bug: Commands exit with code 0 when they fail, and show incorrect error text #2

Blocker: yes. Fix this before the server-monitoring work starts.

Defect

• Most commands use Run. They show the error, but the process exits with code 0.
• Test: fly exec -- php sh -c 'exit 7' with Docker Compose v5.1.1. Result: the CLI shows Error executing command: exit status 7 and exits with code 0.
• The exit codes are not consistent. base start and logs exit with code 1 when they fail. start, stop, restart, wp, exec, base stop, base restart, sites * and update exit with code 0.
• Eight calls give err to color.Red without a format verb. color.Red uses a printf format. Thus, the output shows %!(EXTRA *exec.ExitError=exit status 1).
  • Locations: cmd/base.go:41,53,57,61 and cmd/site.go:43,60,80,84.
  • go vet does not find this defect.
• base stop and base restart show "…successfully" after they show an error.
• The CLI sends errors to stdout, not to stderr. Thus, errors mix with the output in a pipe, for example fly wp … | jq.
• If a user who is not root runs fly update, the CLI shows an error and exits with code 0.

Cost

• Scripts and automation cannot identify a failure.
• The control plane and the event emitter will not get a correct result from each operation.
• New capabilities will copy the current error pattern. The Cursor rules on main tell AI tools to use this pattern.
• Users see incorrect text and incorrect success messages. This decreases their trust in the tool.

Benefit of the fix

• Each command gives a correct exit code. Scripts, the control plane and the event emitter can use this code.
• The CLI gives the exit code of the container command to the caller without change.
• Error messages are clear and go to stderr.
• New capabilities start from a correct error pattern.

Fix

1. Use RunE on each command.
2. Set SilenceUsage: true on the root command. Show the error one time in Execute().
3. Get the exit code from *exec.ExitError. Exit with the same code.
4. Remove os.Exit from the command bodies.
5. Send errors and warnings to stderr.

Completion checks

• fly exec -- php sh -c 'exit 7'; echo $? shows 7.
• Each command exits with a code that is not 0 when its Docker call fails.
• No command shows a success message after a failure.
• Errors go to stderr.
• Regression tests use a fake docker script on PATH.