goutamadwant opened a new pull request, #13179: URL: https://github.com/apache/seatunnel/pull/13179
### Purpose of this pull request Yes, I am contributing this fix to SeaTunnel. Fixes #13178. Keep LLM initialization/inference failures diagnosable without copying rows, prompts, provider payloads or unsafe throwable text into the failure chain. - Reuse existing error categories; retain safe provider/model/status metadata without raw causes or suppressed exceptions. - Check failed HTTP status before reading its body; close responses on success and failure. - Sanitize initialization, request/response parsing and numeric-conversion failures, including invalid metadata and known credential overlap. - Preserve successful CUSTOM fallbacks, strict-Boolean validation and legacy serialization compatibility. - Add unit/Zeta E2E regressions and English/Chinese documentation with migration notes. ### Does this PR introduce _any_ user-facing change? Yes. Error messages and nested causes intentionally change. Failures omit rows and provider content; safe `provider`, bounded `model`, `errorType` and available `httpStatus` remain. Monitoring should use those fields instead of matching raw error text. Options, defaults, successful outputs, existing signatures and invocation attempts are unchanged; no retries are added. This is not general log redaction: explicit close errors, direct constructor validation, testing helpers and configuration/HTTP wire logging are outside scope. ### How was this patch tested? - Reproduced on unchanged dev: 7/7 provider configurations failed the marker regression; the same 7/7 passed after the fix. - Additional regressions reproduced invalid-provider fallback, URL-shaped model metadata and CUSTOM credential overlap before correcting them. - JDK 11 and JDK 17 clean scoped verify: 1,994 tests passed on each, no failures/errors/skips. - Docker-backed Zeta `TestLLMIT` on JDK 11: 14 passed, including existing cases and four OPENAI/CUSTOM HTTP-error/malformed-response checks. - Scoped E2E reactor build and Spotless apply/check passed. Coverage includes all four HTTP adapters and aliases, status/attempt behavior, malformed responses, numeric conversion, causes/suppressed cleanup errors, invalid metadata, failure followed by success and legacy serialization. ```shell # Run with JAVA_HOME pointing to JDK 11, then JDK 17. mvn -B -pl seatunnel-transforms-v2 clean verify -DskipIT=true # Build dependencies on JDK 11, then run the Docker-backed Zeta suite. mvn -B -Pci -pl :seatunnel-transforms-v2-e2e-part-1 -am -DskipTests -DskipIT=true -Dskip.ui=true install RUN_ENGINES=zeta mvn -B -Pci -pl :seatunnel-transforms-v2-e2e-part-1 -DskipUT=true -DskipIT=false -Dit.test=TestLLMIT verify mvn -B -pl seatunnel-transforms-v2,:seatunnel-transforms-v2-e2e-part-1 spotless:apply spotless:check ``` Live provider accounts, Flink/Spark execution and native Windows are unverified. Hosted CI is separate from these local results. ### Check list * [ ] New JAR license/notice updates: not applicable; no dependency or binary package added. * [x] English and Chinese documentation updated. * [x] `incompatible-changes.md` updated for the diagnostic-message/cause contract change. * [ ] Connector registration/distribution checklist: not applicable; existing transform only. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
