Problem

Dagger has the ability to run commands in containers (engine.#Exec). These commands can be long-running (for example a test database). But there is no reliable way to terminate this command later. This effectively makes it impossible to execute long-running commands without hanging DAG execution.

A typical example (from real-world use case #1227):

• Start a test database (engine.#Exec)
• Execute tests against test database (engine.Exec)
• After the tests aere completed, terminate the test database (this is the missing feature)

Solution

Here’s a suggestion to get us started:

package engine

// Terminate a task
#Terminate: {
  // The ID of the task to terminate
  id: #TaskID
}

// Each task produces a unique ID which can be safely referenced
// NOTE: not necessarily the same as our internal IDs
#Exec: {
  // Dagger task ID.
  target: #TaskID
  …
}

Example:

sleep: engine.#Exec & {
  args: [“sleep”, “100”]
}

terminate: engine.#Terminate & {
  target: sleep.id
}

NOTE: in this example, sleep and terminate will run in parallel, causing sleep to be terminated immediately. For a more useful example, we need a way to run terminate after other commands. This requires explicit dependencies between tasks.