`instance serial console` gets a 400 response for some reason i don't understand

[hawkw]

Component

• CLI
• SDK
• Something else
• Not sure

What was the expected behaviour

oxide instance serial console --project <my project> --instance <my instance> should connect to the instance's serial console.

What is the current behaviour and what actions did you take to get there

it does not do that:

eliza@theseus ~/Code/oxide/oxide.rs $ ./cli.sh --debug instance serial console --project eliza --instance my-computer
    Finished `release` profile [optimized] target(s) in 0.20s
     Running `target/release/oxide --debug instance serial console --project eliza --instance my-computer`
{"timestamp":"2026-02-11T22:00:11.323912Z","level":"DEBUG","message":"request succeeded","url":"https://oxide.sys.rack2.eng.oxide.computer/v1/instances/my-computer/serial-console/stream?most_recent=262144&project=eliza","path":"/v1/instances/my-computer/serial-console/stream","operation_id":"instance_serial_console_stream","remote_addr":"172.20.26.6:443","http.request.method":"GET","http.response.content_length":102,"http.response.status_code":400,"start_time":"2026-02-11T22:00:10.962722Z","duration_ms":361,"oxide.request_id":"dd0db86b-0f4c-41b4-b601-5208386399ec","target":"oxide"}
Unexpected Response: Response { url: "https://oxide.sys.rack2.eng.oxide.computer/v1/instances/my-computer/serial-console/stream?most_recent=262144&project=eliza", status: 400, headers: {"content-type": "application/json", "x-request-id": "dd0db86b-0f4c-41b4-b601-5208386399ec", "date": "Wed, 11 Feb 2026 22:00:11 GMT", "content-length": "102"} }

i am allegedly on the latest main branch:

eliza@theseus ~/Code/oxide/oxide.rs $ git status
On branch main
Your branch is up to date with 'origin/main'.

CLI or SDK version

Oxide CLI 0.14.0+20251008.0.0 Built from commit: ae53baa Oxide API: 2026020901.0.0

Operating system

Linux theseus 6.18.1 #1-NixOS SMP PREEMPT_DYNAMIC Fri Dec 12 17:42:47 UTC 2025 x86_64 GNU/Linux

Anything else you would like to add?

No response

[ahl]

It looks like we're getting a 400 response and rather than simply reporting that failure, the generated code is surprised to see that as a response code. This seems to be both carelessness in the OpenAPI document and a lack of precision in code generation.

[ahl]

I hacked up the CLI. This seems to be the actual response from nexus:

Unexpected response (400 Bad Request) from server when attempting to connect to serial console: {
  "request_id": "eb81b7be-e8c7-4881-815a-c7b4b780a05a",
  "message": "expected connection upgrade"
}

That appears to be a dropshot error squirting out.

[ahl]

There seems to be some incompatibility introduced by reqwest 0.13

[ahl]

Confirmed with git bisect that the commit where we upgraded to reqwest 0.13 was where this started: 2b22cf3

[ahl]

This is weird. It seems to have to do with some confluence of moving parts, but forcing HTTP 1.1 seems to (somehow?) fix things. I'm not sure if we were getting lucky previously or what exactly changed.

[hawkw]

This is weird. It seems to have to do with some confluence of moving parts, but forcing HTTP 1.1 seems to (somehow?) fix things. I'm not sure if we were getting lucky previously or what exactly changed.

Cool. Well, knowing that probably makes me feel good and/or bad.

[lifning]

(Though the primary issue is already worked around by forcing HTTP 1.1,) I've applied some carefulness to the OpenAPI document and increased precision to the code generation accordingly.