Summary
When the configurator_node is running and a CallbackGroup configured with SCHED_DEADLINE + CPU affinity restarts repeatedly (e.g., the target application crashes and relaunches), /sys/fs/cgroup/cpuset/<n> directories grow without bound. They are only cleaned up in the destructor when the configurator node itself shuts down.
Details
set_affinity_by_cgroup() (thread_configurator_node.cpp:308) creates a new cpuset directory each time by monotonically incrementing cgroup_num_++. It never reuses or removes old directories.- When a CallbackGroup's
thread_id changes (i.e., the target app restarted), callback_group_callback() (line 462–474) re-applies the configuration via issue_syscalls() → set_affinity_by_cgroup(), creating a new directory while the old one remains.
- After
on_all_configured() (line 552–566), all applied flags are reset, so the next restart cycle will go through the same path again.
- The destructor (line 277–284) calls
rmdir() for all created directories, but this only runs when the configurator node shuts down — not during runtime.
- In long-running deployments where target applications may restart many times, this leads to an ever-growing number of stale cpuset directories under
/sys/fs/cgroup/cpuset/.
Summary
When the
configurator_nodeis running and a CallbackGroup configured withSCHED_DEADLINE+ CPU affinity restarts repeatedly (e.g., the target application crashes and relaunches),/sys/fs/cgroup/cpuset/<n>directories grow without bound. They are only cleaned up in the destructor when the configurator node itself shuts down.Details
set_affinity_by_cgroup()(thread_configurator_node.cpp:308) creates a new cpuset directory each time by monotonically incrementingcgroup_num_++. It never reuses or removes old directories.thread_idchanges (i.e., the target app restarted),callback_group_callback()(line 462–474) re-applies the configuration viaissue_syscalls()→set_affinity_by_cgroup(), creating a new directory while the old one remains.on_all_configured()(line 552–566), allappliedflags are reset, so the next restart cycle will go through the same path again.rmdir()for all created directories, but this only runs when the configurator node shuts down — not during runtime./sys/fs/cgroup/cpuset/.