scale and build do not work well together

[schmunk42]

I have this Dockerfile

mrb:
  build: .
  environment:
    - affinity:container!=/mrb/

I run this against as swarm. The container should spread across nodes and should be built - not a common use-case, I know. But I tried to overcome issues with host-volumes.

I was able to get two containers up and running, but nonetheless building on a swarm and scale should at least trigger a warning, since you can't really ensure that all nodes have the same version of a built image.

docker version
Client:
 Version:      1.9.1
 API version:  1.21
 Go version:   go1.4.3
 Git commit:   a34a1d5
 Built:        Fri Nov 20 12:50:31 UTC 2015
 OS/Arch:      linux/amd64

Server:
 Version:      swarm/1.0.0
 API version:  1.21
 Go version:   go1.5.1
 Git commit:   087e245
 Built:        
 OS/Arch:      linux/amd64

docker-compose version 1.5.2, build 7240ff3

[aanand]

Yes, this is a known issue with Swarm. For the time being, best practice is:

1. Build your image out-of-band with docker build
2. Push it somewhere with docker push
3. Use image to reference it in docker-compose.yml, instead of build

In Compose 1.6.0, you'll be able to use image and build together if you upgrade to the new file format, so it'll be a little better - you'll be able to write your services like this:

foo:
  build: .
  image: username/imagename

You can still let Compose auto-build it for you when developing, and when deploying to a Swarm you can run:

$ docker-compose pull # pull username/imagename onto all nodes
$ docker-compose up -d

[schmunk42]

I still don't get the use-case of using build and image together (there's another issue I can't find at the moment).

But to add my 2 cents here, isn't it totally confusing to use both, since it's very likely you won't end up with the same image when either pulling or building it?

About my use-case with building and scaling: We run reverse-proxies with SSL certs, which are deployed in a swarm. Rather than pushing it "somewhere" (private registry would be required) I'd prefer to rebuild them on all nodes, if I need to - but I want to make sure, that all nodes have an identical image. Using host-volumes is very hard in swarm and only considered as a workaround at the moment.

Maybe a cluster file-system would be the better option here, but that's another piece to the puzzle. The trade-off to re-building images would be OK for, since it's very fast and a well-tested native feature.

Another related issue #2305

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale]

This issue has been automatically closed because it had not recent activity during the stale period.