Custom Entities

So far we’ve covered how to configure and compose entities. There’s a large library of blueprints available, but there are also times when you’ll want to write your own.

For complex use cases, you can write JVM, but for many common situations, some of the highly-configurable blueprints make it easy to write in YAML, including bash and Chef.

Vanilla Software using bash

The following blueprint shows how a simple script can be embedded in the YAML (the | character is special YAML which makes it easier to insert multi-line text):

name: Simple Netcat Server Example
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  name: Simple Netcat Server
  brooklyn.config:
    launch.command: |
      echo hello | nc -l 4321 &
      echo $! > $PID_FILE

This starts a simple nc listener on port 4321 which will respond hello to the first session which connects to it. Test it by running telnet localhost 4321 or opening http://localhost:4321 in a browser.

Note that it only allows you connect once, and after that it fails. This is deliberate! We’ll repair this later in this example. Until then however, in the Applications view you can click the server, go to the Effectors tab, and click restart to bring if back to life.

This is just a simple script, but it shows how any script can be easily embedded here, including a script to download and run other artifacts. Many artifacts are already packaged such that they can be downloaded and launched with a simple script, and VanillaSoftwareProcess can also be used for them.

Downloading Files

We can specify a download.url which downloads an artifact (and automatically unpacking TAR, TGZ, and ZIP archives) before running launch.command relative to where that file is installed (or unpacked), with the default launch.command being ./start.sh.

So if we create a file /tmp/netcat-server.tgz containing just start.sh in the root which consists of the two lines in the previous example, we can instead write our example as:

name: Simple Netcat Example From File
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  name: Simple Netcat Server
  brooklyn.config:
    download.url: file:///tmp/netcat-server.tgz

Port Inferencing

If you’re deploying to a cloud machine, a firewall might block the port 4321. We can tell Brooklyn to open this port explicitly by specifying inboundPorts: [ 4321 ]; however a more idiomatic way is to specify a config ending with .port, such as:

name: Netcat Example with Port Opened
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  name: Simple Netcat Server
    
  brooklyn.config:
    # matching the regex `.*\.port` will cause the port to be opened
    # if in a cloud where configurable security groups are available
    netcat.port: 4321

    launch.command: |
      echo hello | nc -l 4321 &
      echo $! > $PID_FILE

The regex for ports to be opened can be configured using the config inboundPorts.configRegex (which has .*\.port as the default value).

Config keys of type org.apache.brooklyn.api.location.PortRange (aka port) have special behaviour: when configuring, you can use range notation 8000-8100 or 8000+ to tell Brooklyn to find one port matching; this is useful when ports might be in use. In addition, any such config key will be opened, irrespective of whether it matches the inboundPorts.configRegex. To prevent any inferencing of ports to open, you can set the config inboundPorts.autoInfer to false.

Furthermore, the port inferencing capability takes in account static ConfigKey fields that are defined on any Entity sub-class. So, ConfigKey fields that are based on PortRanges type will be also included as required open ports.

Note that in the example above, netcat.port must be specified in a brooklyn.config block. This block can be used to hold any config (including for example the launch.command), but for convenience Brooklyn allows config keys declared on the underlying type to be specified up one level, alongside the type. However config keys which are not declared on the type must be declared in the brooklyn.config block.

Passing custom variables

Blueprint scripts can be parametrised through environment variables, making them reusable in different use-cases. Define the variables in the env block and then reference them using the standard bash notation:

name: Netcat Example with Environment Vars
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  name: Simple Netcat Server

  brooklyn.config:
    launch.command: |
      echo $MESSAGE | nc -l $NETCAT_PORT &
      echo $! > $PID_FILE

    shell.env:
      MESSAGE: hello
      NETCAT_PORT: 4321

Non-string objects in the env map will be serialized to JSON before passing them to the script.

Declaring New Config Keys

We can define config keys to be presented to the user using the brooklyn.parameters block:

name: Netcat Example with Parameter
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  name: Simple Netcat Server
  brooklyn.config:
    launch.command: |
      echo $MESSAGE | nc -l $NETCAT_PORT &
      echo $! > $PID_FILE

    shell.env:
      MESSAGE: $brooklyn:config("message")
      NETCAT_PORT: $brooklyn:attributeWhenReady("netcat.port")
  
  brooklyn.parameters:
  - name: message
    description: a message to send to the caller
    default: hello
  - name: netcat.port
    type: port
    description: the port netcat should run on
    default: 4321+

The example above will allow a user to specify a message to send back and the port where netcat will listen. The metadata on these parameters is available at runtime in the UI and through the API, and is used when populating a catalog.

