compose file version 3.0 removed support for init

[dobbs]

init works fine in version 2.2:

cat > docker-compose.yml <<EOF
version: '2.2'
services:
  xyz:
    image: alpine
    command: sleep 5m
    init: true
EOF
docker-compose up -d

init fails in version 3.0:

cat > docker-compose.yml <<EOF
version: '3.0'
services:
  xyz:
    image: alpine
    command: sleep 5m
    init: true
EOF
docker-compose up -d
ERROR: The Compose file './docker-compose.yml' is invalid because:
Unsupported config option for services.xyz: 'init'

This documentation:
https://docs.docker.com/compose/compose-file/compose-versioning/
reports that init was introduced in version 2.2. There's no mention if it being removed.

Also tested with version 3.1, 3.2, and 3.3... init fails in all of them

[shin-]

It wasn't removed from v3, it was never added to it in the first place.

[dobbs]

Do you know why wasn't it added?

[shin-]

Because v3 is the format for creating Swarm services, and Swarm services do not support custom init yet.

[batjko]

Because v3 is the format for creating Swarm services

That's news to me. I thought it was simply the sequel to v2, i.e. for the same purpose but also supporting additional stuff like swarms.
If v3 files are made for a different purpose than v2, that's not clear in the docs or anywhere else, at all. Shouldn't be docker-compose then, but rather swarm-compose or something, to make that diversion clearer.

[aaronbieber]

This is not an acceptable explanation. There is no way to understand this from the documentation; the version matrix and upgrade instructions omit any mention of difference in purpose from the 2.x to 3.x lines. If what is explained here by @shin- is actually true, the docs must be updated.

[stewartadam]

This just bit me as well. Given v3 is now marked as current and recommended, it should document the init removal there as well: https://docs.docker.com/compose/compose-file/compose-versioning/#version-3

[techieshark]

"Add --init option to docker service create" was recently merged.

Perhaps this issue should be reopened and changed from a question to a feature request for the (in progress) support for init?