From Discord: Challenges in Building a Telemetry Layer from an Effect

[effect-bot]

Summary

Summary

User's Issue:

• The user is trying to implement OpenTelemetry spans in their Effect app using Azure Insights as the exporter.
• They wrapped the layer creation in an Effect.try to handle potential errors.
• They implemented caching using an Option to avoid creating new instances of the telemetry layer on every request.
• Despite this, the telemetry layer does not seem to work as expected when provided to effects.

Debugging Steps:

1. Initial Setup:

   • The user created a telemetry layer using Effect.try and cached it using an Option.
   • They provided this layer to effects using Effect.provide.

2. Observed Behavior:

   • When the telemetry layer is created directly (without effects), it works fine.
   • When created from an effect, the spans do not show up.

3. Discussion Points:

   • The user noted that the telemetry layer gets created once and reused when using their caching mechanism.
   • They observed that without caching, the layer gets created on every request, which is not desired.

4. Suggestions:

   • datner_ suggested using ManagedRuntime to ensure the layer is constructed only once and reused across multiple requests.
   • The user implemented ManagedRuntime and found that it resolves the caching issue but the telemetry layer still does not work.

5. Final Insight:

   • datner_ pointed out that instead of just merging the tracing layer, the entire app should be provided with the tracingLive layer.

Key Takeaways

1. Layer Memoization:

   • Layers in Effect are memoized, meaning they should only construct once even if composed multiple times.
   • Using ManagedRuntime can ensure that layers are constructed once and reused across multiple requests in a long-lived process.

2. Providing Layers:

   • When constructing the application layer, ensure that the entire app is provided with the necessary layers (e.g., tracingLive).

3. Effect Execution:

   • Avoid calling Effect.runPromise repeatedly for short-lived executions. Instead, use a managed runtime to maintain the state of layers.

4. Debugging Layers:

   • If a layer works when created directly but not when created from an effect, it might indicate an issue with how the layer is being provided or constructed.

5. Documentation and DX:

   • The experience highlighted potential areas for improving documentation and developer experience, especially around providing layers and using managed runtimes.

Discord thread

https://discord.com/channels/795981131316985866/1285261463610720297

[datner]

TL;DR

'providing' something might need a bit more clarification from a mental model standpoint