zeromq/libzmq/pull/4422/commits) and report the result(s)? The first result should give an informative error, while the second hopefully actually fixes the issue.

[stfmnsl]

@minrk I am sorry to report that I have the same issue. Brand new computer, up to date Windows 11, Python 3.10.7, latest version of jupyter and all dependencies. I can't start a kernel in jupyter jupyter-lab, it crashes immediately with Bad address (C:\projects\libzmq\src\epoll.cpp:100) which does not exist.
I am happy to engage offline to help debug the issue on my machine.

Processor 12th Gen Intel(R) Core(TM) i7-12700H 2.30 GHz
Installed RAM 32.0 GB (31.7 GB usable)
Device ID 8C5E8BAA-6590-4208-A92D-0DA3829A6FEF
Product ID 00331-20561-79099-AA062
System type 64-bit operating system, x64-based processor
Pen and touch No pen or touch input is available for this display

Edition Windows 11 Pro
Version 21H2
Installed on ‎10/‎22/‎2022
OS build 22000.1165
Experience Windows Feature Experience Pack 1000.22000.1165.0

[minrk]

@stfmnsl thanks for testing, and that's utterly baffling. How did you install pyzmq and which version of pyzmq did you get? Can you test the win_amd64 wheels for your Python from here: https://github.com/zeromq/pyzmq/actions/runs/3090754782

[stfmnsl]

I did a default installation using pip, which downloaded the latest version pyzmq 24.0.1. Then I tried with 25.0.0b1. Same results.

I went into debug to see what was happening:
C:\Users\stephane.monsallier\fastai>python -m pdb test_zmq.py

c:\users\stephane.monsallier\fastai\test_zmq.py(1)()
-> import zmq
(Pdb) break c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py:157
Breakpoint 1 at c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py:157
(Pdb) run
Restarting C:\Users\stephane.monsallier\fastai\test_zmq.py with arguments:
c:\users\stephane.monsallier\fastai\test_zmq.py(1)()
-> import zmq
(Pdb) next
c:\users\stephane.monsallier\fastai\test_zmq.py(2)()
-> ctx = zmq.Context()
(Pdb) next
c:\users\stephane.monsallier\fastai\test_zmq.py(3)()
-> pull = ctx.socket(zmq.PULL)
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(157)init()
-> self._shadow = False
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(159)init()
-> super().init(
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(160)init()
-> ctx_or_socket,
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(161)init()
-> socket_type,
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(162)init()
-> shadow=shadow_address,
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(163)init()
-> copy_threshold=copy_threshold,
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(159)init()
-> super().init(
(Pdb) p shadow
0
(Pdb) p socket_type
<SocketType.PULL: 7>
(Pdb) p copy_threshold
None
(Pdb) p ctx_or_socket
<zmq.Context() at 0x1f5475876a0>
(Pdb) step
Bad address (C:\projects\libzmq\src\epoll.cpp:100)

Not sure that helps. I don't understand why the constructor gets to Bad Address.

I will try the wheels you pointed out.

[stfmnsl]

Same results with the wheel (24.0.0 cp130) but the debug sequence is different:

C:\Users\stephane.monsallier\fastai>python -m pdb test_zmq.py

c:\users\stephane.monsallier\fastai\test_zmq.py(1)()
-> import zmq
(Pdb) next
c:\users\stephane.monsallier\fastai\test_zmq.py(2)()
-> ctx = zmq.Context()
(Pdb) next
c:\users\stephane.monsallier\fastai\test_zmq.py(3)()
-> pull = ctx.socket(zmq.PULL)
(Pdb) step
--Call--
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(277)socket()
-> def socket(self: T, socket_type: int, **kwargs: Any) -> ST:
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(289)socket()
-> if self.closed:
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(291)socket()
-> s: ST = self._socket_class( # set PYTHONTRACEMALLOC=2 to get the calling frame
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(292)socket()
-> self, socket_type, **kwargs
(Pdb) next
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(291)socket()
-> s: ST = self._socket_class( # set PYTHONTRACEMALLOC=2 to get the calling frame
(Pdb) step
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(292)socket()
-> self, socket_type, **kwargs
(Pdb) step
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\context.py(291)socket()
-> s: ST = self._socket_class( # set PYTHONTRACEMALLOC=2 to get the calling frame
(Pdb) step
--Call--
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(93)init()
-> def init(self: "Socket[bytes]", *a, **kw):
(Pdb) step
c:\users\stephane.monsallier\appdata\roaming\python\python310\site-packages\zmq\sugar\socket.py(94)init()
-> super().init(*a, **kw)
(Pdb) step
Bad address (C:\projects\libzmq\src\epoll.cpp:100)

