Refactor Windows perf data helper code

[andrewkroh]

Refactoring of the Windows performance data code. There's more cleanup to come. See #3828.

• Improve generated PDH code to return usable error codes.
• Use the standard error interface in return values.
• Use more restrictive param types in generated functions.
• Create PdhErrno that interprets the PDH error codes and returns their human readable error string.

This PR is based on #3862, so it should be merged first to avoid duplicate code review.

[martinscholz83]

This has to be if value.Err == PDH_CALC_NEGATIVE_DENOMINATOR {

[andrewkroh]

Good catch. It would be nice to have test for this error in the future.

[martinscholz83]

Hi @andrewkroh. You wrote

but the first value will always be 0

This is because the time slot is to small between these two calls. I would prefer the following solution

if err := r.query.Execute(); err != nil {
		return nil, err
	}

	// Get the values.
	values, err := r.query.Values()
	if err != nil {
		return nil, errors.Wrap(err, "failed formatting counter values")
	}

	// Write the values into the map.
	result := common.MapStr{}
	var errs multierror.Errors

	for counterPath, value := range values {
		if value.Err != nil {
			// XXX: Sometimes counters return negative values. We can ignore
			// this error. For an explanation see:
			// https://www.netiq.com/support/kb/doc.php?id=7010545
			if value.Err == PDH_CALC_NEGATIVE_DENOMINATOR {
				result.Put(counterPath, 0)
				continue
			}
			if value.Err == PDH_INVALID_DATA {
				if !r.executed {
					// Most counters require two sample values in order to compute a
					// displayable value. So the first fetch for this counters return PDH_INVALID_DATA.
					// So check if this is the first fetch. If so just move on and dont't put that 0 value
					// into our result. If not this is maybe a real error
					r.executed = true
					continue
				}
			}
			errs = append(errs, value.Err)
			continue

[andrewkroh]

I changed up the error handling a bit to sort of match ^. If an error occurs it will report the metric as zero. But it uses some conditions to determine whether or not the error is reported in the event. There are two special cases, one for PDH_CALC_NEGATIVE_DENOMINATOR and one for PDH_INVALID_DATA on the first fetch.

[andrewkroh]

Feel free to experiment and change it up. I primarily wanted to remove the sleep.

[martinscholz83]

Hi @andrewkroh. These changes are really great!! I have left 2 comments about the first query. One thing i'm very interested in is how you get all these PDH error codes with that gcc command.

[martinscholz83]

Ok, got it. These are define preprocessors coming from pdhmsg.h. And that -D command make them visible.

[martinscholz83]

Hi @andrewkroh. For error PDH_CALC_NEGATIVE_DENOMINATOR i currently write a test. But i think we should first merge your changes to not mix up things.