Skip to content
Permalink
Browse files
add more examples to netcat illustration,
showing port inference, parameters, catalog, and more.
  • Loading branch information
ahgittin committed Jan 21, 2016
1 parent ecbd90e commit 411d3f8a3536de6e321dea14d72cdffeb9938310
Showing 11 changed files with 266 additions and 35 deletions.
@@ -24,15 +24,26 @@ The following blueprint shows how a simple script can be embedded in the YAML
{% endhighlight %}

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`.
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.
We can specify a `download.url` which downloads artifacts (unpacking TAR, TGZ, and ZIP archives)


#### 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 `./start.sh` being the default `launch.command`.
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,
@@ -42,17 +53,130 @@ we can instead write our example as:
{% readj example_yaml/vanilla-bash-netcat-file.yaml %}
{% endhighlight %}

The one requirement of the script is that it store the process ID (PID) in the file

#### 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:

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-port.yaml %}
{% endhighlight %}

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`.

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.


#### Declaring New Config Keys

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

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-port-parameter.yaml %}
{% endhighlight %}

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:

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-catalog.bom %}
{% endhighlight %}

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:

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-reference.yaml %}
{% endhighlight %}

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:

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-cluster.yaml %}
{% endhighlight %}

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 `$PID_FILE`, hence the second line of the script.
This is because Brooklyn wants to monitor the services under management.
(There are other options; you can set `checkRunning.command` and `stop.command` instead,
as documented on the Javadoc of the `VanillaSoftwareProcess` class,
and those scripts will be used instead of checking and stopping the process whose PID is in `$PID_FILE`.)
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 {% include java_link.html class_name="VanillaSoftwareProcess" package_path="org/apache/brooklyn/entity/software/base" project_subpath="software/base" %} class,
and those scripts will be used instead of checking and stopping the process whose PID is in `$PID_FILE`.

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-more-commands.yaml %}
{% endhighlight %}

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 has stopped.
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 be to restart the process:
action. A simple recovery here might just to automatically restart the process:

{% highlight yaml %}
{% readj example_yaml/vanilla-bash-netcat-restarter.yaml %}
@@ -67,6 +191,9 @@ This makes it easy to configure various aspects, such as to delay to see if the
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](example_yaml/vanilla-bash-netcat-w-client.yaml).
This uses `initializers` to define an effector to `sayHiNetcat` on the `Simple Pinger` client,
@@ -100,8 +227,11 @@ so that the `$message` we passed above gets logged and reported back:
period: 1s
command: tail -1 server-input

This is still a simple example, but worth going through carefully.
It shows many of the building blocks used in real-world blueprints,

#### 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,
@@ -0,0 +1,35 @@
brooklyn.catalog:
id: netcat-example
version: "1.0"
item:
type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
name: Simple Netcat Server

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

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")
@@ -0,0 +1,11 @@
name: Netcat Cluster Example
location: localhost
services:
- type: org.apache.brooklyn.entity.group.DynamicCluster
memberSpec:
$brooklyn:entitySpec:
type: netcat-example
message: hello from cluster member
netcat.port: 8000+
initialSize: 3
restartMode: parallel
@@ -0,0 +1,16 @@
name: Netcat Example with Explicit Check and Stop Commands
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
name: Simple Netcat Server
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.

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; }
@@ -0,0 +1,21 @@
name: Netcat Example with Parameter
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
name: Simple Netcat Server
launch.command: |
echo $MESSAGE | nc -l $NETCAT_PORT &
echo $! > $PID_FILE
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+
@@ -0,0 +1,13 @@
name: Netcat Example with Port Opened
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
name: Simple Netcat Server
launch.command: |
echo hello | nc -l 4321 &
echo $! > $PID_FILE
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
@@ -0,0 +1,5 @@
name: Netcat Type Reference Example
location: localhost
services:
- type: netcat-example
message: hello from netcat using a registered type
@@ -1,4 +1,4 @@
name: Simple Netcat Example with Restarter Policy
name: Netcat Example with Restarter Policy
location: localhost
services:
- type: org.apache.brooklyn.entity.software.base.VanillaSoftwareProcess
@@ -1,4 +1,4 @@
name: Simple Netcat with Client
name: Netcat Example with Client

location: localhost

@@ -6,13 +6,3 @@ services:
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.

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; }

# can also define download.url, in which case the launch command defaults to ./start.sh in that (archive) file
@@ -67,22 +67,32 @@ the entity being defined, with these being the most common:
* `description`: short text describing the parameter behaviour/usage, presented
to the user
* `type`: the type of the parameter, one of `string`, `integer`, `long`, `float`,
`double`, `timestamp`, `port`, a fully qualified Java type name. Default is `string`.
* `default`: a default value, converted to the type above
* `constraints`: a list of constraints the parameter should meet, currently
`required` is supported

A shorthand notation is also supported where the name of the parameter is directly
passed as an item in the list. For example:
`double`, `timestamp`, `duration`, `port`, or a fully qualified Java type name;
the default is `string`;
obvious coercion is supported so
`timestamp` accepts most common ISO date formats, `duration` accepts `5m`, and port accepts `8080+`
* `default`: a default value; this will be coerced to the declared `type`
* `constraints`: a list of constraints the parameter should meet;
currently `required` is supported, with the default being not required

A shorthand notation is also supported where just the name of the parameter is supplied
as an item in the list, with the other values being unset or the default.
See `displayName` in the following example for an illustration of this:

~~~ yaml
brooklyn.parameters:
- displayName
- name: user.name
constraints:
- required
# user.age parameter is required, and fully specified
- name: user.age
type: integer
label: Age
description: the age of the user
constraints:
- required
# user.name is optional, and has a default
- name: user.name
default: You
# shorthand notation: displayName will be an optional config of type string with no default
- displayName
~~~

Entities, policies, and initializers may accept additional key-value pairs,

0 comments on commit 411d3f8

Please sign in to comment.