Block a user
21074bbcdb
Review gate — wave-5 suite passed (task review-wave-5): the suite stands as the compatibility instrument and the engine specs flip to stable. All 25 mechanism rows per engine column read in full and verified against core-contract.md §Verification backlog item by item (the appendix map's every claim confirmed by the pinning row/text, incl. the combined-coverage claim for save_offset_tx exactly-once and the five engine-side pins — SQLite open row, watcher-cadence knob, M-1 sweep-rollback: SQLite trigger seam live + the pg frame's in_tx structure code-read; beyond-64 skip-forward; cap-curve); all 25 'Contract stamp:' markers checked against the cited ADR § bodies (ADR-008 §3/§5/§7/§8, 009 §1/§3/§4/§6, 010 §1-§5, 014 §1, 015 §1-§5, 016 §5, 020 §1-§4, 021 §1/§3/§4/§5, 023 §1/§2, 012 §2, 019, 007, 006) — the enqueue_opts_resolution amendment accumulation rides ADR-023 §2 per the convention; all six parked dispositions audited and recorded (M-1 engine-side, skip-forward engine-side twins, cap engine-side, fire-wake parity not demanded, lock busy-path no divergence, reconnect-success test taken); conformance spot-checks clean (no row pins beyond ADR text, determinism posture holds incl. the two documented extensions, cross-references not duplication). Green on both engines this session: workspace build/test/clippy -D warnings/fmt; SQLite column 25/25 server-less twice (isolated + concurrent); pg column 25/25 vs the harness server three full runs (two consecutive + one concurrent with the SQLite column) plus 6 focused --test-threads=6 stress runs of the two development-time flake rows — all clean, the two unreproducible pg failures recorded with the bounded investigation and a wave-6 watch flag in the task's Notes. Docs: engine-sqlite.md and engine-postgres.md frontmatter status draft → stable with dated annotations citing this gate; implementation.md wave-table row 5 and review-rounds entry; suite-opts-backoff-rows.md placeholder remnants removed. Wave 6 (release readiness) decomposition may proceed
5c03fb2130
pg suite-infra hardening (task pg-suite-infra-hardening): the wave-4 review's F-1 defense-in-depth candidates — must_recv_event re-shaped from the 20 ms try_recv polling loop to a parked recv().await under the 15 s timeout wrapper (stream_tests.rs, the sole pg copy — SQLite twin untouched; the poll loop's wake-subscription interaction surface the F-1 flake suspected is out of the hot path, and the deadline assert still bounds the property). The parked form's terminal arms are explicit: Some(Err(e)) keeps the errored-instead-of-idling panic, None gets a receiver-closed panic the poll form never saw. The deadline miss self-diagnoses (the small-honest realization of candidate (b)'s discriminator): read_since(offset, 1000) over rows beyond the receiver's position names the residual class — rows durable undelivered (wake/re-drain class) vs no rows (publish-visibility class); with the parked recv a wake-arrived-but-re-drain-missed arm is structurally implausible (only a read error could miss, and it surfaces Err). Candidate (b)'s literal bridge-side counter skipped, reason recorded in the task (surfacing the bridge wake count across the dyn EventReceiver boundary needs downcast/keyed-global machinery beyond the small-honest bar); the small honest piece did land: wait_wake's Lagged(n) arm now logs (eprintln, house posture matching notify.rs's bridge_loop — and forwarder.rs's 'the receiver bridge's Lagged arm logs / recovers' doc claim now true at both bridges), stream name threaded into wait_wake's signature so the log attributes the lag. Verified: pg stream module 21/21 vs harness server (full module run), tx_publishes_compose_with_the_handle 5 solo re-runs green (determinism re-check under the parked shape), two consecutive full pg harness runs green (121 lib + 25 suite + 9 schema, ~72 s each), cargo test -p alkstore-postgres green server-less (skips clean), workspace cargo test green (12 binaries), clippy -D warnings, fmt clean
0c7744977c
