Code Implementation

Table 1 lists the new classes of the feature.

**Table 1** New classes
Class	Description
Sched_affinity_manager	Scheduling manager interface. The register_thread method is used to register itself with the scheduling manager when a thread starts, and then its scheduling is managed by the scheduling manager. The unregister_thread method is used to deregister a thread from the scheduling manager before the thread is destroyed. The rebalance_group method is used to update the internal status of the scheduling manager and the scheduling status of the existing threads when CPU core parameters are changed. The update_numa_aware method is used to update the internal status of the scheduling manager and the scheduling status of the existing threads when the sched_affinity_numa_aware parameter is changed. The take_group_snapshot method is used to return a snapshot of the internal status of the scheduler manager, in the form of a string. The snapshot can be queried by the user. The get_total_node_number method is used to return the total number of NUMA nodes in the system. The get_cpu_number_per_node method is used to return the number of cores on each NUMA node in the system. The check_cpu_string method is used to check the validity of the input CPU cores.
Sched_affinity_manager_numa	Implements the scheduling manager.
Sched_affinity_manager_dummy	Standby for implementing the scheduling manager. The implementation of all interfaces only returns values that meet the caller's expectation. If Sched_affinity_manager_numa is unavailable (for example, the libnuma dependency does not meet the requirements), Sched_affinity_manager_dummy is enabled.