#Error connecting to remote engine: "error reading server preface: read |0: file already closed"

1 messages · Page 1 of 1 (latest)

halcyon bramble
#

Is anyone else getting these errors, connecting to a remote engine via DOCKER_HOST with an OK but not great internet connection? Actual ssh works fine. I tried on several different remote engines, on completely different providers.

 ✔ starting engine 12.3s
│ │ ✔ create 12.3s
│ │ │ ✔ exec docker ps -a --no-trunc --filter name=^dagger-engine-|^dagger-engine-v0\.18\.1$ --format {{.Names}} 6.4s
│ │ │ ┃ dagger-engine-v0.18.1
│ │ │ ┃ dagger-engine-v0.18.0
│ │ │ ┃ dagger-engine-v0.17.2
│ │ │ ✔ exec docker start dagger-engine-v0.18.1 5.9s
│ │ │ ┃ dagger-engine-v0.18.1
│ ✘ connecting to engine 7.2s
│ ! new client: interrupted
│ │ ✘ moby.buildkit.v1.Control/Info 3.0s
│ │ ! connection error: desc = "error reading server preface: read |0: file already closed"
│ │ ✘ moby.buildkit.v1.Control/Info 0.0s
│ │ ! connection error: desc = "error reading server preface: read |0: file already closed"
│ │ ✘ moby.buildkit.v1.Control/Info 0.0s
│ │ ! connection error: desc = "error reading server preface: read |0: file already closed"
│ │ ✘ moby.buildkit.v1.Control/Info 0.0s
│ │ ! connection error: desc = "error reading server preface: read |0: file already closed"
│ │ ✘ moby.buildkit.v1.Control/Info 0.0s
│ │ ! connection error: desc = "error reading server preface: read |0: file already closed"
rustic atlas
#

I'll try in a bit 🙏

rustic atlas
halcyon bramble
#

actually it's a docker context

#

docker context create --docker=--host=ssh://...`

rustic atlas
halcyon bramble
#

yes.

#

I have 3 different docker contexts targeting 3 different VMs. they all have the same problem

rustic atlas
#

👍

halcyon bramble
#

It looks like some sort of race condition triggered by the latency in my internet connection maybe

rustic atlas
#

was able to repro 🙏

halcyon bramble
rustic atlas
#

doesn't seem related to latency. remotely connecting to the docker engine via TCP works

#

something about SSH and the stdio transport seems like

halcyon bramble
#

what's weird is I used is successfully a lot until yesterday

#

and use it as my primary configuration at home (to a local machine in my garage)

rustic atlas
rustic atlas
#

I'll continue troubleshooting

#

@dull forum seems to be a bug in the underlying buildkit client connection + backoff mechanism when it uses the stdio connection. I was able to repro by setting that MinConnectTimeout to 1 ms locally against my local engine. Thoughts about bumping that timeout to 10s until we can fix upstream?