Shared services

Run Postgres, Redis, Elasticsearch… once, and reach them by name from every workspace of every project.

Instead of every project booting its own Elasticsearch, declare one — once — and every worker of every project reaches it by DNS, at its name. A service is long-lived and independent: it has its own start, stop, restart and logs, and it outlives the workers, and the projects, that use it.

Declaring one

Services in the sidebar, then a preset or a blank form. A service is:

• a slug — its DNS name (postgres);
• an image and tag, and optionally a command override;
• environment variables — literal values, or references to global secrets;
• ports, one of which can be marked as its HTTP port;
• persistent volumes, named, with their mount path;
• a restart policy (unless-stopped by default).

Presets fill the form for Postgres, MySQL, Redis, Elasticsearch, Kibana, MinIO and Mailpit. There is one container per service: a stack is several services — Elasticsearch and Kibana, the latter pointing at http://elasticsearch:9200, which is exactly what the two presets do.

Reaching it from a workspace

Every service container joins a shared Docker network (mp-shared-net) under its slug, and every worker joins it when it is created. So, from any workspace:

psql -h postgres -U postgres
curl http://elasticsearch:9200
redis-cli -h redis

Each worker also gets the address in its environment, so nothing has to hard-code it:

SPUNTO_SERVICES=elasticsearch,postgres
SPUNTO_SVC_ELASTICSEARCH=http://elasticsearch:9200
SPUNTO_SVC_ELASTICSEARCH_HOST=elasticsearch
SPUNTO_SVC_ELASTICSEARCH_PORT=9200
SPUNTO_SVC_POSTGRES=postgres:5432
SPUNTO_SVC_POSTGRES_HOST=postgres
SPUNTO_SVC_POSTGRES_PORT=5432

• SPUNTO_SVC_<SLUG> is a full http:// URL only when the service declares an HTTP port; otherwise it is host:port.
• The slug is uppercased, hyphens become underscores (opensearch-dev → SPUNTO_SVC_OPENSEARCH_DEV).
• Only services that are up are listed.
• The DNS name works at once from every worker; the variables are set when a worker is created, so a service declared later shows up in them after a Rebuild.

A secret of the same name wins over a SPUNTO_SVC_* variable.

In the browser

A service with an HTTP port is browsable through Lite’s own proxy, without publishing anything on the host:

Publishing a port on the host stays possible, per port (“on host” in the form) — for a psql or a redis-cli run outside any container.

Lifecycle

• Editing recreates the container — image, env, ports and volumes are fixed at creation. Its named volumes are kept, so the data survives. A stopped service stays stopped.
• Volumes are named mp-svc-<serviceId>-<name>, so renaming a slug never orphans data. They appear on the Resources page.
• Deleting a service deletes its volumes. That is permanent, and the confirmation says so.
• A container that dies on its own shows as error, with the daemon’s message and its exit code — distinct from one you stopped.
• The card shows live CPU and memory; Logs tails the container.

Why one shared network

Attaching each service to each worker’s own network would isolate better, at the cost of a connect/disconnect for every pair on every spawn. On a single-user machine there is nothing to isolate from: every worker is yours, and services are declared precisely to be shared.