On Mon, 2024-02-05 at 21:30 +0100, Vladislav Bogdanov wrote: > IIRC, there is one issue with that, is that IO load is considered a > CPU load, so on busy storage servers you get throttling with almost > free CPU. I may be wrong that load is calculated from loadavg, which
Yep, it checks the 1-minute average from /proc/loadavg (it also checks the CIB manager separately using utime/stime from /proc/PID/stat) > is a different story at all, as it indicates the number of processes > which are ready to consume the CPU time, including those waiting for > IOs to complete, but that is what my mind recalls. > > I easily get loadavg of 128 on iscsi storage servers with almost free > CPU, no thermal reaction at all. > > Best, > Vlad > > On February 5, 2024 19:22:11 Ken Gaillot <kgail...@redhat.com> wrote: > > > On Mon, 2024-02-05 at 18:08 +0800, hywang via Users wrote: > > > hello, everyone: > > > Is there any way to disable pacemaker throttle mode. If there is, > > > where to find it? > > > Thanks! > > > > > > > > > > You can influence it via the load-threshold and node-action-limit > > cluster options. > > > > The cluster throttles when CPU usage approaches load-threshold > > (defaulting to 80%), and limits the number of simultaneous actions > > on a > > node to node-action-limit (defaulting to twice the number of > > cores). > > > > The node action limit can be overridden per node by setting the > > PCMK_node_action_limit environment variable (typically in > > /etc/sysconfig/pacemaker, /etc/default/pacemaker, etc. depending on > > distro). > > -- > > Ken Gaillot <kgail...@redhat.com> > > > > _______________________________________________ > > Manage your subscription: > > https://lists.clusterlabs.org/mailman/listinfo/users > > > > ClusterLabs home: https://www.clusterlabs.org/ > > > > -- Ken Gaillot <kgail...@redhat.com> _______________________________________________ Manage your subscription: https://lists.clusterlabs.org/mailman/listinfo/users ClusterLabs home: https://www.clusterlabs.org/
