Consuming long running tasks using pika BlockingConnection on EC2 - amazon-ec2

Question

I have this queue system where I use pika's BlockingConnection to consume a rabbitmq queue hosted on https://www.cloudamqp.com/.
Here's how the connection is setup:
params = pika.URLParameters(self.queue_url)
self.connection = pika.BlockingConnection(params)
self.channel = self.connection.channel()
self.channel.queue_declare(queue=self.queue_name, durable=True)
self.channel.basic_qos(prefetch_count=1)
# set up subscription on the queue
self.channel.basic_consume(self._on_new_task,
queue=self.queue_name,
no_ack=False)
self.channel.start_consuming()
def _on_new_task(self, ch, method, properties, body):
logging.info("Processing new scan")
"""
Do long running task here
"""
self.channel.basic_ack(delivery_tag=method.delivery_tag)
# Program crashes here
Processing a task from the queue can take up to 15 minutes. Most of the time, however, processing the task only takes 2-3 minutes, and then everything works fine. When it takes > 5 minutes I get the following errors from pika:
ERROR - pika.adapters.base_connection - Read empty data, calling disconnect
WARNING - pika.adapters.base_connection - Socket closed when connection was open
WARNING - pika.connection - Disconnected from RabbitMQ at squirrel.rmq.cloudamqp.com:5672 (0): Not specified
CRITICAL - pika.adapters.blocking_connection - Connection close detected
appearantly the basic_ack is not handled successfully because the task still remains on the task queue, leading to an infinite loop where the task is never released from the queue.
I've tried to adjust the heartbeat, both to 0, 30 600 etc, but it looks like heartbeats does not work with a blocking connection, since the long running task is blocking the connection to send heartbeats (?)
The program runs inside a docker container. And I get the problem when hosting the container on AWS ECS2 but not when running the container on my home PC. It therefore seems to me that ECS2 is taking down idle connections (?).
This assumption is backed up by the fact that when running
for i in range(1,10):
self.connection.sleep(60)
self.connection.process_data_events()
instead of the long running task, everything works out fine, probably because the heartbeats are sent and keeps the connection alive.
Not sure how to solve this. Is there any point in replacing pika's BlockingConnection with SelectConnection? Or is there any configuration in EC2/ECS2 that does not tear down idle connections?