docs/

reels/

Dem29nMPMwp.mp4

Outline

---

title: From a missing fcntl error to a real Linux setup on Windows

slug: wsl-setup

lang: en

published: 2026-10-31

words: 446

lines: 92

size: 2.6 KB

read: 3 min read

reel: I brought Linux to me, inside Windows

---

From a missing fcntl error to a real Linux setup on Windows

My website backend would not start on Windows. The same code ran fine on the server. The problem was the environment, not the code. This is the exact error, why it happens, and the setup I use now.

The error

output

PS C:\sina_codes\website\backend> uv run gunicorn sina.asgi:application -c gunicorn.conf.py
Traceback (most recent call last):
  File "...\gunicorn\util.py", line 7, in <module>
    import fcntl
ModuleNotFoundError: No module named 'fcntl'

fcntl is part of Python, but only on Unix. Windows builds do not ship it, and gunicorn imports it on line 7 of util.py. No package you install fixes that. The server process itself does not run on Windows.

Servers run Linux, and most tools land on Linux first. Agents like Claude Code also work in the terminal, so the shell they get decides what they can do. I did not want to leave Windows or dual boot, so I brought Linux in.

1. Install WSL

Open PowerShell as administrator:

powershell

wsl --install -d Ubuntu

Restart when it asks, then open Ubuntu from the Start menu and pick a username and password. Check that you are on version 2:

powershell

wsl --version
wsl --list --verbose

2. Give it enough machine

By default WSL takes a share of your RAM and cores. Set them yourself in C:\Users\<you>\.wslconfig. These are my real values, for a laptop with a lot of RAM. Scale them to yours.

ini

[wsl2]
processors=12
memory=20971520000
swap=8388608000

That is about 19.5 GB of RAM and 7.8 GB of swap. Apply it with:

powershell

wsl --shutdown

3. Keep the project inside Linux

This is the step people skip. Clone into your Linux home, not /mnt/c.

bash

cd ~
git clone <your-repo-url>

Files under /mnt/c cross a bridge to Windows on every read. Installs and test runs get slow, and file watchers can miss changes. Inside ~ you get a native Linux filesystem.

4. Open your editor in the same place

Install the WSL extension in VS Code or Cursor, then run this from inside the project:

bash

code .

The editor window stays on Windows. Its terminal, language servers and extensions run in Linux, in the same place as your project and your agent.

5. The same project, this time it starts

bash

curl -LsSf https://astral.sh/uv/install.sh | sh
cd ~/sina_codes/website/backend
uv run gunicorn sina.asgi:application -c gunicorn.conf.py

output

[INFO] Starting gunicorn 26.2.0
[INFO] Using worker: uvicorn_worker.UvicornWorker

Run the agent here too

bash

curl -fsSL https://claude.ai/install.sh | bash
cd ~/your-project
claude

Now the agent runs the same shell as your server, so what works for it has a real chance of working in production.

PreviousRunning small models under a big one NextThree clean code rules for agents, enforced by tools

1	My website backend would not start on Windows. The same code ran fine on the server. The problem was the environment, not the code. This is the exact error, why it happens, and the setup I use now.
2
3	## The error
4
5	```text
6	PS C:\sina_codes\website\backend> uv run gunicorn sina.asgi:application -c gunicorn.conf.py
7	Traceback (most recent call last):
8	File "...\gunicorn\util.py", line 7, in <module>
9	import fcntl
10	ModuleNotFoundError: No module named 'fcntl'
11	```
12
13	`fcntl` is part of Python, but only on Unix. Windows builds do not ship it, and gunicorn imports it on line 7 of `util.py`. No package you install fixes that. The server process itself does not run on Windows.
14
15	Servers run Linux, and most tools land on Linux first. Agents like Claude Code also work in the terminal, so the shell they get decides what they can do. I did not want to leave Windows or dual boot, so I brought Linux in.
16
17	## 1. Install WSL
18
19	Open PowerShell as administrator:
20
21	```powershell
22	wsl --install -d Ubuntu
23	```
24
25	Restart when it asks, then open Ubuntu from the Start menu and pick a username and password. Check that you are on version 2:
26
27	```powershell
28	wsl --version
29	wsl --list --verbose
30	```
31
32	## 2. Give it enough machine
33
34	By default WSL takes a share of your RAM and cores. Set them yourself in `C:\Users\<you>\.wslconfig`. These are my real values, for a laptop with a lot of RAM. Scale them to yours.
35
36	```ini
37	[wsl2]
38	processors=12
39	memory=20971520000
40	swap=8388608000
41	```
42
43	That is about 19.5 GB of RAM and 7.8 GB of swap. Apply it with:
44
45	```powershell
46	wsl --shutdown
47	```
48
49	## 3. Keep the project inside Linux
50
51	This is the step people skip. Clone into your Linux home, not `/mnt/c`.
52
53	```bash
54	cd ~
55	git clone <your-repo-url>
56	```
57
58	Files under `/mnt/c` cross a bridge to Windows on every read. Installs and test runs get slow, and file watchers can miss changes. Inside `~` you get a native Linux filesystem.
59
60	## 4. Open your editor in the same place
61
62	Install the WSL extension in VS Code or Cursor, then run this from inside the project:
63
64	```bash
65	code .
66	```
67
68	The editor window stays on Windows. Its terminal, language servers and extensions run in Linux, in the same place as your project and your agent.
69
70	## 5. The same project, this time it starts
71
72	```bash
73	curl -LsSf https://astral.sh/uv/install.sh \| sh
74	cd ~/sina_codes/website/backend
75	uv run gunicorn sina.asgi:application -c gunicorn.conf.py
76	```
77
78	```text
79	[INFO] Starting gunicorn 26.2.0
80	[INFO] Using worker: uvicorn_worker.UvicornWorker
81	```
82
83	## Run the agent here too
84
85	```bash
86	curl -fsSL https://claude.ai/install.sh \| bash
87	cd ~/your-project
88	claude
89	```
90
91	Now the agent runs the same shell as your server, so what works for it has a real chance of working in production.
92

en3 min read