6.4 Execution Modes and Persistent Workers

Use this when choosing local, sandboxed, persistent-worker, remote, or mixed execution for an action class and verifying where it actually ran.

Use this when considering whether racing local and remote work reduces tail latency enough to justify duplicate work and deployment-specific cancellation evidence.

Use this when deciding whether persistent workers pay off and when worker-key fragmentation, memory growth, or restart churn makes their pools unstable.

Use this when repeated or reordered worker requests produce different outputs, see undeclared files, or contaminate one another through process state or the filesystem.

Use this before enabling multiplex workers to prove that concurrent and cancelled requests remain isolated, correctly attributed, and compatible with the selected sandbox mode.

Use this when increasing --jobs, worker count, or remote capacity fails to improve throughput or causes OOMs and queueing.