Protect the core journeys
Identify must-stay-up endpoints and degrade non-critical features first.
Gate expensive features with feature flags so you can turn them off quickly.
Good shedding preserves revenue flows and status signals even when everything else is noisy.
Apply backpressure
Use queues, rate limits, and token buckets to keep systems from thrashing.
Return friendly fallbacks instead of timeouts to keep customers informed.
Monitor when shedding activates
Send Watch.Dog alerts when shedding thresholds trip and when they clear.
Correlate shedding events with error budget burn to justify capacity or optimizations.