[
https://issues.apache.org/jira/browse/CAMEL-24560?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Claus Ibsen resolved CAMEL-24560.
---------------------------------
Resolution: Fixed
> Change GenAiUsage token fields from Integer to Long
> ---------------------------------------------------
>
> Key: CAMEL-24560
> URL: https://issues.apache.org/jira/browse/CAMEL-24560
> Project: Camel
> Issue Type: New Feature
> Reporter: Omar Atie
> Assignee: Omar Atie
> Priority: Major
> Fix For: 4.23.0
>
>
> h2. Background
> During PR #25337 review, Claus Ibsen noted that OpenAI token counts
> (usage.promptTokens(), usage.completionTokens()) return long, but GenAiUsage
> stores Integer inputTokens/outputTokens, requiring lossy (int) casts in
> OpenAIProducer.
> h2. Problem
> Large token counts could overflow Integer.MAX_VALUE. Casting long to int is a
> silent data-loss risk for observability dashboards and metrics.
> h2. Proposed scope
> * Change GenAiUsage record fields inputTokens and outputTokens from Integer
> to Long
> * Update all call sites (OpenAIProducer, LangChain4j producers,
> SpringAiChatProducer, GenAiObservabilityImpl, Micrometer backend)
> * Update GenAiUsage.of() factory
> * Update tests and span attribute serialization
> * Add upgrade-guide note if any public API/header behavior changes (unlikely
> for internal record, but document if metrics tags change type)
> h2. References
> * Parent: CAMEL-23861
> * components/camel-ai/camel-ai-observability-api/.../GenAiUsage.java
> * apache/camel PR #25337 review (davsclaus)
> h2. Acceptance criteria
> * No (int) casts on OpenAI token usage
> * All existing observability tests pass
> * Span attributes and Micrometer metrics accept long token values correctly
> * API change documented if camel-ai-observability-api is considered public SPI
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
