bamaer commented on PR #9371: URL: https://github.com/apache/hop/pull/9371#issuecomment-6800319244
Thanks @hansva, all addressed in the new commit: - **Endpoint in messages:** no longer uses `URI`. It strips the user info, query and fragment from the text, so `http://user:secret@ollama_host:11434` gives `http://ollama_host:11434` and `myhost:11434/v1?api_key=abc` gives `myhost:11434/v1`. - **Cause message:** URL credentials, key/token/password parameters, bearer tokens and `sk-…` keys are masked. The message is also put on one line and cut at 300 characters. - **Batch size:** `check()` and `init()` share one trimmed parse. Empty means 16 in both, `' 8 '` is valid in both, and the `MIN_VALUE` sentinel is gone. An out-of-range whole number now says it is too large. - **Stop:** a stop during a retry drops the batch, with no error rows. - **Factory:** the existing methods keep langchain4j's default retries. A new overload takes `maxRetries`, and only Embed text passes 0. - **Retries:** only `RetriableException`, or an I/O or timeout cause, is retried. A response that can't be parsed fails at once. - **"at ?":** the `openai` type already showed `https://api.openai.com/v1`. The OpenAI-compatible types with an empty default (Custom, Hugging Face) now show that too. One thing I left as is: without an error hop, the `HopException` keeps the provider exception as its cause, so the raw message can still appear in the log's stack trace. Dropping it would make failures much harder to debug. Happy to change that if you feel it should go. -- 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]