The example also shows how these values can be passed as environment variables to the launch command. The $brooklyn:config(...) function returns the config value supplied or default. For the type port, an attribute sensor is also created to report the actual port used after port inference, and so the $brooklyn:attributeWhenReady(...) function is used. (If $brooklyn:config("netcat.port") had been used, 4321+ would be passed as NETCAT_PORT.)

This gives us quite a bit more power in writing our blueprint:

  • Multiple instances of the server can be launched simultaneously on the same host, as the 4321+ syntax enables Brooklyn to assign them different ports
  • If this type is added to the catalog, a user can configure the message and the port; we’ll show this in the next section

Using the Catalog and Clustering

The Catalog tab allows you to add blueprints which you can refer to in other blueprints. In that tab, click + then YAML, and enter the following:

brooklyn.catalog:
  id: netcat-example
  version: "1.0"
  itemType: entity
  item:
    type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
    name: Simple Netcat Server

    brooklyn.config:
      launch.command: |
        echo $MESSAGE | nc -l $NETCAT_PORT &
        echo $! > $PID_FILE

      shell.env:
        MESSAGE: $brooklyn:config("message")
        NETCAT_PORT: $brooklyn:attributeWhenReady("netcat.port")
      
    brooklyn.parameters:
    - name: message
      description: a message to send to the caller
      default: hello
    - name: netcat.port
      type: port
      description: the port netcat should run on
      default: 4321+

    brooklyn.enrichers:
    - type: org.apache.brooklyn.enricher.stock.Transformer
      brooklyn.config:
        uniqueTag: main-uri-generator
        enricher.sourceSensor: $brooklyn:sensor("host.address")
        enricher.targetSensor: $brooklyn:sensor("main.uri")
        enricher.targetValue:
          $brooklyn:formatString:
          - "http://%s:%s/"
          - $brooklyn:attributeWhenReady("host.address")
          - $brooklyn:attributeWhenReady("netcat.port")

This is the same example as in the previous section, wrapped according to the catalog YAML requirements, with one new block added defining an enricher. An enricher creates a new sensor from other values; in this case it will create a main.uri sensor by populating a printf-style string "http://%s:%s" with the sensor values.

With this added to the catalog, we can reference the type netcat-example when we deploy an application. Return to the Home or Applications tab, click +, and submit this YAML blueprint:

name: Netcat Type Reference Example
location: localhost
services:
- type: netcat-example
  message: hello from netcat using a registered type

This extends the previous blueprint which we registered in the catalog, meaning that we don’t need to include it each time. Here, we’ve elected to supply our own message, but we’ll use the default port. More importantly, we can package it for others to consume – or take items others have built.

We can go further and use this to deploy a cluster, this time giving a custom port as well as a custom message:

name: Netcat Cluster Example
location: localhost
services:
- type: org.apache.brooklyn.entity.group.DynamicCluster
  brooklyn.config:
    dynamiccluster.memberspec:
      $brooklyn:entitySpec:
        type: netcat-example
        message: hello from cluster member
        netcat.port: 8000+
    cluster.initial.size: 3
    dynamiccluster.restartMode: parallel

In either of the above examples, if you explore the tree in the Applications view and look at the Summary tab of any of the server instances, you’ll now see the URL where netcat is running. But remember, netcat will stop after one run, so you’ll only be able to use each link once before you have to restart it. You can also run restart on the cluster, and if you haven’t yet experimented with resize on the cluster you might want to do that.

Determining Successful Launch

One requirement of the launch script is that it store the process ID (PID) in the file pointed to by $PID_FILE, hence the second line of the script. This is because Brooklyn wants to monitor the services under management. You’ll observe this if you connect to one of the netcat services, as the process exits afterwards and Brooklyn sets the entity to an ON_FIRE state. (You can also test this with a killall nc before connecting or issueing a stop command on the server – but not on the example, as stopping an application root causes it to be removed altogether!)

There are other options for determining launch: you can set checkRunning.command and stop.command instead, as documented on the javadoc and config keys of the VanillaSoftwareProcess (src) class, and those scripts will be used instead of checking and stopping the process whose PID is in $PID_FILE.

name: Netcat Example with Explicit Check and Stop Commands
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  name: Simple Netcat Server
  brooklyn.config:
    launch.command: |
      echo hello | nc -l 4321 &
      echo $! > $PID_FILE

    # The following overrides demonstrate the use of a custom shell environment as well as
    # check-running and stop commands. These are optional; default behavior will "do the
    # right thing" with the pid file automatically.

    shell.env:            { CHECK_MARKER: "checkRunning", STOP_MARKER: "stop" }
    checkRunning.command: echo $CHECK_MARKER >> DATE && test -f "$PID_FILE" && ps -p `cat $PID_FILE` >/dev/null
    stop.command:         echo $STOP_MARKER  >> DATE && test -f "$PID_FILE" && { kill -9 `cat $PID_FILE`; rm /tmp/vanilla.pid; }

And indeed, once you’ve run one telnet to the server, you’ll see that the service has gone “on fire” in Brooklyn – because the nc process stops after one run.

Attaching Policies

Besides detecting this failure, Brooklyn policies can be added to the YAML to take appropriate action. A simple recovery here might just to automatically restart the process:

name: Netcat Example with Restarter Policy
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
  id: netcat-server
  name: Simple Netcat Server
  brooklyn.config:
    launch.command: |
      echo hello | nc -l 4321 &
      echo $! > $PID_FILE
  brooklyn.enrichers:
  - type: org.apache.brooklyn.policy.ha.ServiceFailureDetector
    brooklyn.config:
      # wait 15s after service fails before propagating failure
      serviceFailedStabilizationDelay: 15s
  brooklyn.policies:
  - type: org.apache.brooklyn.policy.ha.ServiceRestarter
    brooklyn.config:
      # repeated failures in a time window can cause the restarter to abort,
      # propagating the failure; a time window of 0 will mean it always restarts!
      failOnRecurringFailuresInThisDuration: 0

Autonomic management in Brooklyn often follows the principle that complex behaviours emerge from composing simple policies. The blueprint above uses one policy to triggering a failure sensor when the service is down, and another responds to such failures by restarting the service. This makes it easy to configure various aspects, such as to delay to see if the service itself recovers (which here we’ve set to 15 seconds) or to bail out on multiple failures within a time window (which again we are not doing). Running with this blueprint, you’ll see that the service shows as on fire for 15s after a telnet, before the policy restarts it.

Sensors and Effectors

For an even more interesting way to test it, look at the blueprint defining a netcat server and client. This uses initializers to define an effector to sayHiNetcat on the Simple Pinger client, using env variables to inject the netcat-server location and parameters to pass in per-effector data:

  env:
    TARGET_HOSTNAME: $brooklyn:entity("netcat-server").attributeWhenReady("host.name")
  brooklyn.initializers:
  - type: org.apache.brooklyn.core.effector.ssh.SshCommandEffector
    brooklyn.config:
      name: sayHiNetcat
      description: Echo a small hello string to the netcat entity
      command: |
        echo $message | nc $TARGET_HOSTNAME 4321
      parameters:
        message:
          description: The string to pass to netcat
          defaultValue: hi netcat

This blueprint also uses initializers to define sensors on the netcat-server entity so that the $message we passed above gets logged and reported back:

  launch.command: |
    echo hello | nc -l 4321 >> server-input &
    echo $! > $PID_FILE
  brooklyn.initializers:
  - type: org.apache.brooklyn.core.sensor.ssh.SshCommandSensor
    brooklyn.config:
      name: output.last
      period: 1s
      command: tail -1 server-input
Windows Command Sensor

Like the blueprint above, the following example also uses brooklyn.initializers to define sensors on the entity, this time however it is a windows VM and uses WinRmCommandSensor.

- type: org.apache.brooklyn.entity.software.base.VanillaWindowsProcess
  brooklyn.config:
    launch.command: echo launching
    checkRunning.command: echo running
  brooklyn.initializers:
  - type: org.apache.brooklyn.core.sensor.windows.WinRmCommandSensor
    brooklyn.config:
      name: ip.config
      period: 60s
      command: hostname

Summary

These examples are relatively simple example, but they illustrate many of the building blocks used in real-world blueprints, and how they can often be easily described and combined in Brooklyn YAML blueprints. Next, if you need to drive off-piste, or you want to write tests against these blueprints, have a look at, for example, VanillaBashNetcatYamlTest.java in the Brooklyn codebase, or follow the other references below.