Separate RedDwarf control plane from persistent remote workers #2

Merged

toasterson merged 2 commits from codex/cluster-workers-20261001 into main

2026-10-01 20:50:17 +00:00

toasterson commented

2026-10-01 20:04:10 +00:00

Owner

RedDwarf WI-1: 01a0f6e0-dbd4-7051-b3e0-40ae5c21ad65

Adds control-plane and worker modes while preserving serve and all-in-one agent. One API/database/scheduler owns cluster resources; each worker holds only local identity and persistent IPAM. Atomic node admission rejects duplicate identity/name, CIDR changes and overlapping pod/service networks. HTTP clients validate the origin/CA, use bounded requests, and workers retry transient registration/heartbeat outages. Remote assigned-pod reconciliation uses documented bounded polling.

Validation at 74b7fcc19a: native oibldofgate workspace tests, binary build, formatting, strict all-targets Clippy, SMF validation, API/TLS process smoke and two independent worker processes all passed. Topology checks include concurrent overlapping registration, state-file double start, heartbeat, assigned-node filtering, worker restart/persistent identity and IPAM, control-plane outage/reconnect and deletion. Local tests/lint and topology smoke also passed.

Mock runtime is an explicit native test mode without ZFS/zone/NAT mutations. This is transport proof; real zone images, owned forwarding, API authentication and CI/runner pilot remain separate plan milestones. No new binary target is added. Native SSH checks are not Anima CI: canonical CI is currently unconfigured.

RedDwarf WI-1: 01a0f6e0-dbd4-7051-b3e0-40ae5c21ad65 Adds `control-plane` and `worker` modes while preserving `serve` and all-in-one `agent`. One API/database/scheduler owns cluster resources; each worker holds only local identity and persistent IPAM. Atomic node admission rejects duplicate identity/name, CIDR changes and overlapping pod/service networks. HTTP clients validate the origin/CA, use bounded requests, and workers retry transient registration/heartbeat outages. Remote assigned-pod reconciliation uses documented bounded polling. Validation at 74b7fcc19a93a16a2846116f317ffd3de551e07c: native oibldofgate workspace tests, binary build, formatting, strict all-targets Clippy, SMF validation, API/TLS process smoke and two independent worker processes all passed. Topology checks include concurrent overlapping registration, state-file double start, heartbeat, assigned-node filtering, worker restart/persistent identity and IPAM, control-plane outage/reconnect and deletion. Local tests/lint and topology smoke also passed. Mock runtime is an explicit native test mode without ZFS/zone/NAT mutations. This is transport proof; real zone images, owned forwarding, API authentication and CI/runner pilot remain separate plan milestones. No new binary target is added. Native SSH checks are not Anima CI: canonical CI is currently unconfigured.

toasterson added 2 commits

2026-10-01 20:04:10 +00:00