trim: lean daughters strip monitoring/flow Kustomizations + flip podMonitor toggles #106

Closed
plat wants to merge 0 commits from fix/lean-daughter-observability into main
Owner

Fixes the r76 cold-start failures (runs 72→77): the parent gitops now assumes the monitoring stack, which the lean daughter profile strips.

  • podMonitorEnabled: true on cnpg-operator renders a PodMonitor with no prometheus-operator CRDs on a fresh daughter → the HelmRelease cannot install → forgejo-pg never exists → germinate dies at the [2/2] handoff (run 77, 20m silent stall).
  • The monitoring/monitoring-scrapes Flux Kustomizations point at dirs the trim deletes, and flow's chart exists only in the parent registry (plat/flow is not a seeded repo) → three permanently-failed Kustomizations → gate A1 (ALL Ready) can never pass.

trim_gitops_lean now drops those Kustomization docs, deletes the flow/monitoring-scrapes dirs, and flips podMonitorEnabled/enablePodMonitor to false. Presence-guarded: byte-identical no-op on an r56-vintage tree (verified against both seed vintages + idempotence pass; kustomize build green on both).

The coldstart gate on THIS PR exercises the old (r56) seed vintage; the r76-seed validation happens on #103 after this merges into its branch.

🤖 Generated with Claude Code

Fixes the r76 cold-start failures (runs 72→77): the parent gitops now assumes the monitoring stack, which the lean daughter profile strips. - `podMonitorEnabled: true` on cnpg-operator renders a PodMonitor with no prometheus-operator CRDs on a fresh daughter → the HelmRelease cannot install → `forgejo-pg` never exists → germinate dies at the [2/2] handoff (run 77, 20m silent stall). - The `monitoring`/`monitoring-scrapes` Flux Kustomizations point at dirs the trim deletes, and `flow`'s chart exists only in the parent registry (plat/flow is not a seeded repo) → three permanently-failed Kustomizations → gate A1 (ALL Ready) can never pass. `trim_gitops_lean` now drops those Kustomization docs, deletes the flow/monitoring-scrapes dirs, and flips `podMonitorEnabled`/`enablePodMonitor` to false. Presence-guarded: byte-identical no-op on an r56-vintage tree (verified against both seed vintages + idempotence pass; kustomize build green on both). The coldstart gate on THIS PR exercises the old (r56) seed vintage; the r76-seed validation happens on #103 after this merges into its branch. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
trim: lean daughters strip monitoring/flow Kustomizations + flip podMonitor toggles
Some checks failed
guard-ppt / no-ppt (push) Successful in 7s
guard-ppt / no-ppt (pull_request) Successful in 7s
coldstart / coldstart (pull_request) Failing after 13m53s
cd86af8430
The r76 reseed gate exposed three fresh-daughter blockers, all one root cause:
the parent's gitops now assumes the monitoring stack, which the lean profile
strips.

- cnpg-operator's podMonitorEnabled:true renders a PodMonitor; without the
  prometheus-operator CRDs the HelmRelease cannot install, forgejo-pg never
  exists, and germinate dies at the [2/2] handoff wait (coldstart run 77).
- The monitoring/monitoring-scrapes Flux Kustomizations point at dirs the trim
  deletes, and flow's chart lives only in the parent registry (plat/flow is not
  a seeded repo) — three permanently-failed Kustomizations, so gate A1
  (ALL Ready) can never pass.

trim_gitops_lean now drops those Kustomization docs (chunk-wise, comments ride
along), deletes the flow/monitoring-scrapes dirs, and flips
podMonitorEnabled/enablePodMonitor to false. All edits are presence-guarded:
byte-identical no-op on an old-vintage (r56) tree, verified against both seed
vintages plus an idempotence pass; kustomize build green on both.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
plat closed this pull request 2026-07-22 16:40:36 +00:00
Author
Owner

Superseded: both fix commits landed on main via #103 (whose solo coldstart gate run 102 passed all 13 assertions on the r76 seed). Old-vintage (r56) tolerance was runtime-proven through the full build chain by gate run 90; the remaining old-seed tail failure was host contention, now moot since main's seed is r76.

Superseded: both fix commits landed on main via #103 (whose solo coldstart gate run 102 passed all 13 assertions on the r76 seed). Old-vintage (r56) tolerance was runtime-proven through the full build chain by gate run 90; the remaining old-seed tail failure was host contention, now moot since main's seed is r76.
Some checks failed
guard-ppt / no-ppt (push) Successful in 7s
Required
Details
guard-ppt / no-ppt (pull_request) Successful in 7s
Required
Details
coldstart / coldstart (pull_request) Failing after 13m53s

Pull request closed

Sign in to join this conversation.
No description provided.