docs/nastri-rollout-deadline-1379.md

Rama main · réplica de lectura del repositorio original.

# Sequential rollout readiness budgets

Nastri backend issue #1379 observed setup and earlier services consume the shared worker deployment deadline; Beat then failed before its healthy readiness arrived. The worker now reserves the installation readiness timeout for every regular service without an explicit startup timeout, alongside existing explicit startup and setup-job budgets. The operation and recovery budgets retain bounded cancellation. Each service still has its own readiness timeout and diagnostics; failure is not reclassified as success.

Source base: bc3b46803d3bba40c6438ddb27e190c628d8622a. Read-only Nastri application revision292 had seven regular services without startup_timeout_seconds and a setup job timeout of900seconds. Backend CI's existing ten-minute deployment poll is shorter than that setup allowance; its bounded wait needs separate review after the engine upgrade, while failed-state rejection, digest identity and SHA gates remain mandatory.

Targeted checks ran only on isolated hakopod-lightweight-dev (ubuntu@orb), in ~/cn158-1379-build. Go1.26.8 archive SHA256211ffced9dcb9633a55eac6364816ec0ddd951389a740e88fa8b3337971bdda0 matched official go.dev release index. Worker budget/slow-setup late-Beat checks passed0.217seconds. Kubernetes-client readiness checks passed0.156seconds, including unhealthy readiness timeout under a longer aggregate parent and old-generation pod exclusion. Scoped worker/cluster vet passed. Initial missing pinned templates submodule and killed Oracle dependency vet failures remain in scratch logs; the final source uses exact templates22584117e12c929b4c03a677f41816365cca44e0.

These are component checks with synthetic scheduling and a fake Kubernetes API, not real cluster or running-instance acceptance. Required installer native smoke/host matrix, published alpha62 artifacts, supported upgrade of the actual Narstri control plane, and deployed slow-setup/late-healthy-Beat versus unhealthy-timeout evidence remain outstanding. No customer workloads or production configuration were changed by this implementation.

PR #193 initial CI run38063974406 failed on a nil-client fixture; run38063974806 rejected the unchained source changes before native smoke/host jobs. The nil-client accessor is repaired in d4bfae6b3782a136398398f5f0af59526dc2706b. The alpha62 review chains exact published61 predecessor records while preserving historical native artifact identities. VM compatibility-repair-tests.log passed21 targeted checks, including alpha62 anchor, historical artifact identity, undeclared source and protected-path rejection across all three runtimes. Exact current runtime inventories validate through the production verifier functions. Successful CI38069631268 on exact candidate c3a0f012 passed Go/server/dashboard/SDK. Installer38069631550 passed exact package build, native smoke on amd64/arm64, and all12 fresh/alpha60/61 host upgrade cases; compatibility evidence records now point to that genuine green CI. This candidate remains unmerged and unpublished; successful existing-engine Nastri revision292 means upgrade necessity is not established.