quarkus.langchain4j.timeout as a default configuration for all timeouts

[diegolovison]

Hi, I have been working on #1626 and #1606 on my "private project", and my main difficulty is the timeout granularity.

Today, the quarkus-langchain4j, quarkus, and vert-x have specific timeout configurations that is super good once you have your project working and you know exactly what part you should optimize. But, as someone starting using the quarkus-langchain4j imho, this is super problematic. I spent most of my time trying to setup the timeouts (because of the granularity) instead of working on my project.

My idea is: if the property quarkus.langchain4j.timeout is presented, then all timeouts are set equal to the value of that property. If the property is not set, then each timeout can be configured independently.

This is a good balance and reduces the learning curve. Those "timeout problems" are not because of the library but because of different models and individual computer performance.

[geoand]

I definitely see the value in this, even though one could easily argue it's not semantically correct.

@cescoffier WDYT?

[geoand]

With version 1.1.0 the quarkus.langchain4j.timeout property will apply to MCP invocations as well.

[diegolovison]

Should we close this discussion? Do you have the PR?

[geoand]

Yup, it was mentioned. #1630

[diegolovison]

If I keep finding "timeouts" issues, can I open a PR directly and follow this approach?

[geoand]

You certainly can, but keep in mind what I mentioned above about us beeing weery of using the same approach everywhere as it will lead to unintended changes of the default.