Incremental Runs

Mark a step cache: true and its result is reused whenever its task, args, and upstream results are unchanged. Re-running a workflow then skips the expensive work and reuses the prior result.

queue.workflows
  .define("report")
  .step("fetch", "fetchData")
  .step("crunch", "crunchNumbers", { after: "fetch", cache: true })
  .step("render", "renderReport", { after: "crunch" })
  .submit();

On the second run, if fetch produces the same data, crunch is a cache hit and isn't recomputed; render still runs.

The cache key (dirty set)

A cached step is keyed by a content hash of its task name, its args, and each predecessor's result. So a change anywhere upstream changes the key and re-runs the affected step — only the genuinely-dirty subtree recomputes, the rest is reused.

Results are stored in the queue's shared settings store, so a cache survives across processes and restarts that share the same storage.

TTL

Expire a cache entry with { ttlMs }:

.step("crunch", "crunchNumbers", { after: "fetch", cache: { ttlMs: 3_600_000 } })

Clearing

queue.workflows.clearCache(); // drop every cached step result → next run recomputes