v1.4.0 Plan — Implementation Waves A–F

Status: Complete — pending merge of PR #25 Branch: wave-a-stopgaps (PR #25, single release PR per the next-version release model) Focus: The consolidated issue-implementation plan (docs/plans/issue-implementation-plan.md, v2) — risk-ordered, release-anchored waves covering the GitHub issue backlog (#16–#22), the code-review correctness/security findings, and the telecom-domain gaps.

---

0. Sources

• Plan: docs/plans/issue-implementation-plan.md (v2, post two independent reviews)
• Reviews: docs/reviews/code-review.md, telecom-domain-review.md, ui-ux-review.md, issue-rca.md, issue-tracker-summary.md, implementation-plan-reviews.md
• Design docs: docs/plans/d2-count1-stream-placement-design.md, e2-queued-manual-runs-design.md, f1-counter-delta-design.md (all approved + implemented)
• Verification: docs/reviews/kind-verification.md (live kind-cluster gates, all pass)

1. Progress

• Wave A — stopgaps: #16 mitigation (MALLOC_ARENA_MAX=2, post_batch_cleanup default on, schema-cache LRU, asset-sync skip, sink close), dispatch outcome labeling (#21 stopgap), dead-code removal, #3 closed
• Wave B — correctness core: alert CRUD persistence (A6), watcher delete lifecycle (A7), threaded batch-path rework (B.3: deferred source finalize + bounded in-flight cap), syslog RFC 6587 framing + concurrent TCP (A12), controller RLock, manager-restart adopt guard, Kafka at-least-once default
• Wave C — security (phase 1): TRAM_INTERNAL_AUTH_MODE warn-only, keys plumbed both directions, probe exemptions, constant-time compares, webhook body-size limit, ClickHouse identifier validation, committed chart defaults rotated out
• Wave D — visibility/stats: #17 durable count=1 placements + liveness recovery + config drift, #22 live dashboard stats, #18 json_flatten/explode O(n²) fix, per-slot placement CAS (A13), transport hardening
• Wave E — enhancements: #19 ASN.1 split_path, #21 queued manual runs (E.2 per design), #20 templates UI contract
• Wave F — domain gaps: counter_delta + window_aggregate with durable transform state (F.1), file-done guards (F.2), gNMI reliability/modes + Kafka stop/lag + CORBA dedupe window (F.3), source timezone (F.4)

2. Verification record

• Full backend suite (2130 tests) + ruff + UI build green on the release branch
• Nine independent adversarial reviews (one per wave/feature), all actionable findings fixed pre-commit
• Live kind-cluster verification of every runtime gate: E.2 queue lifecycle, F.1 stateful semantics + guards, D.2 placement lifecycle, #16 RSS soak (13 threaded runs, stable peak, plateau, zero OOMKills)
• CI green (Test + Docker Build Validation)

3. Notable behavioral changes in 1.4.0

• post_batch_cleanup defaults to true; Kafka enable_auto_commit defaults to false
• records_out counts delivered records (per-sink written max), not input chunk size
• Manual runs with no healthy workers return 202 queued (rollback: TRAM_QUEUE_MANUAL_RUNS=0)
• Helm chart ships no default apiKey/authUsers — set explicitly or via envSecret in the same upgrade
• Webhook bodies >10 MiB rejected (413); ClickHouse quoted/bracketed table names rejected
• Stateful transforms rejected with thread_workers > 1, in sink-level transforms, and on broadcast streams

4. Post-merge

• Close GH #16, #17, #18, #19, #20, #21, #22 per the wave mapping in the PR body
• Operational follow-up (next release): flip TRAM_INTERNAL_AUTH_MODE=enforce after keys are confirmed on all clients (C.2)