As with many integrations to Kubernetes, this one also suffers of the issue of node RAM vs container RAM visibility.
With WT/cacheSizeGB we can somehow control the memory consumption but what is with the OS of container (using the file buffer), which seems to see the node’s RAM size. For ex. 64 GB for node and 8GB for the mongod pod. The OS in mongod pod will “see” 64GB available, which most likely will result in a OOM killed event sooner or later.
Is there a workaround for this ?