How can I use SNS to dynamically fetch data from S3 without restarting containers?

You're good on the subscriber count for SNS; it supports a hefty limit of 12.5 million subscribers per topic. However, it’s crucial that your instances consistently perform as expected. You might want to look into AWS services that ensure your instances remain compliant with the latest updates from S3.

I've had some success with using a similar pattern through Netflix Hollow for Java. The trick is to maintain a reference for updates somewhere robust. I opted for Redis after testing a bit with Zookeeper, which was a bit unpopular. Relying on HTTP callbacks for each container could lead to missed updates if one node fails. Instead, consider implementing a polling mechanism using a database or something like Redis, even with a large number of containers—it’s typically more reliable.

I think your reasoning is spot on! Just dive in and see how it goes; that’s really the best way to assess its viability.

You should set up event notifications for changes in the necessary files. At startup, employ a caching mechanism to prevent flooding S3 with GET requests. You could run into throttling issues if you exceed 5,000 GET requests per second. Another option could be to create a head node responsible for fetching data, while other containers pull updates from it. Just be aware, there are plenty of ways to tackle this!