Fan out

Run a body once per item, in parallel.

The body sees each element as item and its position as index. The output is the list of body outputs, in item order, whatever order they finished in. A list is capped at a hundred items.

• Kind: fan_out
• Palette: Flow
• Run it on its own: no, only as part of a graph

When to use

• The same step for every element of a list: build each target, score each row, notify each customer.
• Calling a rate-limited API once per item, paced with rate.

When one item’s work is more than one step, make the body a workflow node, so each item is a child run you can inspect and rerun on its own.

Configuration

Field	Type	Required	Description
body	any	yes	The node body to run for each item.
items	any	yes	The list to iterate: a literal list, or a marker object reading one from the run, {"$expr": "input.items"}.
max_parallel	integer	no	Cap on concurrent iterations.
on_error	string	no	“halt” (default) or “continue” past a failed item. One of "halt", "continue".
rate	object	no	Pace the dispatch: {limit, period}, e.g. 20 per “1m”.

Output

Not declared: what comes out depends on the configuration.

Example

version: hyperspool-graph/v1
nodes:
  - id: build
    type: fan_out
    config:
      items: [linux-amd64, linux-arm64, darwin-arm64]
      max_parallel: 2
      on_error: continue
      body:
        type: shell
        config:
          run: { $template: "make release TARGET={{ item }}" }
  - id: report
    type: shell
    config:
      run: { $template: "echo {{ size(nodes.build.output) }} targets" }
edges:
  - { from: build, to: report }

Two builds run at a time. With on_error: continue a failed target leaves null in its slot and the step still succeeds; the default, halt, fails the step once the builds in flight settle. See Repeating work.

See the node reference for every other node.