Make failure visible.

Generate workload, inject dependency failures, watch Kafka consumer lag and retries build, restore the service, then replay dead-lettered events without duplicate processing.

Java 21Spring BootApache KafkaPostgreSQLMicrometerPrometheusDocker

Runtime topology

Ingress API
    |
    v
  Kafka ------> Retry / DLQ
    |              |
    v              |
 Processor <--------+
    |
    +--> PostgreSQL
    +--> Downstream Simulator

Throughput0events / sec

Consumer lag0offsets behind

P95 latency0milliseconds

Processed0successful

Retry + DLQ0degraded

Success rate100%submitted events

Workload & failure controls

Drive the system into healthy and degraded states.

HEALTHY

Live telemetry

Throughput and Kafka consumer lag.

2s refresh

throughputconsumer lag

Recent events

Latest state persisted in PostgreSQL.

0 events

Event	Status	Attempts	Replays	Latency	Error

Runtime health

Services, broker, database.

live

Recovery state

Queue pressure and terminal outcomes.

Backlog

Retrying

DLQ

Incident timeline

Actions in this browser session.

Engineering contract

What this demo proves.

At-least-once: processors tolerate redelivery.
Failure is testable: latency and hard failures are first-class controls.
Recovery is observable: retries, DLQ, lag, replay and latency are visible.
State is disposable: this is a portfolio lab, not a production datastore.

StreamForge Control Plane · engineering portfolio labSource on GitHub ↗