Run a background job

Runs a queued job, or resumes one waiting for approval, in the current R process and returns when the run ends. It calls authorize first; a refusal errors and leaves the job untouched. It then calls bind; if that fails, or the rebuilt agent doesn't match the setup saved by job_create(), the job fails. A running job is locked, so a second job_run() on it errors.

Usage

job_run(path, bind, authorize, decision = NULL, tool_input = NULL)

Arguments

path

Job directory returned by job_create().

bind

Function that takes the AgentJob and returns the rebuilt Agent, or list(agent = , cleanup = ) where cleanup is a function that job_run() calls when it is done with the agent.

authorize

Function that takes the AgentJob and returns a list of your app's current job_id, owner_id, definition_revision and context_revision for it; the job runs only if they match the saved values. Look them up rather than copying them from the job, and signal an error to refuse.

decision

"approve" or "deny", to resume a job waiting for approval.

tool_input

Optional edited tool arguments (a named list) to approve with.

Value

The updated AgentJob.

Details

A finished job is returned unchanged, and so is a job waiting for approval unless you pass decision. If a previous job_run() stopped partway, for example because its process was killed, the job is marked "indeterminate" and not run again: a tool may already have had effects, so check what happened before creating new work.