Language Model Unavailable when connected to Google Cloud VM via Remote SSH

[spencer411]

This used to work fine but upon updating to the new version the "language model is unavailable" when attempting to use the chat. Note that I have attempted reinstalling the latest version on both the VM and Mac with no success, as well as reinstalling github co-pilot and remote ssh extensions:

• GitHub Copilot Chat v0.23.2
• Remote - SSH v0.116.1
• VS Code Version: 1.96.2 (Universal)
• OS Version: Mac - 15.2 (24C101)
• OS Version (Remote): GCP VM - Ubuntu 22.04.5 LTS
• Logs:

2025-01-02 17:16:58.462 [error] TypeError: fetch failed
at node:internal/deps/undici/undici:13392:13
at processTicksAndRejections (node:internal/process/task_queues:95:5)
at ek._fetch (/home/my_name/.vscode-server/extensions/github.copilot-chat-0.23.2/dist/extension.js:763:5811)
at _p._fetchModels (/home/my_name/.vscode-server/extensions/github.copilot-chat-0.23.2/dist/extension.js:2213:11405)
at _p.getChatModelFromFamily (/home/my_name/.vscode-server/extensions/github.copilot-chat-0.23.2/dist/extension.js:2213:10055)
at WO.getChatEndpoint (/home/my_name/.vscode-server/extensions/github.copilot-chat-0.23.2/dist/extension.js:2213:14751)
at Object.A [as task] (/home/my_name/.vscode-server/extensions/github.copilot-chat-0.23.2/dist/extension.js:488:834)
at xC._processQueue (/home/my_name/.vscode-server/extensions/github.copilot-chat-0.23.2/dist/extension.js:487:1101): [LanguageModelAccess] FAILED to update language models

Steps to Reproduce: Remote SSH to Ubuntu GCP VM and try to use copilot chat feature.

Note I get the following when I attempt to curl -I https://copilot-proxy.githubusercontent.com

HTTP/2 404
content-security-policy: default-src 'none'; sandbox
content-type: text/plain; charset=utf-8
strict-transport-security: max-age=31536000
x-content-type-options: nosniff
content-length: 19
date: Thu, 02 Jan 2025 16:48:27 GMT
x-github-backend: Kubernetes
x-github-request-id: ECD6:DED46:999A9:9EAF0:6776C35B

[kieferrm]

Please have a look at microsoft/vscode#253137.

[spencer411]

I am not part of enterprise or business, it is an individual self-paid subscription.

Do you still see this after F1 > Reload Window and trying again?

• Yes

Try this: F1 > "Set Log Level" > Trace, reload the window, try again. Are there any errors in the Copilot Chat output channel or anywhere else?

• Same error as before: "Language model unavailable".

Follow these steps to capture a network trace of the request: Start the application from commandline like code --user-data-dir=/code-temp-data, replace with a valid path on your system. Confirm if the issue happens in this instance.

• Yes, it happens, Nevertheless I get a message on the command line saying
  "Ignoring option 'user-data-dir': not supported for code." after running "code --user-data-dir=test_directory/code-temp-data". Note I am doing this in the terminal on VS Code via remote ssh, and am running vs-code server version 1.96.2.

If you are unable to repro the issue with step 1) then it confirms that cache was indeed the trigger for this issue. Proceed to the next step

• I did not proceed to the next step because it did repro the issue.

