Choose a compute lifetime
Choose the workload default in the compute step. Tasks can inherit it or use an override. Deployment compute selects a deployment profile; its resources and provisioner come from that profile. Workload and task compute can use supported provisioners, profiles, or custom resources.
The examples leave instance-specific compute choices unset. Select an available mode before saving. Profile names are deployment-specific; there is no universal
standard profile name required by these docs.
Size CPU and memory
CPU is measured in millicores: 1,000 millicores is one CPU. Memory is measured in bytes. A compute profile and custom resource requirements are alternative ways to select resources. More requested resources can increase allocation waits if capacity is limited. First check whether time is spent waiting for a worker, executing SQL, reading storage, or committing output.Control parallel work
parallelism limits concurrent task work within a run. Independent tasks also need available worker capacity. Overlap controls concurrency between runs, which is a separate decision.
For the four-task example, parallelism of one makes the two summaries execute in sequence. More capacity lets both become active when their shared predecessor succeeds.
Set time and attempt limits
Retries create new attempts and can add delay. A schedule interval does not bound execution duration. Read Retry policies and Sensors for their separate clocks.