# Why We Moved to Event Queues

Requests no longer wait on slow consumers
Retries happen without user involvement
Each consumer scales on its own

**Notes:**

Open with the outage from last spring as the motivating example.

# Throughput Before and After

| Metric               | Before | After |
|----------------------|--------|-------|
| Peak jobs per minute |   1200 |  4800 |
| Median latency (ms)  |    850 |   210 |
| Failed jobs per day  |     37 |     4 |

**Notes:**

Numbers come from the two weeks around the cutover.

# What Still Needs Work

Dead-letter handling is manual
Consumer lag alerts are too noisy
Replay tooling needs access controls

**Notes:**

Ask the platform team for owners on each item.