Some deployments of Aleph may wish to implement additional processing stages for content from text documents. A common example of this is automatic translation. This can be done by adding an extra service with custom code and adding it to the document processing pipeline.
All processing in Aleph is coordinated via a set of queues kept in redis, an in-memory data store. When a document is processed, the instruction to handle it is posted on that queue, which is read by multiple independent services. Each service can subscribe to one or many
stages in order to receive tasks, e.g. the
ingest stage (which extracts text from a file), the
analyze stage (which performs NER and language detection) and finally the
index stage (which adds the document to Aleph's search index).
A new stage can be added to this pipeline at runtime, either by extending the code of the aleph worker service, the ingest-file service, or by adding a new service that stands on its own. The last option is the cleanest, and can be based on
servicelayer, the Python package that implements the queue and worker code.
An example implementation of a
servicelayer-based processing service is available in the Aleph repository at
alephdata/translate-service. The code should be short enough to study in full, especially the worker module.
In order to enable the service, you need to add it to the
docker-compose.dev.yml file, and make one of the other services depend on it. An example section is present in the file, but commented out.
Once you have added the service to the compose configuration, you can include the stage in the pipeline by adding the following variable to the
aleph.env configuration file:
You may need to restart the compose environment in order for the environment changes to take effect.
During development, you can restart the container like this:
docker-compose -f docker-compose.dev.yml up \--force-recreate --no-deps translate