Skip to content

docs(context): estate brief — Cygnus single-cluster reality (Lyra decommissioned, GPU→colo)#47

Merged
todie merged 1 commit into
mainfrom
docs/estate-cygnus-single-cluster
Jul 23, 2026
Merged

docs(context): estate brief — Cygnus single-cluster reality (Lyra decommissioned, GPU→colo)#47
todie merged 1 commit into
mainfrom
docs/estate-cygnus-single-cluster

Conversation

@todie

@todie todie commented Jul 23, 2026

Copy link
Copy Markdown
Member

Refresh the cross-harness estate brief constellation section to current ground-truth (2026-07-23 architecture assessment). It still said Lyra is 'retiring / check migration state' and framed the stack as the 'Lyra dev tier'.

  • Cygnus is the SOLE cluster; OPS-468 migration DONE; Lyra decommissioned 2026-07-18 (no dev VKE, no rollback target); dual-target deploymentTarget retired.
  • GPU/inference pivoted to a rented A100→colo path (OPS-549); in-cluster gpu-pool/Dynamo-Triton abandoned.
  • Platform stack reflects live Cygnus: Thanos/Tempo (Jaeger retired)/VictoriaMetrics/Pyrra/Langfuse/Phoenix, SPIRE, Forgejo, LACE; CNPG 3/3; + note that OpenBao/Keycloak/Loki/Tempo remain single-replica SPOFs.

Under the 2400-token budget (~1532). Signed. Re-render (render.sh) refreshes ~/.omp/agent/AGENTS.md + ~/.agents/estate.md.


View with Codesmith Autofix with Codesmith
Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.

…ommissioned, GPU→colo)

The constellation section still said Lyra is "retiring... check migration state"
and framed the platform stack as the "Lyra dev tier". Updated to ground-truth
(2026-07-23 architecture assessment):
- Cygnus is the SOLE cluster; OPS-468 migration DONE; Lyra DECOMMISSIONED
  2026-07-18 (no dev VKE, no rollback target); dual-target deploymentTarget retired.
- GPU/inference pivoted to a rented A100→colo path (OPS-549); in-cluster
  gpu-pool/Dynamo-Triton abandoned.
- Platform stack reflects the live Cygnus reality: Thanos/Tempo(Jaeger retired)/
  VictoriaMetrics/Pyrra/Langfuse/Phoenix, SPIRE, Forgejo, LACE; CNPG 3/3; plus a
  note that OpenBao/Keycloak/Loki/Tempo remain single-replica SPOFs.

Under the 2400-token budget (~1532). Re-render with render.sh to refresh
~/.omp/agent/AGENTS.md + ~/.agents/estate.md.
@todie

todie commented Jul 23, 2026

Copy link
Copy Markdown
Member Author

iac-correctness-critic: APPROVED — all claims live-verified against ~/.kube/cygnus + unsigned-paas docs.

  • Cygnus sole cluster / Lyra decommissioned 2026-07-18 (OPS-468 done): confirmed — single kubeconfig context, no VKE/Lyra namespace or app remnants; matches project memory project_constellation_naming.md.
  • GPU pivot to A100→colo (OPS-549): confirmed — docs/platform-roadmap.md:36, docs/architecture/overview.md:220; live cluster has zero GPU-labeled nodes, no triton/dynamo pods.
  • Platform stack: Thanos (thanos-query/thanos-store-gateway), Tempo (tempo-0; no Jaeger pods), VictoriaMetrics (vmsingle-0), Pyrra, Langfuse+Phoenix namespaces, SPIRE (spire-server+agents), Forgejo, LACE (kagent+hermes-agent) — all present and Running live.
  • SPOF note: OpenBao/Keycloak/Loki/Tempo StatefulSets all confirmed 1/1 (single-replica) live.
  • ArgoCD sync model ("manual for stateful/CRD/credential + small auto-sync set"): confirmed — 12/75 Applications carry syncPolicy.automated, and they're edge/observability apps (cloudflared, traefik, cert-manager, external-dns/secrets, prometheus/grafana, tailscale, priority-classes, gitops-root, lace-deploy, blackbox-exporter), not stateful/credential apps.
  • Token budget: render.sh on the PR content reports ≈1532 tokens, matches the PR body's claim and is under the 2400 gate.

Minor (non-blocking) note: platform-pg CNPG cluster was mid-failover (2/3 ready) at check time — the "3/3" in the doc describes topology/instance-count, not momentary health, so this doesn't contradict the claim.

@todie
todie merged commit 7a27da7 into main Jul 23, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant