Alpenglow Service Dispatch

Yard office · 01

Send a request through the mesh

Everything here is live AWS, not an animation: your submission is published to a custom EventBridge bus, and every lamp that lights on the board below is a hop your event actually made, read back from its trace. All requests are fictional City of Alpenglow business and expire within 48 hours.

Control board · 02

Train sheet · hop-by-hop trace

1. No request selected yet.

Arrivals board · 03

Live activity

Recent requests through the mesh: yours, other visitors', and the scheduled local, a synthetic heartbeat that EventBridge Scheduler publishes every 30 minutes. Click one to replay its trace.

Interlocking tester · 04

Test a routing pattern

EventBridge decides where an event goes by matching JSON patterns against its envelope. Edit the event and the pattern below, then throw the lever: the tester runs a real TestEventPattern call, the exact matcher the bus uses. Your event is also checked against the yard's five live rules, read straight from the deployed rule resources so the tester can never drift from the real interlocking.

Classification · 05

The block order: standard vs FIFO

A standard queue promises delivery, not order; a FIFO queue promises both, and absorbs duplicates. This sends the same cut of 10 numbered cars to both queues in one SendMessageBatch each, with one twist: car 7 is offered to the FIFO queue a second time under the same deduplication id. 21 sends are accepted, 20 cars arrive. Each arrival below is a real Lambda consumer recording its position.

Second section · 06

Replay the archive

Queues and topics forget a message the moment it is consumed. The bus does not have to: an EventBridge archive records every event for 2 days, and a replay runs a past window through the mesh again. Every rule matches the recorded events a second time, and each re-run traces as its own second section on the arrivals board, flagged with the run it repeats. Replay a window that includes a poison message and the rip track fills again; replay yesterday after the nightly sweep and the archive restores what the board forgot.

Reading the vault…

Timetable notes · 07

What this plank proves

• Decoupled, event-driven architecture: the producer knows nothing about the consumers. One published event is matched by up to five EventBridge rules, each fanning out independently to SQS, SNS, and Step Functions.
• Content-based routing you can poke: rules match on the event payload (detail.category, detail.priority), the pattern behind department inboxes, regional routing, and premium-tier handling. The interlocking tester runs your own patterns through the real matcher via TestEventPattern.
• Failure handling that's designed, not hoped for: per-message retry with visibility-timeout backoff, a dead-letter queue after three attempts, readable dead-letter evidence (delivery counts, timestamps, source queue), and an operator redrive path. The poison message demo exercises all of it.
• Replayability: an EventBridge archive records the bus for 2 days and visitor-triggered replays run past traffic through the mesh again, something a queue or topic cannot do once a message is consumed.
• Ordering and exactly-once where they matter: the block-order race shows a standard queue's best-effort delivery next to a FIFO queue's guaranteed order and 5-minute deduplication window, with the same producer batch feeding both.
• Orchestration with retries: an Express Step Functions workflow absorbs a deliberate transient fault via its declarative retry policy. Visible in the trace, invisible to the citizen.
• Durable pub/sub: one SNS topic feeding both a Lambda (push) and an SQS queue (durable pull), the two subscriber shapes real integrations need.
• ~$0 idle: EventBridge, SQS, SNS, Express Step Functions, Lambda, and DynamoDB are all free-tier or fractions of a cent per million at demo volume.