Page Comparison

Table of Contents

Overview

Items that are processed by Squirro go through a pipeline process before they are indexed. In that process a number of built-in pipeline steps are executed. On top of that, custom enrichment steps can be inserted in the form of pipelets. These pipelets are written in the Python programming language. Pipelets can be uploaded to the Squirro server and then be configured in the user interface (Enrichments tab) or through the API.

This reference documentation covers the basic workflow of working with pipelets and the interface that pipelets need to implement.

...

Writing Pipelets

Pipelets are written in Python. They need to inherit from the squirro.sdk.PipeletV1 class and implement the consume method. The simplest possible pipelet looks like this:

from squirro.sdk import PipeletV1
 
class NoopPipelet(PipeletV1):
    def consume(self, item):
        return item

As it name says it does nothing but return the item unchanged.

Modifying Items

The item is a Python dict type and can be modified before it is returned. The available item fields are documented in the Item Format reference. The following example illustrates modifying an item:

from squirro.sdk import PipeletV1
 
class ModifyTitlePipelet(PipeletV1):
    def consume(self, item):
        item['title'] = item.get('title', '') + ' - Hello, World!'
        return item

This pipelet will modify each item it processes, appending the string "Hello, World!" to the title.

Skipping Items

Items can be "skipped" ( i.e. not added to Squirro's search index ) by return None instead of the item for example;

from squirro.sdk import PipeletV1
 
class SkipItemPipelet(PipeletV1):
    def consume(self, item):
        if not item.get('title', '').startswith('[IMPORTANT]')
            return None
        return item

In this example we discard all items where the title does not start with the string "[IMPORTANT]".

Returning multiple items

The pipelet is always called for each item individually. But in some use cases the pipelet should not just return one item but multiple ones. In those cases use the Python yield statement to return each individual item. For example:

from squirro.sdk import PipeletV1
 
class ExtendTitlePipelet(PipeletV1):
    def consume(self, item):
        for i in range(10):
            new_item = dict(item)
            new_item['title'] = '{0} ({1})'.format(item.get('title', ''), i)
            yield new_item

Dependencies

Pipelets are limited in what you can do. For example the print statement is disallowed and you can not import any external libraries except squirro.sdk. If you do need access to external libraries, you need to use the @require decorator. For example to log some output:

from squirro.sdk import PipeletV1, require	

@require('log')
class LoggingPipelet(PipeletV1):
    def consume(self, item):
        self.log.debug('Processing item: %r', item['id'])
        return item

As seen from the example, the @require decorator takes a name of a dependency. That dependency is then made available to the pipelet class.

HTTP requests can be executed by using the requests dependency. The following pipelet shows an example for sentiment detection:

from squirro.sdk import PipeletV1, require

@require('requests')
class SentimentPipelet(PipeletV1):
    def consume(self, item):
        text_content = ' '.join([item.get('title', ''),
                                 item.get('body', '')])
        res = self.requests.post('http://example.com/detect',
                                 data={'text': text_content},
                                 headers={'Accept': 'application/json'})
        sentiment = res.json()['sentiment']
        item.setdefault('keywords', {})['sentiment'] = [sentiment]
        return item

Available Dependencies

The following dependencies can be requested:

...

Dependency

...

Description

...

cache

...

Non-persisted cache.

...

log

...

A logging.Logger instance from Python's standard logging framework.

...

requests

...

Python requests library for to execute HTTP requests.

...

files

...

Python component which provides access to data files on disk.

Default Configuration

When adding enrichments in the user interface, the administrator can pass in configuration to the pipelet. For example the title modification pipelet could accept a custom suffix that is to be added to the title.

The configuration for this is provided in as JSON data structure:

...

This full data structure is then passed in to the pipelet constructor, where it can be retrieved. Usually it's then simply stored in a object variable so it can be used in the consume method again.

from squirro.sdk import PipeletV1

DEFAULT_SUFFIX = ' - Hello, World!'

class ModifyTitlePipelet(PipeletV1):
    def __init__(self, config):
        self.config = config

    def consume(self, item):
        suffix = self.config.get('suffix', DEFAULT_SUFFIX)
        item['title'] = item.get('title', '') + suffix
        return item

In this example, when the suffix hasn't been provided, the default suffix is used.

Custom Configuration

Pipelets can define a custom set of configuration properties, which will be exposed in the UI as a form (instead of a JSON input). This makes the configuration much more user friendly, and is the recommended way. To define its configuration, a pipelet should implement the getArguments() method.

