server_config: address: 0.0.0.0 port: 19530 # the port milvus listen to, default: 19530, range: 1025 ~ 65534 gpu_index: 0 # the gpu milvus use, default: 0, range: 0 ~ gpu number - 1 mode: single # milvus deployment type: single, cluster, read_only db_config: db_path: @MILVUS_DB_PATH@ # milvus data storage path db_slave_path: # secondry data storage path, split by semicolon parallel_reduce: false # use multi-threads to reduce topk result # URI format: dialect://username:password@host:port/database # All parts except dialect are optional, but you MUST include the delimiters # Currently dialect supports mysql or sqlite db_backend_url: sqlite://:@:/ index_building_threshold: 1024 # index building trigger threshold, default: 1024, unit: MB archive_disk_threshold: 0 # triger archive action if storage size exceed this value, 0 means no limit, unit: GB archive_days_threshold: 0 # files older than x days will be archived, 0 means no limit, unit: day insert_buffer_size: 4 # maximum insert buffer size allowed, default: 4, unit: GB, should be at least 1 GB. # the sum of insert_buffer_size and cpu_cache_capacity should be less than total memory, unit: GB metric_config: is_startup: off # if monitoring start: on, off collector: prometheus # metrics collector: prometheus prometheus_config: # following are prometheus configure collect_type: pull # prometheus collect data method port: 8080 # the port prometheus use to fetch metrics push_gateway_ip_address: 127.0.0.1 # push method configure: push gateway ip address push_gateway_port: 9091 # push method configure: push gateway port license_config: # license configure license_path: "@MILVUS_DB_PATH@/system.license" # license file path cache_config: # cache configure cpu_cache_capacity: 16 # how many memory are used as cache, unit: GB, range: 0 ~ less than total memory cpu_cache_free_percent: 0.85 # old data will be erased from cache when cache is full, this value specify how much memory should be kept, range: greater than zero ~ 1.0 insert_cache_immediately: false # insert data will be load into cache immediately for hot query gpu_cache_capacity: 5 # how many memory are used as cache in gpu, unit: GB, RANGE: 0 ~ less than total memory gpu_cache_free_percent: 0.85 # old data will be erased from cache when cache is full, this value specify how much memory should be kept, range: greater than zero ~ 1.0 gpu_ids: 0,1 # gpu id engine_config: nprobe: 10 nlist: 16384 use_blas_threshold: 20 metric_type: L2 # compare vectors by euclidean distance(L2) or inner product(IP), optional: L2 or IP omp_thread_num: 0 # how many compute threads be used by engine, 0 means use all cpu core to compute