Thread Pool

enable_thread_pool

Parameter description: Specifies whether to enable the thread pool function. This parameter is a POSTMASTER parameter. Set it based on instructions provided in Table 1.

Value range: Boolean

• on indicates that the thread pool function is enabled.
• off indicates that the thread pool function is disabled.

Default value: on

thread_pool_attr

Parameter description: Specifies the detailed attributes of the thread pool function. This parameter is valid only when enable_thread_pool is set to on. Only the sysadmin user can access this parameter. This parameter is a POSTMASTER parameter. Set it based on instructions provided in Table 1.

Value range: a string, consisting of one or more characters.

This parameter consists of three parts: thread_num, group_num, and cpubind_info. The meanings of the three parts are as follows:

• thread_num indicates the total number of threads in the thread pool. The value ranges from 0 to 4096. The value 0 indicates that the database automatically configures the number of threads in the thread pool based on the number of CPU cores. If the value is greater than 0, the number of threads in the thread pool is the same as the value of thread_num. You are advised to set the thread pool size based on the hardware configuration. The formula is as follows: Value of thread_num = Number of CPU cores x 3–5. The maximum value of thread_num is 4096.
• group_num indicates the number of thread groups in the thread pool. The value ranges from 0 to 64. The value 0 indicates that the database automatically configures the number of thread groups in the thread pool based on the number of NUMA groups. If the value is greater than 0, the number of thread groups in the thread pool is the same as the value of group_num.
• cpubind_info indicates whether the thread pool is bound to a core. The available configuration modes are as follows: 1. '(nobind)': The thread is not bound to a core. 2. '(allbind)': Use all CPU cores that can be queried in the current system to bind threads. 3. '(nodebind: 1, 2)': Use the CPU cores in NUMA groups 1 and 2 to bind threads. 4. '(cpubind: 0-30)': Use CPU cores 0 to 30 to bind threads. 5. '(numabind: 0-30)': Use CPU cores 0 to 30 in the NUMA group to bind threads. This parameter is case-insensitive.

Default value:

• Independent deployment: '1024,2,(nobind)' (60-core CPU/480 GB memory and 32-core CPU/256 GB memory); '512,2,(nobind)' (16-core CPU/128 GB memory); '256,2,(nobind)'(8-core CPU/64 GB memory); '128,2,(nobind)' (4-core CPU/32 GB memory); '64,2,(nobind)' (4-core CPU/16 GB memory)

thread_pool_stream_attr

Parameter description: Specifies the detailed attributes of the stream thread pool function. This parameter is valid only when enable_thread_pool is set to on and only takes effect on DNs. Only the sysadmin user can access this parameter. This parameter is a POSTMASTER parameter. Set it based on instructions provided in Table 1.

Value range: a string, consisting of one or more characters.

This parameter consists of four parts: 'stream_thread_num, stream_proc_ratio ,group_num ,cpubind_info'. The meanings of the four parts are as follows:

• stream_thread_num indicates the total number of threads in the stream thread pool. The value ranges from 0 to 4096. The value 0 indicates that the database automatically configures the number of threads in the thread pool based on the number of CPU cores. If the value is greater than 0, the number of threads in the thread pool is the same as the value of stream_thread_num. You are advised to set the thread pool size based on the hardware configuration. The formula is as follows: Value of stream_thread_num = Number of CPU cores x 3–5. The maximum value of stream_thread_num is 4096.
• stream_proc_ratio indicates the ratio of proc resources reserved for stream threads. The value is a floating point number. The default value is 0.2. The reserved proc resources are calculated as follows: Value of stream_proc_ratio x Value of stream_thread_num.
• group_num indicates the number of thread groups in the thread pool. The value ranges from 0 to 64. The value 0 indicates that the database automatically configures the number of thread groups in the thread pool based on the number of NUMA groups. If the value is greater than 0, the number of thread groups in the thread pool is the same as the value of group_num. The value of group_num in thread_pool_stream_attr must be the same as that in thread_pool_attr. If they are set to different values, the value of group_num in thread_pool_attr is used.
• cpubind_info indicates whether the thread pool is bound to a core. The available configuration modes are as follows: 1. '(nobind)': The thread is not bound to a core. 2. '(allbind)': Use all CPU cores that can be queried in the current system to bind threads. 3. '(nodebind: 1, 2)': Use the CPU cores in NUMA groups 1 and 2 to bind threads. 4. '(cpubind: 0-30)': Use CPU cores 0 to 30 to bind threads. 5. '(numabind: 0-30)': Use CPU cores 0 to 30 in the NUMA group to bind threads. This parameter is case-insensitive. The value of cpubind_info in thread_pool_stream_attr must be the same as that in thread_pool_attr. If they are set to different values, the value of cpubind_info in thread_pool_attr is used.

Default value:

stream_thread_num: 16

stream_proc_ratio: 0.2

group_num and cpubind_info: For details, see thread_pool_attr.

resilience_threadpool_reject_cond

Parameter description: Specifies the percentage of accumulated sessions in the thread pool for escape from overload. This parameter takes effect only when the GUC parameters enable_thread_pool and use_workload_manager are enabled. This parameter is a SIGHUP parameter. Set it based on instructions provided in Table 1.

Value range: a string, consisting of one or more characters.

This parameter consists of recover_threadpool_percent and overload_threadpool_percent. The meanings of the two parts are as follows:

• recover_threadpool_percent: Percentage of the number of sessions that are recovered to the normal state in the initial number of threads in the thread pool. When the number of accessed sessions is less than the initial number of threads in the thread pool multiplied by the value of this parameter, the escape from overload function is disabled and new connections are allowed. The value ranges from 0 to INT_MAX. The value indicates a percentage.
• overload_threadpool_percent: Percentage of the number of accessed sessions to the initial number of threads in the thread pool when the thread pool is overloaded. If the number of accessed sessions is greater than the initial number of threads in the thread pool multiplied by the value of this parameter, the current thread pool is overloaded. In this case, the escape from overload function is enabled to kill sessions and forbid new connections to access the thread pool. The value ranges from 0 to INT_MAX. The value indicates a percentage.

Default value: '0,0', indicating that the thread pool escape function is disabled.

Example:

resilience_threadpool_reject_cond = '100,200'

When the number of stacked sessions exceeds 200% of the initial number of threads in the thread pool, new connections are forbidden and stacked sessions are killed. When the number of stacked sessions is less than 100% of the initial number of threads in the thread pool, new connections are allowed.

• The number of stacked sessions can be obtained by querying the number of data records in the pg_stat_activity view. A few backend threads need to be filtered out. The initial number of threads in the thread pool can be obtained by querying the thread_pool_attr parameter.
• If this parameter is set to a small value, the thread pool escape from overload process is frequently triggered. As a result, ongoing sessions are forcibly logged out, and new connections fail to be connected for a short period of time. Therefore, exercise caution when setting this parameter based on the actual thread pool usage.
• If the use_workload_manager parameter is disabled and the bypass_workload_manager parameter is enabled, this parameter also takes effect. The bypass_workload_manager parameter is of the SIGHUP type; therefore, after the reload mode is set, you need to restart the database for the setting to take effect.
• The values of recover_threadpool_percent and overload_threadpool_percent can be 0 at the same time. In addition, the value of recover_threadpool_percent must be smaller than that of overload_threadpool_percent. Otherwise, the setting does not take effect.