Documentation
¶
Overview ¶
Package main demonstrates resource-aware admission: one job definition whose memory requirement scales with its input, enqueued twice against a worker that cannot hold both at once.
This is the case that identical worker slots get wrong. "Concurrency 4" says a worker may run four jobs; it says nothing about how big they are. A render of a 2 MiB scene and a render of a 24 MiB scene are the same job to a slot-counting pool, so it starts both, and on a box sized for the small one the large one takes the whole machine down with it.
With a resource model the two are different sizes of the same work. The requirement is computed once, at enqueue, from the declared input; the worker admits against what is actually free; and the large render waits for room instead of being started next to work that is already using it.
Usage:
go run ./_examples/resources
Everything runs in-process: the memory store, the in-memory artifact backend Dispatch's own tests use, and a staging cache in a temp directory. No services to start, nothing to install.
Source Files