I've got a linux 6.6.67 host with ZFS filesystem. The ZFS is known for their cache (ARC) to be reported as memory used by applications. However it behaves much like the linux page cache - i.e. it is freed when memory pressure from applications increases.
This leads to a situation when vm.memory.size[pavailable] is very low, however the performance of the host is quite normal and there is nothing to worry about. I monitor the size of the ARC instead and when it goes close to a minimum value I have a trigger.
I want to change the logic how the vm.memory.size is determined - i.e. I want to subtract from vm.memory.size[used] the current size of zfs.arcstats[size] and add zfs.arcstats[c_min] from a ZFS monitoring template. This metric will show the real memory utilization of the host running ZFS filesystem.
I couldn't find the way to implement it in the docs and would appreciate the advice here.
This leads to a situation when vm.memory.size[pavailable] is very low, however the performance of the host is quite normal and there is nothing to worry about. I monitor the size of the ARC instead and when it goes close to a minimum value I have a trigger.
I want to change the logic how the vm.memory.size is determined - i.e. I want to subtract from vm.memory.size[used] the current size of zfs.arcstats[size] and add zfs.arcstats[c_min] from a ZFS monitoring template. This metric will show the real memory utilization of the host running ZFS filesystem.
I couldn't find the way to implement it in the docs and would appreciate the advice here.
Comment