Resource Management

use_workload_manager

Parameter description: Specifies whether to enable the resource management function. This parameter must be applied on both CNs and DNs.

Type: SIGHUP

Value range: Boolean

• on indicates the resource management function is enabled.
• off indicates the resource management function is disabled.
  

  • If method 2 in Setting GUC Parameters is used to change the parameter value, the new value takes effect only for the threads that are started after the change. In addition, the new value does not take effect for new jobs that are executed by backend threads and reused threads. You can make the new value take effect for these threads by using kill session or restarting the node.
  • After the value of use_workload_manager changes from off to on, the resource management view becomes available, and you can query the storage resource usage collected in the off state. If there are slight errors and the storage resource usage needs to be corrected, run the following command. If data is inserted into the table during the command execution, the statistics may be inaccurate.

    1	select gs_wlm_readjust_user_space(0);

Default value: on

enable_perm_space

Parameter description: Specifies whether to enable the perm space function. This parameter must be applied on both CNs and DNs.

Type: POSTMASTER

Value range: Boolean

• on indicates the perm space function is enabled.
• off indicates the perm space function is disabled.

Default value: on

space_once_adjust_num

Parameter description: In the space control and space statistics functions, specifies the threshold of the number of files processed each time during slow building and fine-grained calibration . This parameter is supported by version 8.1.3 or later clusters.

Type: SIGHUP

Value range: an integer ranging from 0 to INT_MAX

• The value 0 indicates that the slow build and fine-grained calibration functions are disabled.

Default value: 300

The file quantity threshold affects database resources. You are advised to set the threshold to a proper value.

space_readjust_schedule

Parameter description: In the space control and space statistics functions, specifies the space error threshold for triggering automatic calibration. This parameter is supported by version 8.1.3 or later clusters.

Type: SIGHUP

Value range: string

• off indicates that the automatic calibration function is disabled.
• auto indicates that the automatic calibration function is enabled and the error threshold for triggering automatic calibration is 1 GB.
• auto (space size + K/M/G) indicates that the automatic calibration is enabled and the error threshold for triggering automatic calibration is xxx KB/MB/GB (user-defined). For example, auto(200M) indicates that the automatic calibration is enabled and the error threshold for triggering automatic calibration is 200 MB.

Default value: auto

enable_libcomm_schedule

Parameter description: Specifies whether to enable network control. This parameter is supported by version 8.2.1 or later clusters.

Type: POSTMASTER

Value range: Boolean

• on indicates that network control is enabled.

• off indicates that network control is disabled.

Default value: on

max_active_statements

Parameter description: Specifies the maximum global concurrency. This parameter applies to one CN.

The database administrator changes the value of this parameter based on system resources (for example, CPU, I/O, and memory resources) so that the system fully supports the concurrency tasks and avoids too many concurrency tasks resulting in system crash.

Type: SIGHUP

Value range: an integer ranging from –1 to INT_MAX. The values –1 and 0 indicate that the number of concurrent requests is not limited.

Default value: 60

parctl_min_cost

Parameter description: Specifies the minimum estimated cost of a complex job under static resource management. Threshold for dividing simple jobs and complex jobs. A job whose estimated cost is less than the value of this parameter is a simple job, and a job whose estimated cost is larger than or equal to the value of this parameter is a complex job.

Type: SIGHUP

Value range: an integer ranging from –1 to INT_MAX

• If parctl_min_cost is -1, all jobs are simple jobs.
• Jobs whose estimated cost is less than 10 are simple jobs.

Default value: 100000

cgroup_name

Parameter description: Specifies the name of the Cgroup in use. It can be used to change the priorities of jobs in the queue of a Cgroup.

If you set cgroup_name and then session_respool, the Cgroups associated with session_respool take effect. If you reverse the order, Cgroups associated with cgroup_name take effect.

If the Workload Cgroup level is specified during the cgroup_name change, the database does not check the Cgroup level. The level ranges from 1 to 10.

Type: USERSET

You are not advised to set cgroup_name and session_respool at the same time.

Value range: a string

Default value: DefaultClass:Medium

DefaultClass:Medium indicates the Medium Cgroup belonging to the Timeshare Cgroup under the DefaultClass Cgroup.

cpu_collect_timer

Parameter description: Specifies how frequently CPU data is collected during statement execution on DNs.

The database administrator changes the value of this parameter based on system resources (for example, CPU, I/O, and memory resources) so that the system fully supports the concurrency tasks and avoids too many concurrency tasks resulting in system crash.

Type: SIGHUP

Value range: an integer ranging from 1 to INT_MAX. The unit is second.

Default value: 30

enable_cgroup_switch

Parameter description: Specifies whether the database automatically switches to the TopWD group when executing statements by group type.

Type: USERSET

Value range: Boolean

• on: The database automatically switches to the TopWD group when executing statements by group type.
• off: The database does not automatically switch to the TopWD group when executing statements by group type.

Default value: off

memory_tracking_mode

Parameter description: Specifies the memory information recording mode.

