Register now open for the virtual Milvus Community Conf2020!Join us on Oct.17th, 2020.

Milvus Configuration

Configuration overview

The configurations apply to both single server Milvus and all nodes of a distributed Milvus cluster, for example, from whether or not to turn on the monitoring function to more advanced options for performance tuning.

Milvus file structure

After you have successfully started Milvus server, you can see a Milvus file under the path home/$USER/milvus, which contains the following child files:

  • milvus/db (database storage)
  • milvus/logs (log storage)
  • milvus/conf (configuration file)

    • server_config.yaml (service configuration)
    • log_config.conf (log configuration)

Configurations

Many configurations below are intended for tuning Milvus internals. Before changing these settings, welcome to consult Milvus team on GitHub issues or our Slack channel.

Note: If you have edited these configurations, you must restart Milvus server to apply your changes.

$ docker restart <container id>

In the directory home/$USER/milvus/conf, open Milvus service configuration file server_config.yaml.

Section server_config

Parameter Description Type Default
address IP address that Milvus server monitors. String 0.0.0.0
port Port that Milvus server monitors. Port range: 1025 - 65534. Integer 19530
deploy_mode Milvus deployment type. Options are single , cluster_readonly and cluster_writable. DeployMode single
time_zone Use the UTC-x or UTC+x to specify a time zone. For example, use UTC+8 for China Standard Time. Timezone UTC+8

Section db_config

Parameter Description Type Default
primary_path Primary directory used for both the vector data files you want to import, and the metadata. Path /var/lib/milvus
secondary_path A semicolon-separated list of secondary directories used only for the vector data files imported into Milvus. Set this parameter when the data size is too much to fit in the primary_path.
Each file, whether in primary_path or secondary_path, is assigned an equal part of the imported data. Data Size per Directory = Total Data Size / Number of Directories. So make sure the available storage space in these files are enough.
Path
backend_url URL for metadata storage. Use SQLite (for single server Milvus) or MySQL (for distributed cluster) to store the metadata.
The format of dbbackendurl is: dialect://username:password@host:port/database. (dialect can be either mysql or sqlite, depending on which database you use.
Path sqlite://:@:/
insert_buffer_size Maximum buffer size allowed for data insertion. The sum of insert_buffer_size and cpu_cache_capacity (in "Section cache_config" ) should be < total memory. Integer 1 (GB)
preload_table Define if to preload tables into memory after Milvus server restart. Tables can be selected for fully or partially preloading.
To preload all the existing tables, use * ; To preload some tables, list the specific table names, separated by comma. If you choose not to preload any table, keep it empty ( ).
PreloadType

Section metric_config

Parameter Description Type Default
enable_monitor Set to true to enable the monitoring function. Boolean true
collector Connected monitoring system to collect metrics. String Prometheus
port Port to visit Prometheus. Integer 8080

Section cache_config

Parameter Description Type Default
cpu_cache_capacity The size of the CPU memory for caching data for faster query. The sum of cpu_cache_capacity and insert_buffer_size (in "Section db_config) must be less than total CPU memory size. Integer 4 (GB)
cache_insert_data If set to true , the inserted data will be loaded into the cache immediately for hot query.
If you want simultaneous inserting and searching of vector, it is recommended to enable this function.
Boolean false

Section engine_config

Parameter Description Type Default
use_blas_threshold A Milvus performance tuning parameter. The threshold value must be compared with nq to decide if the usage of OpenBLAS library will be triggered.
If nq >= use_blas_threshold , OpenBLAS will be used. The search response times do not fluctuate, but the search speed is relatively slow.
If nq < use_blas_threshold , SSE will be used. The search speed will be enhanced, however with slight fluctuation of search response times. The value should be >= 0.
Integer 20
gpu_search_threshold A Milvus performance tuning parameter. The threshold value must be compared with nq to decide if the search computation will be executed on GPUs only.
If nq >= gpu_search_threshold , the search computation will be executed on GPUs only.
If nq < gpu_search_threshold , the search computation will be executed on both CPUs and GPUs.
Interger 1000

Section gpu_resource_config

Decide in this section whether to enable GPU support/usage in Milvus. GPU support, which uses both CPU and GPUs for optimized resource utilization, is by default enabled to achieve accelerated search performance on very large datasets.

To switch to CPU-only mode, just set enable to false.

Note: In Milvus, the index building and search computation are separate processes, which can be executed on cpu, or gpu or both. You are allowed to assign multiple GPUs to run both processes.

Parameter Description Type Default
enable Decide if to enable GPU usage in Milvus. Boolean true
cache_capacity The size of the GPU memory for caching data for faster query. The size must be less than the total GPU memory size. Integer 1 (GB)
search_resources Define the GPU devices used for search computation in Milvus. Must be in format: gpux. ResourceType gpu0
build_index_resources Define the GPU devices used for index building in Milvus. Must be in format: gpux. ResourceType gpu0
Edit
© 2019 - 2020 Milvus. All rights reserved.