SLO template: queue consumer
For a queue consumer, your users feel delay and loss rather than HTTP errors.
What to measure
The preset starts from a throughput SLI on a counter of consumed records. Replace the metric with your own consumer counter; the wizard checks that the name is a valid Prometheus metric.
A sensible starting SLO
Throughput signals are noisier than availability ones. Start around 99% over 28 to 30 days, then tighten once the data is stable.
Pre-filled in the wizard
These are the exact values the wizard loads when you open this template. Rename them to match your own service and metric.
| Service name | kafka-consumer |
|---|---|
| SLO name | consumer_throughput |
| SLI kind | throughput |
| SLI type | request_based |
| Metric | kafka_consumer_records_consumed_total |
Pitfalls
- A consumer that stops completely raises no errors. Make sure your SLI still moves when nothing is consumed.
- Rebalances and deploys cause short dips: a window shorter than 14 days lets a single one weigh too much, and the wizard warns you about it.