Arch bump: package the full o2r system as a Helm chart #25

Closed
opened 2026-06-02 02:52:24 +00:00 by coilysiren · 2 comments
Owner

Arch bump

Package the full o2r system as a Helm chart so it runs as a real, always-on Kubernetes service rather than make-target demos. Scope: core relay, the Agent Channel layer (Postgres-backed), and the backend extension(s), deployable as one chart with values for peer registry, OTLP endpoint, and channel DB.

Why now

This is the major proto/arch change and it gates external outreach. o2r should be deployable and continuously running (ideally fronting real sessions) before pointing peers at it. It also unblocks standing up more real test peers across machines (a Beelink SER8 in particular) as A2A nodes that talk through a deployed relay instead of localhost.

Acceptance (rough)

  • chart/ with the relay + channels + at least one backend extension, templated values.
  • Runs on the homelab k8s, reachable as a stable service.
  • Real multi-host A2A sessions flow through the deployed relay and land as traces.

Sequencing peer: the existing o2r backlog (baseline chores + #1/#3/#14/#16) burns down around this.

## Arch bump Package the full o2r system as a Helm chart so it runs as a real, always-on Kubernetes service rather than `make`-target demos. Scope: core relay, the Agent Channel layer (Postgres-backed), and the backend extension(s), deployable as one chart with values for peer registry, OTLP endpoint, and channel DB. ## Why now This is the major proto/arch change and it **gates external outreach.** o2r should be deployable and continuously running (ideally fronting real sessions) before pointing peers at it. It also unblocks standing up more real test peers across machines (a Beelink SER8 in particular) as A2A nodes that talk through a deployed relay instead of localhost. ## Acceptance (rough) - `chart/` with the relay + channels + at least one backend extension, templated values. - Runs on the homelab k8s, reachable as a stable service. - Real multi-host A2A sessions flow through the deployed relay and land as traces. Sequencing peer: the existing o2r backlog (baseline chores + #1/#3/#14/#16) burns down around this.
Author
Owner

Design head-start (2026-06-01)

Full design brief for a from-scratch session lives in the vault at Notes/o2r-helm-chart-handoff.md. Key points for anyone picking this up:

Greenfield for deploy. No Dockerfile, no chart, no deploy docs today. Step zero is a container image, then the chart.

Deployable surface to chart:

  • core relay (otel-a2a-relay-core): env OTEL_A2A_RELAY_{HOST,PORT,PEERS,STAR_ENFORCE}, OTEL_EXPORTER_OTLP_ENDPOINT, port 8080.
  • channels (otel-a2a-relay-channels): Postgres-backed FastAPI router, needs a DB.
  • backend extensions (arize-phoenix, tempo-grafana): dev/validation harnesses today, open whether any ship in-cluster.

Precedent to model: coilyco-flight-deck/lunch-money-k8s already has the full chart shape (chart/ with Chart.yaml, values.yaml, values.schema.json, deployment/service/ingress/configmap/secret/serviceaccount/hpa/pdb/networkpolicy templates, tests/test-connection), a Dockerfile, docs/deploy.md, and coily docker/helm-lint/helm-template verbs. Reuse wholesale.

Open design questions: one chart (relay+channels+Postgres) vs relay-only + external DB; bundle Postgres vs require existing; single image vs two; which backend (if any) in-cluster vs backend-agnostic OTLP; how cross-host peers (SER8) register/reach the relay over the tailnet; SSM-to-cluster secret delivery.

Conventions: image at docker.io/coilysiren/otel-a2a-relay; SSM-canonical secrets with existingSecret support; route dev verbs through coily and add chart verbs to .coily/coily.yaml first.

## Design head-start (2026-06-01) Full design brief for a from-scratch session lives in the vault at `Notes/o2r-helm-chart-handoff.md`. Key points for anyone picking this up: **Greenfield for deploy.** No Dockerfile, no chart, no deploy docs today. Step zero is a container image, then the chart. **Deployable surface to chart:** - core relay (`otel-a2a-relay-core`): env `OTEL_A2A_RELAY_{HOST,PORT,PEERS,STAR_ENFORCE}`, `OTEL_EXPORTER_OTLP_ENDPOINT`, port 8080. - channels (`otel-a2a-relay-channels`): Postgres-backed FastAPI router, needs a DB. - backend extensions (arize-phoenix, tempo-grafana): dev/validation harnesses today, open whether any ship in-cluster. **Precedent to model:** `coilyco-flight-deck/lunch-money-k8s` already has the full chart shape (`chart/` with Chart.yaml, values.yaml, values.schema.json, deployment/service/ingress/configmap/secret/serviceaccount/hpa/pdb/networkpolicy templates, tests/test-connection), a `Dockerfile`, `docs/deploy.md`, and coily `docker`/`helm-lint`/`helm-template` verbs. Reuse wholesale. **Open design questions:** one chart (relay+channels+Postgres) vs relay-only + external DB; bundle Postgres vs require existing; single image vs two; which backend (if any) in-cluster vs backend-agnostic OTLP; how cross-host peers (SER8) register/reach the relay over the tailnet; SSM-to-cluster secret delivery. **Conventions:** image at `docker.io/coilysiren/otel-a2a-relay`; SSM-canonical secrets with `existingSecret` support; route dev verbs through coily and add chart verbs to `.coily/coily.yaml` first.
Author
Owner

Closing. The Helm-chart / deployable-service work split out into the otel-a2a-relay-deploy repo; fleet onboarding for it is tracked in #41 (migrated here from coilysiren/inbox).

Closing. The Helm-chart / deployable-service work split out into the `otel-a2a-relay-deploy` repo; fleet onboarding for it is tracked in #41 (migrated here from coilysiren/inbox).
Commenting is not possible because the repository is archived.
No description provided.