use run_interactive for localstack ssh to fix tty issues #9211
Conversation
localstack/cli/localstack.py
Outdated
|
|
||
| if not DOCKER_CLIENT.is_container_running(config.MAIN_CONTAINER_NAME): | ||
| raise CLIError( | ||
| f'Expected a running LocalStack container named "{config.MAIN_CONTAINER_NAME}", but found none' | ||
| ) | ||
| try: | ||
| process = run("docker exec -it %s bash" % config.MAIN_CONTAINER_NAME, tty=True) | ||
| process.wait() | ||
| run_interactive(["docker", "exec", "-it", config.MAIN_CONTAINER_NAME, "bash"]) |
There was a problem hiding this comment.
I would argue that we do not need run_interactive here. It would only be necessary when we want to do something afterwards.
In this case, I would suggest using os.execv* here, as replacing the current process with the new one makes sure every signal is handled as intended.
There was a problem hiding this comment.
fair point. i don't really understand os.exec well and how it works on other platforms, but works fine on my machine this way 👍
There was a problem hiding this comment.
It replaces the current process image with a new process image. Instead of running two processes, we only run one, and all signals will therefore get directly to docker, without any logic on our side :)
There was a problem hiding this comment.
Should be available on unix and windows :)
Motivation
While working on #9210, I noticed that
localstack sshhad some wonky terminal behavior.I couldn't use CTRL+A, CTRL+E, CTRL+R and other control sequences correctly. New lines were also added twice.
It looks like the
tty=Trueoption ofrundoesn't really work correctly.For #8994 I added aUsingrun_interactivemethod that is a much simpler implementation of running an interactive command in a subprocess.os.execlp(based on @dfangl's suggestion), is much simpler. This works much better now:Changes
run_interactiveforlocalstack ssh