The method is expected to return an array of objects defining each property. Inside each object, the fields name, display_label and type are required. Optional field required specifies whether the property is required to be filled by the user. Additionally, a property can be placed in an Advanced section of the configuration, by setting advanced to True on that property.

The type should be one of the following: int, string, bool, password, code.

from squirro.sdk import PipeletV1

class ModifyTitlePipelet(PipeletV1):
    # def __init__, def consume, etc.

    @staticmethod
    def getArguments():
        return [
            {
                'name': 'commands',
                'display_label': 'Source code',
                'type': 'code',
                'required': True,
            },
            {
                'name': 'debug',
                'display_label': 'Log debug output',
                'type': 'bool',
                'help': 'Logs debug output to the server log files',
                'advanced': True,
            },
        ]

Documentation

A pipelet class can be documented using doc-strings. The first sentence (separated by period) is used as a summary in the user interface. All the remaining text is used as a description and is often used to document the expected configuration. The description is parsed as Markdown (using the CommonMark dialect). The 60-second overview serves as a good reference.

from squirro.sdk import PipeletV1

DEFAULT_SUFFIX = ' - Hello, World!'

class ModifyTitlePipelet(PipeletV1):
    """Modify item titles.

    This appends a suffix to the title of each item. When no suffix is
    provided, it appends the default suffix of "- Hello, World!".
    """
    def __init__(self, config):
        self.config = config

    def consume(self, item):
        suffix = self.config.get('suffix', DEFAULT_SUFFIX)
        item['title'] = item.get('title', '') + suffix
        return item

    @staticmethod
    def getArguments():
        return [
            {
                'name': 'suffix',
                'display_label': 'Suffix',
                'type': 'string',
                'default': DEFAULT_SUFFIX,
            }
        ]

Development Workflow

For developing pipelets, Squirro provides the pipelet command line tool as part of the Toolbox.

Develop

The first step is to create the pipelet. In the following examples the pipelet will have been written to a file called pipelet.py in the current directory.

pipelet.py

from squirro.sdk import PipeletV1

class ModifyTitlePipelet(PipeletV1):
    def consume(self, item):
        item['title'] = item.get('title', '') + ' - Hello, World!'
        return item

Validate

On the command line execute the pipelet validate command to verify that there are no errors in the pipelet code. For example this will ensure that no modules are imported that are disallowed from pipelets. See the section on Dependencies for more information.

pipelet validate pipelet.py 

Test

The pipelet consume command can be used to simulate pipelet running. For this purpose, the test items should be present in JSON text files on the disk. In the following example there is a item.json file in the current directory with this contents:

item.json

{
    "title": "Sample",
    "id": "first_item"
}

To test the pipelet with this test file, use:

pipelet consume pipelet.py -i item.json 

This command will output the items that have been returned by the pipelet:

Loading items...
Loading item.json ...
Loaded.
Consuming item first_item
yielded item
{u'id': u'first_item', u'title': u'Sample - Hello, World!'} 

On top of these manual tests, automated tests can be implemented easily using the usual Python tools such as Nose.

Deploy

Once the pipelet is ready, it can be uploaded to the Squirro server. The pipelet upload command achieves that:

pipelet upload --token <your_token> --cluster <cluster> pipelet.py "Hello World"

This will make the pipelet available with the name "Hello World". To update the pipelet code on the server, this command can be re-executed at any time.

To use this in a project use the Pipeline Editor to add the pipelet as an enrichment step to a pipeline.

Using additional files with Pipelets

In many cases, additional files like libraries and pre-trained models must be uploaded and used by a pipelet when it is run. 

To accomplish this, additional files can be uploaded and accessed by the pipelet by following these steps:

...

Specify the additional files in the pipelet upload command. For Example:

Pipelet Upload Command

pipelet upload \
    --data-file 'resource.txt' \
    --cluster <cluster> \
    --token <your_token> \
    'pipelet.py' \
    'TestPipelet'

In this example, we are uploading a file "resource.txt" along with the pipelet.

Access the contents of the file(s) from within the pipelet

Pipelet File

from squirro.sdk import require, PipeletV1

@require('files')
class TestPipelet(PipeletV1):
    def consume(self, item):
        with self.files.get_file('resource.txt') as f:
            data = f.read()

Processing old items

Pipelets are only run for items that are processed in the system after the enrichment has been configured. For information on how to process old items with a pipelet, see Rerunning a Pipelet.

Adding Enrichments

...