[lramos15]

What does F1 -> Copilot Chat Diagnostics

have in it? Can you paste that here

[spencer411]

• Extension Version: 0.23.2 (prod)
• VS Code: vscode/1.96.2
• OS: Linux
• Remote Name: ssh-remote

Network

User Settings:

  "github.copilot.advanced.debug.useElectronFetcher": true,
  "github.copilot.advanced.debug.useNodeFetcher": false,
  "github.copilot.advanced.debug.useNodeFetchFetcher": true

Connecting to https://api.github.com:

• DNS ipv4 Lookup: 140.82.114.6 (2 ms)
• DNS ipv6 Lookup: Error (35 ms): getaddrinfo ENOTFOUND api.github.com
• Proxy URL: None (0 ms)
• Electron fetch: Unavailable
• Node.js https: HTTP 200 (121 ms)
• Node.js fetch (configured): HTTP 200 (157 ms)
• Helix fetch: HTTP 200 (268 ms)

Connecting to https://api.individual.githubcopilot.com/_ping:

• DNS ipv4 Lookup: 140.82.113.22 (2 ms)
• DNS ipv6 Lookup: Error (3 ms): getaddrinfo ENOTFOUND api.individual.githubcopilot.com
• Proxy URL: None (0 ms)
• Electron fetch: Unavailable
• Node.js https: Error (63 ms): read ECONNRESET
• Node.js fetch (configured): Error (86 ms): fetch failed
  read ECONNRESET
• Helix fetch: Error (84 ms): read ECONNRESET

Documentation

In corporate networks: Troubleshooting firewall settings for GitHub Copilot.

[lramos15]

@chrmarti Thoughts? It appears the subscription specific endpoint is giving connection reset while the none subscription specific is not. This is normally what I think you would see with a firewall would you?

@spencer411 Is this just a standard fresh new google cloud VM?

[spencer411]

It is a VM set up by the organization I work for. I did not have issues accessing the LLM before updating copilot, and I do not believe firewall settings have changed. Note the following below when I attempt to ping directly from the VM:

$ ping api.individual.githubcopilot.com
PING glb-db52c2cf8be544.github.com (140.82.113.22) 56(84) bytes of data.
64 bytes from lb-140-82-113-22-iad.github.com (140.82.113.22): icmp_seq=1 ttl=52 time=21.8 ms
64 bytes from lb-140-82-113-22-iad.github.com (140.82.113.22): icmp_seq=2 ttl=52 time=20.4 ms
64 bytes from lb-140-82-113-22-iad.github.com (140.82.113.22): icmp_seq=3 ttl=52 time=20.3 ms
64 bytes from lb-140-82-113-22-iad.github.com (140.82.113.22): icmp_seq=4 ttl=52 time=20.1 ms
etc.

$ curl -v https://api.individual.githubcopilot.com/_ping

• Trying 140.82.113.22:443...
• Connected to api.individual.githubcopilot.com (140.82.113.22) port 443 (#0)
• ALPN, offering h2
• ALPN, offering http/1.1
• CAfile: /etc/ssl/certs/ca-certificates.crt
• CApath: /etc/ssl/certs
• TLSv1.0 (OUT), TLS header, Certificate Status (22):
• TLSv1.3 (OUT), TLS handshake, Client hello (1):
• OpenSSL SSL_connect: Connection reset by peer in connection to api.individual.githubcopilot.com:443
• Closing connection 0
• TLSv1.0 (OUT), TLS header, Unknown (21):
• TLSv1.3 (OUT), TLS alert, decode error (562):
  curl: (35) OpenSSL SSL_connect: Connection reset by peer in connection to api.individual.githubcopilot.com:443

[lramos15]

It is a VM set up by the organization I work for. I did not have issues accessing the LLM before updating copilot, and I do not believe firewall settings have changed. Note the following below when I attempt to ping directly from the VM:

The URL which Copilot accesses changed. So I believe that your org must not have an allow rule for that domain. You can read more about it here

[chrmarti]

@spencer411 It is interesting that it works with curl. If this is indeed a regression, it would help to know which change caused it. We maintain a node.js based CLI tool vscode-bisect that automatically downloads and runs previous VS Code insiders and asks for each build if the issue reproduces. It takes up to 8 steps to find exactly the build that caused the regression.

Steps (on your local machine):

• install node.js (if not yet done)
• npx --yes @vscode/vscode-bisect@latest
• follow the instructions until you found the offending build (its fine to leave the commit empty)
• report back the commit range in this issue

Thanks!