Implement and qualify a concrete group-lease and fenced-history provider

Aperta
#2,904 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
18/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
go, kubernetes, sqlite

Direzione di ricerca

Start by reading parent issues #2666, #2761, and #2760, then inspect the existing lease.Provider / Attempt / Hold contracts and the Backend interface mentioned in this issue. Identify the concrete deployment and storage boundary before assessing implementation work; the issue leaves that selection open. Done means qualification demonstrates obsolete owners cannot append to member histories and the documented support and failure boundaries are explicit.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Outcome

A selected concrete deployment has an implemented, qualified group-lease provider and history-storage composition that atomically fences obsolete writers and fails closed for new execution when authority cannot be established. Claims match evidence for that deployment, not only the in-process memory proof.

Parent: #2666, under #2659. This carries concrete provider/deployment obligations moved out of #2761. #2761 establishes group-scoped contracts and a fixed two-history memory reference; it does not qualify SQLite, independent services or distributed execution.

Scope

  • Review and select a concrete ownership-record/history-storage composition and supported deployment/failure model. Evaluate existing Kubernetes Lease coordination and other existing mechanisms without assuming a Lease token alone fences storage or preselecting Redis, shared filesystems or a particular new service.
  • Implement the lease.Provider / Attempt / Hold contract with authoritative group membership and incarnation identity, stable acquisition retry, renewal/release, and conservative local validity evidence. Integrate with authoritative catalog work under #2666.
  • Establish atomic current-generation/validity plus expected-head append acceptance at the real storage boundary. Separate services/interfaces are permitted only with a concrete protocol preserving that invariant; a lease read followed by unfenced append is insufficient.
  • Qualify restart/incarnation reuse, competing acquisitions, uncertain grant/fence publication, delayed messages, lease-record cleanup/non-reuse, renewal outages, clocks/deadlines, stale release/renewal and ownership loss during active model/tool/storage work. Preserve known acceptance and unresolved external-effect uncertainty.
  • Define ownership discovery/forwarding integration needed by the selected topology. A healthy incumbent is not stolen; location/affinity information is a potentially stale routing hint, never execution authority. Peer forwarding versus an ownership-aware front door remains a design decision, not a required new load-balancer framework.
  • Wire ownership maintenance, final local admission, bounded graceful drain/release and observable unresolved obligations through the actual host/provider composition. Final production activation remains separately approved in #2672.

Acceptance

  • Provider/storage/deployment and clock assumptions, authority linearization, Go/stored/wire compatibility and upgrade consequences are reviewed before implementation; required breaks receive explicit approval.
  • The actual storage rejects obsolete-owner writes to multiple member histories with still-correct expected heads and no successor append, including an unloaded member and restart/competing-process cases appropriate to the supported deployment.
  • Failed/uncertain takeover or fence publication cannot yield usable unproved authority; exact acquisition retries/abandonment cannot create duplicate ownership or affect a successor.
  • Local admission/loss supervision remains responsive during blocked renewal/submission; only provably valid grants permit progress, and a locally retired incarnation never revives.
  • Concrete interruption/clock/cleanup/durability tests demonstrate safe generation retirement and stale-writer rejection. Expected-head CAS or Kubernetes Lease CAS alone is not cited as proof of history fencing.
  • Applicable routing, authorization, resource-cleanup, graceful shutdown and operational diagnostics are exercised through real composition with isolated state and documented residual uncertainty.
  • Deployment support boundaries and residual #2672 assembled qualification are documented; production cutover remains a separate explicit decision.

Dependencies and boundaries

Build on #2761 and #2760 plus catalog/lifecycle delivery under #2666. Coordinate #2665 recovery, #2669 shared controls, #2670 authorization, #2671 delegation and #2672 final integration. Coordinate semantics/test lessons with legacy-store fencing #2064; this issue does not claim to fix that legacy runtime.

The single-host file/SQLite history backend remains independently deliverable behind the existing Backend interface; do not silently require its unfenced adapter to provide distributed execution ownership. No automatic invocation replay, exactly-once external effects, cross-group migration, preserved same-active-tree replica continuation, production cutover or deployment is authorized by filing this issue.

Lingua principale
Go
Stelle
309
Fork
29
Merge medio
15h 56m
PR unite (30g)
225

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di stacklok/mecatl

Tutte le issue di stacklok/mecatl

Issue simili

Altre issue su Go

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.