Stack

Staging host: Tailscale clc-gnosis (100.108.72.127), compose in clc-gnosis-repo.

Service	Image	Publish	Watchtower
postgres 18	`postgres:18-alpine`	`54432:5432`	no
nats	`nats:2` (`-js`)	internal	no
eth-tracker	`eth-tracker:v1.14.0-stable`	`127.0.0.1:5001`	no
eth-indexer	`eth-indexer:latest`	`127.0.0.1:5002`	yes
clc-core	`clc-core:develop`	via Caddy	yes
storage-server	`storage-server:v0.8.0-rc`	via Caddy	yes
caddy	`caddy-docker-proxy:2.10-alpine`	80/443	no
watchtower	`containrrr/watchtower`	none	n/a

Watchtower: --interval 120 --label-enable --cleanup --rolling-restart. Only containers with com.centurylinklabs.watchtower.enable=true.

Hostnames:

gnosis-staging.cosmolocal.credit → clc-core
gnosis-storage.cosmolocal.credit → storage-server

Env mapping

CORE_A__B → config key a.b. A value with a comma or space becomes a list. In core.env only, a value containing \n is passed through unsplit (PEM keys).

Gotchas:

Lists. koanf Strings() does not promote a single env value to []string. Repeat the NATS URL (nats://nats:4222,nats://nats:4222) in core.env / indexer.env. Do not do that for bootstrap.ge_registry (empty element panics). Put the registry in tracker-config.toml.
Tracker reads jetstream.endpoint (singular). The image's endpoints key is unused.
CORE_CHAIN__DIVVI_CONSUMER cannot be empty. Use the zero address.
Do not mount config.toml over Watchtower-updated images. Tracker is pinned, so a mount is safe. clc-core / indexer would freeze on a stale file.

Ordered start

docker compose up -d postgres nats
docker compose logs -f postgres | grep -m1 "database system is ready"
docker compose up -d eth-tracker
docker compose up -d eth-indexer clc-core storage-server
docker compose up -d caddy watchtower

init_db.sql creates chain_data only on an empty volume.