Limit Message Rate

How it fits

Typical backends

Caveats

• FallbackSurplus messages are discarded, not deferred; the data is lost rather than delayed, so add a queue upstream if every message must be delivered.
• Parameter interactiontime_constant sizes the burst allowance — the bucket banks up to message_rate * time_constant tokens, so a larger value lets a longer idle gap burst through at once before steady-rate throttling resumes.
• LimitationDropping is purely timing-driven with no content awareness, so it cannot preferentially keep more important messages.