Docker remains stopped after persistent data-disk provisioning during headless boot

[CFBancroft]

I originally reported this as a follow-up comment on Lima PR #5489 while testing headless autostart. After isolating the behavior further, this appears to be a Colima persistent data-disk provisioning issue, so I am opening it here for proper tracking.

Environment

• 2018 Intel Mac mini (Macmini8,1)
• Intel x86_64
• macOS Sequoia 15.7.9
• Colima 0.10.3
• Docker runtime
• Lima VM started headlessly by a LaunchDaemon
• Dedicated service user: containerserver

The Lima LIMA_HOME autostart problem was handled separately by lima-vm/lima#5489.

After that problem was resolved, Lima successfully started the Colima VM at boot, but Docker and restart-policy containers still did not become available until the first Docker API access.

Observed behavior

During a true headless boot:

1. Lima starts the Colima VM successfully.
2. Docker starts inside the VM.
3. Colima persistent data-disk provisioning runs:
   • systemctl stop docker.service
   • systemctl stop containerd.service
4. Colima bind-mounts the persistent runtime directories.
5. Provisioning completes without explicitly starting Docker again.
6. docker.service remains inactive/dead.
7. docker.socket remains available.
8. The first Docker API access, such as docker ps, socket-activates Docker.
9. Restart-policy containers then start.

Before the first Docker API access, a DokuWiki container with
restart: unless-stopped was not reachable on port 5555.

This means the Lima VM itself is awake and running, but the Docker service does not return automatically after Colima's persistent data-disk provisioning.

Source investigation

In the current Colima source, Docker's DataDisk() includes:

PreMount: []string{
    "systemctl stop docker.service",
    "systemctl stop containerd.service",
},

I did not find a corresponding post-mount start of docker.service.

The Lima disk-mount provisioning runs the PreMount commands and then performs the persistent bind mounts, but there was no post-mount stage to start Docker again.

Experimental fix tested

I added a generic PostMount field to DataDisk:

type DataDisk struct {
    Dirs      []DiskDir
    PreMount  []string
    PostMount []string
    FSType    string
}

For the Docker data disk:

PostMount: []string{
    "systemctl start docker.service",
},

I then extended the Lima disk provisioning in Colima to append the PostMount commands after all persistent bind mounts.

The resulting provisioning order is:

stop Docker/containerd
→ mount persistent runtime directories
→ start Docker
Test result

I built the modified Colima natively on the Intel Mac:

colima version v0.10.3-19-gc3a5f91
git commit: c3a5f9184d83a197184f897a9f07eb3c01b3bc88

The real headless containerserver Colima instance was regenerated with the modified binary.

The resulting lima.yaml contained:

script: systemctl start docker.service

after the persistent bind mounts.

I then tested:

normal reboot — PASS
reboot with no GUI login — PASS
SSH-initiated reboot — PASS
physical monitor disconnected — PASS
no docker ps or other Docker API wake-up — PASS
DokuWiki container on port 5555 returned automatically — PASS

This changed the behavior from:

VM running → Docker inactive until first Docker API access

to:

VM running → persistent mounts complete → Docker starts → restart-policy containers start automatically
Question

Would a generic PostMount stage on DataDisk be an appropriate way to handle this, or would the Colima maintainers prefer Docker to be restarted through a more Docker-specific mechanism after the persistent runtime directories are mounted?

I have kept the working local patch unchanged for further testing if useful.

Tested-by: CFBancroft — 2018 Intel Mac mini (x86_64)

Assisted-by: ChatGPT (GPT-5.6 Sol)