Work that is not a web service.
WHAT THIS COVERS
Data processing Report generation Media handling Machine learning tasks
WHAT DIFFERS FROM SERVICES
They finish They may run for hours They may need far more resources, briefly
WHAT TO USE
Jobs, rather than deployments.
WHAT TO SET
A deadline Retry limits Resource requests reflecting actual need
WHAT TO BE CAREFUL WITH
Jobs being interrupted by node maintenance or scaling.
WHAT TO DO ABOUT IT
Make the work resumable, with progress recorded Or protect the node from disruption while it runs
WHAT RESUMABLE MEANS
Processing in chunks, recording what is complete.
WHY THAT MATTERS MOST
An eight-hour job that restarts from zero is unusable.
WHAT TO CONSIDER FOR VERY LARGE WORK
Splitting it into many small jobs.
WHAT THAT PROVIDES
Parallelism, and failure affecting only one piece.
WHAT TO WATCH
Jobs consuming resources needed by services.
WHAT TO DO ABOUT IT
Separate node pools, or scheduling priorities.
WHAT TO SET
Lower priority for batch work, so services are protected.
WHAT TO MONITOR
Duration, and whether it is growing.
