We’re using flyctl launch to launch one app with one machine, and we’ve tried doing this across a few regions (cdg, lax, lhr) - we’re seeing sporadic failures, with the same error each time:
==> Building image
Searching for image 'registry.fly.io/<redacted>' remotely...
image found: <redacted>
Watch your deployment at https://fly.io/apps/<redacted>/monitoring
Provisioning ips for <redacted>
Dedicated ipv6: <redacted>
Shared ipv4: <redacted>
Add a dedicated ipv4 with: fly ips allocate-v4
Error: Post "https://api.fly.io/graphql": read tcp 10.223.134.252:49840->77.83.143.220:443: read: connection reset by peer
The same exact flyctl launch command sometimes fails and sometimes works. Does anyone have any ideas?
We have investigated further and are seeing intermittent connection reset errors specifically from the MachinesCreateRelease GraphQL query (which we got from watching flyctl launch with LOG_LEVEL=debug).
When POSTing MachinesCreateRelease GraphQL requests directly to https://api.fly.io/graphql we see about 1 in 20 requests go through fine, but most return a socket closed by remote peer.
Fortunately, things resolved later on the same day with no changes from our end. flyctl version gives flyctl v0.2.42 linux/amd64 Commit: 44aedf5b21a71b0c086d4603d2522f71e2908c61 BuildDate: 2024-04-24T22:03:44Z