Heroku memory problems using puma

Question

I checked my logs and since I started using puma (switched from a unicorn that did not have this problem) as my web server on the hero. I have a memory leak problem.

The server itself is inactive and there are no requests in the logs, but memory usage on web dynamos continues to grow to the limit, and then with excess. Any ideas or suggestions on how to understand this?

Answer 1

In the end, I had to switch to the dyno type (Performance Large) with a large amount of RAM to accommodate the memory caching that Ruby / Rails did. I could not find a way to stop it from a maximum of about 2.5 GB, but after that it really leveled off.

Answer 2

I cannot give an answer, but I am investigating the same problem. So far, the following two links have been the most educational for me: https://github.com/puma/puma/issues/342 . A possible workaround (although presumably not tested for Heroku production) is to use the puma-worker-killer gem: https://github.com/schneems/puma_worker_killer . Hope this helps.