Adalennis opened a new issue, #9287:
URL: https://github.com/apache/hop/issues/9287
### Apache Hop version?
2.20.0
### Java version?
Java 21
### Operating system
macOS
### What happened?
Reproduced on 2.20.0 with Ollama (nomic-embed-text), code unchanged on main.
**1. Text made only of some Unicode spaces is embedded**
Follow-up to #8750. `EmbedText.hasNothingToEmbed` uses `trim()` and
`isBlank()`, which don't count these characters, so the rows are sent to the
provider:
| Text | Result |
|---|---|
| 2 × U+00A0 (no-break space) | 768-dim embedding |
| U+202F (narrow no-break space) | 768-dim embedding |
| U+200B (zero-width space) | 768-dim embedding |
| 3 normal spaces | no embedding, no call |
The three embeddings are identical (`[0.022947188, 0.03171978, …]`), so they
carry no meaning, but they cost a call and land in the vector store.
**2. An unresolved batch size silently becomes 16**
Batch size `${EMBED_BATCH_NOT_SET}` (a variable that is not set): `init()`
runs `Const.toInt(resolve(batchSize), 16)`, so the transform runs with 16 and
logs nothing. A batch size of 0 stops the transform with an error, so a value
that is not a number should too.
**3. A provider failure has no details**
Provider pointing at a port where nothing listens
(`http://localhost:11499`), batch size 2, 3 rows:
- With an error hop, every row gets `error_description` =
`java.net.ConnectException`, with no provider name, URL or host.
- Without an error hop, the transform stops with `Error calling embedding
model nomic-embed-text / java.net.ConnectException`.
- langchain4j retries each batch twice (`RetryUtils - A retriable exception
occurred. Remaining retries: 2 of 2`) and logs a full stack trace on each
retry: 4 stack traces for 2 batches.
Expected: the provider name and base URL in the message, and no stack trace
per retry at the default log level.
### Issue Priority
Priority: 3
### Issue Component
Component: Hop Gui
--
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]
