Default: any single healthy node picks up the run. ALL: broadcast to every healthy node in parallel. Specific IDs: route to named nodes only.
TARGET_NODES ALL | (node_1, node_2) | <omitted>This reference topic explains compute node targeting, which controls where a scheduled pipeline run is dispatched for execution. The `TARGET_NODES` clause is specified on a `SCHEDULE` command and accepts three forms: omitted (the default), the keyword `ALL`, or a comma-separated list of quoted node identifiers such as `TARGET_NODES 'warehouse_1', 'warehouse_2'`. When `TARGET_NODES` is omitted, the scheduler assigns the run to any single healthy compute node using round-robin selection, which is suitable for most workloads. Specifying `TARGET_NODES ALL` broadcasts the pipeline run to every healthy node in parallel, which is useful for cache-warming or node-local maintenance tasks. Providing explicit node IDs restricts execution to only those named nodes, and the run will wait until at least one of the listed nodes is healthy before dispatching. The `TARGET_NODES` clause on a `SCHEDULE` command applies to all pipeline runs triggered by that schedule; individual `PIPELINE` declarations do not carry a `TARGET_NODES` clause directly.