Rossel Kafka Laravel Package

Product Decisions This Supports

Framework fit (corrected by 0.7.0): Release notes confirm this is a Symfony bundle — composer type is now symfony-bundle, RosselKafkaBundle extends AbstractBundle, and registration happens via config/bundles.php/Flex. Despite the Laravel framing in earlier drafts, this is not a Laravel package. Any roadmap item targeting Laravel services needs a separate adapter decision; scope this assessment to Symfony/PHP services only.

• Event-driven feature delivery: Publish/consume on Rossel Kafka without custom librdkafka client code — asynchronous workflows, domain events, cross-service integration.
• Symfony 6.4 LTS alignment (new): Official 6.4 support plus Flex auto-registration materially lowers onboarding cost and keeps adopting teams on a supported LTS branch. This is the strongest positive signal in the release and unblocks services still pinned to older Symfony versions.
• Runtime efficiency as a product lever (new): Lazy SSL/CA resolution and a lazy producer mean consume-only apps never open a producer connection, and apps that never touch Kafka pay nothing at boot. Fewer connections and faster cold starts translate to infra savings and lower container memory — usable as a cost argument in build-vs-buy.
• Delivery-semantics decision (new, high priority): Auto-commit is now disabled; offsets commit only after processing. This upgrades the contract to at-least-once, meaning a message can be redelivered on restart. PMs must write idempotency into the acceptance criteria of every consumer (dedupe keys on payments, notifications, inventory) or users will see duplicate side effects.
• Poison-message policy decision (new, high priority): A failing message no longer halts rossel:kafka:listen — it is logged, acknowledged, and skipped. Without a dead-letter path or alerting, failed events silently disappear. Product must decide, per topic, whether silent-skip is acceptable (telemetry, low-value events) or requires a DLQ, retry topic, or replay tooling before go-live.
• Upgrade cost as roadmap work (new): 0.6.2 → 0.7.0 is a breaking release on a minor version. Removed enqueue bundle and default transport, removed MessageHeadersInterface, RdKafkaMessageWrapperInterface, UnsupportedTopicException, logger-free MessageFactory, removed extension/Configuration/services.yaml, ValueError instead of BadRequestException, and broker/producer config now mandatory. Budget a migration spike and a coordinated rollout window per consuming service; do not adopt without pinning versions.
• Reduced dependency surface (new): Dropping enqueue/enqueue-bundle and enqueue/fs shrinks the supply chain and audit scope — a security/compliance talking point, but a hard break for any team that relied on the pre-configured default transport.
• Observability/cost control (new): producer.send_log_messages (default true) lets teams suppress EXEC_* log volume; the one-log-per-message change makes log-based metrics and cost predictable. Pair with a decision on where delivery failures surface.
• Contract ownership moves to consumers (new): Consumed bodies are no longer JSON-decoded/re-encoded — payloads arrive verbatim. Schema decoding, versioning, and validation become explicit consumer responsibilities; document this in the shared event contract.
• Standardization: One integration pattern across teams, paired with conventions for topic naming, schemas, retries, idempotency, DLQ, and ownership.
• Adoption-risk decision (unchanged, now evidenced): 1 star, 0 dependents, single maintainer, still pre-1.0, and breaking changes landing in minor releases. Treat as a strategic internal dependency with a named owner, support model, pinned versions, and an exit plan.

When to Consider This Package

• You are building PHP/Symfony services (not Laravel) that must communicate with Rossel's Kafka infrastructure.
• You need both production and consumption paths and want a maintained abstraction rather than hand-rolled librdkafka handling.
• Rossel has mandated Kafka as the integration backbone, or you are standardizing event-driven patterns across teams and want a single blessed bundle.
• You are on, or moving to, Symfony 6.4 LTS and want Flex auto-registration.
• You can absorb a breaking minor release. Concretely: you can pin the version, can run the UPGRADE-0.7.md migration, and can coordinate restarts across consumers.
• You can commit to idempotent consumers. At-least-once delivery is now the contract; if your domain cannot tolerate duplicates (billing, entitlements, inventory) and you cannot build dedupe, stop here.
• You can supply the failure path yourself. The listener skips and ACKs failures; you need DLQ/retry/alerting around it, or an explicit written acceptance of silent drops per topic.
• Look elsewhere if: you are not on Rossel Kafka; you are a Laravel shop unwilling to maintain an adapter; you use non-PHP services; you need advanced Kafka features the package does not expose (transactions, exactly-once, consumer groups tuning, headers APIs removed in 0.7.0); you depend on the enqueue bundle or the removed extension/config classes; you require broad community support or vendor backing; your scale does not justify Kafka and a queue/database job would do; or you cannot accept lock-in to a low-adoption internal dependency that ships BC breaks on minor versions.

How to Pitch It (Stakeholders)

• Executive summary: This package removes custom Kafka plumbing from Symfony services and now supports Symfony 6.4 LTS with Flex auto-registration, making event-driven delivery cheaper to start and cheaper to run (lazy connections, no producer in consume-only apps, trimmed dependencies). Position it as an internal strategic dependency, not a general open-source bet: single maintainer, pre-1.0, and 0.7.0 ships breaking changes on a minor version.
• For engineering: 0.7.0 is a planned migration, not a drop-in. Remove EnqueueBundle, drop references to the deleted interfaces and extension classes, make broker/producer config explicit, and expect ValueError from MessageType::from(). Two behavioral changes dominate design: consumers must be idempotent (auto-commit off, at-least-once on restart) and failures no longer stop the listener (message logged, ACKed, skipped) — so build dead-letter/retry/alerting alongside any critical topic. Wrap the bundle so Kafka types don't leak across domains; keep payload decoding in your own layer since bodies now arrive verbatim.
• For product: Use it where async messaging creates user-visible value — faster workflows, decoupled releases, resilient integrations — and lead with the runtime and LTS wins. Gate adoption on three things: a short migration spike against UPGRADE-0.7.md, a support/maintenance check with the named owner, and written acceptance of duplicate-delivery and failed-message-skip behavior for each topic. Add idempotency and DLQ requirements to the definition of done for every consuming feature. If Rossel Kafka is not the target platform — or the team is on Laravel — defer and use a standard client or queue instead.