Durable Object returning 405s
I am running into some strange behavior - I deployed a new version of a worker with a bound DO 7.5 hours ago. I tested everything locally and in prod, all seemed well.
About 1.5 hours ago I noticed that all of my DO requests are returning 405s and the service is very much broken. There were two deployments in the interim with identical code - I tried rolling back to the deployment that was working previously and I'm getting the exact same issue.
Could a CF employee help me double-check the health of my DO? Can anyone shed some light on what might be happening here?
1 Reply
Some more context - if I run the service remotely with
wrangler dev --remote I can access the DO just fine. So the DO appears to be fine. It looks like there's something up with my deployed worker's configuration
I just redeployed from the CLI (no code changes) and it's working again. I have no clue why that would fix the problem, and I'm honestly still terrified it's going to break again. Would still appreciate any insight as to what's happening under the hood here and how I can avoid it breaking again. I've made zero changes to the code or config in the meantime, it's an identical deployment.
More context to help debug what's going on here:
- I rolled back to my previous CLI deployment (from yesterday) and it immediately broke the service, 405s from the DO again. I then tried to roll forward to my CLI deployment I just deployed, and it's still broken. Prior to the rollback & forward, that deployment was working no sweat. There is clearly some sort of bug in the deployments system
- Deploying an identical version of the service from CI also causes the problem. Don't have the slightest clue why that would cause the problem
More context - our CI deployments were using wrangler publish, which I just switched to wrangler deploy. That seems to have fixed the problem of the CI deployments breaking the service.
I just deployed via CI and it's still working as expected. So it appears the original cause of the issue was using wrangler publish. Somehow that broke the connection to the DO. Still doesn't explain why rollbacks & forwards were broken
I'm afraid that if I rollback or forward to any other deployment things will break again. Would still appreciate insight on what's going on here under the hood and confirmation that rollbacks will work as expected
@Matt I ran into this problem again yesterday when I connected to logs from the service the DO is bound to - the DO suddenly started only returning 405s. Re-deploying with wrangler deploy fixed the issue, but this is obviously a serious problem. DOs are practically unusable if I can't observe them when I'm trying to debug an issue with a customer on the line. Has this behavior been observed before? Is there anyone at CF looking into this?
The namespace ID of the affected DO is ddd97104edc9415eba92202173ff4735