Bug: Sampler value always lower than sample rate

[sfrom]

Expected Behaviour

Think there is a bug when using log sampling. When I set POWERTOOLS_LOGGER_SAMPLE_RATE environment to 0.9 I would expect in 90 percent of the cases to have all logs elevated to debug level and thus logged even though minimum log level is set to Error using POWERTOOLS_LOG_LEVEL variable.

Current Behaviour

Logs are never elevated to debug. It only happens when I set POWERTOOLS_LOGGER_SAMPLE_RATE to 1.

Code snippet

I am using the logger using the static log configuration in a console application and I set POWERTOOLS_LOG_LEVEL to Error and POWERTOOLS_LOGGER_SAMPLE_RATE to 0.9 using environment variables

[Logging(Service = "HelloWorldService", LoggerOutputCase = LoggerOutputCase.CamelCase, LogEvent = true)]
  static void Main(string[] args)
  {
  
    var logLevel = Environment.GetEnvironmentVariable("POWERTOOLS_LOG_LEVEL");
    Console.WriteLine($"POWERTOOLS_LOG_LEVEL: {logLevel}");
    var sampleRate = Environment.GetEnvironmentVariable("POWERTOOLS_LOGGER_SAMPLE_RATE");
    Console.WriteLine($"POWERTOOLS_SAMPLE_RATE: {sampleRate}");
  
    // This should NOT be logged (Info < Error)
    Logger.LogInformation("This is an info message — should not appear");

  }

Possible Solution

No response

Steps to Reproduce

Just create a console app using the same configuration and environment variables. Information is only loggen when sample rate is set to 1

Powertools for AWS Lambda (.NET) version

2.0.0

AWS Lambda function runtime

dotnet8

Debugging logs

[boring-cyborg]

Thanks for opening your first issue here! We'll come back to you as soon as we can.
In the meantime, check out the #dotnet channel on our Powertools for AWS Lambda Discord: Invite link

[leandrodamascena]

Hi @sfrom, thanks for opening this issue to report this behavior. I think this happens because the sampling decision happens in the environment initialization phase.

Since it's a probability algorithm that uses Random, it can happen that, even using a sampling rate of 0.9, the Random algorithm results in the probability of 0.1 and never calculates it again. And that's why if you use 1.0 you will always have sampling, because this will be true in 100% of the cases.

We had a long discussion about this here and here and we changed the behavior in both runtimes. We need to investigate if we can do the same here in Powertools dotnet.

I'd like to hear your opinions too @hjgraca.

[dcabib]

I've created PR #980 to fix this issue. The fix changes the GetSafeRandom() method to properly normalize random values to the [0,1] range using (double)randomUInt / uint.MaxValue instead of BitConverter.ToDouble(). All 669 core component tests are passing. Please review and merge when ready.