createReplica not returning both parent and child records from hosted feature service

[samhooper]

I'm attempting to use createReplica from the ArcGIS Rest API to download a copy of a hosted feature service, but the results never include records from both a parent layer and its children -- it's either one or the other. The feature service was created automatically when I created a Survey123 survey, and the survey includes a repeat group, hence the parent-child relationship. I'm making calls to the API via the Python script shown below. 

If I don't include the layerQuery parameter, all records in the parent layer and attachments are returned in the result and the child layer exists, but it's completely empty. If I do include the layerQuery parameter with "includeRelated": true and "queryOption": "none", all records in the child layer are returned, but the parent layer is empty. I've tried several different iterations of other parameters, but none of them yield records from both layers. Those iterations include 

• "queryOption": "all" - only the parent layer records are returned
• "dataFormat" of "sqlite", "json", and "filegdb" - behavior is the same (i.e., depends on layerQuery params) 
• "where" and "useFilter" with a query that should evaluate to return everything - only parent layer records are returned

Also of note, if I download the data manually from the Data tab/page of the survey, all of the records are there. I assume this means there's nothing wrong with the feature service itself and there aren't any settings of the feature service I've left unchecked.

UPDATE: If I use query instead of createReplica with layerDefs that should definitely return all records from both layers (i.e., {'0': "CreationDate > TIMESTAMP '1970-1-1 00:00:00'", '1': "CreationDate > TIMESTAMP '1970-1-1 00:00:00'"}), all the features of the parent layer are returned and the child layer is empty. So the problem is not with createReplica. Is there something wrong with the way Survey123 automatically creates relationships such that the REST API ignores related layers/tables?

I've searched for quite a while, and I can't find anything mentioning a similar problem. Usually when that happens it means I'm doing something wrong. The Python script is below. Any help or suggestions would be appreciated.

import os, sys
import requests
import time
import json
import pandas as pd
from sqlalchemy import create_engine
from datetime import datetime


RESULT_TIMEOUT = 900 # seconds in 15 minutes
CHECK_STATUS_INTERVAL = 5


def check_http_error(attempted_action, response):
    try:
        response.raise_for_status()
    except Exception as e:
        raise requests.HTTPError('failed to {action} because {error}'.format(action=attempted_action, error=e))

    # status codes are often valid even if there was an error, so check the response json
    response_json = response.json()
    if 'error' in response_json:
        raise requests.HTTPError('failed to {action} because {error}'.format(action=attempted_action, error=response_json['error']['details']))


def download_data(portal_url, service_url, client_id, client_secret, ssl_cert, out_dir, last_poll_time=None):

    # Get token for REST API
    token_params = {'client_id': client_id, 'client_secret': client_secret, 'grant_type': 'client_credentials'}
    token_response = requests.get(portal_url + '/sharing/rest/oauth2/token/', params=token_params, verify=ssl_cert)
    check_http_error('get token', token_response)
    token = token_response.json()['access_token']

    # Get feature service info
    info_response = requests.get(service_url, params={'f': 'json', 'token': token}, verify=ssl_cert)
    check_http_error('get service info', info_response)
    service_info = info_response.json()

    # Submit POST request to get data
    # Make sure that the layerQueries parameter is given with 'includeRelated' = true. Otherwise, related records will
    # not be included
    layers = [layer_info['id'] for layer_info in service_info['layers']]
    layer_queries = {}
    for layer_id in layers:
        str_id = str(layer_id)
        layer_queries[str_id] = {
            'includeRelated': True,
            'queryOption': 'none'
        }
        if last_poll_time:
            layer_queries[str_id]['queryOption'] = 'useFilter'
            layer_queries[str_id]['where'] = "CreationDate > TIMESTAMP '%s'" % last_poll_time

    create_replica_params = {
        'f': 'json',
        'token': token,
        'layers': layers,
        'geometry': '-180,-90,180,90',
        'geometryType': 'esriGeometryEnvelope',
        'inSR': 4326,
        'dataFormat': 'sqlite',
        'returnAttachments': True,
        'returnAttachmentsDatabyURL': False,
        'async': True,
        'syncModel': 'none',
        'layerQueries': json.dumps(layer_queries)
    }
    replica_response = requests.post('%s/createReplica' % service_url, params=create_replica_params, verify=ssl_cert)
    check_http_error('create replica', replica_response)
    status_url = replica_response.json()['statusUrl']

