{"id":107376,"date":"2026-09-19T13:40:37","date_gmt":"2026-09-19T05:40:37","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/107376.html"},"modified":"2026-09-19T13:40:37","modified_gmt":"2026-09-19T05:40:37","slug":"milvus-%e7%94%9f%e4%ba%a7%e7%8e%af%e5%a2%83%e9%83%a8%e7%bd%b2%ef%bc%8c%e4%bc%98%e5%8c%96%ef%bc%8c%e6%97%a5%e5%b8%b8%e7%bb%b4%e6%8a%a4%e4%b8%ad%e9%81%87%e5%88%b0%e7%9a%84%e9%97%ae%e9%a2%98","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/107376.html","title":{"rendered":"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898"},"content":{"rendered":"

Milvus \u751f\u4ea7\u73af\u5883\u8fd0\u7ef4\u6df1\u5ea6\u6307\u5357<\/h2>\n

\u4e00\u3001\u67b6\u6784\u5168\u666f<\/h3>\n

\u6574\u4e2a\u96c6\u7fa4\u7531\u56db\u5927\u5c42\u7ec4\u6210，\u6bcf\u5c42\u72ec\u7acb\u6269\u5c55：<\/p>\n

Access Layer Access Layer
\n Proxy Proxy Proxy \u2192 \u8d1f\u8f7d\u5747\u8861 \u2192 \u5ba2\u6237\u7aef SDK
\n \u2502 \u2502 \u2502
\nCoordinator Layer
\n RootCoord QueryCoord DataCoord IndexCoord（\u5355 Active）
\n \u2502 \u2502 \u2502 \u2502
\nWorker Layer
\n QueryNode DataNode IndexNode（\u53ef\u6c34\u5e73\u6269\u5c55，\u6709\u72b6\u6001）
\n \u2502 \u2502
\nStorage Layer
\n MinIO\/S3（\u5bf9\u8c61\u5b58\u50a8） + etcd（\u5143\u6570\u636e） + Kafka\/Pulsar（\u6d88\u606f\u961f\u5217）<\/p>\n

\u5404\u7ec4\u4ef6\u804c\u8d23\u4e0e\u74f6\u9888<\/h4>\n\n\u7ec4\u4ef6\u89d2\u8272\u5173\u952e\u74f6\u9888\u5e38\u89c1\u6545\u969c<\/tr>\n\n\n\n\n\n\n\n\n
Proxy<\/td>\n\u8bf7\u6c42\u5165\u53e3，\u9274\u6743\u3001\u8def\u7531\u3001\u9650\u6d41<\/td>\n\u8fde\u63a5\u6570\u3001QPS<\/td>\n\u8fde\u63a5\u6c60\u8017\u5c3d\u3001OOM<\/td>\n<\/tr>\n
RootCoord<\/td>\nDDL\u3001TSO \u65f6\u95f4\u6233\u5206\u914d<\/td>\n\u5355\u70b9（Active-Standby）<\/td>\nTSO \u5ef6\u8fdf\u5bfc\u81f4\u5168\u96c6\u7fa4\u963b\u585e<\/td>\n<\/tr>\n
QueryCoord<\/td>\nQueryNode \u62d3\u6251\u7ba1\u7406\u3001\u8d1f\u8f7d\u5747\u8861<\/td>\n\u5355\u70b9<\/td>\n\u9891\u9053\u5206\u914d\u4e0d\u5747<\/td>\n<\/tr>\n
DataCoord<\/td>\nSegment \u7ba1\u7406\u3001Compaction \u8c03\u5ea6<\/td>\n\u5355\u70b9<\/td>\n\u5c0f\u6587\u4ef6\u8fc7\u591a\u3001Compaction \u79ef\u538b<\/td>\n<\/tr>\n
QueryNode<\/td>\n\u67e5\u8be2\u6267\u884c\u3001\u5411\u91cf\u68c0\u7d22<\/td>\nCPU\u3001\u5185\u5b58\u3001\u78c1\u76d8 I\/O<\/td>\nOOM\u3001Segment \u52a0\u8f7d\u6162<\/td>\n<\/tr>\n
DataNode<\/td>\n\u6570\u636e\u5199\u5165\u3001Flush \u5230 S3<\/td>\n\u5185\u5b58（\u5199\u5165\u7f13\u51b2）\u3001\u7f51\u7edc<\/td>\nFlush \u8d85\u65f6\u3001\u80cc\u538b<\/td>\n<\/tr>\n
IndexNode<\/td>\n\u7d22\u5f15\u6784\u5efa（CPU\/GPU \u5bc6\u96c6\u578b）<\/td>\nCPU\/GPU\u3001\u5185\u5b58<\/td>\n\u5927 Segment \u7d22\u5f15\u6784\u5efa OOM<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u4e8c\u3001\u751f\u4ea7\u90e8\u7f72（Kubernetes + Operator）<\/h3>\n

