Claus Ibsen created CAMEL-26216:
-----------------------------------
Summary: camel-pinecone - QUERY without the TopK header fails with
NullPointerException, and the pinecone:rag transformer always gives an empty
list
Key: CAMEL-26216
URL: https://issues.apache.org/jira/browse/CAMEL-26216
Project: Camel
Issue Type: Bug
Components: camel-pinecone
Reporter: Claus Ibsen
Fix For: 4.24.0
Found while adding the catalog body types of pinecone (CAMEL-24961, PR #28402).
h4. 1. QUERY and QUERY_BY_ID without CamelPineconeQueryTopK fail with
NullPointerException
{{PineconeVectorDbProducer.query}} (main 2a465aa4814c, line 302) and
{{queryById}} (line 329) do {{int topK = in.getHeader(QUERY_TOP_K,
Integer.class)}}; without the header the unboxing fails with a
{{NullPointerException}}. There is no default (camel-weaviate uses 10,
camel-pgvector 3). Checked on 4.24.0-SNAPSHOT with pinecone-local.
Proposed: use a default (for example 10) or an endpoint option, and fail with a
clear message otherwise.
h4. 2. The pinecone:rag transformer never finds the query results
{{PineconeReverseEmbeddingsDataTypeTransformer}} (line 38) reads the body as a
{{List<ScoredVectorWithUnsignedIndices>}}, but QUERY leaves a
{{QueryResponseWithUnsignedIndices}} as the body (line 313), so
{{getBody(List.class)}} gives null and the transformer sets an empty list. A
route QUERY then {{transform(new DataType("pinecone:rag"))}} always gets no
text. Read in the code, not run.
Proposed: take {{getMatchesList()}} when the body is a
{{QueryResponseWithUnsignedIndices}}.
h4. 3. A vector as a List of Double fails with ClassCastException
{{upsert}} (line 218), {{update}} and {{query}} pass the body List to the
Pinecone SDK, which expects {{List<Float>}}; a List of Double (what a JSON
array unmarshals to) fails with a raw {{ClassCastException}}. camel-weaviate
and camel-pgvector have the same problem. Checked on 4.24.0-SNAPSHOT with
pinecone-local.
Proposed: convert each element with {{Number.floatValue()}}.
h4. 4. UPSERT without CamelPineconeIndexId fails
{{upsert}} (line 213) passes a null id to {{index.upsert}} when the header is
not set, and the SDK fails with {{PineconeValidationException}}. Checked on
4.24.0-SNAPSHOT with pinecone-local.
Proposed: generate an id (as the {{pinecone:embeddings}} transformer does with
a random UUID) or fail with a message that names the header.
h4. 5. camel-test-infra pinecone maps only port 5080
{{PineconeLocalContainerInfraService}} (line 64) and {{PineconeIndexContainer}}
(line 47) map only port 5080, the control plane of pinecone-local; the indexes
it creates are served on 5081 and up, so data-plane calls (upsert, query,
fetch) from the test cannot reach the index. Checked on 4.24.0-SNAPSHOT:
pinecone-local was started by hand with ports 5080-5085 to run the data-plane
operations.
Proposed: also map the index ports (5081-5085), or document that only the
control plane can be tested.
_Claude Code on behalf of davsclaus_
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
