Fix issues with echobot invite link in cmdeploy

[mrdaemon]

The entire stack is setup to support localhost as a value for ssh_host, returning LocalExec when that is the case, but this last step made a new explicit SSHExec connection to config.mail_domain.

This changes that to simply use the subprocess module when ssh_host is local.

This also fixes the issue where the connection was made to config.mail_domain instead of the supplied ssh_host value, ensuring that remains consistent.

Additionally, the entire process will be skipped if --dry-run is used with cmdeploy, allowing a dry-run to complete without error.

Absolutely looking for feedback, as I mostly fixed this to get a deployment running in my environment, if more is needed, or a different approach is preferable, I'd be glad to rework this.

[hpk42]

lgtm, although it'd probably be good to revisit all the places where the split local/remote execution takes place, and see if the if-branching can be avoided.

[mrdaemon]

I originally tried this but LocalExec and SSHExec don't seem to have the same semantics, with LocalExec only having a 'logged()' method, so I left it alone.

I wouldn't mind adding the required methods to LocalExec so it could just be used in place of SSHExec and avoiding the crappy if/else branching, if that sounds ok

[missytake]

Suggested change

	echobot_cmd = "cat /var/lib/echobot/invite-link.txt"
	if ssh_host in ["localhost", "@local", "@docker"]:
	result = (
	subprocess.check_output(echobot_cmd, shell=True)
	.decode()
	.strip()
	)
	print(result)
	else:
	echo_sshexec = get_sshexec(ssh_host, verbose=args.verbose)
	print(
	echo_sshexec(
	call=remote.rshell.shell,
	kwargs=dict(command=echobot_cmd),
	)
	invite_path = "/var/lib/echobot/invite-link.txt"
	if ssh_host in ["localhost", "@local", "@docker"]:
	with open(invite_path) as f:
	print(f.readline())
	else:
	echo_sshexec = get_sshexec(ssh_host, verbose=args.verbose)
	print(
	echo_sshexec(
	call=remote.rshell.shell,
	kwargs=dict(command=f"cat {invite_path}"),
	)

Thanks for the fix, just stumbled upon it ourselves. We don't even have to open a subprocess if we're on the same machine :) let's just do it in python.

[j4n]

This is actually too fast now and requires a time.sleep (5 seconds works for me) before the file exists. (Somewhat related to #641).

Suggested change

	echobot_cmd = "cat /var/lib/echobot/invite-link.txt"
	if ssh_host in ["localhost", "@local", "@docker"]:
	result = (
	subprocess.check_output(echobot_cmd, shell=True)
	.decode()
	.strip()
	)
	print(result)
	else:
	echo_sshexec = get_sshexec(ssh_host, verbose=args.verbose)
	print(
	echo_sshexec(
	call=remote.rshell.shell,
	kwargs=dict(command=echobot_cmd),
	)
	invite_path = Path("/var/lib/echobot/invite-link.txt")
	if ssh_host in ["localhost", "@local", "@docker"]:
	sleep(5) # wait a bit for echobot to settle down
	with invite_path.open() as f:
	print(f.readline())
	else:
	sshexec = SSHExec(args.config.mail_domain, verbose=args.verbose)
	print(
	localexec(
	call=remote.rshell.shell,
	kwargs=dict(command=f"cat {invite_path}"),
	)
	)

(still needs the from time import sleep at the top)

[hpk42]

sleep(5) is a questionable hack to avoid. Why is it that "cmdeploy run" finishes but echobot is not running? Is it possible to write a final "deployer" that waits e.g. for echobot to be up? Not sure if that's easily possible with pyinfra (or a systemctl command).

[hpk42]

maybe "systemctl is-active --wait echobot.service" or something similar could ensure that "cmdeploy run" returns only after echobot runs?

[missytake]

systemctl is-active --wait echobot.service will only wait if state is initializing or starting, and we have either failed or active with active not necessarily meaning that configuration worked :/ I'll try something with Requires=: #746

Maybe we should move this question out of this PR.