\u96c6\u7fa4\u89c4\u6a21\u89c4\u5212<\/h4>\n\n\u6d41\u91cf\u7b49\u7ea7QPSProxyQueryNodeDataNode\u5355\u8282\u70b9\u89c4\u683c<\/tr>\n\n\n\n\n\n
\u4f4e<\/td>\n1K<\/td>\n1<\/td>\n1<\/td>\n1<\/td>\n4C\/32G<\/td>\n<\/tr>\n
\u4e2d<\/td>\n5K<\/td>\n2<\/td>\n2<\/td>\n2<\/td>\n8C\/64G + T4<\/td>\n<\/tr>\n
\u9ad8<\/td>\n10K<\/td>\n2<\/td>\n3<\/td>\n2<\/td>\n16C\/128G<\/td>\n<\/tr>\n
\u8d85\u5927<\/td>\n50K<\/td>\n4<\/td>\n8<\/td>\n4<\/td>\n32C\/256G<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u5173\u952e\u914d\u7f6e<\/h4>\n

apiVersion<\/span>:<\/span> milvus.io\/v1beta1
\nkind<\/span>:<\/span> Milvus
\nmetadata<\/span>:<\/span>
\n name<\/span>:<\/span> prod–<\/span>milvus
\n namespace<\/span>:<\/span> milvus
\nspec<\/span>:<\/span>
\n mode<\/span>:<\/span> cluster
\n components<\/span>:<\/span>
\n image<\/span>:<\/span> milvusdb\/milvus:<\/span>v2.5.4
\n proxy<\/span>:<\/span>
\n replicas<\/span>:<\/span> 3<\/span>
\n serviceType<\/span>:<\/span> ClusterIP
\n queryNode<\/span>:<\/span>
\n replicas<\/span>:<\/span> 3<\/span>
\n resources<\/span>:<\/span>
\n requests<\/span>:<\/span>
\n memory<\/span>:<\/span> 32Gi
\n cpu<\/span>:<\/span> 8<\/span>
\n limits<\/span>:<\/span>
\n memory<\/span>:<\/span> 48Gi
\n cpu<\/span>:<\/span> 16<\/span>
\n dataNode<\/span>:<\/span>
\n replicas<\/span>:<\/span> 2<\/span>
\n resources<\/span>:<\/span>
\n requests<\/span>:<\/span>
\n memory<\/span>:<\/span> 16Gi
\n cpu<\/span>:<\/span> 4<\/span>
\n limits<\/span>:<\/span>
\n memory<\/span>:<\/span> 32Gi
\n cpu<\/span>:<\/span> 8<\/span>
\n indexNode<\/span>:<\/span>
\n replicas<\/span>:<\/span> 1<\/span>
\n resources<\/span>:<\/span>
\n requests<\/span>:<\/span>
\n memory<\/span>:<\/span> 16Gi
\n cpu<\/span>:<\/span> 8<\/span>
\n limits<\/span>:<\/span>
\n memory<\/span>:<\/span> 32Gi
\n cpu<\/span>:<\/span> 16<\/span>
\n dependencies<\/span>:<\/span>
\n etcd<\/span>:<\/span>
\n external<\/span>:<\/span> true<\/span>
\n endpoints<\/span>:<\/span>
\n –<\/span> etcd–<\/span>0.etcd–<\/span>headless:<\/span>2379<\/span>
\n storage<\/span>:<\/span>
\n external<\/span>:<\/span> true<\/span>
\n endpoint<\/span>:<\/span> s3.amazonaws.com
\n bucket<\/span>:<\/span> milvus–<\/span>prod
\n accessKey<\/span>:<\/span> "${AWS_ACCESS_KEY}"<\/span>
\n secretKey<\/span>:<\/span> "${AWS_SECRET_KEY}"<\/span>
\n msgStream<\/span>:<\/span>
\n external<\/span>:<\/span> true<\/span>
\n kafka<\/span>:<\/span>
\n broker<\/span>:<\/span> kafka–<\/span>broker:<\/span>9092<\/span>
\n config<\/span>:<\/span>
\n # \u5173\u952e\u4f18\u5316\u53c2\u6570<\/span>
\n queryNode<\/span>:<\/span>
\n cache<\/span>:<\/span>
\n cacheSize<\/span>:<\/span> 32<\/span> # QueryNode \u7f13\u5b58\u5927\u5c0f，\u5355\u4f4d GB<\/span>
\n warmup<\/span>:<\/span> async # \u5f02\u6b65\u9884\u70ed，\u907f\u514d\u542f\u52a8\u963b\u585e<\/span>
\n dataCoord<\/span>:<\/span>
\n segment<\/span>:<\/span>
\n maxSize<\/span>:<\/span> 512<\/span> # \u5355 Segment \u6700\u5927 512MB<\/span>
\n maxLifetime<\/span>:<\/span> 86400<\/span> # 24 \u5c0f\u65f6\u5f3a\u5236 Flush<\/span>
\n common<\/span>:<\/span>
\n security<\/span>:<\/span>
\n authorizationEnabled<\/span>:<\/span> true<\/span>
\n tlsMode<\/span>:<\/span> 1<\/span>
\n retentionDuration<\/span>:<\/span> 43200<\/span> # Binlog \u4fdd\u7559 12 \u5c0f\u65f6<\/span><\/p>\n

