Hi Matthew, Further to Cindy's update I've created a separate patch proposal for this issue also. Layer 1, the trigger: the attach waits for a grace period while holding cgroup_mutex. - Code: the proposal-2 patch. The wait moves in front of the mutex, so only the attaching process waits. - Config: favordynmods (boot parameter cgroup_favordynmods=true, or a remount with the full option list). On 7.0 this removes the grace period from the cgroup.procs path entirely. Costs: one extra rwsem read in fork, exec and exit for every process on the host, and it cannot be turned off again without a reboot.
Layer 2, the cause: the grace period takes minutes because rcu_sched and the nocb kthreads can be placed on the FIFO-hogged isolated CPUs. - Code: proposal-1 patch 2, which keeps unbound kthreads off nohz_full CPUs. - Config: add domain to isolcpus, i.e. isolcpus=domain,nohz,18-95,108-178. That excludes the isolated CPUs from the kthread placement mask. A userspace stop-gap is to taskset rcu_sched, rcuog and rcuop to housekeeping CPUs, which the current isolate service does not do. ** Attachment added: "lp2165410-proposal-2.tar.gz" https://bugs.launchpad.net/ubuntu/+source/linux-aws/+bug/2165410/+attachment/6008680/+files/lp2165410-proposal-2.tar.gz -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2165410 Title: Ubuntu LTS 26.04 linux-aws: systemd enters D state and blocks SSH during cpuset migration on nohz_full CPUs To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux-aws/+bug/2165410/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
