https://docs.all-hands.dev/ logo
Join Slack
Powered by
My canvas -> cloud connection just broke again....
# proj-agent-canvas
w
My canvas -> cloud connection just broke again. removing and re-adding the backend fixed it I logged into SaaS on mobile this morning, dunno if that triggered it. Tried to repro again and no luck
h
Tim ran into this as well and pulled some data from datadog https://github.com/OpenHands/agent-canvas/issues/849 . I think @green-greece-46378 might have looked at this but hadn't found the root cause
g
I briefly tried to reproduce but was unable to. I will try again today.
b
I wonder if it's after each Cloud release for some reason?
🤔 1
Another release just went out. I wonder if you can reproduce it now. Just curious.
h
Seems like a release is related
g
Ok. I can work with that
So looking at @fierce-account-55183 ticket, it looks like the offline token is getting deleted:
Copy code
keycloak.exceptions.KeycloakPostError: 400: b'{"error":"invalid_grant","error_description":"Offline user session not found"}'
I deleted the my offline token and got the broken connection as expected. Logging in to generate an offline token again fixed it. The issue to to figure why the offline token is getting deleted. @fierce-account-55183, the ticket (https://github.com/OpenHands/agent-canvas/issues/849) doesn't really tell me how to reproduce the issue. Can you reproduce it consistently? If anyone gets into this state, I would like to look into keycloak while you are in a bad state.
f
After logging out and logging back in again in cloud, I can no longer reproduce the issue - I'll let you know if I see it again
g
@better-park-34863, can you point me to any info/logs on the release? No idea why it would have anything to do with this, but seems like our only lead at this point.
Maybe we can try something similar on staging?
b
What kind of info do you need?
g
Maybe logs on what pods were updated? Not really sure.
w
if you have the time of the release, you can look in datadog around that time
g
Will do. How often does it happen for you @wooden-umbrella-84098?
w
only this once
haven't been able to repro
g
As workaround for now, if we get into this state I can pop something up to initiate a cloud login.
@hundreds-twilight-32833, does this look like an MVP candidate or a post follow-on: https://github.com/OpenHands/agent-canvas/issues/911
h
Hard to say with a scope on a fix. I would assume this is the same on cloud so it may not be a major issue just yet. If we did some minimal validation of keys per a few major providers it would be a simple change...maybe? For anthropic as an example:
Copy code
<input
  type="text"
  name="apiKey"
  required
  pattern="^sk-.+"
  title="Value must start with sk-"
/>
g
So I've got a PR for agent-server that is close: https://github.com/OpenHands/software-agent-sdk/pull/3410
that needs to go out first.
Then there would be a PR for agent-canvas
I've tested locally and everything seems to work.
w
I don't think this is critical for MVP
👍 1
2 Views
Open in Slack
PreviousNext
Powered by