Load shedding intentionally rejects requests to protect system health during overload.
Why shed load:
Accepting more requests than you can handle makes everything slow. Better to reject some requests fast than serve all requests slowly.
Strategies:
Random rejection: Drop X% of requests when overloaded
Priority-based: Reject low-priority requests first
Client-based: Protect paying customers, shed free tier
Netflix approach:
Prioritized load shedding. Critical requests (playback) protected. Optional features (recommendations) shed first.
Return Service Unavailable with Retry-After header.