[kalsi-avneet]

No description provided.

[pbiering]

Thank you

[schildbach]

I'm testing this PR (via master) and I get this error:

radicale-1  | [2026-11-07 14:54:25 +0000] [1] [INFO] Starting Radicale python=3.15.2 radicale=3.9.1.dev0 vobject=0.9.9 passlib=1.9.4 defusedxml=0.7.1 hypercorn=0.18.0 cheroot=n/a requests=n/a bcrypt=5.0.0 argon2-cffi=n/a pika=n/a ldap=n/a ldap3=n/a pam=n/a with hypercorn on 0.0.0.0:5232, [::]:5232
radicale-1  | [2026-11-07 14:54:25 +0000] [1] [INFO] Running on http://0.0.0.0:5232 (CTRL + C to quit)
radicale-1  | [2026-11-07 14:54:25 +0000] [1] [INFO] Running on http://0.0.0.0:5232 (CTRL + C to quit)
radicale-1  | [2026-11-07 14:54:25 +0000] [1] [INFO] Stopping Radicale
radicale-1  | [2026-11-07 14:54:25 +0000] [1] [CRITICAL] An exception occurred during server startup: [Errno 98] Address in use
radicale-1 exited with code 1

Is it possible that the other (old) HTTP server is also still listening on port 5232?

My entire config:

# cat config 
[server]
type = hypercorn

[auth]
type = htpasswd
htpasswd_filename = /etc/radicale/users
htpasswd_encryption = plain

[schildbach]

I cross-checked and before starting the container, there is no port 5232 listening:

# netstat -lnpt | grep 5232
(no output)

[pbiering]

See https://github.com/Kozea/Radicale/blob/master/DOCUMENTATION.md#hosts, h2 is not supporting separate bind for IPv4 and IPv6.

[schildbach]

Ok, I tried both

[server]
type = hypercorn
hosts = [::]:5232

and

[server]
type = hypercorn
hosts = 0.0.0.0:5232

but none of the two changes anything. Same "Address in use" error.

[schildbach]

In fact, I haven't even got IPv6 enabled on the docker bridge network. Will try that now.

[pbiering]

if it works with "builtin" or "cheroot" than it looks like something is incompatible inside "hypercorn"

IPv6 on the docker bridge network should not be required for an IPv4-only bind.

[schildbach]

No matter what I try (any permutation of IPv6 available or not, and wether I bind to 0.0.0.0 or [::]), hypercorn always wants to bind to 0.0.0.0:5232 twice and fails. Switching back to "builtin" makes it work again, at the expense of using vulnerable HTTP/1.1 again.

I didn't try the sysctl net.ipv6.bindv6only = 0 as such a system-wide change is out of question for me.

[pbiering]

It's a problem in hypercorn, unrelated to IPv6. I already wondered in the past why this 2nd bind message appeared and tried to traced down (a little) the hypercorn code, no clue so far why from 1x __init__.py/serve 2x run.py/worker_serve are called.