milvus/cpp/conf/server_config.template

99 lines
4.0 KiB
Plaintext

server_config:
address: 0.0.0.0 # milvus server ip address
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://:@:/
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
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
cache_config:
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: # gpu id
- 0
- 1
engine_config:
use_blas_threshold: 20
omp_thread_num: 0 # how many compute threads be used by engine, 0 means use all cpu core to compute
resource_config:
# resource list, length: 0~N
# please set a DISK resource and a CPU resource least, or system will not return query result.
#
# example:
# resource_name: # resource name, just using in connections below
# type: DISK # resource type, optional: DISK/CPU/GPU
# memory: 256 # memory size, unit: GB
# device_id: 0
# enable_loader: true # if is enable loader, optional: true, false
# enable_executor: false # if is enable executor, optional: true, false
resources:
ssda:
type: DISK
memory: 2048
device_id: 0
enable_loader: true
enable_executor: false
cpu:
type: CPU
memory: 64
device_id: 0
enable_loader: true
enable_executor: false
gpu0:
type: GPU
memory: 6
device_id: 0
enable_loader: true
enable_executor: true
gpu_resource_num: 2
pinned_memory: 300
temp_memory: 300
# gtx1660:
# type: GPU
# memory: 6
# device_id: 1
# enable_loader: true
# enable_executor: true
# connection list, length: 0~N
# format: -${resource_name}===${resource_name}
connections:
io:
speed: 500
endpoint: ssda===cpu
pcie:
speed: 11000
endpoint: cpu===gpu0
# - cpu===gtx1660