Project: Tracing without Performance

[HazAT]

Project Board

See our project board to track this initiative across all SDKs

Web Frontend SDKs

Beta Give feedback

1. Tracing without Performance sentry-javascript#8352
   23 of 25
   Feature: Spans Package: core Status: Backlog +3
   
2. Tracing without Performance sentry-electron#688
   Feature: performance Type: Enhancement +2
   

Web Backend SDKs

Beta Give feedback

1. Tracing without Performance sentry-go#656
   5 of 5
   Status: Backlog Type: Feature +2
   
2. Tracing without Performance sentry-php#1525
   5 of 5
   Status: In Progress Type: Feature Type: Improvement +3
   
3. Tracing without Performance sentry-python#2128
   Enhancement +1
   
4. [java] Tracing without Performance sentry-java#2768
   Platform: Android Platform: Java +2
   
5. Tracing without Performance sentry-laravel#713
   Status: Backlog Type: Improvement +2
   
6. Tracing without Performance sentry-symfony#735
   Status: Backlog Type: Improvement +2
   
7. Tracing without Performance sentry-dotnet#2447
   Feature +1
   
8. Tracing without Performance sentry-ruby#2056
   5 of 7
   enhancement +1
   
9. Tracing without Performance sentry-unity#1395
   Platform: Unity +1

Mobile SDKs

Beta Give feedback

1. Tracing without Performance sentry-cocoa#3136
   Platform: Cocoa +1
   
2. Tracing without Performance sentry-dart#1541
   Distributed Tracing Platform: Dart +2
   
3. Tracing without Performance sentry-react-native#3166
   2 of 3
   Platform: React-Native +1

Description

Internal Notion Page:
https://www.notion.so/sentry/Tracing-without-performance-efab307eb7f64e71a04f09dc72722530

To always have access to a trace and span ID, add a new internal PropagationContext property to the scope, an object holding a traceId and spanId and an optional dynamicSamplingContext.

Populate the traceId and spanId with valid, random IDs on construction.

You may update the traceId and dynamicSamplingContext from the request headers during an incoming request or if the process was exposed to a SENTRY_TRACE and/or SENTRY_BAGGAGE environment variable if performance is disabled.

The Dynamic Sampling Context should be lazily constructed once needed, using a new fromOptions function that constructs the DSC based on the client options if performance is disabled.

When captureEvent is called, you may use these values on the scope as a fallback to construct the trace context and envelope header item if no transaction exists in the SDK.

For outgoing HTTP requests, use the values in the scope as a fallback if no transaction is present. The sentry-trace will contain a trailing -0 indicating an unsampled transaction.

Task List for each SDK

Beta Give feedback

1. All events are always sent through the /envelope endpoint if performance is enabled. (If possible just always use /envelope)
2. Add a mechanism to continue a trace from environment variables.
3. All error, transaction and check-in events sent to Sentry always contain a Dynamic Sampling Context (trace) envelope header item.
4. All error, transaction and check-in sent to Sentry always contain a trace context.
5. All outgoing HTTP requests always contain a sentry-trace and baggage header if the destination is allow-listed in [tracePropagationTargets](https://develop.sentry.dev/sdk/performance/#tracepropagationtargets).
6. Create detailed docs about how to setup distributed tracing
7. Set the replay context (Python & PHP only) https://develop.sentry.dev/sdk/event-payloads/contexts/#replay-context

[sl0thentr0py]

So I am fine with doing something quick right now to unblock replay without needing majors. But eventually, we should solve this properly in majors by decoupling Propagation from start_transaction and having a separate well-defined abstraction as discussed with @cleptric and @AbhiPrasad ealrier.

[HazAT]

@sl0thentr0py full ack - eventually, with Starfish and potentially having just Spans, we are going to revamp this and make it way more straightforward and also internally less complex.

[HazAT]

ref: getsentry/sentry#45529

[bruno-garcia]

(consider enableTracing)

I recall proposing this earlier when it was being introduced. My understanding is that today this means tracesSampleRate=1.0 which means turning that on by itself will cause someone to burn through their quota quickly.
Does it mean the documentation would need to be:

enableTracing=true;
tracesSampleRate=0

?

[HazAT]

@bruno-garcia yes or a new flag - or just tracing origin setting 🙃

[bruno-garcia]

For the Replay 'Connect to Backend Errors' feature to light up, we need getsentry/team-webplatform-meta#45 which doesn't depend on this work. Through getsentry/team-webplatform-meta#45 we at least get errors on backend errors when Performance monitoring is turned on.