Skip to content

refactor(sync): durable settings-sync LWW (#37 CAND-3)#50

Merged
ElbertePlinio merged 2 commits into
mainfrom
refactor/atomic-lww-sync
Jul 19, 2026
Merged

refactor(sync): durable settings-sync LWW (#37 CAND-3)#50
ElbertePlinio merged 2 commits into
mainfrom
refactor/atomic-lww-sync

Conversation

@ElbertePlinio

Copy link
Copy Markdown
Member

Summary

Issue #37, PR 3 (durable sync last-writer-wins).

The settings-sync last-writer-wins decision used to be client-driven
choreography: a conditional .update() guarded by updated_at <, a
fallback ignoreDuplicates .upsert() on a miss, and a third read to
learn which row won. That reconstructed a durable concurrency decision
out of retries instead of making it one durable operation.

  • Adds public.settings_sync_lww_write (new forward migration
    supabase/migrations/20260720000000_settings_sync_lww.sql, additive
    only per chore: reconcile hosted Supabase migration history #35): locks the (user_id, field_group) row, compares the
    incoming updated_at, and performs whichever of insert/update/no-op
    wins, returning the now-authoritative row in the same call. security invoker, so it stays inside the table's existing RLS boundary
    (auth.uid() = user_id), same pattern as public.credit_balance_cents.
  • pushGroup/deleteGroup (packages/sync/src/index.ts) keep all
    validation/sanitization and now just call the durable RPC. written/
    stale result semantics are unchanged; obsolete retry choreography is
    removed.
  • New local-Postgres contract lane
    (packages/sync/test/lww.contract.test.ts +
    lww-fixtures.ts/postgres-sync-client.ts), following PR test(billing): checkout lifecycle contract tests against local Postgres #47's
    pattern: skips cleanly with a console message when no local Postgres
    is reachable, wired into test:supabase. Covers real concurrent
    push/push, first-insert/first-insert, and push/delete races, equal-
    timestamp ties (first writer wins), and older data racing in after a
    newer tombstone (no resurrection).
  • pgTAP coverage (supabase/tests/database/settings_sync_lww.test.sql)
    adds microsecond canonicalization, the existing future-timestamp
    check constraint, and per-user isolation.

Validation

  • bun run typecheck
  • bun run test — 353/353
  • supabase db reset — all migrations apply cleanly (additive-only
    confirmed)
  • PICKFORGE_ALLOW_LOCAL_DB_TESTS=1 bun run test:supabase against
    local Postgres: pgTAP (155 tests incl. new lane), welcome-credits
    concurrency, the CAND-1 checkout-lifecycle contract lane (20/20,
    still green), and the new sync LWW contract lane (8/8) — all pass
  • supabase db advisors --local --type security --fail-on warn — no
    issues

Closes part of #37 (CAND-3).

Replace the settings-sync last-writer-wins choreography (conditional
`.update()` guarded by `updated_at <`, falling back to an
`ignoreDuplicates` `.upsert()`, falling back to a third read to learn
the outcome) with one durable `settings_sync_lww_write` SQL function
that locks the (user_id, field_group) row, compares/writes/tombstones,
and always returns the authoritative row in the same call.

pushGroup/deleteGroup keep all validation/sanitization and now just
call the durable RPC; `written`/`stale` result semantics are
unchanged. The migration is additive only (#35): it adds the function,
nothing else.

Adds a local-Postgres contract lane (packages/sync/test/lww.contract.test.ts)
following PR #47's pattern (skips cleanly without Postgres, wired into
`test:supabase`), covering real concurrent push/push, first-insert/
first-insert, and push/delete races, equal-timestamp ties, and older
data racing in after a newer tombstone. pgTAP coverage
(supabase/tests/database/settings_sync_lww.test.sql) adds microsecond
canonicalization, the existing future-timestamp check constraint, and
per-user isolation.

