We are using polkadot-v0.9.37 and we are experiencing some memory-leak-like issue on the polkadot/kusama nodes. 
We also host polkadot/kusama nodes on other providers (for example AWS), but it did not seems to have the same pattern.
On GCP we used n1-standard-16 (16 vCPU 60GB Mem) for hosting.
Here's our launch parameters
--base-path=/chain-data
--rpc-cors=all
--port=30333
--rpc-port=9933
--ws-port=9944
--ws-external
--rpc-external
--node-key=key_here
--pruning=archive
--rpc-methods=Safe
--name=name_here
--chain=kusama
--db-cache=3500
--in-peers=10000
--out-peers=50
Please kindly advise if there's any fix for this issue. Thank you!
