This page defines key terms used in E-MapReduce (EMR) Serverless Spark to help you build a clear mental model of the service before diving into the documentation.
Workspaces and resources
| Term | Definition |
|---|
| workspace | The basic unit for business development in EMR Serverless Spark. A workspace bundles jobs, computing resources, and permissions together — and keeps them isolated from other workspaces. |
| resource queue | A pool of computing capacity measured in compute units (CUs). Each CU maps to a specific vCore and memory configuration, and you can allocate one or more CUs to a Spark driver or executor node. Each compute node gets between 20 GiB and 160 GiB of local storage. The number of CUs a job consumes depends on its computation complexity and data distribution — check the job list to see actual CU usage. For pricing details, see Billing. |
| session resource | A Spark session deployed within a workspace. Sessions provide the base compute needed to run SQL statements and work in the notebook environment. Deploy a session into a resource queue, then adjust its engine version, queue assignment, and Spark parameters as your workload changes. |
Access control
| Term | Definition |
|---|
| user | An access control entity. Add a RAM user as a workspace member and grant them permissions to manage jobs and resources within that workspace. |
| role | An access control entity that groups permissions. A single user can assume multiple roles, and multiple users can assume the same role. All users who assume a role have the same permissions. |
Job execution
| Term | Definition |
|---|
| workflow | An ordered sequence of jobs. Jobs in a workflow depend on each other and are run in a specific order. Use workflows to model multi-step data pipelines with defined execution order. |
| job run | A single execution of a workflow. Each time a workflow runs, the job orchestration system assigns it a job run ID. |
| publish | The action of promoting a draft job file to production. Publishing is required after you finish editing a draft — it prevents in-progress changes from affecting live job schedules and keeps the development and production environments separate. |