Important

The project although maintained it will not be further developed as NiFi workflows are a better alternative. Please see NiFi and its relevant workflows.

Feel free to ask questions on the github issue tracker or on our discourse website which is frequently used by our development team!

Introduction

This simple application implements an ingestion process to:

• retrieve the documents from a specified ElasticSearch source,
• send the selected content from these documents to NLP REST service receiving back the annotations from the text,
• send the annotations with selected metadata to specified ElasticSearch sink.

The ingestion parameters (source, sink, fields mapping, etc.) can be set in config.yml file (in config directory).

Usage

Firstly install the Python libraries specified in requirements.txt.

To run: python main.py --config config/config.yml

Configuration

The ingestion process properties are configured in config.yml.

Available options:

ElasticSearch data store

Data Source

Entries under source key specify the ElasticSearch source.

Data sink

Entries under sink key specify the ElasticSearch sink.

Security and SSL certificates

This only applies when ElasticSearch cluster is using X-Pack / Open Distro and requires secure connections with using SSL certificates. Under the key source and/or sink can be optionally specified security entry that will contain necessary configuration -- these are: ' ca-file-path -- path to CA certificate file (PEM) used solely with SSL, (DO NOT USE THIS along with the other ca/client parameters mentioned underneath, use it solely)

• ca-certs-path -- the path to CA certificates file (PEM),
• client-cert-path -- the path to client certificate file (PEM),
• client-key-path -- the path to client key (PEM).

Extra ES sink/source connection options

Entires under the key extra_params are optional, useful for test cases or deployments where we only use internal resources:

• use_ssl -- use ssl connection

• verify_certs -- verify SSL certificates

• credentials

  • username and password can be used to provide connection credentials
  • use-api-key if this is enabled the username and password fields will be used as api_id and api_key

NLP service

• endpoint-url, url that points othe REST api annotation service endpoint

• endpoint-request-mode , this is either left empty, or in case of use with the GATE NLP Annie annotation service it should be set to gate-nlp

• use-bulk-indexing ingest in bulk mode (1000 docs / bulk chunk),

• credentials

  • username and password can be used to provide connection credentials

Fields mapping

Entries under mapping key define the mapping of the document fields for the ingestion.

The sub-entry index-ingest-mode defines:

• same-index: False , set to True if you wish to ingest annotations into the same index
• use-nested-objects : 'False` , set to True if you wish to ingest into the same index with nested object type, useful but beware of search query speed impact
• es-nested-object-schema-mapping : for medcat annotations use medcat-separate-index or medcat-nested-object , for GATE-nlp use gate-nlp-separate-index or gate-nlp-nested-object

The sub-entry source defines the field names that contain:

• text-field - the free text to be processed,
• docid-field - the unique identifier of the document,
• persis-fields - a list of fields to be send back to the sink alongside each annotation entry.

The sub-entry batch defines the possible portion of documents to be processed according to the date. The used fields are:

• date-field - the name of the field containing the date,
• date-format - the format of the date/time used by ElasticSearch to specify the time window by the user (below),
• python-date-format - the format of the date/time used by Python to specify the time window by the user (below),
• interval - the number of days to be used for incremental batch processing in processing time window,
• date-start and date-end - the time window to be processed,
• threads - the number of processing threads to speed up the ingestion.

The sub-entry sink specifies additional options during sending the processed annotations:

• split-index-by-field - the name of the field in the returned annotations the value of which will be used as a prefix for the index name (e.g., used to send annotations of different types to separate indices). If you don't want this functionality simply leave the field empty, otherwise , to split by annotation type use type

The sub-entry nlp specifies additional options during processing the documents with NLP:

• skip-processed-doc-check - whether to skip checking for already processed documents in ElasticSearch,
• annotation-id-field - the name of field containing the annotation id returned from the NLP app.

Missing

• tests
• API specs