\u4e3a\u4ec0\u4e48\u5916\u90e8\u4f9d\u8d56\u8981\u72ec\u7acb\u90e8\u7f72<\/h4>\n
    \n
  • etcd：Milvus \u5185\u5d4c\u7684 etcd \u4e0d\u505a\u6301\u4e45\u5316，\u91cd\u542f\u540e\u5168\u96c6\u7fa4\u5143\u6570\u636e\u4e22\u5931\u3002\u751f\u4ea7\u73af\u5883\u5fc5\u987b\u5916\u90e8\u72ec\u7acb\u90e8\u7f72 3 \u8282\u70b9 etcd \u96c6\u7fa4\u3002<\/li>\n
  • MinIO\/S3：\u5185\u90e8 MinIO \u65e0\u6cd5\u505a\u9ad8\u53ef\u7528\u3002\u751f\u4ea7\u73af\u5883\u8d70 AWS S3 \/ \u963f\u91cc\u4e91 OSS \/ Ceph\u3002<\/li>\n
  • Kafka\/Pulsar：\u5185\u5d4c RocksMQ \u65e0\u6cd5\u5904\u7406 > 10MB\/s \u7684\u5199\u5165\u541e\u5410，\u5fc5\u987b\u5207\u5916\u90e8 Kafka\u3002<\/li>\n<\/ul>\n

    \u4e09\u3001\u7d22\u5f15\u9009\u578b\u4e0e\u8c03\u53c2<\/h3>\n

    \u7d22\u5f15\u5bf9\u6bd4\u77e9\u9635<\/h4>\n\n\u7d22\u5f15\u7c7b\u578b\u7cbe\u5ea6\u901f\u5ea6\u5185\u5b58\u9002\u7528\u573a\u666f\u6784\u5efa\u65f6\u95f4<\/tr>\n\n\n\n\n\n\n\n
    IVF_FLAT<\/td>\n\u9ad8<\/td>\n\u4e2d<\/td>\n\u4f4e<\/td>\n\u901a\u7528，\u7cbe\u5ea6\u4f18\u5148<\/td>\n\u5feb<\/td>\n<\/tr>\n
    IVF_SQ8<\/td>\n\u4e2d<\/td>\n\u5feb<\/td>\n\u66f4\u4f4e<\/td>\n\u5185\u5b58\u654f\u611f<\/td>\n\u5feb<\/td>\n<\/tr>\n
    IVF_PQ<\/td>\n\u4f4e<\/td>\n\u5feb<\/td>\n\u6781\u4f4e<\/td>\n\u8d85\u5927\u89c4\u6a21（> \u4ebf\u7ea7）<\/td>\n\u4e2d<\/td>\n<\/tr>\n
    HNSW<\/td>\n\u6781\u9ad8<\/td>\n\u6700\u5feb<\/td>\n\u9ad8<\/td>\n\u4f4e\u5ef6\u8fdf，\u5c0f\u89c4\u6a21（< 1 \u4ebf）<\/td>\n\u6162<\/td>\n<\/tr>\n
    DiskANN<\/td>\n\u9ad8<\/td>\n\u4e2d<\/td>\n\u6781\u4f4e<\/td>\n\u78c1\u76d8\u5b58\u50a8，\u6210\u672c\u654f\u611f<\/td>\n\u6162<\/td>\n<\/tr>\n
    GPU_IVF_FLAT<\/td>\n\u9ad8<\/td>\n\u6781\u5feb<\/td>\nGPU \u663e\u5b58<\/td>\n\u9ad8\u541e\u5410，\u4f4e\u5ef6\u8fdf<\/td>\n\u5feb<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u8c03\u53c2\u539f\u5219<\/h4>\n

    # HNSW \u53c2\u6570（\u5c0f\u89c4\u6a21，\u7cbe\u5ea6\u4f18\u5148）<\/span>
    \nindex_params =<\/span> {<\/span>
    \n "index_type"<\/span>:<\/span> "HNSW"<\/span>,<\/span>
    \n "metric_type"<\/span>:<\/span> "COSINE"<\/span>,<\/span>
    \n "params"<\/span>:<\/span> {<\/span>
    \n "M"<\/span>:<\/span> 32<\/span>,<\/span> # \u8282\u70b9\u8fde\u63a5\u6570，\u8d8a\u5927\u7cbe\u5ea6\u8d8a\u9ad8，\u5185\u5b58\u8d8a\u5927<\/span>
    \n "efConstruction"<\/span>:<\/span> 200<\/span>,<\/span> # \u6784\u5efa\u65f6\u641c\u7d22\u5bbd\u5ea6<\/span>
    \n "ef"<\/span>:<\/span> 64<\/span> # \u67e5\u8be2\u65f6\u641c\u7d22\u5bbd\u5ea6（\u5355\u72ec\u8bbe\u7f6e）<\/span>
    \n }<\/span>
    \n}<\/span><\/p>\n

    # IVF_FLAT \u53c2\u6570（\u901a\u7528，\u5747\u8861）<\/span>
    \nindex_params =<\/span> {<\/span>
    \n "index_type"<\/span>:<\/span> "IVF_FLAT"<\/span>,<\/span>
    \n "metric_type"<\/span>:<\/span> "COSINE"<\/span>,<\/span>
    \n "params"<\/span>:<\/span> {<\/span>
    \n "nlist"<\/span>:<\/span> 2048<\/span>,<\/span> # \u805a\u7c7b\u6570 = 4 \u00d7 sqrt(N)，N \u4e3a\u5411\u91cf\u6570<\/span>
    \n "nprobe"<\/span>:<\/span> 16<\/span> # \u67e5\u8be2\u65f6\u641c\u7d22\u7684\u805a\u7c7b\u6570，\u8d8a\u5927\u7cbe\u5ea6\u8d8a\u9ad8<\/span>
    \n }<\/span>
    \n}<\/span><\/p>\n

    \u9ec4\u91d1\u6cd5\u5219：nlist = 4 \u00d7 sqrt(N)，nprobe \u9ed8\u8ba4 16-32\u3002<\/p>\n

    \u56db\u3001\u76d1\u63a7\u4f53\u7cfb<\/h3>\n

    \u5fc5\u987b\u76d1\u63a7\u7684\u6307\u6807<\/h4>\n\n\u5c42\u7ea7\u6307\u6807\u9608\u503c\u62a5\u8b66<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n\n
    Proxy<\/td>\n\u8bf7\u6c42\u5ef6\u8fdf p99<\/td>\n> 100ms<\/td>\n\u4e0a\u6e38\u544a\u8b66<\/td>\n<\/tr>\n
    Proxy<\/td>\nQPS<\/td>\n\u964d\u5e45 > 30%<\/td>\n\u670d\u52a1\u5f02\u5e38<\/td>\n<\/tr>\n
    QueryNode<\/td>\nSegment \u52a0\u8f7d\u6570<\/td>\n\u5355\u8282\u70b9 > 500<\/td>\n\u6570\u636e\u503e\u659c<\/td>\n<\/tr>\n
    QueryNode<\/td>\n\u5185\u5b58\u4f7f\u7528\u7387<\/td>\n> 85%<\/td>\nOOM \u98ce\u9669<\/td>\n<\/tr>\n
    QueryNode<\/td>\n\u7f13\u5b58\u547d\u4e2d\u7387<\/td>\n< 80%<\/td>\n\u7f13\u5b58\u4e0d\u8db3<\/td>\n<\/tr>\n
    DataNode<\/td>\n\u5199\u5165\u5ef6\u8fdf p99<\/td>\n> 500ms<\/td>\nFlush \u963b\u585e<\/td>\n<\/tr>\n
    DataNode<\/td>\nFlush \u961f\u5217\u957f\u5ea6<\/td>\n> 1000<\/td>\n\u80cc\u538b<\/td>\n<\/tr>\n
    IndexNode<\/td>\n\u7d22\u5f15\u6784\u5efa\u961f\u5217<\/td>\n> 10<\/td>\n\u7d22\u5f15\u79ef\u538b<\/td>\n<\/tr>\n
    IndexNode<\/td>\n\u6784\u5efa\u901f\u5ea6<\/td>\n< 10MB\/s<\/td>\nI\/O \u74f6\u9888<\/td>\n<\/tr>\n
    etcd<\/td>\n\u8bf7\u6c42\u5ef6\u8fdf p99<\/td>\n> 100ms<\/td>\n\u5143\u6570\u636e\u6545\u969c<\/td>\n<\/tr>\n
    S3<\/td>\n\u8bf7\u6c42\u5ef6\u8fdf p99<\/td>\n> 500ms<\/td>\n\u5b58\u50a8\u74f6\u9888<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    Prometheus + Grafana \u90e8\u7f72<\/h4>\n

    # \u4f7f\u7528\u5b98\u65b9 Grafana \u4eea\u8868\u76d8<\/span>
    \ncd<\/span> deployments\/monitor\/grafana
    \ndocker-compose<\/span> up -d<\/span><\/p>\n

    Milvus \u66b4\u9732 \/metrics \u7aef\u70b9，Prometheus \u81ea\u52a8\u6293\u53d6\u5373\u53ef\u3002<\/p>\n

    \u4e94\u3001\u9ad8\u9891\u6545\u969c\u4e0e\u6392\u67e5<\/h3>\n

    1. \u7d22\u5f15\u6784\u5efa\u5931\u8d25（ErrorCode 21: BuildIndexError）<\/h4>\n

    \u73b0\u8c61：fail to build index: out of memory<\/p>\n

    \u6839\u56e0：\u5927 Segment \u7d22\u5f15\u6784\u5efa\u65f6\u5185\u5b58\u8d85\u8fc7 IndexNode \u9650\u5236\u3002<\/p>\n

    \u6392\u67e5\u94fe\u8def：<\/p>\n

    # 1. \u67e5\u770b IndexNode \u5185\u5b58<\/span>
    \nkubectl top<\/span> pod -n<\/span> milvus |<\/span> grep<\/span> indexnode<\/p>\n

    # 2. \u67e5\u770b\u5f53\u524d\u6784\u5efa\u4efb\u52a1<\/span>
    \n# \u8fde\u63a5 Milvus \u540e\u6267\u884c<\/span>
    \nmilvus_cli index -l<\/span><\/p>\n

    # 3. \u67e5\u770b Segment \u5927\u5c0f<\/span>
    \n# \u5728 DataCoord \u65e5\u5fd7\u4e2d grep "sealed segment"<\/span><\/p>\n

    \u4fee\u590d：<\/p>\n

      \n
    • \u964d\u4f4e dataCoord.segment.maxSize \u4ece 1024MB \u5230 512MB（\u51cf\u5c0f Segment \u53ef\u964d\u4f4e\u7d22\u5f15\u6784\u5efa\u5cf0\u503c\u5185\u5b58）<\/li>\n
    • \u589e\u5927 IndexNode \u5185\u5b58\u9650\u5236<\/li>\n
    • \u6362\u7528 IVF_FLAT \u66ff\u4ee3 HNSW（HNSW \u6784\u5efa\u65f6\u5185\u5b58\u662f IVF \u7684 3-5 \u500d）<\/li>\n<\/ul>\n

      2. \u67e5\u8be2\u5ef6\u8fdf\u7a81\u7136\u98d9\u5347（p99 > 500ms）<\/h4>\n

      \u6392\u67e5\u51b3\u7b56\u6811：<\/p>\n

      \u67e5\u8be2\u53d8\u6162
      \n\u251c\u2500 \u6240\u6709\u67e5\u8be2\u90fd\u6162 \u2192 Proxy\/QueryCoord \u74f6\u9888
      \n\u2502 \u251c\u2500 \u68c0\u67e5 Proxy \u8fde\u63a5\u6570：netstat -an | grep 19530 | wc -l
      \n\u2502 \u2514\u2500 \u68c0\u67e5 QueryCoord \u9891\u9053\u5206\u914d：\u662f\u5426\u5927\u91cf\u9891\u9053\u96c6\u4e2d\u5728\u5355\u4e2a QueryNode
      \n\u251c\u2500 \u90e8\u5206 Collection \u6162 \u2192 \u6570\u636e\u503e\u659c
      \n\u2502 \u251c\u2500 \u68c0\u67e5 Segment \u5206\u5e03：QueryNode \u4e0a Segment \u6570\u91cf\u662f\u5426\u5747\u8861
      \n\u2502 \u2514\u2500 \u624b\u52a8\u89e6\u53d1 LoadBalance：\u901a\u8fc7 MilvusOperator \u914d\u7f6e
      \n\u2514\u2500 \u7279\u5b9a\u67e5\u8be2\u6162 \u2192 \u7d22\u5f15\u6216\u53c2\u6570\u95ee\u9898
      \n \u251c\u2500 \u68c0\u67e5 nprobe\/ef \u662f\u5426\u5408\u7406
      \n \u2514\u2500 \u68c0\u67e5\u8fc7\u6ee4\u6761\u4ef6\u662f\u5426\u547d\u4e2d\u4e86\u672a\u5efa\u7d22\u5f15\u7684\u6807\u91cf\u5b57\u6bb5<\/p>\n

      \u4fee\u590d：<\/p>\n

        \n
      • \u7ed9\u8fc7\u6ee4\u5b57\u6bb5\u521b\u5efa\u6807\u91cf\u7d22\u5f15<\/li>\n
      • \u589e\u52a0 QueryNode \u526f\u672c\u6570<\/li>\n
      • \u51cf\u5c0f top_k \u503c<\/li>\n
      • \u4e3a Collection \u521b\u5efa\u5206\u533a（Partition），\u7528\u5206\u533a\u952e\u8fc7\u6ee4\u66ff\u4ee3\u5168\u91cf\u626b\u63cf<\/li>\n<\/ul>\n

        3. \u5185\u5b58\u6ea2\u51fa（OOM）<\/h4>\n

        \u53d1\u751f\u6761\u4ef6：<\/p>\n

          \n
        • QueryNode \u540c\u65f6\u52a0\u8f7d\u7684 Segment \u8fc7\u591a<\/li>\n
        • HNSW \u7d22\u5f15\u5185\u5b58\u5360\u7528\u8fc7\u5927<\/li>\n
        • \u5927\u91cf\u5e76\u53d1\u67e5\u8be2\u7684\u4e34\u65f6\u5185\u5b58<\/li>\n<\/ul>\n

          \u4fee\u590d：<\/p>\n

          queryNode<\/span>:<\/span>
          \n memoryLimit<\/span>:<\/span> 48Gi # \u4e0a\u9650<\/span>
          \n cache<\/span>:<\/span>
          \n cacheSize<\/span>:<\/span> 24<\/span> # \u7f13\u5b58\u4e0a\u9650，\u8bbe\u4e3a Pod \u5185\u5b58\u7684 50%-60%<\/span>
          \n warmup<\/span>:<\/span> async
          \n grouping<\/span>:<\/span>
          \n maxNQ<\/span>:<\/span> 1000<\/span> # \u5355\u6b21\u67e5\u8be2\u6700\u5927\u5411\u91cf\u6570，\u9632 OOM<\/span><\/p>\n

          4. Compaction \u79ef\u538b（\u5c0f\u6587\u4ef6\u8fc7\u591a）<\/h4>\n

          \u73b0\u8c61：\/metrics \u4e2d compaction_pending \u6301\u7eed\u589e\u957f，\u67e5\u8be2\u53d8\u6162\u3002<\/p>\n

          \u6839\u56e0：\u9891\u7e41\u5c0f\u6279\u91cf\u63d2\u5165\u4ea7\u751f\u5927\u91cf\u5c0f Segment，Compaction \u5408\u5e76\u901f\u5ea6\u8ddf\u4e0d\u4e0a\u3002<\/p>\n

          \u4fee\u590d：<\/p>\n

          # 1. \u589e\u5927 Compaction \u5e76\u53d1<\/span>
          \n# milvus.yaml<\/span>
          \ndataCoord:
          \n compaction:
          \n maxParallel: 4<\/span> # \u9ed8\u8ba4 3<\/span><\/p>\n

          # 2. \u6279\u91cf\u5199\u5165，\u6bcf\u6b21\u81f3\u5c11 1000 \u6761<\/span>
          \n# 3. \u89e6\u53d1\u624b\u52a8 Compaction<\/span><\/p>\n

          collection.<\/span>compact(<\/span>)<\/span><\/p>\n

          5. TSO \u5ef6\u8fdf（\u5168\u96c6\u7fa4\u5361\u6b7b）<\/h4>\n

          \u73b0\u8c61：\u6240\u6709\u64cd\u4f5c\u8d85\u65f6，\u4f46\u5404\u7ec4\u4ef6 Pod \u5747\u4e3a Running\u3002<\/p>\n

          \u6839\u56e0：RootCoord \u5355\u70b9\u6545\u969c\u6216 etcd \u54cd\u5e94\u6162，\u5bfc\u81f4 TSO \u5206\u914d\u5ef6\u8fdf\u3002<\/p>\n

          \u6392\u67e5：<\/p>\n

          # \u68c0\u67e5 RootCoord \u662f\u5426\u5b58\u6d3b<\/span>
          \nkubectl logs -n<\/span> milvus deployment\/rootcoord –tail<\/span>=<\/span>50<\/span> |<\/span> grep<\/span> -i<\/span> "error\\\\|tso"<\/span><\/p>\n

          # \u68c0\u67e5 etcd \u5ef6\u8fdf<\/span>
          \netcdctl endpoint health<\/p>\n

          \u4fee\u590d：\u786e\u4fdd etcd \u4f7f\u7528 SSD，RootCoord \u4e3a\u9ad8\u53ef\u7528\u90e8\u7f72（Active-Standby \u6a21\u5f0f，Milvus Operator \u81ea\u52a8\u7ba1\u7406）\u3002<\/p>\n

          \u516d\u3001\u5347\u7ea7\u7b56\u7565<\/h3>\n

          2.5 \u2192 2.6 \u6eda\u52a8\u5347\u7ea7（\u96f6\u505c\u673a）<\/h4>\n

          \u4e25\u683c\u6309\u4ee5\u4e0b\u987a\u5e8f\u6267\u884c $TRAE_REF：<\/p>\n

          1. \u5907\u4efd\u6570\u636e（milvus-backup）
          \n2. \u62c9\u65b0 Streaming Node
          \n3. \u8fc1\u79fb Delegator \u5230\u65b0 Streaming Node
          \n4. \u6eda\u52a8\u5347\u7ea7 MixCoord
          \n5. \u6eda\u52a8\u5347\u7ea7 QueryNode
          \n6. \u6eda\u52a8\u5347\u7ea7 DataNode
          \n7. \u6eda\u52a8\u5347\u7ea7 IndexNode
          \n8. \u6eda\u52a8\u5347\u7ea7 Proxy
          \n9. \u9a8c\u8bc1\u6240\u6709\u7ec4\u4ef6\u6b63\u5e38<\/p>\n

          \u5347\u7ea7\u524d\u5fc5\u987b\u5907\u4efd<\/h4>\n

          # \u5b89\u88c5<\/span>
          \npip install<\/span> milvus-backup<\/p>\n

          # \u5168\u91cf\u5907\u4efd<\/span>
          \nmilvus-backup backup \\\\<\/span>
          \n –host<\/span> 127.0<\/span>.0.1 \\\\<\/span>
          \n –collection<\/span> "*"<\/span> \\\\<\/span>
          \n –backup-path \/backup\/20250824<\/p>\n

          \u4e03\u3001\u5b89\u5168\u52a0\u56fa<\/h3>\n

          common<\/span>:<\/span>
          \n security<\/span>:<\/span>
          \n authorizationEnabled<\/span>:<\/span> true<\/span>
          \n defaultRootPassword<\/span>:<\/span> "<\u5f3a\u5bc6\u7801>"<\/span>
          \n tlsMode<\/span>:<\/span> 1<\/span> # \u5355\u5411 TLS<\/span><\/p>\n

          \u521b\u5efa RBAC \u7528\u6237：<\/p>\n

          from<\/span> pymilvus import<\/span> MilvusClient<\/p>\n

          client =<\/span> MilvusClient(<\/span>uri=<\/span>"https:\/\/milvus:19530"<\/span>,<\/span> user=<\/span>"root"<\/span>,<\/span> password=<\/span>"xxx"<\/span>)<\/span><\/p>\n

          # \u521b\u5efa\u53ea\u8bfb\u7528\u6237<\/span>
          \nclient.<\/span>create_user(<\/span>user_name=<\/span>"reader"<\/span>,<\/span> password=<\/span>"xxx"<\/span>)<\/span>
          \nclient.<\/span>grant_role(<\/span>user_name=<\/span>"reader"<\/span>,<\/span> role_name=<\/span>"public"<\/span>)<\/span><\/p>\n

          # \u521b\u5efa\u8bfb\u5199\u7528\u6237<\/span>
          \nclient.<\/span>create_user(<\/span>user_name=<\/span>"writer"<\/span>,<\/span> password=<\/span>"xxx"<\/span>)<\/span>
          \nclient.<\/span>grant_role(<\/span>user_name=<\/span>"writer"<\/span>,<\/span> role_name=<\/span>"public"<\/span>)<\/span><\/p>\n

          \u516b\u3001\u8fd0\u7ef4\u68c0\u67e5\u6e05\u5355<\/h3>\n\n\u9891\u7387\u68c0\u67e5\u9879\u547d\u4ee4\/\u5de5\u5177<\/tr>\n\n\n\n\n\n\n\n\n\n\n\n
          \u6bcf\u65e5<\/td>\n\u5404\u7ec4\u4ef6 Pod \u72b6\u6001<\/td>\nkubectl get pods -n milvus<\/td>\n<\/tr>\n
          \u6bcf\u65e5<\/td>\nQueryNode \u5185\u5b58\u4f7f\u7528<\/td>\nkubectl top pods -n milvus<\/td>\n<\/tr>\n
          \u6bcf\u65e5<\/td>\nCompaction \u961f\u5217<\/td>\nGrafana \u9762\u677f<\/td>\n<\/tr>\n
          \u6bcf\u5468<\/td>\n\u78c1\u76d8\u4f7f\u7528\u7387（S3\/MinIO）<\/td>\nS3 \u63a7\u5236\u53f0<\/td>\n<\/tr>\n
          \u6bcf\u5468<\/td>\n\u7d22\u5f15\u6784\u5efa\u961f\u5217<\/td>\nGrafana \u9762\u677f<\/td>\n<\/tr>\n
          \u6bcf\u6708<\/td>\n\u6162\u67e5\u8be2\u5206\u6790<\/td>\ngrep "slow query" \/var\/logs\/milvus\/<\/td>\n<\/tr>\n
          \u6bcf\u6708<\/td>\netcd \u5feb\u7167\u5907\u4efd<\/td>\netcdctl snapshot save<\/td>\n<\/tr>\n
          \u6bcf\u6708<\/td>\n\u5b8c\u6574\u5907\u4efd<\/td>\nmilvus-backup<\/td>\n<\/tr>\n
          \u5347\u7ea7\u524d<\/td>\n\u7248\u672c\u517c\u5bb9\u6027\u68c0\u67e5<\/td>\n\u5b98\u65b9 release notes<\/td>\n<\/tr>\n
          \u5347\u7ea7\u524d<\/td>\n\u6570\u636e\u5907\u4efd<\/td>\nmilvus-backup<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

          Milvus \u751f\u4ea7\u8fd0\u7ef4\u7684\u6838\u5fc3\u662f\u4e09\u6761\u7ebf**：\u8d44\u6e90\u7ebf（\u5185\u5b58\u548c CPU \u9884\u7559\u8db3\u591f）\u3001\u6570\u636e\u7ebf（Segment \u7ba1\u7406\u3001Compaction\u3001\u7d22\u5f15\u6784\u5efa）\u3001\u76d1\u63a7\u7ebf（p99 \u5ef6\u8fdf\u3001OOM \u9884\u8b66\u3001\u961f\u5217\u79ef\u538b）\u3002\u4e09\u6761\u7ebf\u90fd\u5b88\u4f4f，\u57fa\u672c\u4e0d\u51fa\u5927\u95ee\u9898\u3002**<\/p>\n","protected":false},"excerpt":{"rendered":"

          Milvus \u751f\u4ea7\u73af\u5883\u8fd0\u7ef4\u6df1\u5ea6\u6307\u5357
          \n\u4e00\u3001\u67b6\u6784\u5168\u666f
          \n\u6574\u4e2a\u96c6\u7fa4\u7531\u56db\u5927\u5c42\u7ec4\u6210，\u6bcf\u5c42\u72ec\u7acb\u6269\u5c55：
          \nAccess Layer Access LayerProxy Proxy Proxy \u2192 \u8d1f\u8f7d\u5747\u8861 \u2192 \u5ba2\u6237\u7aef SDK\u2502 \u2502 \u2502
          \nCoordinator LayerRootCoord QueryCoord DataCoord IndexCoord（\u5355 Active）\u2502 <\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[75,1741],"topic":[],"class_list":["post-107376","post","type-post","status-publish","format-standard","hentry","category-server","tag-llm","tag-milvus"],"yoast_head":"\nMilvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/107376.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"Milvus \u751f\u4ea7\u73af\u5883\u8fd0\u7ef4\u6df1\u5ea6\u6307\u5357 \u4e00\u3001\u67b6\u6784\u5168\u666f \u6574\u4e2a\u96c6\u7fa4\u7531\u56db\u5927\u5c42\u7ec4\u6210，\u6bcf\u5c42\u72ec\u7acb\u6269\u5c55： Access Layer Access LayerProxy Proxy Proxy \u2192 \u8d1f\u8f7d\u5747\u8861 \u2192 \u5ba2\u6237\u7aef SDK\u2502 \u2502 \u2502 Coordinator LayerRootCoord QueryCoord DataCoord IndexCoord（\u5355 Active）\u2502\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/107376.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-19T05:40:37+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/107376.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/107376.html\",\"name\":\"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-09-19T05:40:37+00:00\",\"dateModified\":\"2026-09-19T05:40:37+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/107376.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/107376.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/107376.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/107376.html","og_locale":"zh_CN","og_type":"article","og_title":"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"Milvus \u751f\u4ea7\u73af\u5883\u8fd0\u7ef4\u6df1\u5ea6\u6307\u5357 \u4e00\u3001\u67b6\u6784\u5168\u666f \u6574\u4e2a\u96c6\u7fa4\u7531\u56db\u5927\u5c42\u7ec4\u6210，\u6bcf\u5c42\u72ec\u7acb\u6269\u5c55： Access Layer Access LayerProxy Proxy Proxy \u2192 \u8d1f\u8f7d\u5747\u8861 \u2192 \u5ba2\u6237\u7aef SDK\u2502 \u2502 \u2502 Coordinator LayerRootCoord QueryCoord DataCoord IndexCoord（\u5355 Active）\u2502","og_url":"https:\/\/www.wsisp.com\/helps\/107376.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-09-19T05:40:37+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"4 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/107376.html","url":"https:\/\/www.wsisp.com\/helps\/107376.html","name":"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-09-19T05:40:37+00:00","dateModified":"2026-09-19T05:40:37+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/107376.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/107376.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/107376.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"Milvus \u751f\u4ea7\u73af\u5883\u90e8\u7f72\uff0c\u4f18\u5316\uff0c\u65e5\u5e38\u7ef4\u62a4\u4e2d\u9047\u5230\u7684\u95ee\u9898"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/107376","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=107376"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/107376\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=107376"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=107376"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=107376"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=107376"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}