SQLite commit-error-arm coverage (task sqlite-commit-error-arm): the wave-3 review gate's deferred test landed — failed_commit_replenishes_the_writer_slot drives a failing COMMIT through the engine's commit path and pins the review's code-read fix end-to-end: the error surfaces as the opaque Database carrying the SQLITE_FULL-shaped source chain, the failed connection is dropped and the writer slot replenished via the handle's reopen closure (the next begin_tx proceeds within a bounded timeout, no parking — the store-wide-livelock posture), no partial-commit residue (the dropped connection's uncommitted writes read back None), the post-failure commit is a real clean commit (the fault disarms on consumption), and auto-commit notify works afterward. Injection is a cfg(test) commit-fault seam in seam.rs — a per-store Arc<AtomicBool> arm (born disarmed, armed via arm_commit_fault, take() disarms on first consumption so exactly one commit faults) whose fabricated rusqlite SqliteFailure feeds the production commit error arm rather than replicating it; the PRAGMA max_page_count route was probed live against both WAL and DELETE journal modes first and rejected: SQLite checks the page-count limit at page-allocation time, so the squeeze always fails the growth statement (SQLITE_FULL/DiskFull on the first INSERT) and leaves no transaction active for COMMIT to fail — the arm is unreachable through PRAGMA-space (also probed: the pragma is per-connection, so pre-begin arming on the writer conn would have ridden into the tx conn; the route failed on error placement, not delivery). Mechanism choice and probes documented in the seam doc comment and the task Notes. Cross-test safety is per-store scoping; parallel stores never see the arm. Replay-proofed live: with the error arm's writer_reopen replenish temporarily removed the test fails (begin_tx parks past the 5 s timeout — the stranding the review identified), reverted it passes. Plumbing follows the pg-fix-forwarder-reconnect cfg(test) precedent: fields on SqliteStore/SqliteTxHandle and a begin param are cfg-gated, production builds compile the plain path. The waves-1-2 review's optional watcher reconnect-success add rides here (taken — recorded in Notes): reconnect_success_resumes_wake_delivery drives run_poll_loop through its existing open_conn_fn seam (same instrument as the W-1 failure test), with the db file present throughout because the vanished-file route cannot reach the success body (file reappearance trips the dead-man's identity switch first): initial open + first two reconnects fail by injection, the third reconnect succeeds, and a subsequent commit wakes on_change — the success arm's data_version re-baseline and restored delivery pinned. Watcher shape untouched. Verified: cargo test -p alkstore-sqlite green server-less (191 lib + 25 suite), workspace cargo test 399/0, clippy -D warnings, fmt clean
36023914b2
Contract-suite rows (task suite-opts-backoff-rows): the backoff-curve equivalence row and the deferred enqueue-opts clock legs. backoff_curve_equivalence — the equal-jitter exponential pinned as the range, not a jitter label (ADR-010 §3): claim → retry(err, None) cycles on a backoff_base_s = 2 queue land the ranges [1,2], [2,4], [4,8] across attempts 1–3, each computed delay read back through get_job's resolved run_at minus a clock read taken before the retry — the lower bound race-free (the retry's internal clock read cannot precede the suite's, so the observed value over-reads the delay, never under-reads) and the upper bound carrying a one-second straddle tolerance for the integer-second stamp crossing a wall second; identical bounds on both engines is the equivalence pin (ADR-012 §2, the row body shared). The explicit-delay override legs ride the same row: retry(err, Some(5)) honored verbatim ([5, 6] under the same straddle tolerance) and retry(err, Some(0)) resolving ready-now per the boundary-total rule (ADR-023 §2), claimed within a bounded wait. enqueue_opts_resolution extended in place with the wave-5 legs its deferral note named — the run_at-alone literal leg (a future run_at is the row's ready time verbatim, a deterministic equality with no clock read involved; a past run_at stores the literal too and is claimable now through the run_at <= now predicate within a bounded wait, the run_at-ASC ordering making the observation unambiguous among the row's other legs) and the neither-field leg (ready at the enqueue instant, abs_diff(now) <= 5 tolerance-bounded proximity, never a tight timing assert); the deferral note replaced by the completion statement, ADR-023 §2 stamp accumulated per the convention (ADR-020 §1 governs the resolutions). Cap-leg disposition recorded in Notes for the review gate: the 1-hour cap is not suite-pinnable (capped attempts need minute-scale waits) and stays pinned engine-side on both engines' unit tests per the wave-3/4 reviews. Ready-now waits are bounded claim_one poll loops (deadline asserts only, sequential single-store drive), with the one draft defect the finding surfaced (a re-claim after the bounded wait consumed nothing; the helper returns the claimed handle) noted. Wired into both engines' suite targets (SQLite tokio test, pg harness_row!), 25 rows per column up from 24. Verified: SQLite column green server-less (25/25 suite, 189 lib), pg column green vs harness (postgres/poc@:15432 — 25/25 suite, 121 lib + 9 schema), 3 solo re-runs of each new/extended row per engine (determinism), server-less pg skips cleanly via the reachability gate, cargo test -p alkstore-sqlite -p alkstore-postgres green, clippy -D warnings, fmt clean
2a2ad7e11f
Contract-suite wake row (task suite-wake-rows): wake_receiver_shapes — the wake contract's pinnable core, tolerance-bounded to state outcomes (never delivery counts or latencies): a pre-attached listener receives a wake after a committed notify on its channel through all three recv forms (recv/try_recv/recv_timeout), every wake's channel field matching the listened channel (the one piece of semantic content a wake carries, ADR-008 §3); a three-notify burst floors at one wake — never an exact per-notify count (coalescing the documented engine asymmetry: SQLite's 1-slot feed vs pg per-notify, the row pins the floor not the shape); a listener attached after a commit never sees that commit's notify — recv_timeout idles a 400 ms absence window (a 300 ms pre-settle sleep closes SQLite's watcher baseline race: the last_version baseline is store-open-captured, so an unconsumed commit's version change would fire one late tick at the new subscriber indistinguishable from replay; pg's gap-commit no-replay hole and SQLite's burst coalescing both legal under the pin); and the channel-scoped leg — a foreign-channel notify never surfaces a wake naming it (SQLite's same-commit overtrigger may deliver but carries only the listened channel; the pg LISTEN fanout skips foreign channels; the reserved reconnect straggler tolerated), with a same-channel positive control proving the silence is scoping not a dead listener. The failure-surface close arms (SQLite watcher-death recv()->None, pg synthetic reconnect-wake) stay pinned engine-side and in receiver_close_and_save_arms's disposal leg — cross-referenced in the doc comment, not re-pinned. Stamped ADR-006 + ADR-008 §3. Wired into both engines' suite targets (SQLite tokio test, pg harness_row!), 24 rows per column up from 23. Dispositions in Notes: the backlog row stays in core-contract.md's inventory (flushes at review-wave-5's stable gate, like the other discharged rows), the absence windows are single recv_timeout calls (the documented Ok(None) idle arm as the bounded wait), and the scoping leg's observable is the channel field not absence (SQLite overtrigger makes an absence-only pin vacuous there). Verified: SQLite column green server-less, pg column green vs harness (postgres/poc@:15432), 3 solo re-runs of the row per engine (determinism), cargo test -p alkstore-sqlite -p alkstore-postgres green (SQLite 189+24, pg 121+24+9), workspace cargo test green (12 binaries), clippy -D warnings, fmt clean
98de4a49cd
Contract-suite lock rows (task suite-lock-rows): two version-stamped rows discharging the lock backlog rows — lock_ttl_expiry_and_reacquisition (the ADR-008 §7 guarantee row: a held lock excludes a second acquirer (contender None); after a 1 s TTL the exclusion lapses silently — no revocation event, no error — and a second owner acquires; the original holder's post-expiry renew is refused (false, the lost-it arm); the second owner's release frees the name for a third acquirer, which consumes cleanly; tolerance-sleep posture, state outcomes only; ADR-008 §7 + ADR-019 §1, duration-guard legs cross-referenced to duration_refusal_on_non_positive_ttl) and concurrent_try_lock_loser_is_a_value (the contention posture: four sequential contenders — two owners, repeated — against a held lock all land the clean None value, never Database, never a busy-throw; the backlog row's SQLite busy-path open question answered: no divergence from the pg reference; release-then-contend cycle proves the loser path leaves no state blocking a later acquire, granted to a former loser and consumed cleanly; ADR-008 §5 + §7 + ADR-019 §1) — wired into both engines' suite targets (SQLite tokio tests, pg harness_row!s), 23 rows per column up from 21. Dispositions in Notes: no divergence found (no fix/ADR call needed), ADR-023 §2 excluded from stamps (its guard property is the duration-refusal row's, cross-referenced), the backlog parenthetical re-acquire-does-not-refresh-TTL stays engine-pinned, renew-refusal asserted after the second owner's re-acquisition (strongest form), contenders distinct-owner only (same-owner re-acquire grants per the inherited substrate shape), and one unreproducible first-cold-run pg failure recorded (message lost to truncation; 13 subsequent clean runs incl. concurrent SQLite+pg — no timing hazard identified, the lapse assertions are post-sleep state outcomes). Drive-by: alkstore-contract-suite's tokio dep gained the macros feature — a pre-existing compile break in the crate's own tests/suite_harness.rs (since 7f749ac) failed the harness target and the workspace test gate on HEAD; test-side non-event (ADR-017 §2 class 4). Verified: cargo test -p alkstore-sqlite -p alkstore-postgres green (SQLite 121+23+9, pg 189+23 vs harness server on :15432, server-less pg skips clean), workspace cargo test green (12 binaries), clippy -D warnings, fmt clean
