executor.md

Functions are executed within containers. In the following, we will describe how incoming requests are served within containers, assuming that a warm container for the function exists.

Each function container must run an Executor server, which listens for HTTP requests on port 8080 (by default).

When a function request is scheduled for local execution within a warm container, an invocation request is sent to the Executor as follows:

• URL: <container IP>:<executor port>/invoke

• Method: POST

• Body: an executor.InvocationRequest (JSON-encoded)

• Response (on success): an executor.InvocationResult (JSON-encoded)

An InvocationRequest has the following fields:

type InvocationRequest struct {
	Command      []string
	Params       map[string]interface{}
	Handler      string
	HandlerDir   string
	ReturnOutput bool
}

• Command (runtime-dependent; optional, depending on the Executor implementation): the command that the Executor has to run upon reception of a new request. E.g., for a Python runtime, it may be set as python /entrypoint.py.

• Params: user-specified function parameters.

• Handler (runtime-dependent): identifier of the function to be executed. E.g., for Python runtimes, <module_name>.<function_name>.

• HandlerDir: directory where the function code has been copied.

• ReturnOutput: whether function standard output and error should be returned.

The following object is returned upon function completion (or failure):

type InvocationResult struct {
	Success  bool
	Result   string
	Output   string
}

• Success: whether the function has been successfully executed.

• Result: what the function returned.

• Output: function combined std. output and error (if captured)