How we keep a noisy neighbor from starving your jobs: fairness in a multi-tenant queue
Blog post from Inngest
Inngest’s multi-tenant durable execution queue schedules hundreds of millions of jobs while ensuring that high-volume customers cannot starve others and that queued work survives failures. Inspired by Apple’s QuiCK system, it uses a red-black tree ordered by job due time, with a sliding time window that selects ready work and reschedules jobs blocked by concurrency or throttle limits. The platform separates scheduling data in the queue from execution state in a state store, allowing retries to resume without repeating completed steps. Although Valkey has supported the queue, its memory limits, recovery risks, and single-threaded operation make true per-tenant fairness difficult at scale, particularly when many jobs are blocked or tenants have many distinct keys. Inngest is therefore migrating toward FoundationDB, whose disk-based, replicated, multi-threaded architecture can process due work across tenant lanes in parallel. The article argues that simple queues may suit lower-volume or non-fairness-sensitive workloads, but durable multi-tenant workflow scheduling requires careful data-layout, storage, fairness, retry, and flow-control design.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.