Improved error handling with external commands

[hostcc]

• Invoking any external command as metrics source now properly handles a failure from any command invoked including piped ones (if any). Specifically that includes rightmost commands in those pipes don't mask an error running the leftmost ones.

  To implement that the change adds invoke_shell_cmd() function that sets -o pipefail shell option prior to any command executed (some shells presumably have argument processing ordering so using Popen() with args being list and that option amended would have no effect).

  Initial scenario I started the change with: vcgencmd fails on non-Raspberry hardware but that error doesn't propagate leading to then failing attempt to parse error output as valid.

• Added tls_insecure configuration option allowing to skip verification of TLS certificate broker provides, should be turned on under specific circumstances only

[ironsheep]

Thank you for contributing these changes. They'll really help make this script more robust! -Stephen

[ironsheep]

Oh, I'd like to add documentation on setting MQTT_SENSOR_HOSTNAME in the daemons environment. Can you provide examples of how you are doing it?

[ironsheep]

@hostcc I'm getting an error with the invoke_shell_cmd() using sh as the shell.
--> sh: 0: Illegal option -o pipefail

If I change this to using bash vs. sh then it all works.

Thoughts?

[hostcc]

@hostcc I'm getting an error with the invoke_shell_cmd() using sh as the shell. --> sh: 0: Illegal option -o pipefail

If I change this to using bash vs. sh then it all works.

Thoughts?

Apologies for that, @ironsheep - it indeed might be too bash specific. Will try to address today

[hostcc]

Oh, I'd like to add documentation on setting MQTT_SENSOR_HOSTNAME in the daemons environment. Can you provide examples of how you are doing it?

Sure sir - will do a PR amending the documentation. Basically, it allows to override the hostname discovered from the system, containerized setup are't always allowing to customize that easily

[hostcc]

@hostcc I'm getting an error with the invoke_shell_cmd() using sh as the shell. --> sh: 0: Illegal option -o pipefail If I change this to using bash vs. sh then it all works. Thoughts?

Apologies for that, @ironsheep - it indeed might be too bash specific. Will try to address today

@ironsheep , the solution you introduced in https://github.com/ironsheep/RPi-Reporter-MQTT2HA-Daemon/commit/259e42c1faa349ebcdcfc7106751c5326cd394f1#diff-075607ba753db7556565789706ad37a90055655b26016827c9f79343ddee723bR389 is a good one - it immediate solves the issue, agree. Although it implies the users have bash installed (should be applicable to most of the systems?), could be mentioned in documentation if users report problems

[ironsheep]

@hostcc I was thinking of detecting the presence of bash and using it when it is present. Then, if it is not present, falling back to using sh without the new option would keep older systems working as well as they did in the past, while systems with bash would be improved.

I was thinking of logging a warning message to indicate that the script would be more robust if bash is installed. but I don't want to keep dumping into systems logs. Maybe I only do this if -d is enabled? Or maybe dumping to logs only at startup is really ok?