pnp 0.23.0

Pull 'n' Push

Pull 'n' Push

Pulls data from sources and pushes it to sinks with optional transformations in between.

1. Installation
2. Getting started
3. Console runner
4. Building Blocks
4.1. Pull
4.2. Push
4.3. Selector
4.4. Dependencies
4.5. Envelope (0.7.1+)
4.6. Payload unwrapping
5. Useful hints
5.1. Configuration checking
5.2. Logging (0.11.0+)
5.3. dictmentor (0.11.0+)
5.4. Advanced selector expressions (0.12.0+)
5.5. UDF Throttle (0.15.0+)
5.6. Docker images
6. Plugins

1. Installation

pip install pnp

Installation with extras:

pip install pnp[fswatcher,faceR]

Please consult the component documentation to see if a component requires an extra or not.

2. Getting started

Define pulls to fetch / pull data from source systems. Define one push or multiple pushes per pull to transfer the pulled data anywhere else (you only need a plugin that knows how to handle the target). You configure your pipeline in yaml:

- name: hello-world
  pull:
    plugin: pnp.plugins.pull.simple.Repeat
    args:
      interval: 1s
      repeat: "Hello World"
  push:
    - plugin: pnp.plugins.push.simple.Echo

Copy this configuration and create the file helloworld.yaml. Run it:

pnp helloworld.yaml

This example yields the string 'Hello World' every second.

Tip: You can validate your config without actually executing it with

   pnp --check helloworld.yaml

3. Console runner

> pnp --help
Pull 'n' Push

Usage:
  pnp [(--engine=<engine>)] [(-v | --verbose)] [--log=<log_conf>] <configuration>
  pnp (-c | --check) <configuration>
  pnp (-h | --help)
  pnp --version

Options:
  -c --check        Only check configuration and do not run it.
  --engine=<engine> Override engine from configuration file (thread, process, sequential, async).
  -v --verbose      Switches log level to debug.
  --log=<log_conf>  Specify logging configuration to load.
  -h --help         Show this screen.
  --version         Show version.

4. Building Blocks

Below the basic building blocks of pull 'n' push are explained in more detail

4.1. Pull

As stated before pulls fetch data from various source systems and/or apis. Please see the section plugins for shipped implemented pulls. To instantiate a pull by configuration file you only have to provide it's fully qualified name and the argument that should be passed.

- name: example
  pull:
    plugin: pnp.plugins.pull.mqtt.Subscribe
    args:
      host: localhost
      port: 1883
      topic: test/#

The above snippet will create a pull that listens on the topic test/# on a mqtt broker. The output of the pull is a dictionary that contains the topic, levels and the actual payload.

# When the message 'Here i am' arrives on the topic 'test/foo/bar' then the output will look like that:
{'topic': 'test/foo/bar', 'levels': ['test', 'foo', 'bar'], 'payload': 'Here i am'}

4.2. Push

A pull passes its data to multiple pushes to transfer/transform the data. For example a push might save sensor data to influx or dump a file to the file system.

- name: example
  pull:
    plugin: pnp.plugins.pull.mqtt.Subscribe
    args:
      host: localhost
      port: 1883
      topic: test/#
  push:
    - plugin: pnp.plugins.push.fs.FileDump
      args:
        directory: "/tmp"
        binary_mode: false
    - plugin: pnp.plugins.push.simple.Echo

The above snippet adds two pushes to the already known pull. The first push takes the incoming data and dumps it into the specified directory as a textfile. The second push just prints out the incoming data.

4.3. Selector

Sometimes the output of a pull needs to be transformed before the specified push can handle it. Selectors to the rescue. Given our input we decide to just dump the payload and print out the first level of the topic.

- name: example
  pull:
    plugin: pnp.plugins.pull.mqtt.Subscribe
    args:
      host: localhost
      port: 1883
      topic: test/#
  push:
    - plugin: pnp.plugins.push.fs.FileDump
      selector: data.payload
      args:
        directory: "/tmp"
        binary_mode: false
    - plugin: pnp.plugins.push.simple.Echo
      selector: data.levels[0]

Easy as that. We can reference our incoming data via data or payload.

4.4. Dependencies

By default any pushes will execute in parallel (not completely true) when new incoming data is available. But now it would be nice if we could chain pushes together. So that the output of one push becomes the input of the next push. The good thing is: Yes we can.

Back to our example let's assume we want to print out the path to the created file dump after the dump is created.

- name: example
  pull:
    plugin: pnp.plugins.pull.mqtt.Subscribe
    args:
      host: localhost
      port: 1883
      topic: test/#
  push:
    - plugin: pnp.plugins.push.fs.FileDump
      selector: data.payload
      args:
        directory: "/tmp"
        binary_mode: false
      deps:
        - plugin: pnp.plugins.push.simple.Echo
    - plugin: pnp.plugins.push.simple.Echo
      selector: data.levels[0]

As you can see we just add a dependant push to the previous one.

