is_sampled logic is flawed

[Anonymous-Coward]

If the X-B3-Sampled header is missing from the incoming request - for example when you just call the app from a browser - no span whatsoever will be sampled. This effectively means the sampling rate will be ignored.

I'm not enough of a Python programmer to submit a merge request, but here's what I think would work:

• create a new field unmarkedRequests in init
• in each _before_request
  • if the header X-B3-TraceId is missing, increment unmarkedRequests
  • is_sampled = (unmarkedRequests >= 100 / sample_rate) or X-B3-Sampled is present and "1"
  • if unmarkedRequests >= 100 / sample_rate then set unmarkedRequests to 0

I tested with this code added in _before_request(self), and it seems to work right:

        if (not headers.get('X-B3-Sampled')):
            self._unmarked = self._unmarked + 1;
            
        is_sampled = (self._unmarked >= 100 / self._sample_rate) or is_sampled;
        if (self._unmarked >= 1 / self._sample_rate):
            self._unmarked = 0;

Why this way: if requests come from a source which has already set sampling info, we don't need to deal with them. Otherwise, we need to record a proportion of sample_rate of them - which translates roughly to one request every 100 / sample_rate requests.