OpenAI-compatible passthrough에서 caller가 보낸 provider-native thinking field가 catalog 기본 thinking budget으로 덮이지 않아야 한다. provider가 거부한 extension field도 Edge가 선판단하지 않고 provider status/body로 relay되는지 테스트 근거를 남긴다.
provider 라우트는 raw tunnel 패스스루, 그 외 라우트는 정규화된
RunEvent 경로로 응답을 결정한다. caller metadata는 임의 컨텍스트이며
응답 경로/shape를 선택하지 않는다 (SDD S01).
- chat_handler.go: response_mode parse/switch 로직 제거, 라우트 기반 분기로 단순화
- responses_handler.go: tunnelResponsesPassthroughSideband 함수 및 response_mode switch 제거
- stream.go: 관련 response_mode 라벨 사용 정리
- server_test.go: response_mode 관련 테스트 케이스 제거 및 정리
- usage_metrics_test.go: 사용되지 않는 테스트 대목 제거
- docs/openai-usage-grafana.md: response_mode 라벨 설명을 passthrough/normalized로 수정
- seulgivibe OpenAI-compatible provider 완료 후 archive로 이동
- 모델 그룹 혼합 제공자 디스패치 milestone 진행
- agent-spec: OpenAI-compatible surface, provider pool config refresh 갱신
- Go: mapper.go/config.go provider config 매핑 개선
- agent-task: 새 subtask plan/code-review 문서 추가
- Implement stream.go with sideband passthrough support
- Update chat_handler.go with streaming handler changes
- Add run_dispatch.go with new dispatch service logic
- Add server tests for streaming and sideband functionality
- Archive old plan/code-review docs for cloud-G07
Remove all *_preview zap.String/Any fields from OpenAI-compatible log
calls (chat_handler, responses_handler, stream) so that prompt, content,
reasoning, source, and delta values are never written in plain to
operational logs.
Replace preview fields with non-content metadata:
- prompt_len, content_len, reasoning_len, delta_len, source_len
- message_count, content_len, finish_reason, tool_call_count
Also remove the now-unused previewString helper from types.go.
Add log_safety_test.go with regression tests that:
- verify no *_preview fields exist on any log line
- verify operational metadata fields are retained for observability
- verify logOpenAICompatStreamOutput does not emit delta_preview
vLLM-MLX provider-pool에서 strict output이 thinking을 끄지 않도록 하고, queue_timeout_ms=0을 IOP queue timeout 없음으로 해석해야 한다. long reasoning/long-context 요청은 IOP queue timeout이 아니라 caller cancellation과 backend timeout 정책으로 제어한다.
- Update control-plane edge wire and edge-node runtime wire contracts
- Refactor model queue service with admission control
- Update chat handler and responses handler for edge
- Modify run dispatch and status provider logic
- Add/modify runtime proto definitions
- Move G07 status logs to archive
- Add config refresh classification and test coverage
- Update OpenAI chat/responses/stream handlers with config refresh support
- Add comprehensive server tests for config refresh
- Update local dev guide with config refresh information
- Extend config package with ConfigRefresh field
- Add stream trace logging to edge server streamChatCompletion
- Add stream trace logging to node adapter openai_compat provider
- Enable trace via IOP_OPENAI_COMPAT_TRACE_STREAM env var or iop_trace_stream metadata
- Track content, reasoning, tool_calls chunks with think tag detection
- reasoning_delta 이벤트 전파: edge console와 node sink에서 explicit 케이스로 처리
- 각 줄에 [node-reasoning] prefix를 붙여 reasoning 출력을 구분
- 빈 delta는 무시하고, 멀티라인 content도 각 줄에 prefix 적용
- 관련单元测试 추가
- Add classify.go/classify_test.go for config-based priority routing
- Add model_queue.go/test with priority queue support
- Update run_dispatch.go to use priority routing
- Add edge.yaml config for model priority settings
- Add config.go/config_test.go for priority config parsing
- Remove G06 priority routing plan/review (merged into main flow)
- Add archive for m-node-provider-first-config-surface task
- Move node-resource-model-unification milestone and SDD to archive
- Update PHASE.md for operational-observability-provider-management
- Refactor edge node registration and model queue logic
- Add status provider tests for model queue validation
- Update edge config and Go config structs
