Backpressure patterns for bursty ingest

Backpressure is not a bug. It is a control mechanism.

Incidents happen when pressure behavior is implicit and unowned.

Decision question

Should your ingest path prioritize lossless throughput or controlled degradation during bursts?

Core design options

1. Lossless-first Preserve events with stronger queueing and replay semantics.
2. Latency-first Shed non-critical work to protect user-facing response time.
3. Tiered model Apply differentiated policies by workload criticality.

Recommended default

Use tiered backpressure unless strict regulatory requirements force lossless handling everywhere.

Execution pattern

• classify workloads by business criticality
• define queue depth and timeout thresholds per class
• implement explicit shed/retry/dead-letter policies
• instrument pressure propagation at every boundary
• run burst simulations before production rollout

Failure mode to avoid

Global throttling rules that apply equally to all workloads usually turn one noisy source into broad degradation.

KPI target example

• no priority incident from ingest bursts for one quarter
• critical-path latency SLO maintained during 2x peak load events
• recovery to steady-state under 20 minutes after burst exhaustion

If burst traffic is already driving incidents, start with a direct conversation with Stratorys.