ReferenceGlossary
Load shedding
Rejecting some requests on purpose when a server is overloaded, so the requests it keeps finish in time instead of all of them timing out. Good shedding rejects early, before the work uses resources, and drops the least important or cheapest work first.
Reliability
In Quorum#
With Priority load shedding researched, you can turn on load shedding on an App Server. Above 90% load it drops whole request kinds, lowest value first (bots, then Realtime, then page views and up), until its load is back at 90%, so a $25 checkout survives an overload. Dropped requests count as errors and spend the error budget, but the requests it keeps stay fast.
Read next#
Backpressure, Circuit breaker, Rate limiting, Metastable failure