Introduction

ComposeFlux is a GitOps tool for managing Docker Compose stacks on home servers. It watches a Git repository and automatically deploys stacks when changes are detected.

Goals

• Manage a few Docker Compose stacks on home servers
• No complex orchestration, clustering, or remote agents
• Local operation only - each server runs its own instance
• Just Git + Docker Compose + Secrets Manager

How Sync Works

ComposeFlux runs a Git sync loop in daemon mode (run command). It performs an initial sync at startup, then checks the remote Git repository for changes and syncs again when updates are detected.

1. Pulls latest commits and tracks changed file paths
2. Fetches secrets from secrets manager
3. Loads environment variables from stack.yml (if present)
4. Discovers compose stacks (one level deep in STACK_PATH)
5. Builds dependency file set for each stack (compose files, include blocks, env files, mounted configs, secrets, build context)
6. Deploys stacks that have file updates or are missing from Docker (respects startup_order)
7. Prunes stacks deleted from Git

Optionally, a separate cron-scheduled image update check (IMAGE_UPDATE_SCHEDULE) pulls new images and redeploys stacks when a new image digest is detected.

Two additional background loops run independently:

• Health reconciliation — checks all managed stacks on HEALTH_RECONCILE_INTERVAL (disabled by default) and redeploys any that are stopped or have exited/dead containers
• Docker resource prune — prunes unused images, volumes, and build cache on PRUNE_INTERVAL (default: 24h), but only when all managed stacks are healthy (see Periodic Docker Resource Pruning)

Git Diff & Dependency Change Detection

ComposeFlux uses a Git diff and dependency-tree-based approach to decide whether a stack needs redeploying:

• Git Diff Path Matching: ComposeFlux tracks modified, added, or deleted file paths between Git commits.
• Dependency Tree Resolution: Each stack's Compose project resolves all related file dependencies, including:
• Compose files and include directives
• Environment files (env_file)
• Mounted configuration files (configs) and secrets (secrets)
• Host bind mounts (volumes)
• Local build context and Dockerfiles (build)
• Base / Overlay Support: Changes in shared base directories (e.g., base/app1 included by overlays/prod/app1) are automatically mapped to dependent stacks.
• Targeted Redeployment: A stack is redeployed only if any changed file in Git overlaps with its dependency file set, or if the stack is missing from Docker.

Image Update Exclusion

Exclude stacks from automatic image updates by adding the composeflux.image-update.exclude: "true" label to any service. If any service has this label, the entire stack is skipped.

Example:

services:
  db:
    image: postgres:15
    labels:
      composeflux.image-update.exclude: "true"

Notes: If ANY service has the label, the entire stack is excluded

Stack Configuration

Optional configuration file in the Git repository within the STACK_PATH directory that allows you to:

• Control deployment order (e.g., deploy Traefik first for proxy/certificates)
• Share environment variables across all stacks

The configuration file should be placed at <repo>/<STACK_PATH>/stack.yml.

Directory structure:

your-stacks-repo/
└── stacks/              ← STACK_PATH
    ├── stack.yml        ← Config file here
    ├── traefik/
    │   └── compose.yml
    ├── nextcloud/
    │   └── compose.yml
    └── jellyfin/
        └── compose.yml

Example:

# Only list stacks that need specific order
# Everything else deploys in whatever order
startup_order:
  - traefik # Must match the directory name in STACK_PATH

# Common variables available to all stacks
envs:
  DOMAIN: homeserver.local
  TZ: America/New_York
  ENVIRONMENT: production

With this configuration, Traefik deploys first, then the rest of the stacks deploy in any order.

Important Notes:

• Scoped to STACK_PATH only - doesn't affect other directories
• Names in startup_order must match directory names exactly
• No need to list all stacks - only ones requiring specific order
• Do not set a custom name: in your compose.yml. The Docker Compose project name must match the stack directory name.

