Zilla Architecture
Architecture
Zilla is a cloud-native gateway that bridges diverse client protocols to Apache Kafka. Clients use familiar protocols — HTTP, MQTT, gRPC, WebSocket — while Zilla handles Kafka translation transparently.


How It Works
Zilla routes traffic through a declarative graph of bindings — pipeline stages that decode, transform, and re-encode protocol frames. Each binding has a network side and an application side, allowing protocol translation without exposing Kafka internals to clients.
Traffic flows as streams with a unified frame shape: a BEGIN frame to open, one or more DATA frames, and an END or ABORT frame to close. WINDOW frames manage backpressure; protocol-specific metadata attaches as extensions (e.g. HttpBeginEx, MqttBeginEx, KafkaBeginEx).
Module Types
Zilla is composed of seven extensible module types:
| Module | Role | Example |
|---|---|---|
| Bindings | Protocol pipeline stages — servers, clients, mappings, caches | http_server, kafka_cache_client, mqtt_server |
| Guards | Identity verification and access control | jwt |
| Vaults | Cryptographic key and certificate storage | filesystem, aws_secrets |
| Catalogs | API spec retrieval for schema-driven message definitions | apicurio, confluent, aws_glue |
| Models | Message validation and format conversion | avro, json |
| Metrics | Telemetry collection | stream |
| Exporters | Integration with Prometheus, OpenTelemetry, and others | prometheus, otlp, stdout |
Scaling
Vertical: One worker thread per CPU core, each running in isolated single-threaded context. Lock-free data structures minimize cross-core contention. Large messages are fragmented to avoid blocking smaller ones. TLS math offloads to secondary cores.
Horizontal: Zilla nodes are stateless. Multiple instances independently hydrate their Kafka caches from the same topics with no coordination overhead, enabling scale-out to millions of concurrent connections.
Observability
Zilla includes always-on binary tracing through passive work queue observation — no performance impact in production. Metrics and structured log events are exported to Prometheus, OpenTelemetry, AWS CloudWatch, or stdout.
Ready to deploy?
Jump to the Quickstart guide to get Zilla running in minutes.

