ckanext-resource_indexer

Discover more results in the Dataset search by searching through the content of resources.

This extension indexes the content of files attached to resources. In this way user has more chances to find the relevant results when using site search.

The process if indexation can be customized for each file format via resource indexers. The following formats are supported out of the box:

• Plain text
• PDF
• JSON

Structure

• Installation
• Configuration
• Indexers
  • Register own indexer
  • Built-in indexers

Installation

1. Install the package as a CKAN extension

   pip install ckanext-resource-indexer
   

2. Add resource_indexer to the list of enabled plugins
3. Optional. Enable built-in indexers by adding the following items to the list of enabled plugins:
   • plain_resource_indexer
   • pdf_resource_indexer
   • json_resource_indexer

Configuration

# Make an attempt to index remote files(fetch into tmp folder
# using URL)
# (optional, default: false).
ckanext.resource_indexer.allow_remote = 1

# Tiemeout for the attempt to download remote file
# (optional, default: 2).
ckanext.resource_indexer.remote_timeout = 10

# The size treshold(MB) for remote resources
# (optional, default: 4).
ckanext.resource_indexer.max_remote_size = 4

# List of resource formats(lowercase) that should be
# indexed.
# (optional, default: None)
ckanext.resource_indexer.indexable_formats = txt pdf

# Store the data extracted from resource inside specified field in the index.
# If empty, store data inside the general-purpose `text` field.
# (optional, default: text)
ckanext.resoruce_indexer.index_field = extras_res_attachment

# Boost matches by resource's content. Set values greater that 1 in order
# to promote such matches and value between 0 and 1 in order to put such
# matches further in search results. Works only when using custom index
# field(`ckanext.resoruce_indexer.index_field`)
# (optional, default: 1)
ckanext.resoruce_indexer.search_boost = 0.5

##### Indexer specific option ###############

### Plain
# Space-separated list of formats that can be indexed as a plain text
# (optional, default: txt csv json yaml yml html)
ckanext.resource_indexer.plain.indexable_formats = xml txt csv

### PDF
# Change a text from a single page before it added to the index
# (optional, default: builtins:str)
ckanext.resoruce_indexer.pdf.page_processor = custom.module:value_processor

### JSON
# Index JSON files as plain text(in addition to indexing as mapping)
# (optional, default: false)
ckanext.resoruce_indexer.json.add_as_plain = true

# Change a key before it's used for patching the package dictionary
# (optional, default: builtins:str)
ckanext.resoruce_indexer.json.key_processor = custom.module:key_processor

# Change a value before it's used for patching the package dictionary
# (optional, default: builtins:str)
ckanext.resoruce_indexer.json.value_processor = custom.module:value_processor

Indexers

In order to extract the data from resources, this extension uses Indexers. These are CKAN plugins implementing IResourceIndexer interface.

For every resource with the format specified by ckanext.resource_indexer.indexable_formats config option, an appropriate indexer is searched. If no indexers were found(or resource format is missing from the ckanext.resource_indexer.indexable_formats config option), the resource is skipped.

:information_source: Indexation can be temporarily disabled using one of the following approaches:

• Set environment variable CKANEXT_RESOURCE_INDEXER_BYPASS(any non-empty value), and the plugin won't interfer into standard dataset indexation process.
• Use ckanext.resource_indexer.utils.disabled_indexation context manager:

  with disabled_indexation():
      here_indexation_does_not_happen()
  
  here_indexation_happens()
  

Every indexer has weight(priority). Indexer with the highest weight will be used to index the resource.

Indexation consists of two steps:

• meaningful data segments extracted from the resource
• these data segments are merged into the package dictionary consumed by the search engine(Solr) for indexing

It means, that the format of extracted segments must be compatible with the merging logic from the second step. But other than that, there are no particular requirements for the format of extracted data.

Data extraction happens locally. If the resource was uploaded to the local filesystem, data is extracted directly from the resource's file. If the resource is stored remotely(either uploaded to the cloud or linked via remote URL), it can be temporarily downloaded to the local filesystem and removed after processing. By default, non-local resources are ignored, but this can be changed via ckanext.resource_indexer.allow_remote config option.

Register own indexer

Implement ckanext.resource_indexer.interface.IResourceIndexer by providing following methods:

class CustomIndexerPlugin(plugins.SingletonPlugin):
    plugins.implements(IResourceIndexer)

    def get_resource_indexer_weight(self, resource: dict[str, Any]) -> int:
        """Define priority of the indexer

        Args:
            resource: resource's details

        Returns:
            the weight of the indexer
            Expected values:
               0: skip handler
               10: use handler if no other handlers found
               20: use handler as a default one for the resource
               30: use handler as an optimal one for the resource
               40: use handler as a special-case handler for the resource
               50: ignore all the other handlers and use this one instead
        """
        return Weight.fallback

    def extract_indexable_chunks(self, path: str) -> Any:
        """Extract indexable data from the resource

        The result can have any form as long as it can be merged into the
        package dictionary by implementation of `merge_chunk_into_index`.

        Args:
            path: path to resource file

        Returns:
            all meaningfuld pieces of data with no type assumption

        """
        return []

    def merge_chunks_into_index(self, pkg_dict: dict[str, Any], chunks: Any):
        """Merge data into the package dictionary.


        Args:
            pkg_dict: package that is going to be indexed
            chunks: collection of data fragments extracted from resource

        Returns:
            all meaningfuld pieces of data with no type assumption
        """
        pass

Built-in indexers

Plain indexer

Index formats specified by ckanext.resource_indexer.indexable_formats if they fall into the value of ckanext.resource_indexer.plain.indexable_formats config option, unless other handler with a non-fallback weight(>10) found.

Resources are indexed as-is. File is read and sent to the index without any additional changes.

Enable it by adding plain_resource_indexer to the list of enabled plugins.

PDF indexer

Extract and index text from the PDF file.

In order to enable it:

• install current extension with the pdf extra:

  pip install 'ckanext-resource-indexer[pdf]'
  

• add pdf_resource_indexer to the list of enabled plugins and
• install system packages for PDF processing. This will be different depending on your system. Examples:

  • CentOS

    yum install -y pulseaudio-libs-devel \
       gcc-c++ pkgconfig \
       python3-devel \
       libxml2-devel libxslt-devel \
       poppler poppler-utils poppler-cpp-devel
    

  • Debian

    apt install build-essential libpoppler-cpp-dev pkg-config python3-dev
    

  • macOS

    brew install pkg-config poppler python
    

If PDF content requires preprocessing, specify function that converts text from a every separate as a ckanext.resoruce_indexer.pdf.page_processor. It uses standard import-string format: module.import.path:function

JSON indexer

Read a dictionary from the JSON file, convert all non-string values into strings(i.e, no nested values allowed), and apply it as a patch to the indexed dataset.

Optionally, if ckanext.resoruce_indexer.json.add_as_plain flag enabled, index the content of the file as a plain-text(similar to the plain indexer)

If key or value requires preprocessing, specify function that converts data as a ckanext.resoruce_indexer.json.key_processor or ckanext.resoruce_indexer.json.value_processor. It uses standard import-string format: module.import.path:function

Enable it by adding json_resource_indexer to the list of enabled plugins.