Use ping api instead of system info for fingerprinting

[towe75]

Overview

All nomad drivers must implement a fingerprint mechanism. Back at the day i went with the system info api call and it worked well.

But it turned out that this is a rather heavy invocation, going even down to the package manager level (dpkg, rpm) to gather informations. The driver builds a fingerprint each 30 seconds and thus it wastes quite some cpu cycles and has an impact on the overall machine load.

Some numbers

A simple perf stat on my machine shows the difference between the two podman cli equivalents:

 Performance counter stats for 'podman info':

            240.46 msec task-clock                #    0.721 CPUs utilized
             1,481      context-switches          #    6.159 K/sec
               141      cpu-migrations            #  586.383 /sec
            18,851      page-faults               #   78.397 K/sec

       0.333681791 seconds time elapsed

       0.098392000 seconds user
       0.172141000 seconds sys

 Performance counter stats for 'podman version':

             60.49 msec task-clock                #    0.426 CPUs utilized
               720      context-switches          #   11.903 K/sec
                74      cpu-migrations            #    1.223 K/sec
             4,449      page-faults               #   73.549 K/sec

       0.141960043 seconds time elapsed

       0.036505000 seconds user
       0.047359000 seconds sys

What the PR does

This PR changes the mechanism to use the _ping api call which is not so resource intensive. System info is now usually only called once.

Conclusion

You will not see a big performance different on a crowded host but it certainly makes the situation better on a idle machine.

[lgfa29]

Minor nit-picking, but feel free to ignore them 🙂