1d05df200e
Contract-suite stream rows (task suite-stream-rows): two version-stamped rows discharging ADR-015's backlog legs — stream_ordering_equivalence (a keyed/unkeyed interleaved publish sequence of 6 reads back in the same order on every read form: whole + cursor-paginated + mid-stream read_since, read_from_consumer fresh and from a mid checkpoint, and a subscriber's attach drain; offsets strictly increasing per stream — per-stream relative order, absolute values explicitly not cross-pinned (pg bigserial vs SQLite AUTOINCREMENT); key round-trips exactly None/Some on every read form including the explicit-None keyed publish form; stream carries the name; created_at tolerance-bounded informational, never an ordering assertion; ADR-015 §4/§3/§1, byte-exactness and tx-seam legs cross-referenced to the payload-round-trip and keyed-tx-atomicity rows) and trim_to_semantics (the full ADR-015 §5 row: exact-boundary trim — the horizon's own row deletes, horizon+1 survives, repeated trim 0; survivors keep offsets; reads from a trimmed-away region resume at the horizon's first remaining row; a below-horizon saved checkpoint stays a get_offset-visible position marker with read_from_consumer and a fresh subscribe both resuming at the horizon, never a renumbered past; a pre-trim subscriber's above-horizon checkpoint keeps its place; a pre-attached listener idles across the trim in a 400 ms bounded window — no dedicated wake, SQLite's spurious watcher hint contract-legal and delivering nothing; ADR-015 §5/ADR-019 §6, negative-horizon/immutability legs cross-referenced to extent_clamp_semantics). Wired into both engines' suite targets (SQLite tokio tests, pg harness_row!s). Dispositions in Notes: the attach-drain pattern leads with a blocking recv() before the try_recv drain (engines deliver the attach read asynchronously); the pg LISTEN channel is mechanism-named and database-wide so concurrent suite rows' wakes cross schemas — safe by construction (wakes re-drain own-schema storage only, delivering nothing), no "events" rename needed. Verified: sqlite suite 21/21, pg suite 21/21 vs harness (postgres/poc@:15432, 7 consecutive full runs), 5 focused --test-threads=6 runs of the two rows per engine, workspace build/test green, clippy -D warnings, fmt clean
360e71e51e
Contract-suite tx commit-atomicity rows (task suite-tx-commit-atomicity-rows): the N-5 panic-probe admission in properties.rs's module doc (spawned with_tx task joined via JoinError::is_panic — catch_unwind around an async closure cannot see the panic point across an await; post-panic assertions read-only until convergence, single-task drive, no race window; ADR-017 §2 class 4) and three version-stamped rows: outbox_enqueue_tx_commit_atomicity (rollback drops the backing-queue job with the business write — get_job_tx-gone + run_once claims nothing; commit makes the job claimable exactly when the business write commits, real delivery through the RecordingDelivery closure with job-id identity, derived __alkstore_outbox:mail queue, exact payload, 60/5/5 stamps, consumed-after-ack; ADR-014 §1, ADR-010 §3a, ADR-021 §4, ADR-007), publish_with_key_tx_commit_atomicity (rollback drops the keyed event with the business write, key round-tripping inside the tx; commit surfaces it to read_since and a post-commit subscriber attach with the key round-tripping; ADR-015 §2/§4, ADR-021 §4, ADR-007), and with_tx_panicking_closure_rolls_back (N-5's probe against real engines: the panicking closure writes all four kinds then panics mid-flight; panic surfaces via the join; no-ghost reads converge with begin_tx granting every iteration; pre-panic listener silence on the notified channel; ghost never claimable; fresh with_tx commits through the same seam — both engines green; ADR-007, ADR-021 §4) — wired into both engines' suite targets (SQLite tokio tests, pg harness_row!s). Dispositions recorded in Notes: the 60/5/5 stamp inspection rides the delivery handle's job() because get_job cannot target the reserved derived backing queue through the contract surface (exemplar row pins the rejection; engine-side raw-row probes stayed put), the panic row's notify leg is a windowed silence (SQLite's wake overtriggers on any commit — the fully cross-engine notify-ghost pin rides the engines' rollback-ghosts twins, the suite's drop-rollback row made the same call), the closure tail routes through a #[cold] mid_flight_panic -> Error helper (a bare Err(panic!