Guide to Typical Multi-Host Deployment

This section is all about deploying FindFace Multi in a multi-host environment.

Tip

If, after having read this section, you still have questions, do not hesitate to contact our experts at support@ntechlab.com.

Important

This section doesn’t cover the Video Recorder deployment. You can find a step-by-step instruction on this subject here.

The reasons for deploying FindFace Multi in a multi-host environment are the following:

  • The necessity to distribute the video processing high load.

  • The necessity to process video streams from a group of cameras in the place of their physical location.

    Note

    The most common use cases where such need comes to the fore are hotel/retail chains, several security checkpoints in the same building, etc.

    See also

    Allocate findface-video-worker to Camera Group

  • The necessity to distribute the feature vector extraction high load.

  • Large number of objects to search through, that requires implementation of a distributed object database.

Before you start the deployment, outline your system architecture, depending on its load and allotted resources (see Requirements). The most common distributed scheme is as follows:

  • One principal server with the following components: findface-ntls, findface-multi-legacy, findface-sf-api, findface-video-manager, findface-upload, findface-video-worker, findface-extraction-api, findface-tarantool-server, and third-parties.

  • Several additional video processing servers with installed findface-video-worker.

  • (If needed) Several additional extraction servers with installed findface-extraction-api.

  • (If needed) Additional database servers with multiple Tarantool shards.

This section describes the most common distributed deployment. In high load systems, it may also be necessary to distribute the API processing (findface-sf-api and findface-video-manager) across several additional servers. This procedure requires a high level of expertise and some extra coding. Please do not hesitate to contact our experts for help (support@ntechlab.com).

Important

Installing new FindFace Multi components into a directory with already deployed FindFace Multi components will overwrite the contents of the installation directory and the docker-compose.yaml file. If you need to install a combination of components on the selected server, it is recommended to install all required components at once.

To deploy FindFace Multi in a multi-host environment, follow the steps below:

Deploy Principal Server

To deploy the principal server as part of a distributed architecture, do the following:

  1. On the designated physical server, install FindFace Multi from the installer as follows (don’t forget to prepare a server prior to the FindFace Multi deployment):

    • Product to install: FindFace Multi.

    • Installation type: Single Server. FindFace Multi will be installed and configured to interact with additional remote findface-video-worker instances.

    • Type of the findface-video-worker acceleration (on the principal server): CPU or GPU, subject to your hardware configuration.

    • Type of the findface-extraction-api acceleration (on the principal server): CPU or GPU, subject to your hardware configuration.

    Once the installation is complete, the following output will be shown in the console:

    #############################################################################
#                       Installation is complete                            #
#############################################################################
- all configuration and data is stored in /opt/findface-multi
- upload your license to http://172.168.1.9/#/license/
- user interface: http://172.168.1.9/
  superuser:      admin
  documentation:  http://172.168.1.9/doc/

  2. Upload the FindFace Multi license file via the main web interface http://<Host_IP_address>/#/license. To access the web interface, use the superuser credentials, specified during installation.

    Note

    The host IP address in FindFace Multi web interface URL is the IP address that you have specified as the external address during installation.

    Important

    Do not disclose the superuser (Super Administrator) credentials to others. To administer the system, create a new user with the administrator privileges. Whatever the role, Super Administrator cannot be deprived of its rights.

Deploy Video Processing Servers

On an additional video processing server, install a findface-video-worker instance following the step-by-step instructions. Answer the installer questions as follows:

  • Product to install: FindFace Video Worker.

  • Type of the findface-video-worker acceleration: CPU or GPU, subject to your hardware configuration.

  • The domain name or IP address that is used to access the findface-ntls and findface-video-manager services.

