Skip to main content

The configuration record

Every ignition-stack create writes the usual runnable tree — docker-compose.yaml, .env, the seed tree, a Makefile, and POST-SETUP.md — plus one more file: .ignition-stack/config.json, the resolved configuration the stack was built from.

That record is a single source of truth. It holds the fully-expanded topology — every gateway, its ports and edition, the database, the selected services, the network layout — after the resolver has made implicit dependencies explicit. It is exactly what the generator consumed, so reproducing it reproduces the stack.

Re-running and cloning with create -f​

Pass the record back to create to recreate or clone the stack:

# Recreate the same stack (after wiping it, for example)
ignition-stack create demo -f demo/.ignition-stack/config.json

# Clone it under a new name — the positional name overrides the recorded one
ignition-stack create demo-copy -f demo/.ignition-stack/config.json

The positional name argument always overrides the name in the file, so the same config file can produce as many distinct compose projects as you need. Everything else — architecture, database, services, editions, ports — comes from the file.

Editing the file before re-running is a supported workflow. The format is the same YAML or JSON schema that create --dry-run dumps, so you can adjust any field and rebuild from the edited copy.

The declarative dump/build loop​

create --dry-run dumps the resolved config to stdout without writing anything to disk:

ignition-stack create demo --arch scale-out --dry-run > arch.yml

That dump is byte-identical to the .ignition-stack/config.json the generator would write, so you can inspect it, version it, and rebuild from it:

ignition-stack create demo -f arch.yml

See Declarative config for the full dump/edit/rebuild guide.

Why it is always written​

The record is small, lives in a dot-directory out of the way of the hand-readable files, and costs nothing to ignore if you don't need it. Making it unconditional means every project is re-runnable and every project can be dumped — there is no mode to choose at create time and no project that is stuck without a record.

make wipe does not need the record — it scopes teardown to the compose project name baked into the Makefile — so a project stays tearable even if you delete .ignition-stack/.