[stfmnsl]

I downloaded the latest version of libzmq (8d4f4efe) and compiled it. I manually replaced libzmq into pyzmq.lib.
I don't see the error message anymore (Bad address (C:\projects\libzmq\src\epoll.cpp:100)).
I can run successfully the small test:

import zmq
ctx = zmq.Context()
pull = ctx.socket(zmq.PULL)
url = "tcp://127.0.0.1:5555"
pull.bind(url)
push = ctx.socket(zmq.PUSH)
push.connect(url)
push.send(b'test')
pull.recv()
push.close()
pull.close()
ctx.term()

But I have different errors when trying to run jupyter:
OSError: [WinError 10038] An operation was attempted on something that is not a socket

[KianSOH]

I wish to add here that for people who have issues logging in with Kernel of Jupyter Lab/Notebook, based on the simple advise of installing / downgrading, using the ANACONDA commands [For people with ANACONDA installed]

conda install pytest pyzmq==19.0.2

This will fix the problem of non linking up with the Kernel! I basically looked through all advise online, and this one works for me!

[KianSOH]

I wish to add another findings to this frustrating event of Jupyter Notebook (Python), not able to run in browsers (Chrome, EDGE, FireFox, etc) and in Visual Studio Code. [The stupid error codes, and not able to run Jupyter Notebook locally at Visual Studio Code]

I found out that the real culprit is due to the VPN settings that most people will use.

Next time when you find yourself facing this frustrating event perform the following steps.

1. go to windows command prompt, and run as administrator
2. type the following command "netsh winsock reset"
3. reboot your computer.

This should be a permanent solution, but when the VPN starts to mess up the network settings, all you need to do is to perform the above command. [pyzmq is not the true culprit, I found out that I didn't actually install version 19.0.2 since my version of Python is 3.9 and above]

[TengPan2011]

The VPN issue sounds about right, I had a PC, freshly installed spyder on it. It works, but right after a VPN was installed, everything started failing, including fresh installations of vscode, new conda envs, non conda venvs, etc, it all traces back down to pyzmq and the epoll bad address issue.
@KianSOH 's solution worked perfectly.
1.create a new conda environment based on python 3.8,
2.conda install pytest pyzmq==19.0.2
3.Install spyder/jupyter whatever.
I could not believe VPNs would be intefering with jupyter.

[alonbl]

Building libzmq with the following configuration resolves the issue:

-DPOLLER=select -DZMQ_HAVE_IPC=OFF -DZMQ_CV_IMPL=stl11

[SunXinZH]

The VPN issue sounds about right, I had a PC, freshly installed spyder on it. It works, but right after a VPN was installed, everything started failing, including fresh installations of vscode, new conda envs, non conda venvs, etc, it all traces back down to pyzmq and the epoll bad address issue. @KianSOH 's solution worked perfectly. 1.create a new conda environment based on python 3.8, 2.conda install pytest pyzmq==19.0.2 3.Install spyder/jupyter whatever. I could not believe VPNs would be intefering with jupyter.

I have the same problem. After uninstalling VPN, it is ok.
I think some VPN software will forward network requests and localhost's real address is changed

[alejandrozavaletam]

I wish to add another findings to this frustrating event of Jupyter Notebook (Python), not able to run in browsers (Chrome, EDGE, FireFox, etc) and in Visual Studio Code. [The stupid error codes, and not able to run Jupyter Notebook locally at Visual Studio Code]

I found out that the real culprit is due to the VPN settings that most people will use.

Next time when you find yourself facing this frustrating event perform the following steps.

1. go to windows command prompt, and run as administrator
2. type the following command "netsh winsock reset"
3. reboot your computer.

This should be a permanent solution, but when the VPN starts to mess up the network settings, all you need to do is to perform the above command. [pyzmq is not the true culprit, I found out that I didn't actually install version 19.0.2 since my version of Python is 3.9 and above]

Just another confirmation of the same issue with Jupyter and "bad address" message. After looking all around without a proper solution and reading this thread, this was the only solution. The previous approach downgrading pyzmq was not possible and gave other errors. Precisely, I'm using a VPN, w11, conda 23.7.2 and typical python 3.10 and jupyter installation

[KianSOH]

Glad to know that my days of frustration with this silly problem have helped many online! It took me a few days of reading online and researching on the problem, until I come across some nice bloke who posted the solution online in youtube.

I tried the simple solution, and Jupyter started to work in all browsers.

1. go to windows command prompt, and run as administrator
2. type the following command "netsh winsock reset"
3. reboot your computer.

I had actually reinstalled conda numerous times, and also tried with different versions of python, and nothing works... damn frustrating, then I came across this thread with the same error message, and I tried the pyzmq downgrade, it worked intermittently... the true cause of the problem is the VPN.

