Roadmap

What's shipped, what's planned, what's explicitly out of scope. Every roadmap item is either implemented or has a triage tier; items below the cut are demand-gated.

This page is a snapshot — for the live planning detail (per-domain breakdown, design notes), the source is docs/kafka-gap-analysis/ in the repo (gitignored internal planning, but the structure is public).

---

Shipped

The current capability matrix:

Surface	Status
Postgres / MySQL stores with SKIP LOCKED claim	✅
Polling, notify-waker, streaming, binlog relays	✅
Idempotent + transactional Kafka producer	✅
kafkajs + @confluentinc/kafka-javascript drivers	✅
mTLS, SASL/PLAIN, SCRAM, OAUTHBEARER	✅
AWS MSK IAM helper (@eventferry/kafka-iam)	✅
Producer tuning passthrough (lingerMs, batchSize, …)	✅
OpenTelemetry tracing hook + W3C propagation	✅
Lifecycle hooks (onConnect, onPublish, onError, …)	✅
Error classification (PublishErrorKind taxonomy)	✅
Explicit per-message partition override	✅
DLQ enrichment + routing	✅
Backpressure + quota multiplier	✅
Schema Registry serializer (Avro / Protobuf / JSON)	✅
Subject naming strategies (Topic / Record / TopicRecord)	✅
Schema Registry basic + bearer auth	✅
Avro key serialization	✅
autoRegister: false for production registries	✅
Admin operations (publisher.admin(), ensureTopics)	✅
validateTopicsOnConnect startup check	✅
publisher.healthCheck() reachability probe	✅
decode + extractTraceContext consumer helpers	✅
defineOutbox(registry) typed events both sides	✅
Power-user escape hatches (rawProducerConfig, customPartitioner, …)	✅
Compression level (compressionLevel)	✅
librdkafka stats hook (onStats)	✅
Producer-fenced restart + onProducerFenced hook	✅
Concurrent fence-recovery dedup	✅
TLS CA pinning + SNI documentation	✅

---

Planned

Targeted for upcoming releases. Each item is a single-PR-sized addition unless flagged otherwise.

Reliability

• onMessageRetried hook — fires per per-record retry attempt. Currently only onFailed(record, err, willRetry) fires once at failure; observability stacks want a per-retry counter.
• Configurable per-topic retry policy — retry: { perTopic: { "orders.created": { maxAttempts: 10 } } }. Today retry is global.

Schema / contracts

• Compatibility mode probe — at connect(), verify each configured topic's compatibility mode against the registry and warn if NONE. Catches "no schema enforcement" misconfigurations early.
• Reference schema support — pass-through to upstream for Protobuf nested types; documented examples.

Observability

• OpenTelemetry messaging metrics — separate from spans, the new OTel messaging.client.published.messages counter. Wait until OTel SDK stabilizes the API.
• Prometheus metrics helper — small optional package (@eventferry/kafka-metrics) that registers the standard signals as prom-client gauges + counters.

Admin

• describeAcls probe at startup — optional. Verify the configured principal can Write on each known topic. Catches silent SASL auth that fails per-record.

Cross-cluster / DR

• DR-aware bootstrap — pass a secondary cluster's brokers + auth, automatic failover when primary is unreachable. Niche; demand-gated.

---

Out of scope (explicit)

These will not ship from eventferry. They're either out of the publisher-focused scope, or well-served by existing tools.

Surface	Why out
Kafka Streams runtime	Different problem domain entirely.
Kafka Connect runtime	Different operational model. Connect is the right tool for many-to-many CDC; eventferry is for app-emitted domain events.
ksqlDB integration	Stream processing is downstream of publishing.
Consumer group management	eventferry is publisher-only by design.
JMX metrics exporter	Node ecosystem uses Prometheus / OTel; we ship hooks, you wire your stack.
Built-in scheduler for purgeDone	Use cron / a sidecar / k8s CronJob. We don't want a scheduler in the publisher's process.

---

Demand-gated (open an issue with your use case)

We'll build these when someone needs them and is willing to pilot:

• SASL/GSSAPI (Kerberos) — enterprise on-prem only. librdkafka supports it natively, so the work is exposing the right config on the confluent driver. Has not been a blocker yet.
• Apicurio Registry adapter — separate package @eventferry/schema-registry-apicurio that conforms to the same Serializer interface. Real demand from Red Hat shops.
• AWS Glue Schema Registry — different wire format, separate package @eventferry/schema-registry-aws. Ship only with demand signal.
• MirrorMaker / DR-aware failover — multi-cluster bootstrap + topic-rename handling. Real but niche.
• MongoDB / DynamoDB / Cassandra stores — outbox pattern works on any transactional store. The store interface is intentionally minimal so adapters are small; we'd ship MongoDB first if asked.
• NATS / RabbitMQ / Pulsar publishers — same shape as KafkaPublisher. We'd ship NATS first based on community signal.
• DELEGATION_TOKEN auth — niche, big-platform-only.

---

How to influence the roadmap

• Open an issue with your use case and what you've already tried. Concrete user stories move items up the queue.
• Submit a PR with a design doc for non-trivial additions. We treat docs-first design as the norm — see docs/superpowers/specs/ in the repo for the format.
• Sponsor specific work if you need an item shipped on a timeline. Open an issue tagged sponsor with your shape.

---

Versioning policy

Bump	Trigger
Patch	Bug fix, docs, internal refactor. No public API change.
Minor	Additive public API. New option with a backward-compatible default.
Major	Renamed / removed public API, changed default, dropped Node major. Rare.

A major bump in any single package does NOT force majors elsewhere — independent SemVer. See Migrations and Upgrades for the schema-diff policy across versions.

---

What's next

• Operational rollout → Operations Guide
• Migration path between versions → Migrations and Upgrades
• Symptom-driven help → Troubleshooting
• FAQ → FAQ