Multi-Server Setup

ComposeFlux runs locally on each server - there's no central controller or remote agents:

Server 1 (homeserver-1)          Server 2 (homeserver-2)
┌─────────────────────┐          ┌─────────────────────┐
│ ComposeFlux         │          │ ComposeFlux         │
│ → stacks/server-1/  │          │ → stacks/server-2/  │
└─────────────────────┘          └─────────────────────┘
         ↓                                ↓
    ┌────────────────────────────────────────┐
    │   Git Repository (shared)              │
    │   your-stacks-repo/                    │
    │   └── stacks/                          │
    │       ├── server-1/   ← Server 1 stacks│
    │       │   ├── app1/                    │
    │       │   └── app2/                    │
    │       └── server-2/   ← Server 2 stacks│
    │           ├── app3/                    │
    │           └── app4/                    │
    └────────────────────────────────────────┘

Example Configuration:

• Server 1: STACK_PATH=stacks/server-1
• Server 2: STACK_PATH=stacks/server-2

Each ComposeFlux instance only manages stacks in its configured directory.

Proactive Stack Health Reconciliation

In addition to Git-triggered syncs, ComposeFlux periodically checks all managed stacks and redeploys any that are unhealthy. This catches stacks that stopped, crashed, or were manually shut down between git ticks — without relying solely on Docker restart policies.

A container is considered healthy if:

• Its state is running, OR
• Its state is exited with exit code 0 and it has the composeflux.init: "true" label

Everything else (dead, paused, exited without the init label, non-zero exit) is unhealthy. A stack is unhealthy if any of its containers are unhealthy.

restarting containers are unhealthy — they are not running. If you want Docker's own restart policy to handle recovery without ComposeFlux intervening, use the Suspend Label to pause health reconciliation for that stack.

Init containers: If your stack uses init containers (short-lived containers that run setup tasks and exit), mark them with the composeflux.init: "true" label so ComposeFlux treats a clean exit (code 0) as healthy:

services:
  migrate:
    image: flyway:latest
    labels:
      composeflux.init: "true"
  app:
    image: myapp:latest
    depends_on:
      migrate:
        condition: service_completed_successfully

Without this label, an exited container (even with exit code 0) is treated as unhealthy and triggers a redeploy.

Recovery action: ComposeFlux calls docker compose up using the existing project config (no git pull). The git ticker continues to handle source drift independently.

Max attempts: After 3 consecutive deploy failures for a stack, health reconciliation skips that stack and logs a warning. The counter resets on the next successful git sync or successful image update.

Configure the check interval with HEALTH_RECONCILE_INTERVAL (default: disabled). Set to e.g. 5m to enable.

Health Suspend Label

You can pause reconciliation for a specific stack by adding the composeflux.health.suspend: "true" label to any service in the stack's compose file:

services:
  db:
    image: postgres:15
    labels:
      composeflux.health.suspend: "true"

Commit the change — ComposeFlux will redeploy with the label applied. To resume reconciliation, remove the label and commit again.

When any container in a stack has this label:

• The health reconciliation loop skips that stack entirely
• The Docker resource prune loop aborts and skips pruning for the entire run

This is useful during maintenance operations — for example, labelling a database service as suspended before stopping it for a backup (docker stop postgres) without triggering an immediate reconcile that would restart it.

Periodic Docker Resource Pruning

When PRUNE_INTERVAL is set, ComposeFlux runs a periodic prune cycle (default: every 24h, configurable via PRUNE_INTERVAL) to reclaim disk space from unused Docker resources. Set PRUNE_INTERVAL=0 to disable pruning entirely.

What is pruned: dangling (untagged) images, volumes, build cache. Containers and networks are not pruned.

Safety guard: The prune cycle only runs when all composeflux-managed stacks are healthy. If any stack is stopped, degraded, or has the composeflux.health.suspend=true label set, the prune cycle is skipped for that interval and a warning is logged.

Blog Posts

To learn more about the motivation behind ComposeFlux and see it in action:

• GitOps for Homeservers (Part 1) — My Homeservers, Ansible, and the Pain Points
• GitOps for Homeservers (Part 2) — Searching for the Right Tool
• GitOps for Homeservers (Part 3) — ComposeFlux: A Lightweight GitOps Tool
• How I Manage My Homeservers with GitOps and Docker Compose (Medium)

Limitations

• Nested stack discovery (only scans one level deep)
• Multi-server orchestration (no central controller)
• Rolling updates or zero-downtime deployments
• Built-in monitoring or alerting

Stack Discovery is One Level Deep:

stacks/
├── app1/            ← Discovered ✓
│   └── compose.yml
├── app2/            ← Discovered ✓
│   └── compose.yml
└── nested/
    └── app3/        ← NOT discovered ✗
        └── compose.yml

💡 Use a flat structure. For multi-server setups, create separate directories per server.