Issue #37 CAND-3.
@ElbertePlinio
ElbertePlinio merged commit e97f425 into main Jul 19, 2026
2 checks passed
@ElbertePlinio
ElbertePlinio deleted the refactor/atomic-lww-sync branch July 19, 2026 19:27
ElbertePlinio added a commit that referenced this pull request Jul 19, 2026
…cation (#51)

Issue #37 (CAND-4): operator-router's idempotency only became durable
AFTER provider work. The handler checked `credit_ledger` for a completed
route, invoked the OpenAI-compatible provider, and only then wrote a
debit row keyed by the idempotency key — two concurrent requests for the
same key could both find nothing completed, both call the provider, and
the losing response was simply discarded after the winner's debit
landed (wasted provider spend with no compensating record).

New `router_attempts` table and three service-role-only RPCs
(`router_attempt_claim` / `router_attempt_complete` / `router_attempt_fail`,
supabase/migrations/20260722000000_router_attempt_claim.sql) move the
durable decision in front of the provider call: at most one caller ever
owns "go call the provider" for a given (user_id, idempotency_key) at a
time. A claim that is neither completed nor explicitly failed is
recoverable once its lease elapses. This is additive only: existing
`credit_ledger` rows written before this migration remain authoritative
for old completed routes (`findDebitedRouteResult`), independent of
whether a `router_attempts` row exists for that key.

`createOperatorRouterHandler` now claims before invoking the provider,
completes the claim with the provider outcome, then debits — releasing
the claim immediately on a definitive provider failure, and skipping
straight to a debit retry (no re-invocation) when a claim is found
already completed.

- `packages/edge-shared/src/index.ts`: `claimRouteAttempt`,
  `completeRouteAttempt`, `failRouteAttempt`, `findDebitedRouteResult`
  own the RPC/route-metadata encode-decode boundary; `debitCredits` is
  unchanged. `createOperatorRouterHandler` takes `serviceSupabase`
  instead of separate `findCompletedRoute`/`debit` callbacks, and
  returns 409 `attempt_in_progress` while another claim is live.
- `supabase/functions/operator-router/index.ts`: wires the caller's
  service-role client straight through; drops the adapter-owned
  `findCompletedRoute` query and `debit` callback.
- `packages/edge-shared/test/edge-shared.test.ts`: replaces the
  `findCompletedRoute`/`debit` fakes with a `FakeRouterServiceSupabase`
  reimplementing the claim/complete/fail/debit RPC contract, covering
  the full claim -> provider -> complete -> debit choreography, live
  claims, lease recovery, and debit-retry-without-re-invoking-the-
  provider.
- `supabase/tests/database/router_attempt_claim.test.sql`: pgTAP
  coverage for claim/in_progress/completed outcomes, idempotent
  completion replay, failure recovery, lease-based recovery, per-user
  isolation, privilege grants, and input validation.
- `packages/edge-shared/test/router-attempt.contract.test.ts` (+
  `router-fixtures.ts`, `postgres-router-client.ts`): a local-Postgres
  contract lane, following #47/#50's convention, proving the durable
  claim/complete/fail/debit functions against real Postgres, including
  two genuinely concurrent claims for the same key resolving to
  exactly one claim, and a full `createOperatorRouterHandler` run
  proving the provider is invoked and the ledger is debited exactly
  once across two concurrent same-key requests. Wired into
  `bun run test:supabase`.
- Bumps `@pickforge/*` 0.11.0 -> 0.12.0 and re-pins every Deno
  function's `npm:` import, per the existing publish-then-pin
  convention.

Validated: `bun run typecheck`; `bun run test` (366/366); `bun run
build`; `supabase db reset` (all migrations apply cleanly); `bun run
test:supabase` against real local Postgres — pgTAP (178 tests,
including the new router_attempt_claim lane), welcome-credits
concurrency, the CAND-1 checkout-lifecycle contract lane (20/20) and
CAND-3 sync LWW contract lane (8/8) stay green, and the new router
attempt contract lane (8/8) all pass; `supabase db advisors --local
--type security --fail-on warn` clean. `deno check` against the real
npm registry isn't possible pre-publish (same documented gap as #48:
published `@pickforge/*` is still behind main) — confirmed the
pre-existing gap is unchanged (this diff's remaining `deno check`
errors are a strict subset of main's) and validated the adapter's
typing directly against local `src` via `tsc` instead.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant