Hi, I'm still working on vm balancing/scheduling, and need some new metrics.
As suggested by Thomas, instead sending values to new rrd, and then reparse rrd later in balancer, we broadcast: - last minute && last 5 min average cpu/ram of qemu/lxc/node - pressure average of cpu/mem/io for qemu/lxc/node Theses values are broadcast each minute in 1 kvstore key "balancer-stats" Size of broadcasted "balancer-stats" is around : 500bytes * number vms. Size of in memory hash for average values is around : 6k (host+mem) * number vms. (for last minute, we keep each X seconds iteration, for last 5min we keep last 5x1min average) - last ksm value This value is broadcast each iteration is a separate kvstore key "ksm" pve-common: Alexandre Derumier (2): cgroup: get_pressure_stat: use controllers in get_path cgroup: get_pressure_stat: add cpu full pressure src/PVE/CGroup.pm | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) qemu-server: Alexandre Derumier (3): vmstatus: add hostcpu value vmstatus: add hostmem value vmstatus: add pressure stats PVE/QemuServer.pm | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) pve-container: Alexandre Derumier (1): vmstatus: add pressure stats src/PVE/LXC.pm | 2 ++ 1 file changed, 2 insertions(+) Alexandre Derumier (4): pvestatd: add broadcast_balancer_stats pvestatd: qemu/lxc/node : add pressure stats pvestatd: qemu/lxc/node : add hostcpu/hostmem average stats pvestatd: node : broadcast ksm value PVE/Service/pvestatd.pm | 154 +++++++++++++++++++++++++++++++++++++++- 1 file changed, 153 insertions(+), 1 deletion(-) -- 2.30.2 _______________________________________________ pve-devel mailing list pve-devel@lists.proxmox.com https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