4.5. Envelope (0.7.1+)

Using envelopes it is possible to change the behaviour of pushes during runtime. Best examples are the pnp.plugins.push.fs.FileDump and pnp.plugins.push.mqtt.Publish plugins, where you can override the actual file_name and extension of the file to dump resp. the topic where the message should be published.

Easier to understand by a practical example:

- name: envelope
  pull:
    plugin: pnp.plugins.pull.simple.Count
    args:
      interval: 1s
  push:
    plugin: pnp.plugins.push.fs.FileDump
    selector:
      file_name: "lambda data: str(data)"
      extension: ".cnt"
      data: "lambda data: data"
    args:
      directory: "/tmp/counter"
      file_name: "counter"  # Overridden by envelope
      extension: ".txt"  #  Overridden by envelope
      binary_mode: false  # text mode

The push dumps multiple files (0.cnt, 1.cnt, 2.cnt, ...) for each counter value emitted by the pull, instead of dumping one file 'counter.txt' which is overridden each time a new counter is emitted.

How does this work: If the emitted or transformed payload (via selector) contains the key data or payload the pipeline assumes that this is the actual payload and all other keys represent the so called envelope.

4.6. Payload unwrapping

By default any payload that is provided to a push will be "as-is". If the payload is an iterable, it is possible to unwrap each individual item of the iterable and provide that single item to the push instead of the whole list. Yes, now you can perform for each loops with pushes.

- name: unwrapping
  pull:
    plugin: pnp.plugins.pull.simple.Repeat
    args:
      interval: 1s
      repeat:
        - 1
        - 2
        - 3
  push:
    - plugin: pnp.plugins.push.simple.Echo
      unwrap: true

Hint: Selector expressions are applied after unwrapping. So the selector is applied to each individual item. If you need the selector to augment your list, use a push.simple.Nop with unwrap = False and a dependent push.

- name: unwrapping
  pull:
    plugin: pnp.plugins.pull.simple.Repeat
    args:
      interval: 1s
      repeat:
        - 1
        - 2
        - 3
  push:
    - plugin: pnp.plugins.push.simple.Nop
      selector: "data + [4, 5, 6]"
      unwrap: false  # Which is the default
      deps:
        - plugin: pnp.plugins.push.simple.Echo
          unwrap: true

5. Useful hints

5.1. Configuration checking

You can check your pnp configuration file by starting pnp with the -c | --check flag set. This will only run the initializer but not execute the configuration.

pnp --check <pnp_configuration>

5.2. Logging (0.11.0+)

You can use different logging configurations in two ways:

# Specify when starting pnp
pnp --log=<logging_configuration> <pnp_configuration>
# Specify by environment variable
export PNP_LOG_CONF=<logging_configuration>
pnp <pnp_configuration>

A simple logging configuration that will log severe errors to a separate rotating log file looks like this:

version: 1
disable_existing_loggers: False

formatters:
    simple:
        format: "%(asctime)s - %(name)s - %(levelname)s - %(message)s"

handlers:
    console:
        class: logging.StreamHandler
        level: DEBUG
        formatter: simple
        stream: ext://sys.stdout

    error_file_handler:
        class: logging.handlers.RotatingFileHandler
        level: ERROR
        formatter: simple
        filename: errors.log
        maxBytes: 10485760 # 10MB
        backupCount: 20
        encoding: utf8

root:
    level: INFO
    handlers: [console, error_file_handler]

A simple slack logging confiuration that will log warnings and errors to a slack channel looks like this;

version: 1
disable_existing_loggers: False

formatters:
    simple:
        format: "%(asctime)s - %(name)s - %(levelname)s - %(message)s"

handlers:
    console:
        class: logging.StreamHandler
        formatter: simple
        stream: ext://sys.stdout

    slack:
        level: WARNING  # Do not use DEBUG - This will result in a recursion (cause slacker is using urllib which uses logging)
        api_key: '<your_api_key>'  # Retrieve from api.slack.com
        class: pnp.logging.SlackHandler  # Do not change
        channel: '#alerts'  # The channel to use
        ping_level: ERROR  # Ping users when the message has this severity
        ping_users:  # Ping these users (can be real name, display name, internal name, ...)
          - dede

root:
    level: INFO
    handlers:
        - slack
        - console

5.3. dictmentor (0.11.0+)

You can augment the configuration by extensions from the dictmentor package. Please see DictMentor for further reference.

The DictMentor instance will be instantiated with the following code and thus the following extensions:

from dictmentor import DictMentor, ext
return DictMentor(
    ext.Environment(fail_on_unset=True),
    ext.ExternalResource(base_path=os.path.dirname(config_path)),
    ext.ExternalYamlResource(base_path=os.path.dirname(config_path))
)

Example:

---
# Uses the dictmentor package to augment the configuration by dictmentor extensions.
# Make sure to export the environment variable to echo:
# export MESSAGE="Hello World"

- name: dictmentor
  pull:
    plugin: pnp.plugins.pull.simple.Repeat
    args:
      wait: 1
      repeat: "{{env::MESSAGE}}"
  push:
    - external: echo.pull
    - external: nop.pull

# Contents of echo.pull
plugin: pnp.plugins.push.simple.Echo

# Contents of nop.pull
plugin: pnp.plugins.push.simple.Nop

5.4. Advanced selector expressions (0.12.0+)

Instead of string-only selector expressions, you may now use complex dictionary and/or list constructs in your yaml to define a selector expression. If you use a dictionary or a list make sure to provide "real" selectors as a lambda expression, so the evaluator can decide if this is a string literal or an expression to evaluate.

The configuration below will repeat {'hello': 'Hello', 'words': ['World', 'Moon', 'Mars']}.

- name: selector
  pull:
    plugin: pnp.plugins.pull.simple.Repeat
    args:
      wait: 1
      repeat: "Hello World Moon Mars"
  push:
    - plugin: pnp.plugins.push.simple.Echo
      selector:
        hello: "lambda payload: payload.split(' ')[0]"
        words:
          - "lambda payload: payload.split(' ')[1]"
          - "lambda payload: payload.split(' ')[2]"
          - "lambda payload: payload.split(' ')[3]"

Before the implementation of the advanced selector feature your epxressions would have probably looked similiar to this:

dict(hello=payload.split(' ')[0], words=[payload.split(' ')[1], payload.split(' ')[2], payload.split(' ')[3]])

The first one is more readable, isn't it?

Additional example:

---
- name: selector
  pull:
    plugin: pnp.plugins.pull.simple.Repeat
    args:
      interval: 1s
      repeat: "Hello World"
  push:
    - plugin: pnp.plugins.push.simple.Echo
      # Returns: 'World'
      selector: "str(payload.split(' ')[0])"  # no complex structure. Evaluator assumes that this is an expression -> you do not need a lambda
    - plugin: pnp.plugins.push.simple.Echo
      selector:  # Returns {'header': 'this is a header', 'data': 'World', 'Hello': 'World'}
        header: this is a header  # Just string literals
        data: "lambda data: data.split(' ')[1]"  # Value is lambda and therefore evaluated
        "lambda data: str(data.split(' ')[0])": "lambda data: data.split(' ')[1]"  # Both are lambdas and therefore evaluated
    - plugin: pnp.plugins.push.simple.Echo
      selector:  # Returns ['foo', 'bar', 'Hello', 'World']
        - foo  # String literal
        - bar  # String literal
        - "lambda d: d.split(' ')[0]"  # Lambda -> evaluate the expression
        - "lambda d: d.split(' ')[1]"  # Lambda -> evaluate the expression

5.5. UDF Throttle (0.15.0+)

Consider the following situation: You have a selector that uses a udf to fetch a state from an external system. The state won't change so often, but your selector will fetch the state every time a pull transports a payload to the push. You want to decrease the load on the external system and you want to increase throughput. Throttle to the rescue. Specifying throttle when instantiating your udf will manage that your udf will only call the external system once and cache the result. Subsequent calls will either return the cached result or call the external system again when a specified time has passed since the last call that actually fetched a result from the external system.

Example:

udfs:
  - name: count  # Instantiate a Counter user defined function
    plugin: pnp.plugins.udf.simple.Counter
    args:  # The presence of args tells pnp to instantiate a Counter - important because it has a state (the actual count)
      init: 1
      # Will only call the counter if 10 seconds passed between current call and last call.
      # In the meantime a cached result will be returned.
      throttle: 5s
tasks:
  - name: hello-world
    pull:
      plugin: pnp.plugins.pull.simple.Repeat
      args:
        interval: 1s
        repeat: "Hello World"
    push:
      - plugin: pnp.plugins.push.simple.Echo
        selector:
          counter: "lambda d: count()"

5.6. Docker images

# Mount the task and logging configuration when starting up the container
docker run --rm \
    -v /path/to/pnp/config/01_hello_world.yaml:/config/config.yaml \
    -v /path/to/logging/config/file.logging:/config/logging.yaml \
    hazard/pnp:latest

6. Plugins

A complete list of plugins is available here