This is what I've come up with for a class that handles parsing AWS Textract output. So far it only gets the AWS Textract output into a .txt file.

import json
import logging as log
import os
import sys
import subprocess
import time

  format='%(asctime)s [%(levelname)s]: %(message)s',
  datefmt='%Y-%m-%d %H:%M:%S',

class InitialParser:

  def __init__(self, filename, s3_bucket, version=1):
    self.filename = filename
    self.s3_bucket = s3_bucket
    self.version = version

    self.file = None
    self.filepath = None
    self.data = None
    self.output_filename = self.filename.replace('.pdf', '.txt')

  def run(self):
    log.info('Running initial parser on file {}'.format(self.filename))
    if self.run_textract():
      log.info('Textract process successful')
      log.info('Textract process failed, exiting')

  def load_file_data(self):
    self.filepath = os.path.join(os.getcwd(), '../extracted/{}'.format(filename))
      with open(self.filepath, 'r') as file:
        self.file = file
        self.data = json.load(file)
    except FileNotFoundError as e:
      log.error('Could not find parsing file: {}'.format(e))

  def run_textract(self):
    Runs AWS Textract commands on the given document and sends the output
    to a .txt file.

    @param s3_bucket: the S3 Bucket where the document is located.
    job_id = ''
    # note: adding "Version":<str> to the AWS object below breaks the command
    aws_object = json.dumps({"S3Object":{"Bucket":self.s3_bucket,"Name":self.filename}}).replace(' ', '')  # can't have any spaces lol picky AWS CLI

    start_textract_command = "aws textract start-document-text-detection --document-location '{}'".format(aws_object)
    get_textract_output_command = 'aws textract get-document-text-detection --job-id '

      job_id = '"{}"'.format(json.loads(subprocess.check_output([start_textract_command], shell=True, stderr=subprocess.STDOUT).decode('utf-8'))['JobId'])
    except subprocess.CalledProcessError as e:
      if 'InvalidS3ObjectException' in e.output.decode('utf-8'):
        log.error('InvalidS3ObjectException (could not fetch object metadata from S3).\n Check the document name, AWS CLI configuration region (run `aws configure list`), permissions, and the S3 Bucket name & region.')
      elif 'ProvisionedThroughputExceededException' in e.output.decode('utf-8'):
        log.error('ProvisionedThroughputExceededException (provisioned rate exceeded). You\'re doing that too much.')
        log.error('Starting Textract failed. Error: {}'.format(e.output.decode('utf-8')))

    time.sleep(10)  # wait for Textract to do its' thing

    if job_id != '':
        subprocess.call(['touch {}'.format(self.output_filename)], shell=True)
        subprocess.call(['{} > {}'.format(get_textract_output_command+job_id, self.output_filename)], shell=True, stderr=subprocess.STDOUT)
        return True
      except subprocess.CalledProcessError as e:
      return False

if __name__ == '__main__':
  initial_parser = InitialParser(

I'd like a better way of handling the wait period for Textract to finish other than waiting 10 seconds, and it may take longer than that in some instances.

I do know that the output file will have "JobStatus": "IN PROGRESS" so I could repeatedly search the output .txt file for "SUCCEEDED", but that could be slow if the file has 40,000+ lines when it's actually done (it will)

aws_object = json.dumps({"S3Object":{"Bucket":self.s3_bucket,"Name":self.filename}}).replace(' ', '')  # can't have any spaces lol picky AWS CLI
start_textract_command = "aws textract start-document-text-detection --document-location '{}'".format(aws_object)

This is not AWS which is picky, rather you’re not building the command line properly. If you provide a single string and start your subprocess using shell=True, each space will mark a new argument. Instead, you’d be better off building a list of arguments and let subprocess quote them properly:

start_textract_command = ['aws', 'textract', 'start-document-text-detection', '--document-location', json.dumps(…)]

and then just run subprocess.check_output(start_textract_command, stderr=subprocess.STDOUT).

Speaking of which, instead of using the older high level API, you should switch to subprocess.run:

process = subprocess.run(start_textract_command, capture_output=True, encoding='utf-8')
except CalledProcessError:
    if 'InvalidS3ObjectException' in process.stderr:
    job_id = json.loads(process.stdout)['JobId']
    with open(self.output_filename, 'wb') as output_file:
       process = subprocess.run(['aws', 'textract', 'get-document-text-detection', '--job-id', str(job_id)], stdout=output_file, stderr=subprocess.PIPE)
    if process.returncode:

Now for the time.sleep part, AWS Textract provides two modes of operations: synchronous and asynchronous. You can start by using the synchronous detect-document-text operation if it fits your need and you won't have to deal with the timing at all.

Otherwise, if you need to stick to start-document-text-detection, the completion of the process is published as a notification. I’m not at all versed in how SNS works but they have a tutorial to get you started on using Amazon SNS so you can create a channel, specify it in your textract job and wait for the completion event.

  • \$\begingroup\$ thanks for the tips, this really cleans up the code. \$\endgroup\$ – ChumiestBucket Apr 24 '19 at 13:07
  • \$\begingroup\$ just wanted to add that the capture_output arg for subprocess.run does not exist in Python 3.6.X and using stdout=subprocess.PIPE will accomplish the same thing. \$\endgroup\$ – ChumiestBucket Apr 24 '19 at 14:56
  • \$\begingroup\$ @star_trac You should also add stderr=PIPE, to account for the error case as well. \$\endgroup\$ – 301_Moved_Permanently Apr 24 '19 at 17:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.