[aszekMosek]

Same issue here. Unfortunately any solution that involves rebooting is not an option since I am in a docker contained started by a CI pipeline. I am not able to stay on version 19.0.2 forever either. What other options are there? I don't even want to run jupyter, only to execute a notebook file from a Python script, something along the lines of:

import sys
import nbformat
from nbconvert.preprocessors import ExecutePreprocessor as EP

with open(sys.argv[1]) as f:
    nb = nbformat.read(f, as_version=4)
    ep = EP(timeout=400, kernel_name='python3')
    try:
        ep.preprocess(nb, {'metadata' : { 'path' : '.' }})
        with open('executed_notebook.ipynb', mode='wt', encoding='utf8') as fout:
            nbformat.write(nb, fout)

If nothing else helps then I could extract the Python code from the notebook and run it with standalone python, but any more straightforward ideas will be appreciated. (I am surprised such a simple operation as this has to involve libzmq at all.)

[minrk]

I am surprised such a simple operation as this has to involve libzmq at all

executing any code from a notebook uses libzmq because it uses the Jupyter protocol to communicate with the kernel, which is a different process (can be a different language altogether), otherwise it wouldn't be a realistic execution of the notebook.

If you want to export the notebook to a script before running it with a plain Python interpreter, that may be an option (nbconvert --to script).

I think you might be hitting a quite different issue, though, as this is specifically about an unusual Windows-specific networking configuration issue, and I assume your CI pipeline in docker is not a Windows container? I'm not aware of any non-Windows users seeing issues like this, so more information (and a different Issue with repro steps) would be required.

[aszekMosek]

@minrk It is a windows docker (I also get it on physical windows machines). I make a miniconda environment listed below and in it run the notebook execution code mentioned in my last comment.

dependencies:
  - python==3.11
  - numpy
  - scipy
  - matplotlib
  - nb_conda_kernels
  - nbconvert
  - nbformat
  - pandas
  - statsmodels

With python==3.7 and additional pyzmq=19.0.2 it is still OK.

[minrk]

ah, got it. Then it is the same issue. This is a libzmq bug that pyzmq can't do anything about, but there is a chance the brand new libzmq 4.3.5 build fixes it, if you want to try adding zeromq==4.3.5 to your env to test env.

[aszekMosek]

@minrk Thanks. I will try when pyzmq is upgraded to allow it. Forcing zeromq==4.3.5 didn't work for me without breaking it all even worse.

[minrk]

There are now updated wheels with zeromq-4.3.5.

If anyone who can still reproduce this issue can test the wheels available here and report back, that would be a huge help.

pyzmq now has control over the cmake options of the libzmq it bundles in wheels, so there is some hope of resolving it if those wheels don't work yet.

I'm sorry I've never been able to personally reproduce this issue, so I can't verify the fix myself!

[aszekMosek]

I installed the new wheel in an existing environment which is specified as in one of my previous comments but with python==3.9. The installation did not affect the existing zeromq==4.3.4 I had in that environment but maybe now it doesn't matter. After that it seems to resolve the problem, ie. I can run all the notebooks I previously only could run with pyzmq==19.0.2 and the error in epool.cpp:100 does not come up any more.

The only new thing I have is this warning which is probably a different story (irrelevant for me):

\lib\site-packages\zmq\_future.py:700: RuntimeWarning: Proactor event loop does not implement add
_reader family of methods required for zmq. Registering an additional selector thread for add_reader support via tornado. Use `asyncio.set_event_loop_policy(WindowsSelectorEventLoopPolicy())` to avoid this warni
ng.

and in one case (not deterministically reproducible) I got

Assertion failed: pfd.revents & POLLIN (C:\Users\runneradmin\AppData\Local\Temp\tmp46punxxq\build\_deps\bundled_libzmq-src\src\signaler.cp
p:238)

but it seems the execution continued regardless and completed as success.

I will keep you informed if anything should change after more thorough testing.

Many thanks for looking into this and for your help!

[minrk]

The proactor warning is not an issue, but could probably use some better communication (#1830)

The signaler error is concerning. Do keep us updated on if that continues to be a problem.

[ademyankov]

I read all the posts here and on stack overflow and I wasn't able to find a solution that works for me.

I use Windows 10, Python 3.12.3 (installed via the app store). No VS Code or notebook whatever... I simply run my py script that communicates to my C++ code (running libzmq 4.3.5).

Bad file descriptor (C:\Users\runneradmin\AppData\Local\Temp\tmpkywtyyea\build\_deps\bundled_libzmq-src\src\epoll.cpp:73)

I tried to uninstall pyzmq and install an older version (tried different versions) but I wasn't able to do it. Every time I get the same error.

