Mongod killed by oom_killer

Hello! I have a cluster with the following characteristics:
Hardware: AMD Ryzen 9 7950X3D 16-Core Processor with 128GB of RAM.
cluster configuration: mongodb community edition 4.4.24,
the number of shards is 7, the number of replicas in each shard is 3, the Wired Tiger cache size is 64GB.
problem: During operation, RAM consumption increases randomly, which leads to OOM_kill being triggered and the primary replica of the shard stops. The same thing happens with the secondary replica that won the election. The problematic primary shard contains a database containing only non-sharded collections, and its size is many times smaller than the cache, namely about 5 GB. What can cause such an increase in mapi consumption and how to get rid of it?

Do you see anything interesting from mongodb log messages?

Nothing interesting. There were no mistakes. Just a few slow query.