#Depot + Dagger Not Working

1 messages · Page 1 of 1 (latest)

hybrid pond
#

Hi - I've reached out to Depot as well, but figured I'd cross post here just in case anyone has experienced this.

I opened a Depot account, connected my GH Org and Dagger Cloud Org, and am running this GHA workflow:

  test:
    runs-on: depot-ubuntu-latest,dagger=0.15.2
    steps:      
      - run: dagger version
      - run: printenv
      - run: dagger -m github.com/shykes/daggerverse/hello@v0.3.0 call hello --greeting="bonjour" --name="monde"

Dagger is preinstalled, but it is not picking up any dagger cloud token nor a running engine. It pulls the engine image and starts one on the runner. I've tried various version of dagger up to 0.16.1, and various runner labels of differing sizes, OS and architectures. I've also read both Dagger and Depots documentation and watched the related youtube videos. I can't figure out what I'm doing wrong.

wet sorrel
#

Agreed, same issue you're seeing 👆

#

using a standing engine and shipping telemetry to connected Dagger Cloud org

cc @terse blaze @crimson ember

hybrid pond
#

Thanks for the quick response!

terse blaze
#

Hey thanks for flagging this @hybrid pond and @wet sorrel, we have a fix deployed now! We had an unintentional port conflict between GitHub's new caching API and Dagger remote engines that was preventing the remote engines from connecting, that's resolved now.

@hybrid pond I'll ping you directly, I can't find your message to us anywhere, so want to make sure we don't have anything broken on the support side

wet sorrel
#

@gilded rain 👆

hybrid pond
#

I can confirm it is now working for me! Thanks again for the rapid response 🙏 Are these timings expected/normal? I did four runs from the same git ref, same workflow/dagger function call:

  1. fully uncached
  • 14s for dagger client to connect to engine (surprising)
  • 4m46s function runtime (about what I would expect)
  1. fully cached
  • .5s for dagger client to connect to engine (as expected)
  • 6s function runtime (as expected)
  1. fully cached
  • 90s for dagger client to connect to engine (surprising)
  • 7s function runtime (as expected)
  1. fully cached
  • 54s for dagger client to connect to engine (surprising)
  • 5s function runtime (as expected)

In my experience running ephemeral dagger engines w/ distributed caching, the connect timings are a result of the dagger engine pulling our tokens cache volumes from dagger cloud on startup. Happy to take the timings discussion to a more appropriate thread/channel, lmk.

gilded rain
terse blaze
#

That may be an issue on our side, though 90s/54s seem much longer than expected. 14s I could believe if we needed to set up the runner infra / cache volume for the first time. For the connection error, is that at the very beginning? It might align with the time the engine machine is launching.

I can take a look from our side (I'm about to be OOO for the day though) and let you know what I find, unless you discover something else beforehand

warm mulch
#

I am frequently observing hangs of dagger call with Dagger Cloud for around 2-3 minutes with the following logs:

178 : │ │ │ moby.buildkit.v1.Control/Info
178 : │ │ │ moby.buildkit.v1.Control/Info ERROR [0.0s]
178 : │ │ │ ! connection error: desc = "error reading server preface: read tcp 10.20.109.222:45644->10.20.109.222:979: use of closed network connection"

repeated for a really long time.

I'm using Dagger Cloud with Depot. This is not normal right?