just did a refresh on our internal VM on a convers...
# proj-agent-canvas
w
just did a refresh on our internal VM on a conversation that appears frozen šŸ˜•
whole server appears locked up
CC @quick-window-91705
q
oh nooo maybe too much load? investigating…
w
how big is the VM?
q
Copy code
Machine type: e2-standard-2
CPU/RAM: 2 vCPU, 8 GB RAM
Boot disk: 20 GB
Disk type: pd-balanced
Image: Debian 12
Hmm load doesn't look terrible
Copy code
rohit_openhands_dev@dev-remote-vm-vm:~$ echo "CPU: $(nproc) cores" && free -h
CPU: 2 cores
               total        used        free      shared  buff/cache   available
Mem:           7.8Gi       2.8Gi       3.6Gi       900Ki       1.6Gi       4.9Gi
Swap:             0B          0B          0B
rohit_openhands_dev@dev-remote-vm-vm:~$ uptime
 21:04:13 up 2 days,  1:43, 26 users,  load average: 0.00, 0.05, 0.08
Odd that its saying an llm isn't configured, the LLM page has one configured
w
nod seems like a weird failure mode
q
Copy code
litellm.AuthenticationError: AuthenticationError: OpenAIException - {
  "error": {
    "message": "Incorrect API key provided: None. You can find your API key at <https://platform.openai.com/account/api-keys>.",
    "type": "invalid_request_error",
    "param": null,
    "code": "invalid_api_key"
  }
}
Seems to be this? The API key somehow became invalid (I'm using our cloud proxy) and its affecting conversation start/reload I can't start a new convo even though the llm model exists and is selected
Hmm the api key was wiped out somehow it seems
Should work now
We should fix that failure mode though, its so confusing it makes it seem as though I have a valid LLM and locks all conversation related paths And I'm not sure how we ended up in a state where a profile exists but no api key alongside it
c
Which of the cloud proxies were you using?
q
The prod one from app.all-hands.dev
c
That's
openhands/
so I bet the bug is related to the changes to that. I’ll have the agent make a pass trying to figure out the issue
q
Hmm its been working basically all day until now, and the profiles api key was completely unset But would appreciate the investigation nevertheless šŸ™
Ooh maybe it is related, I think the provider had changed from
openhands/
to
openai/
somehow as well, but unclear if that was someone else meddling with the vm
c
Oh, there was just today a PR fixing openai subscription via agent-server
That may be what happened: if you test sub, it will not have an api key