Fix job containers unable to reach DinD daemon for direct docker commands

docker_host: "-" was disabling docker access in job containers entirely,
so steps running `docker build`/`docker` directly had no daemon to talk
to. Point docker_host at the dind sibling and add a host-gateway entry
so the "dind" hostname resolves inside job containers, which run nested
on dind's own internal bridge network rather than the outer compose one.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XFsukF6FzwBxJtRGzbt28F
This commit is contained in:
2026-07-03 15:56:03 +07:00
co-authored by Claude Sonnet 5
parent 6ca26d2309
commit 88d5c6d2f5
+8 -2
View File
@@ -32,12 +32,18 @@ container:
# Talk to the DinD daemon, not the host. # Talk to the DinD daemon, not the host.
network: "" network: ""
privileged: false privileged: false
options: "" # Job containers run *inside* the dind engine, on its own internal bridge network,
# so the "dind" hostname (only known on the outer compose network) doesn't resolve
# there. host-gateway maps it to dind's own bridge-gateway IP, where its dockerd
# listens (started with --host=tcp://0.0.0.0:2375), so job steps can reach it.
options: "--add-host=dind:host-gateway"
workdir_parent: "" workdir_parent: ""
# Allow workflows to mount caches from the DinD volume. # Allow workflows to mount caches from the DinD volume.
valid_volumes: valid_volumes:
- "**" - "**"
docker_host: "-" # Pass DOCKER_HOST into job/service containers so steps that call `docker` directly
# (e.g. `docker build`) can reach the DinD daemon. Paired with the --add-host above.
docker_host: "tcp://dind:2375"
force_pull: false force_pull: false
host: host: