Agents

Agents are additional components of the Connectware that can be deployed and started individually, usually somewhere outside of the central installation of the Connectware.

Agents are the solution to the requirement for “computing on the edge”. They allow data-processing at a very early stage of the pipeline and will secure non-secure hardware communication protocols right at the machine. Additionally, agents can be used to partition workloads of protocol translations, preprocessing, or data forwarding across multiple container instances.

For agents to work only a single pre-condition must be met: agents must be in a network that allows them to connect to the (central) Connectware instance.

Note

Agents solely use MQTT for communication and hence act as network clients. They do not require any inbound traffic and no special network configuration for inbound connectivity is needed.

This page describes how to run individual agent instances outside of the central Connectware, although any of those deployment possibilities can also be applied inside the central Connetware as well. Both is possible. The following documentation will focus on the docker-based agent setup.

Installation

To run an agent, one single Docker container has to be started. The container image to be used is identical to the Connectware’s protocol-mapper image. The image tag (i.e. the version) of that container must be the exact same version as the Connectware, otherwise the agent cannot start to communicate with the Connectware.. In the examples below the image tag (version) is stored as IMAGE_TAG= 1.1.0

Orchestration

Starting the container is accomplished by the chosen orchestration tool. The container can be started either by executing a docker run command, or by using docker-compose with a corresponding docker-compose.yml file, or by creating a deployment in a Kubernetes cluster, see examples below.

In either case the agent must be configured through environment variables made available to the container.

The following environment variables are required:

  • CYBUS_AGENT_MODE=distributed (switches protocol-mapper to run in agent mode)

  • CYBUS_AGENT_NAME=<agentName> (unique name identifying this agent instance)

  • CYBUS_MQTT_HOST=<connectwareHost> (the IP/hostname of the Connectware)

The following environment variables are optional:

  • CYBUS_MQTT_USERNAME=<userName> Username for authentication in the Connectware (default is <agentName>)

  • CYBUS_MQTT_SCHEME=<scheme> (mqtts or mqtt, default: mqtt)

  • CYBUS_MQTT_PORT=<port> (default: 1883)

Additionally, it is recommended to persist the data (credentials) of the agent, guaranteeing unsupervised and automatic availability, even after power cycles. To achieve this persistency, a docker volume mount must be added for the container. The volume must be mounted to the Docker–internal /data directory. The agent will automatically store the credentials on that volume and re-use them on subsequent start-up.

Note: If multiple agents will be run on the same computer, each agent must use its own volume mount, as otherwise agents would overwrite each other’s data, including username and password credentials.

As an alternative to locally generated credentials, it is also possible to specify a password for this protocol-mapper agent directly by an additional environment variable CYBUS_PROTOCOL_MAPPER_PASSWORD. This might be useful in scenarios where the agent deployment is controlled by the same orchestration tool as the central Connectware instance. In most other cases it is recommended to let the agent generate local credentials, stored in a volume, and authorized to the Connectware through the client registration workflow.

It may also be useful to set the hostname property of the docker container to the local host’s hostname, because this name will be displayed in the Admin-UI overview page. If it is not set, the local docker container ID will be displayed instead (e.g. d172c8c3667b), which might look rather confusing in this overview page. Depending on the operating system, the suitable value for the hostname may be available as an environment variable ${HOSTNAME} or it must be specified manually, as noted in the example below.

In case of mutual TLS, first you need to enable it for the Broker:

  • CYBUS_BROKER_USE_MUTUAL_TLS="yes" (default: no)

Some additional configuration is necessary on the agent:

  • CYBUS_MQTT_SCHEME should be set to mqtts

  • CYBUS_MQTT_PORT should target the appropriate port of the Broker, which listens for mqtts messages on 8883 by default

  • USE_MUTUAL_TLS=true (default: false)

  • You should mount your certificates as regular volumes. By default Connectware will look for the certifcates specified in these environment variables:
    • AGENT_KEY=<key> (default: /connectware/certs/client/tls.key)

    • AGENT_CERT=<cert> (default: /connectware/certs/client/tls.crt)

    • CA=<caChain> (default: /connectware/certs/ca/ca-chain.pem)

Example 1: docker-compose

version: '2.0'
services:
  protocol-mapper-agent:
    image: registry.cybus.io/cybus/protocol-mapper:${IMAGE_TAG}
    environment:
      CYBUS_AGENT_MODE: distributed
      CYBUS_AGENT_NAME: myAgent
      CYBUS_MQTT_HOST: 10.11.12.13
    volumes:
      - protocol-mapper-agent:/data
    restart: unless-stopped
    network_mode: host
    hostname: <some-suitable-hostname>
volumes:
  protocol-mapper-agent:

To start, create a new directory (e.g. myAgent). Inside, create a new file called docker-compose.yml with the above content and finally run:

docker-compose up -d

Again it is noted that if multiple agents should run on the same computer, each agent must have defined its own docker volume, otherwise different agents would overwrite each other’s persisted data, including username and password credentials.

Example 2: docker run

Using the environment variable IMAGE_TAG= 1.1.0, the following command will run the agent directly (but with the comment lines removed):

# The docker run command
docker run \
    # Make it run as an agent
  -e CYBUS_AGENT_MODE=distributed \
    # Set the name of the agent. This must match the agentName used in the commissioning file.
  -e CYBUS_AGENT_NAME=myAgent \
    # The IP address where the Connectware is running.
  -e CYBUS_MQTT_HOST=10.11.12.13 \
    # Attach a volume so the agent can store data across restarts
  -v /tmp/cybus:/data \
    # Upon failures, restart the container
  --restart unless-stopped \
    # Run it in host mode networking
  --net=host \
    # Some suitable hostname of the container.
  --hostname=${HOSTNAME} \
    # The image to use
  registry.cybus.io/cybus/protocol-mapper:${IMAGE_TAG}

This will start an agent named myAgent that persists data to /tmp/cybus on the host OS.

Example 3: Kubernetes

By using the following yaml manifest file, you can run an agent on a Kubernetes cluster. You only need to configure the same parameters used by the others types of deployments.

Additionally you will need to create a Kubernetes secret to store the Cybus registry credentials needed to fetch the Connectware docker images.

You can create a secrets like this, using the secret name cybus-docker-registry:

kubectl create secret docker-registry cybus-docker-registry \
        --docker-server=registry.cybus.io \
        --docker-username=john.doe \
        --docker-password=abcd123 \
        --docker-email=john.doe@example.org \
        --namespace=default

The username and password can either be your Cybus Portal username and password, or when using the Connectware license key, the username is license and the password is your Connectware license key.

The manifest file can be created from this example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
apiVersion: apps/v1
kind: Deployment
metadata:
  # The id for this kubernetes object
  name: connectware-agent-deployment
  labels:
    # Attach labels for easy group selection
    app: protocol-mapper-agent
spec:
  replicas: 1
  selector:
    matchLabels:
      app: protocol-mapper-agent
  template:
    metadata:
      labels:
        app: protocol-mapper-agent
    spec:
      # Configure volumes used by the agent to store data across restarts
      volumes:
      - name: agent-data-volume
        emptyDir: {}
      containers:
        - name: protocol-mapper-agent
          # Container image. Watch out: Must use identical version tag to current Connectware version!
          image: registry.cybus.io/cybus/protocol-mapper:1.0.32
          env:
            # Make the protocol-mapper runs as an agent
          - name: CYBUS_AGENT_MODE
            value: "distributed"
            # Set the name of the agent. This must match the agentName used in the commissioning file.
          - name: CYBUS_AGENT_NAME
            value: "myAgent"
            # The IP address or hostname where the Connectware is running
          - name: CYBUS_MQTT_HOST
            value: "10.11.12.13"
          ports:
          # Expose the ports used by the agent.
          - containerPort: 443
          # Configure mount points
          volumeMounts:
          - mountPath: /data
            name: agent-data-volume
      # Kubernetes secret used to pull the images from the Cybus registry
      imagePullSecrets:
        - name: cybus-docker-registry

Manifest file for download: agent-deployment.yaml

Example 4: Mutual TLS Agent

version: '2.0'
services:
  protocol-mapper-agent:
    image: registry.cybus.io/cybus/protocol-mapper:${IMAGE_TAG}
    environment:
      CYBUS_AGENT_MODE: distributed
      CYBUS_AGENT_NAME: myAgent
      CYBUS_MQTT_SCHEME: mqtts
      CYBUS_MQTT_HOST: 10.11.12.13
      CYBUS_MQTT_PORT: '8883'
      USE_MUTUAL_TLS: true
    volumes:
      - protocol-mapper-agent:/data
      - /mycerts:/connectware/certs/client
      - /myca:/connectware/certs/ca
    restart: unless-stopped
    network_mode: host
    hostname: <some-suitable-hostname>
volumes:
  protocol-mapper-agent:

Network requirements

For the configured agent container, no incoming TCP/IP port permissions are needed. All communication will be handled by outgoing connections.

In order for the agent container to reach the central Connectware, the following incoming TCP/IP ports are needed for the central Connectware. In other words, the agent must be able to initiate connections to the central Connectware via the following ports:

  • 443/tcp (https), and

  • 1883/tcp (mqtt) or 8883/tcp (mqtts), whichever is chosen in CYBUS_MQTT_SCHEME

Registration

Registration describes the process of authorizing an agent to connect to the Connectware.

In the recommended setup where the agent has persistency enabled (see above), this step is needed only once upon initial deployment. To be precise, this registration is needed only in one of these cases:

  • the agent is started the very first time, or

  • the persisted credentials got deleted, or

  • the agent could not login to the Connectware with its cached credentials.

In all other cases the registration process is not needed and the agent will seamlessly start working with the persisted credential from the storage volume. (Again note: If multiple agents should run on the same computer, each agent must use its own volume mount, as otherwise agents would overwrite each other’s credentials.)

In the initial deployment, the registration is performed as follows: First the agent container must be started. Then, the console log output of the container should look as follows:

{"level":30,"time":1590665714502,"msg":"Starting in distributed (edge) mode","pid":6,"hostname":"someHost","v":1}
{"level":30,"time":1590665714568,"msg":"Asking for Connectware registration at 1.2.3.4 with newly generated credentials...","pid":6,"hostname":"someHost","v":1}
{"level":30,"time":1590665714568,"msg":"The registration process must be finished using the web-interface.","pid":6,"hostname":"someHost","v":1}
{"level":30,"time":1590665714568,"msg":"Navigate to \"User Management -> Client Registry\", click unlock and authorize this client.","pid":6,"hostname":"someHost","v":1}
{"level":30,"time":1590665714568,"msg":"Username: cybus-sample-agent, pairing ID (should match in UI): a82123","pid":6,"hostname":"someHost","v":1}

If there is instead a message mentioning 409 Conflict, there is a registration conflict existing that needs to be resolved, see User Registration Conflict below.

To finalize the registration process, log into the Connectware Admin-UI and authorize the client as described in the chapter Client Registry.

Note

In addition to the agent name, you can also compare the Pairing ID to verify that you are granting access to the correct agent instance.

Usage

Agents are able to provide three resource types:

The endpoint resource can also use the rules property to enable direct data-manipulation and pre-processing on the edge using the rule engine (edge processing).

In order to start such resources on the agent, you must set the agentName property at each Connection in your commissioning file. All Endpoint resources using that connection will automatically also run on the the same agent as well without setting any further properties.

Important

The value of the agentName property (as available for Cybus::Connection resources) must match exactly the value of the CYBUS_AGENT_NAME environment variable used during agent startup.

You can find an example commissioning file here.

Troubleshooting

As for docker logs (also see below): The logs can not be accessed from within the Connectware. Instead, the logs must be collected and made accessible by the orchestration tool.

Connection loss

If there is a connection loss between agent and central Connectware, the data connectivity is broken and messages will be lost. In particular, due to the resource management system, any connection from the agent to a machine will stop receiving data as long as the connection to the central Connectware is broken.

Nevertheless, short-term temporary connection outages (such as less than a few seconds) can be covered by using the MQTT QoS (quality of service) level. This QoS level must be set on a per-endpoint basis, because it must be set for each published package and for each subscription individually. To set this for the MQTT connection from the agent to the central Connectware, set the qos property of each involved Endpoint to the value 1 (one) instead of zero. (QoS level 2 is most likely not useful in the industry context and is not recommended here.) Setting a non-zero QoS level will ensure that no in-flight MQTT message is getting lost if there is a temporary network outage.

However, as soon as the MQTT connection loss between agent and Connectware gets noticed by the agent, it will stop receiving data from the machine, and the Connectware in turn will mark the respective resources as deviated.

If you are interested in how to set up different solutions for a distributed data collection system that could also tolerate temporary network fragmentation, please contact our support team at support@cybus.io.

User Registration Conflict

If the user registration as described above is not accepted by the Connectware but lets the agent print a log message 409 Conflict, most likely there is already an existing user with the given agent name in the Connectware, but with a different password. The easiest solution is to go to the “User Management” page in the Admin-UI and delete that particular user, see Delete User.

The full message at the agent’s log in such cases is the following:

{"level":30,"time":1590665714502,"msg":"Failed to register agent. Response: 409 Conflict. A conflicting registration might be pending, or a user with the same username myAgent is already existing (which you must delete first)."}

Monitoring

Status

The connectivity status of the agents can be monitored by navigating to System > Agents in the UI.

../_images/agents.png

In the table of currently known agents, for each agent the current status (online / offline) is shown, as well as the version number of this particular agent container, and the hostname as configured when starting the agent container.

Agents in offline status can be removed from this list by pressing the Delete icon in the upper right corner after selecting that row in the table. Note: This merely removes the agent from the list, but it does not remove a registered user from the user database! To remove a registered agent user from the user database, go to the “User Management” page in the Admin-UI and delete that particular user, see Delete User.

Docker Logs

The docker logs (or: debug output) of the agent containers can not be monitored from within the Connectware. Instead, those logs need to be monitored by the chosen orchestration tool (above) that is also responsible for the container deployment.