PS > pip install pyzmq==25.0.2
Defaulting to user installation because normal site-packages is not writeable
Collecting pyzmq==25.0.2
  Downloading pyzmq-25.0.2.tar.gz (1.2 MB)
     ---------------------------------------- 1.2/1.2 MB 3.7 MB/s eta 0:00:00
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... done
Building wheels for collected packages: pyzmq
  Building wheel for pyzmq (pyproject.toml) ... error
  error: subprocess-exited-with-error

  × Building wheel for pyzmq (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [417 lines of output]
      C:\Users\ademyankov\AppData\Local\Temp\pip-build-env-e5nik6y2\overlay\Lib\site-packages\setuptools\_distutils\dist.py:266: UserWarning: Unknown distribution option: 'cffi_modules'
        warnings.warn(msg)

This is the end of the error logs:

      Generating code
      Finished generating code
      building 'zmq.backend.cython._device' extension
      creating build\temp.win-amd64-cpython-312\Release\zmq
      creating build\temp.win-amd64-cpython-312\Release\zmq\backend
      creating build\temp.win-amd64-cpython-312\Release\zmq\backend\cython
      "C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.39.33519\bin\HostX86\x64\cl.exe" /c /nologo /O2 /W3 /GL /DNDEBUG /MD -DWINVER=0x0602 -D_WIN32_WINNT=0x0602 -Ibundled\zeromq\include -Izmq\utils "-IC:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include" "-IC:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\Include" "-IC:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.39.33519\include" "-IC:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.39.33519\ATLMFC\include" "-IC:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Auxiliary\VS\include" "-IC:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\ucrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\um" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\shared" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\winrt" "-IC:\Program Files (x86)\Windows Kits\10\\include\10.0.22621.0\\cppwinrt" /Tczmq\backend\cython\_device.c /Fobuild\temp.win-amd64-cpython-312\Release\zmq\backend\cython\_device.obj
      _device.c
      zmq\backend\cython\_device.c(4372): error C2039: 'curexc_traceback': is not a member of '_ts'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/pystate.h(115): note: see declaration of '_ts'
      zmq\backend\cython\_device.c(4375): error C2039: 'curexc_traceback': is not a member of '_ts'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/pystate.h(115): note: see declaration of '_ts'
      zmq\backend\cython\_device.c(5408): error C2039: 'ob_digit': is not a member of '_longobject'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/longintrepr.h(87): note: see declaration of '_longobject'
      zmq\backend\cython\_device.c(5463): error C2039: 'ob_digit': is not a member of '_longobject'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/longintrepr.h(87): note: see declaration of '_longobject'
      zmq\backend\cython\_device.c(5680): error C2039: 'ob_digit': is not a member of '_longobject'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/longintrepr.h(87): note: see declaration of '_longobject'
      zmq\backend\cython\_device.c(5735): error C2039: 'ob_digit': is not a member of '_longobject'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/longintrepr.h(87): note: see declaration of '_longobject'
      zmq\backend\cython\_device.c(6086): error C2039: 'ob_digit': is not a member of '_longobject'
      C:\Program Files\WindowsApps\PythonSoftwareFoundation.Python.3.12_3.12.1008.0_x64__qbz5n2kfra8p0\include\cpython/longintrepr.h(87): note: see declaration of '_longobject'
      DEBUG EXT_SUFFIX '.cp312-win_amd64' '.cp312-win_amd64.pyd'
      ************************************************
      DEBUG EXT_SUFFIX '.cp312-win_amd64' '.cp312-win_amd64.pyd'
      ************************************************
      error: command 'C:\\Program Files\\Microsoft Visual Studio\\2022\\Professional\\VC\\Tools\\MSVC\\14.39.33519\\bin\\HostX86\\x64\\cl.exe' failed with exit code 2
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR: Failed building wheel for pyzmq
Failed to build pyzmq
ERROR: Could not build wheels for pyzmq, which is required to install pyproject.toml-based projects

All was working before with the older version of pyzmq. But now I can't even downgrade the version.

Any advice people?

[minrk]

@ademyankov I think you want pyzmq 25.12, not 25.0.2. 25.1.1 introduced support for Python 3.12. It is probably best to do

pip install 'pyzmq<26'

to get the latest pre-26 version, which will be 25.1.2.

[ademyankov]

@minrk Ah, man! Thank you so much. It works like a charm!

[minrk]

I think this is actually fixed in pyzmq 26.2, and possibly the root cause fixed upstream in zeromq/libzmq#4732

Only took me three and a half years to find a way to reproduce this!