Parallel Processing

Run independent work at the same time — fan out, then bring the results back together.

By default a workflow follows its connections in order. When steps don't depend on each other, run them in parallel to finish faster.

Fork & Join

  • Fork splits the flow into multiple branches that all execute concurrently.
  • Join waits for the branches and merges their results before continuing.

Use this when you have several independent tasks — call three APIs at once, process multiple inputs together, or generate several variations side by side. See the Fork node, which also covers Join.

Loop over many items

Hand a list — rows from a Sheet or files from a folder — to a Loop to process each item through the same steps. Combined with multiple servers, items can be spread across machines.

Scale across machines

Pair parallel branches with a compute cluster: each branch can run on a different machine, so heavy data-science workloads finish in a fraction of the time.

Tips

  • Only parallelize steps that are truly independent — shared state between branches causes surprises.
  • Use Join to synchronize before any step that needs all the results.

Related