Replication spike leads to OOM

[roman-khimov]

Expected Behavior

No OOM.

Current Behavior

OOM.

Possible Solution

Optimize memory consumption, throw away requests that can't be processed in time.

Steps to Reproduce (for bugs)

Not really known, but we need a lot of replications (or large objects?) to the same node with limited RAM (like 4G).

Context

Testnet.

Regression

No.

Your Environment

• Version used: 0.52.0
• Server setup and configuration:
• Operating System and version (uname -a):

[roman-khimov]

Solved by less argressive policer settings, these machines can't handle 30 big objects at once and it's expected. However there are still things that can be optimized and we need replication to be more effective.