Qiao-yq opened a new pull request, #1594:
URL: https://github.com/apache/dubbo-admin/pull/1594

   ## Status
   
   Draft. This PR is blocked by and must be merged after 
[#1592](https://github.com/apache/dubbo-admin/pull/1592), which establishes the 
Session ownership and access-isolation foundation used by this work.
   
   ## Summary
   
   Implements the shared sensitive-data controls for Phase 1 of Issue #1539.
   
   - add versioned data classification, sanitization, and safe logging 
primitives
   - preserve canonical conversation messages while creating filtered 
model-input views
   - keep normal logs and OpenTelemetry spans metadata-only
   - enforce sensitive-content rejection and metadata allowlists for RAG indexes
   - filter RAG query, embedding, and reranking provider inputs
   - prevent provider, tool, stream, and storage errors from leaking through 
logs or HTTP/SSE responses
   - document the security boundaries and deferred encryption/retention 
decisions
   
   ## Scope
   
   This PR implements PR 2 of the Issue #1539 Phase 1 plan: unified security 
and sensitive-data controls.
   
   Canonical conversation data remains unchanged. Filtering is applied only to 
ephemeral provider views, operational telemetry, and derived index boundaries. 
Baseline protections are always enabled and cannot be disabled by feature flags.
   
   RAG indexing rejects an entire document when its content, external ID, or 
allowed metadata contains detected sensitive data. Query understanding, 
embeddings, and external reranking receive filtered temporary inputs, including 
when their lower-level APIs are called directly.
   
   ## Out of scope
   
   - application-managed field encryption, key rotation, and ciphertext 
migration
   - retention policy and deletion propagation, which belong to Phase 1 PR 3
   - production identity adapters and resource-level authorization beyond #1592
   - a privileged payload-capture or diagnostic channel
   
   ## Compatibility and behavior changes
   
   - tracing `capture_content` now accepts only `none`; existing `truncated` or 
`full` configurations fail validation
   - RAG indexing may reject documents that contain credentials or sensitive 
identifiers instead of silently modifying indexed evidence
   - HTTP and SSE failures return stable public messages while detailed errors 
remain server-side as structured error types
   - canonical Session messages are not rewritten or persisted in sanitized form
   
   ## Validation
   
   - `cd ai && go test ./...`
   - `cd ai && go vet ./...`
   - `git diff --check`
   - `gofmt -d` on changed Go files
   - focused regression tests for model/RAG provider boundaries, sensitive 
document IDs, bounded encoded-secret detection, safe Finding paths, and 
HTTP/SSE error conversion
   
   ## Affected areas
   
   - [x] Docs
   - [ ] Installation
   - [x] User Experience
   - [ ] Dubboctl
   - [x] Console
   - [x] Core Component
   
   ## Characteristics
   
   - [x] Includes tests
   
   Part of #1539.
   
   Blocked by #1592.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to