tests(sharness): socat version

[guillaumemichel]

Checklist

• This is a bug report, not a question. Ask questions on discuss.ipfs.tech.
• I have searched on the issue tracker for my bug.
• I am running the latest kubo version or have an issue updating.

Installation method

built from source

Version

ipfs version 0.34.0-dev

Config

not relevant

Description

Following 3c9cc3f, sharness tests are failing in CI.

I have tried to run them locally and they also fail for the same tests. I have isolated the problem as coming from socat. The sharness tests are passing fine with socat 1.7.4.4, but not with socat 1.8.0.0 and above. There are just 4 tests using socat that are failing, the rest of the sharness tests using socat are passing.

IIUC after 3c9cc3f, we use a more recent ubuntu image in the CI, which comes with socat 1.8.0.0. Before the ubuntu image was probably shipped with an older socat.

I tried to specify the version when installing socat as a dependency, but it didn't work, probably because versions prior to 1.8.0.0 are tagged as unstable by ubuntu noble (ubuntu:latest).

Failures are reported for socat 1.8.0.0, but apparently they were fixed in 1.8.0.1. I am running the latest 1.8.0.2 and the tests still failed. There was probably a change of behaviour between socat 1.7.4.4 and socat 1.8.0.0.

When trying to debug I got the following outputs when running the command with both socat versions in the according test

socat -d -d - tcp:localhost:$SWARM_PORT,connect-timeout=1 >swarmnc <../t0060-data/mss-plaintext

1.7.4.4 (SUCCESSFUL):
2025/01/29 13:00:45 socat[174418] N reading from and writing to stdio
2025/01/29 13:00:45 socat[174418] N opening connection to AF=2 127.0.0.1:40201
2025/01/29 13:00:45 socat[174418] N starting data transfer loop with FDs [0,1] and [9,9]
2025/01/29 13:00:45 socat[174418] N socket 1 (fd 0) is at EOF
2025/01/29 13:00:45 socat[174418] W read(9, 0x5747a4d95000, 8192): Connection reset by peer
2025/01/29 13:00:45 socat[174418] N socket 2 to socket 1 is in error
2025/01/29 13:00:45 socat[174418] N socket 2 (fd 9) is at EOF
2025/01/29 13:00:45 socat[174418] N exiting with status 0

1.8.0.2 (FAILURE):
2025/01/29 13:01:04 socat[175148] N reading from and writing to stdio
2025/01/29 13:01:04 socat[175148] N opening connection to AF=2 127.0.0.1:36491
2025/01/29 13:01:04 socat[175148] N starting data transfer loop with FDs [0,1] and [9,9]
2025/01/29 13:01:04 socat[175148] N write(9, 0x5a0e4ce8f000, 38) completed
2025/01/29 13:01:04 socat[175148] N socket 1 (fd 0) is at EOF
2025/01/29 13:01:04 socat[175148] N write(1, 0x5a0e4ce8f000, 20) completed
2025/01/29 13:01:04 socat[175148] N write(1, 0x5a0e4ce8f000, 18) completed
2025/01/29 13:01:04 socat[175148] N write(1, 0x5a0e4ce8f000, 79) completed
2025/01/29 13:01:04 socat[175148] E read(9, 0x5a0e4ce8f000, 8192): Connection reset by peer
2025/01/29 13:01:04 socat[175148] N exit(1)

The newer version is making more write. Also 1.7.4.4 has a warning W on connection reset by peer, where as 1.8.0.2 has an error E. I didn't find what is the cause of this change. 1.7.4.4 is probably ending on a FIN and 1.8.0.2 on a RST.

It seems like the quick fix is to install socat 1.7.4.4 from source in sharness.yml. Any cleaner suggestions are welcome.

cc: @galargh

[galargh]

3c9cc3f is not exactly relevant here. It just happened to coincide with the changes to the base images used on self-hosted runners.

Last weekend, we transitioned from using Ubuntu 22.04 to 24.04 on self-hosted runners. A similar change on GitHub hosted runners (ubuntu-latest label), happened earlier in the month.

I was able to make the sharness tests pass again on 24.04 by passing the -s option to socat. I created a PR with that change here: #10683

Let me know if that works for you 😄

[guillaumemichel]

That's perfect, thanks a lot @galargh!