()) tail trips unreachable_code under -D warnings), and the post-panic convergence loop is the suite-side bounded wait (state outcomes only, begin_tx doubling as the seam-still-grants probe). Verified: sqlite suite 19/19, pg suite 19/19 vs harness twice (postgres/poc@:15432) + solo re-runs of each new row per engine (determinism), workspace build/test green, clippy -D warnings, fmt clean
7f749ac673
Contract-suite scheduler rows (task suite-scheduler-rows): the determinism-posture extension in properties.rs's module doc (runner-driving rows admitted — spawned run_schedules(stop) tasks on a core StopToken against state outcomes only, elapsed-boundary-band tolerances, never timing-value assertions, never two-task race windows; ADR-017 §2 class 4) and three version-stamped rows: scheduler_boundary_fires (fired jobs are ordinary claimable work with ScheduleOpts over the plain-queue derived defaults 300/9/5/none + payload exact, clean-stop Ok(()), fired count inside the elapsed-boundary band — no double-fire per boundary while one leader runs; ADR-009 §3/§4, ADR-020 §3, ADR-019 §4), scheduler_bounded_catchup (runner-less downtime proves no fire without a runner, ≥3 elapsed boundaries replay boundary-by-boundary bounded below the 64-cap and inside the band; ADR-009 §4), scheduler_leadership_discipline (two spawned runners on one store: exactly one Ok(())/Err(LeadershipLost) pair by value, no duplicated fires inside the band; ADR-009 §1/§6, ADR-019 §4) — wired into both engines' suite targets (SQLite tests, pg harness_row!s), suite tokio dep added for the runner rows. Dispositions recorded in Notes: the beyond-cap skip-forward leg stays pinned engine-side (both engines' scheduler tests already backdate next_fire_at directly — catch_up_replays_up_to_the_cap_then_skips_forward twins), and the pg fire-wake parity gap is not demanded by these rows' shapes (claim-polling observation only; the one-call wake_tx disposition recorded for a later task). Verified: sqlite suite 16/16, pg suite 16/16 vs harness twice + solo re-runs of the three rows per engine (determinism), workspace build/test green, clippy -D warnings, fmt clean
5c9ae6a7fc
Contract-suite queue-depth rows (task suite-queue-depth-rows): the job-handle validity predicate row (in-window heartbeat/ack landing, late-heartbeat/post-lapse-ack/retry/fail refusals past a 1 s stamp with the row untouched, ack_batch per-id predicate live-1/lapsed-0/nonexistent-0, fail(None)→"failed" and retry-at-budget→"max attempts exceeded"), the ADR-010 depth row (reclaim consumes an attempt with claimed_at/deadline refreshed and the original holder refusing, reclaim-exhaustion dead-lettering with the pre-claim sweep — get_job-visible "max attempts exceeded"+died_at, cancel unconditional delete with the not-an-interrupt refusal shape plus pending/dead/missing arms), and the no-stranded-rows sweep row (both states move with "expired", unexpired/never-expiring untouched, retention TTL enforcing with the moved+deleted sum, None=forever) — each version-stamped per the suite convention (ADR-010 §1–§5, ADR-019 §3, ADR-008 §5), wired into both engines' suite targets (SQLite tokio tests, pg harness_row!s). M-1's retention-failure pin dispositioned engine-side per the task's honest call: the storage-level DELETE failure is not injectable through the contract surface, so it lands in the SQLite substrate's trigger seam (sweep_rolls_back_the_retention_half_with_the_move — the move half rolls back with the retention half), with the pg twin verified structurally (both halves inside in_tx's frame) and the disposition recorded in the task Notes. Verified: sqlite suite 13/13, pg suite 13/13 vs harness twice (postgres/poc@:15432), substrate sweep-rollback tests 4/4, workspace build/test green, clippy -D warnings, fmt clean
250511d480
decompose wave 5 — contract suite: audit-first split of the verification backlog (engines' columns already discharge most rows; map in review-wave-5's appendix), seven mechanism-grouped suite-row tasks (queue depth, scheduler, tx commit-atomicity, streams, locks, wakes, opts/backoff equivalence), two engine-side hardening tasks (sqlite commit-error arm, pg F-1 defense-in-depth), and the review-wave-5 gate that flips the engine specs to stable
