This message was deleted.

The Memory consumption of the Longhorn pods that manage nodes is quite stable and it's not high. CPU is more relevant in terms of sizing for Longhorn. It's more likely that other workloads are the ones exhausting the RAM. The problem is that once the RAM is exhausted the OS will launch OOM Killer and it'll kill the process without caring if they are pods from Longhorn or any other workload. Enabling monitoring is a key best practice for production so you can have alerts if your worker nodes are running out of RAM and it'll also help you identify which workloads are consuming a lot of memory. You can also have a look at the comments here as it gives you some clues to have more control over those situations: https://github.com/longhorn/longhorn/issues/1487