95% memory usage scyllaDB even when idle

ScyllaDB
1

I have 3 node scylla cluster with 8 core , 32 Gigs RAM.

I am trying to generate data for some testing. Upon running my data generation scripts, memory shot up to 95%. Even after I have stopped the script, memory is still at 95% even when idle.

I am assuming that somewhere data is cached by scylla.

How to debug this further and how to bring down the memory?

If it is indeed being cached, what is the impact of such high memory usage?

2

TL;DR: it’s a feature, not a bug

ScyllaDB is designed to use as much of each node’s resources, such as disk, memory, and network, as possible.
For RAM, it will allocate most of it, splitting the space internally between cache, memetable, etc.

Cache data will be served Reads faster and reduce storage access.

1 Like
3

@tzach Thanks for the answer.

Questions-

  1. In Cassandra, the behavior is not the same. Running the same workload, Cassandra memory usage is lower.
  2. How to limit the memory usage ?
  3. What data from cache is evicted to maintain 95% mem usage ?
4
  1. yes, ScyllaDB has a different architecture, which maximizes the node resources. This is one of the reasons for ScyllaDB performance
  2. Why do you want to limit the memory in the first place?
    ScyllaDB assumes it’s the only application running on a node; if that’s the case, there is no reason to do that.
    If you do choose to limit the memory, see here: Scylla Memory Usage | ScyllaDB Docs
  3. cache is evicted based on last usage