9a00fb8a7e
Review gate — wave-4 fix batch passed: all seven pg-fix resolutions verified against review 002's failure mechanisms (code-read + live gates), one minor doc mismatch fixed inline (outbox wake comment's 'scheduler's fire-wake' claim — the tick fires issue no wake and schedule() can never target a reserved backing queue); no pinned posture regressed; pg suite green vs harness twice + compose determinism 20/20 re-verified; wave 5 may decompose (task review-wave-4-fixes)
49face898d
docs alignment: v1 TLS posture owned by deployment.md, QueueOpts numeric consumer-obligation notes (task pg-fix-docs-alignment, review 002 Finding 6 remainder)
40625090f6
pg queue/tx/notify dedupe + doc truth-telling: job_from_row has one owner (queue.rs, tx.rs imports it) and get_job_tx reuses live_columns()/dead_columns() instead of inlining the 18-column lists — the contract-pinned Job shape now has exactly one decode owner (ADR-012 §2); sweep_expired's doc states the moved + retention-deleted sum in the in_tx multi-statement frame (the sum the SQLite twin returns — the doc was the only liar); notify.rs's closed-store listen error routes through the shared database_error helper; bridge_capacity() is a const with its rationale doc carried. No behavior change — identical SQL strings, error shapes, and capacity. (task pg-fix-dedupe-cleanup, review 002 minor notes + Finding 6 queue bullet)
ace94f0e13
pg forwarder: generation-tagged commands — a reconnect drops its predecessor's queued commands before replaying, closing the stale-UNLISTEN window (a stale UNLISTEN replayed after the snapshot's re-issued LISTENs silently cancelled a re-registered channel, its wakes lost until the next reconnect); the reconcile decision is the pure helper (stale LISTEN and UNLISTEN dropped, current-generation commands replay in queue order, dropped stale LISTENs ack the transient mid-LISTEN error) pinned by a four-combination server-less unit test (replay-proven against the neutered shape); 'queued commands replay harmlessly' doc corrected and the post-reconnect LISTEN-set invariant (server LISTEN set = registry snapshot, dead-generation commands can't undo it) stated in the module + loop docs (task pg-fix-stale-unlisten, review 002 Finding 4)
86719a39cc
pg open path: validate max_size > 0 at entry (typed Database before any round trip — 'max_size: 0' previously hung open forever at the bootstrap checkout, deadpool 0.13/0.14 neither validates nor defaults timeouts) and append the engine's '-c synchronous_commit' SET to the DSN's parse-carried options (was: setter replaced them, a silent override); pins: the zero-guard test (bounded by inner timeout, server-less-capable — fires pre-connect) and the DSN-options coexistence test (consumer SHOW statement_timeout + engine SHOW synchronous_commit, both settings) (task pg-fix-open-path, review 002 Finding 3 + options note)
9a5d1f4705
pg tx producer paths wake commit-atomically: publish_with_key_tx/enqueue_tx/outbox_enqueue_tx issue pg_notify on their mechanism-named channel (stream name / queue name / derived backing queue) inside the caller's tx — empty payload, best-effort log-and-swallow, no double-wake on the auto-commit paths; shared tx::wake_tx owner + module-doc section pinning the semantics; new tx_tests harness test pinning pre-commit silence, commit delivery on all three channels, and rollback silence; F-1 engine arm retired in review-wave-4 + implementation.md records (tx_publishes_compose_with_the_handle deterministic: 20 solo runs green; pg suite 116/116 x2 vs harness) (task pg-fix-tx-wake, review 002 Finding 2)
7ae426a01d
pg forwarder: the reconnect arm retries failed connects until success or shutdown — one failed connect no longer kills the loop permanently; every loop exit path releases the fanout sender via the shared-slot drop guard (receivers-terminal-iff-loop-gone); reconnect-config test seam + failed-connect pins: six-cycle exhausted-backoff server-less unit, pre-flip abort, guard drop, and the harness end-to-end unreachable-outage → full-recovery test (bug replay-proven against the old shape) (task pg-fix-forwarder-reconnect, review 002 Finding 1)
e067357de9
Wave-4 fix-batch decomposition: seven pg-fix tasks + review-wave-4-fixes gate from the general review (002) — forwarder reconnect retry (Finding 1, with the failed-connect test seam), tx pg_notify wakes retiring F-1's engine arm (Finding 2), max_size/DSN-options open path (Finding 3), stale-UNLISTEN generation drop (Finding 4), scheduler quarantine/retry (Finding 5), decode dedupe + cleanups, doc alignment; wave 5 gates on the two HIGH fixes landing