草稿
[WIP]feat: expose retrieval config and structured observability #1
DoraA_Mengjie创建于 24 天前
[WIP]feat: expose retrieval config and structured observability #1
草稿
合并受阻
23 天前 强制推送 18 个提交:61017ba7-build: configure Huawei Cloud Cargo mirror,831d5b7c-!9 build: configure Huawei Cloud Cargo mirror,02b86583-feat: add graph build pipeline,e801ae00-feat: add graph benchmark CLI options,1bf12e15-feat: enable graph mode in memory-bench,6b565e2f-feat: pass graph benchmark flags through evaluation runners,bd66c4a5-docs: document graph benchmark usage,9878a6c3-feat: wire graph memory benchmark evaluation,19c36fd5-feat: broaden default graph registry,476f3b4f-feat: add deterministic graph evidence retrieval,66bc6426-feat: add graph build audit and bounded concurrency,cc0551fd-feat: add source provenance and graph query planning,c5ddc6a2-feat: add bounded rank fusion for graph retrieval,9cb4c60a-feat: align atomic memory provenance for graph ingestion,e35d2848-feat: integrate graph memory with MCP service,bbf6ba5d-feat: align graph and rerank evaluation across datasets,ee049780-!11 feat: integrate graph memory into retrieval, MCP, and benchmark evaluation,16a50521-feat: configure retrieval and runtime observability
23 天前 强制推送 18 个提交:61017ba7-build: configure Huawei Cloud Cargo mirror,831d5b7c-!9 build: configure Huawei Cloud Cargo mirror,02b86583-feat: add graph build pipeline,e801ae00-feat: add graph benchmark CLI options,1bf12e15-feat: enable graph mode in memory-bench,6b565e2f-feat: pass graph benchmark flags through evaluation runners,bd66c4a5-docs: document graph benchmark usage,9878a6c3-feat: wire graph memory benchmark evaluation,19c36fd5-feat: broaden default graph registry,476f3b4f-feat: add deterministic graph evidence retrieval,66bc6426-feat: add graph build audit and bounded concurrency,cc0551fd-feat: add source provenance and graph query planning,c5ddc6a2-feat: add bounded rank fusion for graph retrieval,9cb4c60a-feat: align atomic memory provenance for graph ingestion,e35d2848-feat: integrate graph memory with MCP service,bbf6ba5d-feat: align graph and rerank evaluation across datasets,ee049780-!11 feat: integrate graph memory into retrieval, MCP, and benchmark evaluation,16a50521-feat: configure retrieval and runtime observability
23 天前 强制推送 1 个提交:3e2793c6-feat: add RPM retrieval configuration and staged observability
23 天前 强制推送 1 个提交:3e2793c6-feat: add RPM retrieval configuration and staged observability
22 天前 推送 1 个提交:11f51ca6-fix: address retrieval observability review feedback
22 天前 推送 1 个提交:50e7ce4c-fix: address PR 13 review findings
变更说明
这是基于 openEuler RAM-A PR11 (
ae85f03) 的依赖 PR,用于补齐 RPM/MCP 服务的检索配置和运行时可观测性。检索配置
retrieval配置,支持mode、candidate_k、embedding_weight和bm25_weight。retrieval.rerank,支持 provider、model、credential env、base URL、input_k、timeout 和 fail-open。features.graph_memory和graph_memory.retrieval控制,避免出现第二个图记忆开关来源。OpenRouter-compatible 规格
/rerankwire protocol,也支持兼容该协议的本地 HTTP rerank 服务。api_key_env: null时不发送 Authorization header,适用于部署在受信任网络内的无鉴权服务。可观测性
RUST_LOG控制级别。request_id贯穿 MCP tool、memory ingest/search、provider retry 和异步 graph build。兼容性
retrieval时继续使用 hybrid0.7/0.3,rerank 默认关闭。验证
cargo fmt --all -- --checkcargo clippy -p memory-mcp -p memory-core -p memory-pipeline -p memory-cases --all-targets -- -D warningsgit diff --check依赖关系
本 PR 以 PR11 为基线。PR11 合入后应 rebase 到最新
openeuler/master,再将目标分支切换为master。