Aias00 opened a new issue, #7906:
URL: https://github.com/apache/shenyu/issues/7906
### Behavior and reproduction
A response-transformation rule builds a prompt containing the original
request body, but reads that body after the downstream HTTP chain has already
consumed it. With a normal one-shot POST body, the model receives an empty/null
request body, or the second subscription fails, even though the backend
received the real payload.
Reproduced at `fda742c48e7f553c9964cf74b36732326d0fde00` with the real
`AiResponseTransformerTemplate` and Spring codecs. A one-shot request publisher
emits `{"customer":"alice"}` once. Consume it as the downstream HTTP client
does, then assemble the response prompt. Observed: backend body is the full
JSON, request-body subscription count is **2**, and the prompt contains no
`alice` (its request body is null). A replayable MockServerHttpRequest body
masks this lifecycle problem.
### Location and cause
- [Capture the original request, then invoke the downstream
chain](https://github.com/apache/shenyu/blob/fda742c48e7f553c9964cf74b36732326d0fde00/shenyu-plugin/shenyu-plugin-ai/shenyu-plugin-ai-response-transformer/src/main/java/org/apache/shenyu/plugin/ai/transformer/response/AiResponseTransformerPlugin.java#L132)
- [Assemble the prompt during response
writing](https://github.com/apache/shenyu/blob/fda742c48e7f553c9964cf74b36732326d0fde00/shenyu-plugin/shenyu-plugin-ai/shenyu-plugin-ai-response-transformer/src/main/java/org/apache/shenyu/plugin/ai/transformer/response/AiResponseTransformerPlugin.java#L337)
- [Subscribe to
originalRequest.getBody()](https://github.com/apache/shenyu/blob/fda742c48e7f553c9964cf74b36732326d0fde00/shenyu-plugin/shenyu-plugin-ai/shenyu-plugin-ai-response-transformer/src/main/java/org/apache/shenyu/plugin/ai/transformer/response/template/AiResponseTransformerTemplate.java#L157)
There is no replayable request-body decorator or snapshot installed by this
plugin before downstream consumption. The response transformer retaining a
ServerHttpRequest object does not retain its consumed body.
### Expected behavior / suggested repair
Capture a bounded request-body snapshot before forwarding and preserve a
replayable stream for the downstream consumer. Build the prompt from that
snapshot; do not subscribe to the original transport body twice. Test both the
backend body and model prompt through one-shot POST/PUT bodies, with empty-body
and cancellation controls.
### Scope / related issues
Functional correctness, medium severity. Distinct from #6773 (unset response
status), #6795 (stream response buffering), and #6757 (writeAndFlushWith
interception). No code fix is included in this audit issue.
--
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]