    # Query was set as asynchronous, so check the status at a set interval
    for i in range(0, RESULT_TIMEOUT, CHECK_STATUS_INTERVAL):
        status_response = requests.get(status_url, params={'f': 'json', 'token': token}, verify=ssl_cert)
        check_http_error('check status after %s iterations', status_response)
        status_json = status_response.json()

        # If the resultUrl isn't empty, that means the result is ready to download
        result_url = status_json['resultUrl']
        if result_url != '':
            break
        elif status_json['status'] in ('CompletedWithErrors', 'Failed'):
            raise requests.HTTPError('the query failed for an unspecified reason')
        else:
            if i + CHECK_STATUS_INTERVAL >= RESULT_TIMEOUT: # timeout exceeded
                raise requests.exceptions.ConnectTimeout(
                    'Asynchronous query exceeded RESULT_TIMEOUT of %.1f minutes' % (RESULT_TIMEOUT/60.0)
                )
            time.sleep(CHECK_STATUS_INTERVAL)

    # Write the result to disk
    result_response = requests.get(result_url, params={'f': 'json', 'token': token}, verify=ssl_cert)
    check_http_error('get createReplica result', result_response)
    service_name = service_info['serviceDescription']
    sqlite_path = os.path.join(out_dir, '{0}_{1}.db'.format(service_name, datetime.now().strftime('%Y%m%d-%H%M%S')))
    with open(sqlite_path, 'wb') as f:
        f.write(result_response.content)

    # Get attachments (stored as bytes in Blob dtype column of the sqlite DB)
    engine = create_engine('sqlite:///' + sqlite_path)
    with engine.connect() as conn, conn.begin():
        attachment_tables = pd.read_sql("SELECT name FROM sqlite_master WHERE name LIKE('%__ATTACH');", conn)\
                                    .squeeze(axis=1)

        attachments_dir = os.path.join(out_dir, 'attachments')
        for table in attachment_tables:
            df = pd.read_sql_table(table, conn)
            # Create a separate dir for saving attachments (if it doesn't already exist). Check if it should be created
            # here because the attachment table might exist but the table might be empty
            if len(df) and not os.path.isdir(attachments_dir):
                os.mkdir(attachments_dir)

            for _, row in df.iterrows():
                # Write the attachment with a name that is unique so if another submission comes in with an attachment
                # that has the same name, the original doesn't get written over
                name, extension = os.path.splitext(row.ATT_NAME)
                attachment_id = row.GLOBALID.replace('{', '').replace('}', '') #for some annoying reason, ids have braces
                attachment_path = os.path.join(attachments_dir, '{name}_{id}{ext}'
                                               .format(name=name, id=attachment_id, ext=extension))
                with open(attachment_path, 'wb') as attachment_pointer:
                    attachment_pointer.write(row.DATA)

                # write a JSON file with some metadata so the attachment can be related back to this DB file
                row_dict = row.drop('DATA').to_dict() # drop 'DATA' field because it contains non-serializable bytes
                row_dict['sqlite_path'] = sqlite_path
                json_path = attachment_path.rstrip(extension) + '.json'
                with open(json_path, 'w') as json_pointer:
                    json.dump(row_dict, json_pointer, indent=4)


if __name__ == '__main__':
    sys.exit(download_data(*sys.argv[1:]))‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍

[samhooper · Accepted solution]

Well, it turns out the it was something funky with the feature service itself. The repeat group at one point contained a geopoint question, which turned the associated table into a layer. I just recreated the survey (and the feature service, which now stores the related records in a table), and everything works fine.