After that, the installation process will automatically begin. The answers will be saved to a file /tmp/<findface-installer-*>.json. Use this file to install FindFace Video Worker on other hosts without having to answer the questions again, by executing:

sudo ./<findface-*>.run -f /tmp/<findface-installer-*>.json

Note

If findface-ntls and/or findface-video-manager are installed on different hosts, specify their IP addresses in the /opt/findface-multi/configs/findface-video-worker/findface-video-worker.yaml configuration file after the installation. Consider findface-ntls is installed on the principal server 172.168.1.9 and findface-video-manager is installed on an additional server 172.168.1.10, then configure the /opt/findface-multi/configs/findface-video-worker/findface-video-worker.yaml file accordingly.

sudo vi /opt/findface-multi/configs/findface-video-worker/findface-video-worker.yaml

In the ntls_addr parameter, specify the findface-ntls host IP address.

ntls_addr: 172.168.1.9:3133

In the mgr → static parameter, specify the findface-video-manager host IP address, which provides findface-video-worker with settings and the video stream list.

static: 172.168.1.10:18811

Deploy Extraction Servers

  1. On an additional extraction server, install a findface-extraction-api instance from the console installer. Answer the installer questions as follows:

    • Product to install: FindFace Multi.

    • Installation type: Fully customized installation. Refer to the Fully Customized Installation section to see the next following installer questions.

    • Once the installer asks you to select FindFace Multi components to install, specify the findface-extraction-api and findface-data components. To make a selection, first, deselect all the listed components by entering -* in the command line, then select findface-extraction-api and findface-data by entering their sequence number (keyword). Enter done to save your selection and proceed to another step.

    • Type of findface-extraction-api acceleration: CPU or GPU.

    • Neural network models to install: CPU or GPU model for an object detection and object attribute recognition. Be sure to choose the right acceleration type for each model, matching the acceleration type of findface-extraction-api: CPU or GPU. Be aware that findface-extraction-api on CPU can work only with CPU-models, while findface-extraction-api on GPU supports both CPU- and GPU-models.

      Tip

      See Neural Networks Summary, Enable Face and Face Attribute Recognition, Enable Vehicle and Vehicle Attribute Recognition, Enable Body and Body Attribute Recognition for details.

    • Detectors and attributes to install: you can install detectors and attributes at once during the installation process or continue with the face detector at default settings. Depending on your choice, the system will ask you additional questions about detectors and attributes to install or pass on to the installation of the neural network models.

    • The system will invite you to edit configuration files. Modify the /opt/findface-multi/configs/findface-extraction-api/findface-extraction-api.yaml configuration file: in the license_ntls_server parameter, specify the IP address of the findface-ntls server.

    After that, the installation process will automatically begin. The answers will be saved to a file /tmp/<findface-installer-*>.json. Use this file to install findface-extraction-api on other hosts without having to answer the questions again.

    sudo ./<findface-*>.run -f /tmp/<findface-installer-*>.json

  2. To move the principal findface-extraction-api instance to another host, in the /opt/findface-multi/configs/findface-sf-api/findface-sf-api.yaml configuration file, specify the IP address of the extraction server host. E.g., if the extraction server host is 172.168.1.11, specify extraction-api: http://172.168.1.11:18666.

    listen: :18411
extraction-api:
  timeouts:
    connect: 5s
    response_header: 30s
    overall: 35s
    idle_connection: 10s
  max-idle-conns-per-host: 20
  keepalive: 24h0m0s
  trace: false
  url: http://127.0.0.1:18666
  extraction-api: http://172.168.1.11:18666

  3. On each extraction server, configure the /opt/findface-multi/docker-compose.yaml file.

    • Specify the ports for the findface-extraction-api service:

      findface-extraction-api:
  command: [--config=/etc/findface-extraction-api.yml]
  depends_on: [findface-ntls]
  image: docker.int.ntl/ntech/universe/extraction-api-cpu:ffserver-12.241211.2
  logging: {driver: journald}
  ports: ['18666:18666']
  networks: [product-network]
  restart: always
  volumes: ['./configs/findface-extraction-api/findface-extraction-api.yaml:/etc/findface-extraction-api.yml:ro',
    './models:/usr/share/findface-data/models:ro', './cache/findface-extraction-api/models:/var/cache/findface/models_cache']

    • Restart all FindFace Multi containers.

      cd /opt/findface-multi/

sudo docker-compose down
sudo docker-compose up -d

Important

Starting the GPU-accelerated findface-extraction-api service for the first time after deployment may take a considerable amount of time due to the caching process (approximately two hours). During this time, object detection in videos and photos, as well as feature vector extraction, will be unavailable.

To make sure that the findface-extraction-api service is up and running, view the service log.

docker logs findface-multi-findface-extraction-api-1 --tail 30 -f

After all the extraction servers are deployed, distribute load across them by using a load balancer.

Distribute Load across Extraction Servers

To distribute load across several extraction servers, you need to set up load balancing. The following step-by-step instructions demonstrate how to set up nginx load balancing in a round-robin fashion for 3 findface-extraction-api instances located on different physical hosts: one on the FindFace Multi principal server (172.168.1.9), and 2 on additional remote servers (172.168.1.10, 172.168.1.11). Should you have more extraction servers in your system, load-balance them by analogy.

Tip

You can use any load balancer according to your preference. Please refer to the relevant official documentation for guidance.

To set up load balancing, do the following:

  1. Designate the FindFace Multi principal server (recommended) or any other server with running findface-sf-api service as a gateway to all the extraction servers.

    Important

    You will have to specify the gateway server IP address when configuring the FindFace Multi network.

  2. On the designated server with the installed findface-sf-api instance, create a nginx folder that contains the extapi.conf file in the /opt/findface-multi/configs/ directory. Make sure that the extapi.conf file includes the information like in the example below. In the upstream directive (upstream extapibackends), substitute the exemplary IP addresses with the actual IP addresses of the extraction servers. In the server directive, specify the gateway server listening port as listen. You will have to enter this port when configuring the FindFace Multi network.

    upstream extapibackends {
        server 172.168.1.9:18666; ## ``findface-extraction-api`` on principal server
        server 172.168.1.10:18666; ## 1st additional extraction server
        server 127.168.1.11:18666; ## 2nd additional extraction server
}
server {
        listen 18667;
        server_name extapi;
        client_max_body_size 64m;
        location / {
                proxy_pass http://extapibackends;
                proxy_next_upstream error;
        }
        access_log /var/log/nginx/extapi.access_log;
        error_log /var/log/nginx/extapi.error_log;
}

  3. Define the Nginx service in the docker-compose.yaml file. To do that, add the container with Nginx image to the docker-compose.yaml file:

    sudo vi /opt/findface-multi/docker-compose.yaml

nginx:
  image: nginx:latest
  ports:
    - 18667:18667
  volumes:
    - ./configs/nginx/extapi.conf:/etc/nginx/conf.d/default.conf:ro

  4. In the findface-sf-api.yaml configuration file, specify the distributor address:

    sudo vi /opt/findface-multi/configs/findface-sf-api/findface-sf-api.yaml

...
extraction-api: http://172.168.1.9:18667

  5. Restart all FindFace Multi containers.

    cd /opt/findface-multi/

sudo docker-compose down
sudo docker-compose up -d

The load balancing is now successfully set up. Be sure to specify the actual gateway server IP address and listening port, when configuring the FindFace Multi network.

Deploy Additional Database Servers

The findface-tarantool-server component connects the Tarantool-based feature vector database and the findface-sf-api component, transferring search results from the database to findface-sf-api for further processing.

To increase search speed, you can allocate several additional servers to the feature vector database and create multiple findface-tarantool-server shards on each additional server. The concurrent functioning of multiple shards will lead to a remarkable increase in performance, as each shard can handle up to approximately 10,000,000 feature vectors.

To deploy additional database servers, do the following:

  1. Install the findface-tarantool-server component on the first designated server. Answer the installer questions as follows:

    • Product to install: FindFace Multi.

    • Installation type: Fully customized installation. Refer to the Fully Customized Installation section to see the next following installer questions.

    • Once the installer asks you to select FindFace Multi components to install, specify the findface-tarantool-server component. To make a selection, first, deselect all the listed components by entering -* in the command line, then select findface-tarantool-server by entering its sequence number (keyword). Enter done to save your selection and proceed to another step.

    • Number of tntapi instances: the default number of tntapi instances is 8. Specify the required number of instances according to your system configuration.

    • Detectors and attributes to install: you can install detectors and attributes at once during the installation process or continue with the face detector at default settings. Depending on your choice, the system will ask you additional questions about detectors and attributes to install or pass on to the installation of the neural network models.

    • Edit configuration files: the system will invite you to edit configuration files. You can agree or skip to the next step.

    After that, the installation process will automatically begin.

    As a result of the installation, the findface-tarantool-server shards will be automatically installed in the amount of N = min(max(min(mem_mb // 2000, cpu_cores), 1), 16 * cpu_cores). I.e., it is equal to the RAM size in MB divided by 2000, or the number of CPU physical cores (but at least one shard), or the number of CPU physical cores multiplied by 16 if the first obtained value is greater.

  2. Use the created /tmp/<findface-installer-*>.json file to install findface-tarantool-server on other servers without answering the questions again. To do so, execute:

    sudo ./<findface-*>.run -f /tmp/<findface-installer-*>.json

  3. Be sure to specify the IP addresses and ports of the shards later on when configuring the FindFace Multi network. To learn the port numbers, execute on each database server:

    sudo cat /opt/findface-multi/docker-compose.yaml | grep -E "CFG_LISTEN_PORT"

    You will get the following result:

    CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,

    The CFG_LISTEN_PORT number is 8101 (the same for all the shards), and is configured for deployment in a bridge network. On the next step, configure the CFG_LISTEN_PORT for each shard.

  4. On the designated server with the installed findface-tarantool-server component, modify the configuration of each shard in the /opt/findface-multi/docker-compose.yaml file. Specify the findface-ntls license server address in the CFG_NTLS parameter. For each shard, except for the first one, add 1 to the external port, e.g., 8101, 8102, 8103, etc.

    sudo vi /opt/findface-multi/docker-compose.yaml

findface-tarantool-server-shard-001:
  depends_on: [findface-ntls]
  environment: {CFG_EXTRA_LUA: loadfile("/tnt_schema.lua")(), CFG_LISTEN_HOST: 0.0.0.0,
    CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
    TT_CHECKPOINT_INTERVAL: '14400', TT_FORCE_RECOVERY: 'true', TT_LISTEN: '0.0.0.0:32001',
    TT_MEMTX_DIR: snapshots, TT_MEMTX_MEMORY: '2147483648', TT_WAL_DIR: xlogs, TT_WORK_DIR: /var/lib/tarantool/FindFace}
  image: docker.int.ntl/ntech/universe/tntapi:ffserver-12.241211.2
  logging: {driver: journald}
  networks: [product-network]
  restart: always
  ports: ['8101:8101']
  volumes: ['./data/findface-tarantool-server/shard-001:/var/lib/tarantool/FindFace',
    './configs/findface-tarantool-server/tnt-schema.lua:/tnt_schema.lua:ro']
findface-tarantool-server-shard-002:
  depends_on: [findface-ntls]
  environment: {CFG_EXTRA_LUA: loadfile("/tnt_schema.lua")(), CFG_LISTEN_HOST: 0.0.0.0,
    CFG_LISTEN_PORT: '8101', CFG_NTLS: 'findface-ntls:3133', TT_CHECKPOINT_COUNT: 3,
    TT_CHECKPOINT_INTERVAL: '14400', TT_FORCE_RECOVERY: 'true', TT_LISTEN: '0.0.0.0:32001',
    TT_MEMTX_DIR: snapshots, TT_MEMTX_MEMORY: '2147483648', TT_WAL_DIR: xlogs, TT_WORK_DIR: /var/lib/tarantool/FindFace}
  image: docker.int.ntl/ntech/universe/tntapi:ffserver-12.241211.2
  logging: {driver: journald}
  networks: [product-network]
  restart: always
  ports: ['8102:8101']
  volumes: ['./data/findface-tarantool-server/shard-002:/var/lib/tarantool/FindFace',
    './configs/findface-tarantool-server/tnt-schema.lua:/tnt_schema.lua:ro']

  5. Restart the containers.

    cd /opt/findface-multi/
sudo docker-compose down
sudo docker-compose up -d

  6. In the /opt/findface-multi/configs/findface-sf-api/findface-sf-api.yaml configuration file, specify shards.

    sudo vi /opt/findface-multi/configs/findface-sf-api/findface-sf-api.yaml

...
  shards:
  - master: http://172.168.1.11:8101/v2/
    slave: ''
  - master: http://172.168.1.11:8102/v2/
    slave: ''

  7. To apply migrations, restart FindFace Multi containers.

    cd /opt/findface-multi/

sudo docker-compose restart

Configure Network

After all the FindFace Multi components are deployed, configure their interaction over the network. Do the following:

  1. Open the /opt/findface-multi/configs/findface-sf-api/findface-sf-api.yaml configuration file:

    sudo vi /opt/findface-multi/configs/findface-sf-api/findface-sf-api.yaml

    Specify the following parameters:

    Parameter

    Description

    extraction-api → extraction-api

    IP address and listening port of the gateway extraction server with set up load balancing.

    storage-api → shards → master

    IP address and port of the findface-tarantool-server master shard. Specify each shard by analogy.

    upload_url

    WebDAV Nginx path to send original images, thumbnails and normalized object images to the findface-upload service.

    		 
    ...
extraction-api:
  extraction-api: http://172.168.1.9:18667

...
webdav:
  upload-url: http://findface-upload:3333/uploads/

...
storage-api:
  ...
  shards:
  - master: http://172.168.1.9:8101/v2/
    slave: ''
  - master: http://172.168.1.9:8102/v2/
    slave: ''
  - master: http://172.168.1.12:8101/v2/
    slave: ''
  - master: http://172.168.1.12:8102/v2/
    slave: ''
  - master: http://172.168.1.13:8102/v2/
    slave: ''
  - master: http://172.168.1.13:8102/v2/
    slave: ''

    Restart the findface-multi-findface-sf-api-1 container.

    sudo docker restart findface-multi-findface-sf-api-1

  2. Open the /opt/findface-multi/configs/findface-multi-legacy/findface-multi-legacy.py configuration file.

    sudo vi /opt/findface-multi/configs/findface-multi-legacy/findface-multi-legacy.py

    Specify the following parameters:

    Parameter

    Description

    SERVICE_EXTERNAL_ADDRESS

    FindFace Multi IP address or URL prioritized for webhooks. Once this parameter not specified, the system uses EXTERNAL_ADDRESS for these purposes. To use webhooks, be sure to specify at least one of those parameters: SERVICE_EXTERNAL_ADDRESS, EXTERNAL_ADDRESS.

    EXTERNAL_ADDRESS

    (Optional) IP address or URL that can be used to access the FindFace Multi web interface. Once this parameter not specified, the system auto-detects it as the external IP address. To access FindFace Multi, you can use both the auto-detected and specified IP addresses.

    VIDEO_DETECTOR_TOKEN

    To authorize the video object detection module, come up with a token and specify it here.

    VIDEO_MANAGER_ADDRESS

    IP address of the findface-video-manager host.

    NTLS_HTTP_URL

    IP address of the findface-ntls host.

    ROUTER_URL

    External IP address of the findface-multi-legacy host that will receive detected objects from the findface-video-worker instance(s).

    SF_API_ADDRESS

    IP address of the findface-sf-api host.

    		 
    sudo vi /opt/findface-multi/configs/findface-multi-legacy/findface-multi-legacy.py

...
SERVICE_EXTERNAL_ADDRESS = 'http://172.168.1.9'
...
EXTERNAL_ADDRESS = 'http://172.168.1.9'


...
FFSECURITY = {
    'VIDEO_DETECTOR_TOKEN': '7ce2679adfc4d74edcf508bea4d67208',
    ...
    'NTLS_HTTP_URL': 'http://findface-ntls:3185',
    ...
    'ROUTER_URL': 'http://172.168.1.9',
    ...
    'VIDEO_MANAGER_ADDRESS': 'http://findface-video-manager:18810',
    'SF_API_ADDRESS': 'http://findface-sf-api:18411',
    ...
}

    Restart all FindFace Multi containers.

    cd /opt/findface-multi/

sudo docker-compose restart

The FindFace Multi components interaction is now set up.

Important

To preserve the FindFace Multi compatibility with the installation environment, we highly recommend you to disable the Ubuntu automatic update. In this case, you will be able to update your OS manually, fully controlling which packages to update.

To disable the Ubuntu automatic update, execute the following commands:

sudo apt-get remove unattended-upgrades
sudo systemctl stop apt-daily.timer
sudo systemctl disable apt-daily.timer
sudo systemctl disable apt-daily.service
sudo systemctl daemon-reload

Important

The FindFace Multi services log a large amount of data, which can eventually lead to disc overload. To prevent this from happening, we advise you to disable rsyslog due to its suboptimal log rotation scheme and use the appropriately configured systemd-journal service instead. See Logging for the step-by-step instructions.