Use float for sample rand #4677
Conversation
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## master #4677 +/- ##
==========================================
- Coverage 85.04% 84.97% -0.08%
==========================================
Files 156 156
Lines 15859 15851 -8
Branches 2684 2687 +3
==========================================
- Hits 13488 13469 -19
- Misses 1591 1597 +6
- Partials 780 785 +5
|
| ) | ||
| sample_rand_scaled = rng.randrange(int(lower * 1000000), int(upper * 1000000)) | ||
|
|
||
| return sample_rand_scaled / 1000000 |
There was a problem hiding this comment.
Bug: Sampling Context Propagation Fails for Tiny Rates
The _generate_sample_rand function's integer scaling for randrange arguments (int(value * 1e6)) can cause the start and stop values to become equal for very narrow intervals (less than 1e-6). This leads randrange to raise a ValueError ("empty range"). This error is caught in PropagationContext._fill_sample_rand, which then silently skips setting the sample_rand. As a result, dynamic sampling context propagation fails for traces with small sample_rates (e.g., when parent_sampled=True with a tiny sample_rate), leading to inconsistent sampling decisions across services.
Closes #4270