Set sample rate per error type

[zsofiaserenyi]

I'm currently using Sentry iOS SDK version 8.9.4 and I'm looking to transition from Crashlytics to Sentry. However, the quantity of custom errors (non-fatal errors) is quite high, making a direct 1:1 migration challenging. According to the information in the documentation, adjusting the sampleRate seems to impact all error types uniformly.

My goal is to receive notifications for all crashes without a percentage-based limitation, while still applying a sampling rate to the custom errors to manage their volume. Is there a method/best practice to accomplish this within the Sentry iOS SDK?🤔

[armcknight]

Hi @zsofiaserenyi, thanks for writing in, this is a good question. We have sampling mechanisms for three things, events, traces and profiles. For each, you can set a numerical sample rate. For traces/profiles, you can also set a custom block you can add arbitrary logic in where you could make such a decision (see tracesSampler and profilesSampler), but there's no equivalent API for event-level custom sampling logic like eventsSampler 🤔 This seems like it'd be a great feature addition, I'm going to ask about this with the team.

[armcknight]

@zsofiaserenyi Quick followup here, you should be able to get the same functionality by using SentryOptions.beforeSend, please check that out and let me know if it'd work for your purposes!

[zsofiaserenyi]

Hi Andrew, thank you very much for getting back to me so quickly, I appreciate your prompt help with my issue.:)

Based on the documentation and my current understanding beforeSend can be used to:

• add additional info to the event
• drop events

So I could do something like this to drop all non-fatal errors:

  options.beforeSend = { event in
      event.level == .error ? nil : event
  }

However, this is not the desired outcome because I would like to send some % of the non-fatal errors.

So I could also do something like this to only send 5% of the non-fatal errors:

  options.beforeSend = { event in
      guard event.level == .error else { return event }
      return Int.random(in: 0 ... 100) <= 5 ? event : nil
  }

This means setting up a simple way to select only some of the errors locally.

I'm searching for a solution that doesn't require such logic from my side as my assumption is that the sampleRate option within the SDK possesses a more sophisticated capability than what I can achieve through the above method. 🤔

Let me know if I've overlooked any facets of the beforeSend functionality and if there are other ways to achieve my goal.🙏

[philipphofmann]

What you described with

  options.beforeSend = { event in
      guard event.level == .error else { return event }
      return Int.random(in: 0 ... 100) <= 5 ? event : nil
  }

is the current way to go.

As pointed out by @armcknight there is no API for custom sampling of events. @armcknight, did you get any feedback on such an API? I agree with you that such an API could make sense.

[acfriedman]

Hi @philipphofmann

I'd suspect we would need to set the sampleRate to 1.0 if we do this custom sampling with beforeSend? If not, we would be applying a sample to a sample, right?

[philipphofmann]

Sorry for the late reply, @acfriedman; I was on PTO for a month. Yes, indeed, you need to set the sampleRate to 1.0 for your use case.