PDF to ALTO

Service that listens for incoming messages from SQS. On receipt of a message this service will download reference PDF, extract ALTO file per-page and upload to specified S3 bucket.

If COMPLETED_TOPIC_ARN env var specified a notification will be raised.

Messages Format

The incoming message is in the shape:

{
  "pdfLocation": "https://www.hq.nasa.gov/alsj/a17/A17_FlightPlan.pdf",
  "pdfIdentifier": "a17_flightplan",
  "outputLocation": "s3://pdf-to-alto/a17_flightplan_alto"
}

Where

pdfLocation - the URL where PDF can be downloaded from.
pdfIdentifier - unique identifier for PDF. where i is 0-based page index. If omitted a random uuid will be used.
outputLocation - s3 location where final ALTO files will be output. With or without preceding s3:// and no trailing /.

(See sample.json)

The completed notification message echos back the original message with "numberOfFiles" property added.

The generated alto file will be placed in outputLocation. The format of each file will depend on value of PREPEND_ID envvar. This format will be (where i is the page number):

If true: f"{pdfIdentifier}-{i:04d}.xml"
else f"{i:04d}.xml",

Technology

This is a Python script that utilises the following libraries:

pdfalto - C lib used to generate ALTO files.
PyMuPDF - Python lib used to query PDF object for page dimensions.
lxml - Used to parse ALTO files and update scaled values.
requests - Used to download PDF files.

Configuration

The following environment variables can be used to configure the app:

Env Var	Description	Default
DOWNLOAD_CHUNK_SIZE	Chunk size for downloading PDF	2048
WORKING_FOLDER	Local working folder for storing generated files	./work
REMOVE_WORK_DIR	Whether to clean up working dir on completion	True
RESCALE_ALTO	Whether to rescale generated ALTO to page	True
MONITOR_SLEEP_SECS	How long to sleep long polling operations if no messages received	30
AWS_REGION	AWS region being used	eu-west-1
INCOMING_QUEUE	The name of the SQS queue to monitor for incoming messages. Required
COMPLETED_TOPIC_ARN	The ARN of a topic to post completion notifications to
LOCALSTACK	If using LocalStack	False
LOCALSTACK_ADDRESS	Address for LocalStack instance	http://localhost:4566

(See .env.dist for sample .env file)

Running Locally

There is a multi-stage Dockerfile that builds the pdfalto binary and copies it to a new stage.

docker-compose.yml will build and start the main Python app alongside a LocalStack instance.

# build and start image using LocalStack
docker-compose up

# build image
docker build --tag pdf-to-alto:local .

# run docker image and listen to queue
docker run --env-file .env -it --rm --name pdf-to-alto pdf-to-alto:local

# run docker image to process 1 single api
docker run -it --rm --name pdf-to-alto \
  pdf-to-alto:local \
  opt/app/app/pdf_processor.py https://text.example/test.pdf my-pdf-identifier s3://pdf-bucket/alto

Note: Building pdfalto from source takes a few minutes

LocalStack

The docker-compose.local.yml file will spin up a LocalStack instance and configure a few resource for local testing:

An S3 bucket titled "pdf-to-alto"
An SNS topic "incoming-topic" with an SQS subscription to "incoming"
An SNS topic "completed-topic" with an SQS subscription to "completed"

docker-compose -f docker-compose.local.yml up

To use LocalStack set the LOCALSTACK and LOCALSTACK_ADDRESS env vars (see above).

When using the aws-cli with LocalStack the --endpoint-url needs to be specified. Below are some handy commands to use when testing: