PATTERN Cited by 1 source
Per-recipient concurrency limiting¶
Pattern¶
Cap how many deliveries to a given recipient can be in-flight at once, tracked in a shared cluster-wide counter so the limit holds regardless of which pod picks up the next message. When a recipient slows down, its in-flight deliveries pile up against its own cap, and further deliveries are deferred rather than dispatched.
What it catches¶
Per-tenant rate limiting catches volume (too many requests). Concurrency limiting catches latency: a recipient receiving requests at a perfectly normal rate but taking 10 seconds to answer each one never trips a rate limit. Left unchecked, that recipient can quietly tie up shared delivery capacity one slow response at a time.
Mechanism¶
- Limit is scoped per recipient, applied selectively where a concurrency budget is configured.
- Each reservation carries a short expiry — a pod crash or dropped release doesn't leave capacity permanently stuck.
- When healthy: deliveries flow untouched. When slow: in-flight deliveries hit the cap → further deliveries deferred.
Relationship to blast radius¶
This is the mechanism that makes "contained blast radius" work in delivery systems. A slow recipient's impact is bounded to its own concurrency cap — it cannot starve the rest of the platform (Source: sources/2026-07-29-atlassian-inside-the-events-rail).
Seen in¶
- sources/2026-07-29-atlassian-inside-the-events-rail — per-recipient concurrency limiting as the fourth fairness layer in Atlassian's Events Rail