Represents a function in a computation graph.

structure

Structure

Each function in a Handoff DAG has these key components:

• :id - A unique identifier for the function (any term, typically an atom)
• :args - A list of other function IDs whose results this function depends on
• :code - The actual function to execute, receives results of dependencies as input
• :results - Storage for function output after execution
• :node - Optional node assignment for distributed execution
• :cost - Optional resource requirements map (e.g., %{cpu: 2, memory: 1000})
• :extra_args - Additional arguments provided at execution time
• :type - Type of the function, :regular or :inline
• :argument_inclusion - One of :variadic or :as_list (defaults to :variadic)
  • :variadic - Pass the list of N arguments as the first N arguments to the function
  • :as_list - Pass arguments as a list in the first argument to the function

examples

Examples

# A simple computation function with no dependencies
%Handoff.Function{
  id: :generate_data,
  args: [],
  code: &Enum.random/1,
  extra_args: [1..100]
}

# A function that depends on another function's result.
# The result of :generate_data (e.g., an integer) will be passed as the first argument
# to &*/2. The second argument for the multiplication (2) comes from extra_args.
%Handoff.Function{
  id: :process_data,
  args: [:generate_data],
  code: &*/2,
  extra_args: [2]
  # If :generate_data produces X, this executes X * 2.
}

# A function with resource requirements for distributed execution.
# The result of :process_data will be passed as the first argument to IO.inspect/2.
%Handoff.Function{
  id: :inspect_result,
  args: [:process_data],
  code: &IO.inspect/2,
  extra_args: [label: "Inspect result"],
  cost: %{cpu: 1, memory: 500} # Adjusted cost for a simple inspect
}

resource-costs

Resource Costs

The :cost field allows you to specify resource requirements for distributed execution:

• :cpu - Number of CPU cores required
• :memory - Memory in MB required
• :gpu - GPU units required (if applicable)
• Custom resource types can also be defined

These costs are used by allocators to determine which nodes can run which functions.