feat(compose): add logosdeliverynode docker-compose stack

[NagyZoltanPeter]

What

Adds apps/logos_delivery_node/compose/ — a docker-compose project that runs the logosdeliverynode image as a service node with a Postgres store and a Prometheus + Grafana monitoring stack. Ported from logos-messaging/logos-delivery-compose and adapted for logosdeliverynode instead of wakunode2.

Services

Service	Purpose
logos-messaging-node	logosdeliverynode (built from the repo Dockerfile, MAKE_TARGET=logosdeliverynode, -d:postgres; overridable via LOGOS_IMAGE)
postgres + postgres-exporter	message store + metrics
prometheus + grafana	monitoring (Grafana provisioned with the node dashboard)
certbot	optional Let's Encrypt certs for WebSocket-Secure

Configuration

• PRESET (default logos.dev) and ENTRY_LAYER (default kernel) are env-configurable via .env; empty PRESET disables the preset.
• --entry-layer is always emitted (the binary's own default is channels, not kernel).
• REST + admin enabled; Postgres store enabled with size-based retention (STORAGE_SIZE).

Notable decisions / fixes

• .dockerignore: exclude /nimbledeps and /build. A populated host nimbledeps/ leaked into the build context and had its package submodules stripped by **/vendor/*, breaking the in-container make build-deps. Excluding it lets the container resolve deps fresh (context dropped ~4.6 GB → ~370 kB) and also speeds up the existing wakunode2 docker builds.
• Postgres readiness: node depends_on is gated on postgres: condition: service_healthy to avoid a startup crash-loop on connection-refused.
• WSS: enabled only when DOMAIN is explicitly set (dropped the reference's reverse-DNS auto-guess, which could deadlock the node waiting for a cert that certbot never issues).
• Mix: --mix=true is passed explicitly — the logos.dev/logos.test presets set the mix flag but do not mount the mix protocol on their own (also affects wakunode2 --preset=logos.dev; worth a separate issue).
• Branding: Grafana logos replaced with the official Logos mark; dashboard file named logos-delivery-monitoring.json.
• Omitted for now (per scope): RLN settings/files, setup_wizard, RLN keystore tooling.

Verification

• docker compose build produces a working image; binary boots and joins the network.
• Brought the full stack up on both logos.dev/kernel and logos.test/messaging:
  • node reaches nodeHealth: READY, connects to bootstrap peers, discv5 up, Postgres store migrates.
  • Postgres messages table accumulates live traffic.
  • /admin/v1/peers + libp2p_peers reconcile; Grafana scrape targets up.
  • Messaging REST round-trip (entry-layer=messaging): POST /messaging/v1/messages → requestId; GET /messaging/v1/events/send/{id} shows propagated+sent; GET /messaging/v1/events/received returns the message — all correlated with send_service.nim log lines by matching requestId/msgHash.

🤖 Generated with Claude Code

[github-actions]

You can find the images built from this PR at

quay.io/wakuorg/nwaku-pr:4057
quay.io/wakuorg/nwaku-pr:4057-logosdeliverynode

Built from d75f250

[fcecin]

Great!

Got Fable 5 to review so maybe this helps:

# Review: PR #4057 — feat(compose): add logosdeliverynode docker-compose stack

**Verdict: LGTM after two small fixes.** The port from logos-delivery-compose is faithful
(most files are byte-identical to the reference), the adaptation in run_node.sh /
docker-compose.yml matches the PR description exactly, and the stack was clearly
live-verified. Two things should be fixed before merge, both quick.

## Must fix

### 1. Dead dashboard panels after the Prometheus job rename

The dashboard JSON is a verbatim copy of the reference `nwaku-monitoring.json`, but this PR
renames the Prometheus scrape job from `nwaku` to `logos-messaging-node` (and thus the
instance label from `nwaku:8003` to `logos-messaging-node:8003`). Two panels still pin the
old names and will show "No data":

- panel **"Start Times (UTC)"**: `process_start_time_seconds{job="nwaku"}`
- panel **"Version"**: `waku_version{instance="nwaku:8003"}`

Fix: search-replace `nwaku` -> `logos-messaging-node` inside
`monitoring/configuration/dashboards/logos-delivery-monitoring.json` (4 occurrences total).
Everything else in the dashboard is either unfiltered or keyed on the `$Instance` template
variable (which is derived from the postgres-exporter job and unaffected).

### 2. Missing `target:` in the compose build

The `build:` block for `logos-messaging-node` has no `target:`, so `docker compose build`
builds the **last** stage of the root Dockerfile — `debug-with-heaptrack`. That means:

- heaptrack is cloned from KDE and compiled with cmake for nothing (extra build time), and
- the network-facing node image ships gdb + heaptrack.

It happens to work only because the compose service overrides `entrypoint: sh` (the stage's
own entrypoint is heaptrack). Add `target: prod` now — or `target: logosdeliverynode` once
PR #4059 (which adds that stage) merges.

## Notes (non-blocking)

- **`.env.example` DOMAIN comment is stale**: it says "Leave empty to auto-detect from
  reverse DNS", but this PR deliberately removed the reverse-DNS auto-guess (run_node.sh
  documents why — the cert-wait deadlock). The README already has the correct wording
  ("empty disables WSS"); align .env.example with it.
- **PR body is stale on `.dockerignore`**: it says "/nimbledeps and /build" are excluded,
  but the follow-up commit correctly kept `/build` in the context (docker/binaries/
  Dockerfile.bn.amd64 ADDs the prebuilt binaries from ./build/). Final state excludes only
  /nimbledeps — worth updating the description.
- The `container-image.yml` serialize commit is byte-identical to the one in PR #4058 —
  merges cleanly in either order, whichever lands second just contributes an empty change.
- Inherited verbatim from the reference repo (fine to leave, listing for awareness):
  - run_certbot.sh defines `EMAIL` but the certbot call hardcodes `--email admin@DOMAIN`;
  - chkhealth.sh passes a stray `GET` argument to curl (works by accident — curl treats it
    as a failing extra URL and the exit status of the last transfer wins);
  - set_storage_retention.sh's sudo check (`[ "$(... | grep -c ...)" ]`) is always true, so
    it uses sudo whenever ./postgresql exists;
  - Grafana runs with anonymous **Admin** on 0.0.0.0:3000 (the README does say so);
  - the certbot service always binds host port 80, even with DOMAIN unset.

[Ivansete-status]

LGTM!
I haven't looked at the changes in detail but I like the idea. We might need to deprecata/archive the former docker-compose repo: https://github.com/logos-messaging/logos-delivery-compose.

[Ivansete-status]

I think is more explicit to use the term docker for that folder's name.

Suggested change

	cd apps/logos_delivery_node/compose
	cd apps/logos_delivery_node/docker