Fix race condition in Command.IsRunning() check

[circleci-app]

Summary

Fixes a race condition in the task agent command execution that was causing test failures in TestOrchestrator/error: task agent encountered fatal error.

Problem

The test was failing with an unexpected error message:

error on shutdown: task agent process is still running, which could interrupt the task...

This was caused by a race condition where IsRunning() was called in the shutdown phase before the isCompleted flag was set, even though Wait() had already returned.

Root Cause

The isCompleted flag was being set in the defer of the wait() function:

defer func() {
    _ = cmd.Cancel()
    c.isCompleted.Store(cmd.ProcessState != nil)
}()

err := cmd.Wait()

The defer runs after wait() returns, but there's no guarantee about timing relative to the calling goroutine. This created a race where:

1. Process exits and Wait() returns an error
2. executeAgent() returns the error to the main goroutine
3. shutdown() checks IsRunning()
4. The defer hasn't completed yet, so isCompleted is still false
5. IsRunning() incorrectly returns true
6. An "still running" error is generated

Solution

Set isCompleted = true immediately after Wait() returns, before any error handling:

err := cmd.Wait()
c.isCompleted.Store(true)  // Set immediately, before error handling

This ensures IsRunning() returns false as soon as Wait() completes, eliminating the race condition.

Testing

• All existing tests pass on Linux
• The specific failing test now passes: TestOrchestrator/error: task agent encountered fatal error
• No behavior changes for successful command execution paths

Original failing job

View this task in the CircleCI web app →