Skip to content

Dual Runtimes: Docker & Podman (Rootless), Config Mounting (files) & Priorities ​

Starry natively orchestrates containers across both major Linux engines: Docker Engine and Podman (in both rootful and unprivileged rootless namespaces).

Communication operates directly over Unix Domain Sockets using pure .NET SocketsHttpHandler with zero third-party SDK dependencies.


1. Engine Comparison: Docker vs Podman ​

FeatureDocker EnginePodman (Rootful / Rootless)
ArchitectureClient-server daemon dockerdDaemonless, socket activated via systemd.socket
Socket Location/var/run/docker.sock/run/podman/podman.sock (rootful)
/run/user/$UID/podman/podman.sock (rootless)
PrivilegesRequires root / docker groupFull unprivileged Rootless user-namespace support
P2P Streamingdocker save | zstd | ssh docker loadpodman save | zstd | ssh podman load

Starry automatically detects the active socket and adapts volume mounts and networking configurations accordingly.


2. Dynamic Config Mounting (files) & SIGHUP Reloads ​

Services frequently require configuration files (e.g. nginx.conf, prometheus.yml, settings.toml) that evolve independently from container images.

Starry provides a first-class files mapping section in service manifests:

yaml
services:
  web:
    image: "nginx:alpine"
    node: "bocchi"
    priority: 80
    files:
      /etc/nginx/conf.d/default.conf: |
        server {
            listen 80;
            server_name _;
            location / {
                proxy_pass http://api.internal:8080;
            }
        }

Hot-Reload Lifecycle: ​

  1. Starry computes the SHA-256 checksum for each mounted file.
  2. Files are persisted under /var/lib/starry/files/<service>/... and mounted read-only into the container.
  3. When file content changes, Starry atomically rewrites the host file.
  4. If the container is already running, Starry dispatches a SIGHUP signal (POST /containers/{id}/kill?signal=SIGHUP). Processes like Nginx or Prometheus reload their configuration without dropping connections or restarting the container!

3. Numeric Priority Scheduling (priority: int) ​

To guarantee that infrastructure dependencies start in the correct order, services specify an integer priority from 0 to 100:

Priority TierCategoryExamples
0 – 9Registries & CorePrivate Docker Registry
10 – 29Databases & StoragePostgreSQL, Redis, ClickHouse, RabbitMQ
30 – 69ApplicationsBackend APIs, job workers
70 – 100Ingress ProxiesNginx, Caddy, Traefik

Starry sorts deployment execution ascending by priority: databases are confirmed running before backends attempt to connect!

Released under the MIT License.