ChatObservationProperties.class missing logCompletion field and corresponding setter method.

[iAMSagar44]

The ChatObservationProperties.class is missing the logCompletion field and its corresponding setter method.

As per the documentation, the following properties have changed in RC1 -

spring.ai.chat.observations.include-prompt → spring.ai.chat.observations.log-prompt
spring.ai.chat.observations.include-completion → spring.ai.chat.observations.log-completion

This is resulting in the following error -

"message": "'spring.ai.chat.observations.log-completion' is an unknown property. Did you mean 'spring.ai.chat.observations.log-prompt'?"

A similar error is noticed for the below property too, due to the missing field (logQueryResponse) in VectorStoreObservationProperties.class
spring.ai.vectorstore.observations.log-query-response

Spring AI version - 1.0.0-RC1

[quaff]

It should be fixed by #3156

[iAMSagar44]

Thanks @quaff. I also noticed that with RC-1 release the chat prompts are no longer shown in a tracing service (like Zipkin). Prior to the RC-1 release, I was able to see the prompts in tracing services like Zipkin, but with the RC-1 release, they have switched observability from tracing to logging.
Is there a way I can switch back to tracing?

[tzolov]

Thanks for catching this.
IMO, the issue isn't as critical as it appear. You can continue using the properties as documented (e.g., log-xyz).
The discrepancy exists because ChatObservationProperties isn't used programmatically. The affected properties are only used for conditional auto-configurations (like this), where the values are correctly implemented (e.g., log-completion vs. include-completion).
Tests are working fine with the current implementation, but would likely fail if camelCase (e.g., logCompletion) were used instead of log-completion.

We need to fix this issue, but, IMO, the observability features should still work as documented in the meantime.

[iAMSagar44]

@tzolov

Thanks for clarifying.

Prior to the RC-1 release, I was able to see the prompts in tracing services like Zipkin, but with the RC-1 release, it looks like observability has switched from tracing to logging.
Is there a way I can switch back to tracing?
Validating chat memory and prompts across multiple AI agents is easier with tracing services like Zipkin instead of relying on console logs.

[markpollack]

@jonatan-ivanov @shakuzen @ThomasVitale - thoughts?

I think we could provide some additional configuration or example code to use an alternative configuration that reproduces the old behavior.

[jonatan-ivanov]

Is there a way I can switch back to tracing?

See the discussions in the PR for the background: #3003
If you really want to attach possibly large amount of data to spans, as a workaround, you can copy this (you need to hardcode the key): https://github.com/spring-projects/spring-ai/blob/72a2862bcea951d465d87734ef32b3cc1ab91a4e/spring-ai-client-chat/src/main/java/org/springframework/ai/chat/client/observation/ChatClientPromptContentObservationFilter.java
If you create a bean from this or add @Component, it should work.

Validating chat memory and prompts across multiple AI agents is easier with tracing services like Zipkin instead of relying on console logs.

Those are not necessarily console logs. They are logs going to wherever you want them to go.
The logs contain the traceId and the spanId so if you have the span in Zipkin, you can look them up in your log backend.

Could you please also tell us what is your use-case to add this kind of data (large amount of text) to spans?

[codefromthecrypt]

@making has been making zipkin-otel which you can drop in an image and it can accept the logs as annotations https://github.com/openzipkin-contrib/zipkin-otel?tab=readme-ov-file#docker-images

[tzolov]

@iAMSagar44 , I've tried to factor out the handlers and filters (removed in RC1) into a standalone project: https://github.com/tzolov/observability-extensions
One option is to add the project to the spring-ai-community