Type: USERSET

Value range:

• none: Memory statistics is not collected.
• normal: Only memory statistics is collected in real time and no file is generated.
• executor: The statistics file is generated, containing the context information about all allocated memory used by the execution layer.
• fullexec: The generated file includes the information about all memory contexts requested by the execution layer.

Default value: none

memory_detail_tracking

Parameter description: Specifies the sequence number of the memory background information distributed in the needed thread and plannodeid of the query where the current thread is located.

Type: USERSET

Value range: a string

Default value: empty

It is recommended that you retain the default value for this parameter.

enable_resource_track

Parameter description: Specifies whether the real-time resource monitoring function is enabled. This parameter must be applied on both CNs and DNs.

Type: SIGHUP

Value range: Boolean

• on indicates the resource monitoring function is enabled.
• off indicates the resource monitoring function is disabled.

Default value: on

enable_resource_record

Parameter description: Specifies whether resource monitoring records are archived. When this parameter is enabled, records that have been executed are archived to the corresponding INFO views (GS_WLM_SESSION_INFO and GS_WLM_OPERAROR_INFO). This parameter must be applied on both CNs and DNs.

Type: SIGHUP

Value range: Boolean

• on indicates that the resource monitoring records are archived.
• off indicates that the resource monitoring records are not archived.

Default value: on

The default value of this parameter is on for a new cluster. In upgrade scenarios, the default value of this parameter is the same as that of the source version.

enable_track_record_subsql

Parameter description: Specifies whether to enable the function of recording and archiving sub-statements. When this function is enabled, sub-statements in stored procedures and anonymous blocks are recorded and archived to the corresponding INFO table (GS_WLM_SESSION_INFO). This parameter is a session-level parameter. It can be configured and take effect in the session connected to the CN and affects only the statements in the session. It can also be configured on both the CN and DN and take effect globally.

Type: USERSET

Value range: Boolean

• on indicates that the sub-statement resource monitoring records are archived.
• off indicates that the sub-statement resource monitoring records are not archived.

Default value: on

enable_user_metric_persistent

Parameter description: Specifies whether the user historical resource monitoring dumping function is enabled. When this function is enabled, data in the PG_TOTAL_USER_RESOURCE_INFO view is periodically sampled and saved to the GS_WLM_USER_RESOURCE_HISTORY system catalog, and data in the GS_RESPOOL_RESOURCE_INFO view is periodically sampled and saved to the GS_RESPOOL_RESOURCE_HISTORY system catalog.

Type: SIGHUP

Value range: Boolean

• on indicates that the user historical resource monitoring dumping function is enabled.
• off indicates that the user historical resource monitoring dumping function is disabled.

Default value: on

user_metric_retention_time

Parameter description: Specifies the retention time of the user historical resource monitoring data. This parameter is valid only when enable_user_metric_persistent is set to on.

Type: SIGHUP

Value range: an integer ranging from 0 to 3650. The unit is day.

• If this parameter is set to 0, user historical resource monitoring data is permanently stored.
• If the value is greater than 0, user historical resource monitoring data is stored for the specified number of days.

Default value: 7

enable_instance_metric_persistent

Parameter description: Specifies whether the instance resource monitoring dumping function is enabled. When this function is enabled, the instance monitoring data is saved to the system catalog GS_WLM_INSTANCE_HISTORY.

Type: SIGHUP

Value range: Boolean

• on indicates that the instance resource monitoring dumping function is enabled.
• off: Specifies that the instance resource monitoring dumping function is disabled.

Default value: on

instance_metric_retention_time

Parameter description: Specifies the retention time of the instance historical resource monitoring data. This parameter is valid only when enable_instance_metric_persistent is set to on.

Type: SIGHUP

Value range: an integer ranging from 0 to 3650. The unit is day.

• If this parameter is set to 0, instance historical resource monitoring data is permanently stored.
• If the value is greater than 0, the instance historical resource monitoring data is stored for the specified number of days.

Default value: 7

resource_track_level

Parameter description: Specifies the resource monitoring level of the current session. This parameter is valid only when enable_resource_track is set to on.

Type: USERSET

Value range: enumerated values

• none: Resources are not monitored.
• query: enables query-level resource monitoring. If this function is enabled, the plan information (similar to the output information of EXPLAIN) of SQL statements will be recorded in top SQL statements.
• perf: enables the perf-level resource monitoring. If this function is enabled, the plan information (similar to the output information of EXPLAIN ANALYZE) that contains the actual execution time and the number of execution rows will be recorded in the top SQL.
• operator: enables the operator-level resource monitoring. If this function is enabled, not only the information including the actual execution time and number of execution rows is recorded in the top SQL statement, but also the operator-level execution information is updated to the top SQL statement.

Default value: query

time_track_strategy

Parameter description: Specifies the policy used to collect the operator execution time of the current session. This parameter is supported by version 8.2.1 or later clusters.

Type: USERSET

Value range: enumerated values

• tsc: Use Time-Stamp Counter (TSC) to collect the operator execution time. This method is applicable to perf-level top SQL statements and EXPLAIN and applies only to non-vectorized operators. In other scenarios, the time function is still used.
• vector: Disable the collection of the execution time of the non-vectorized operators in the top SQL statements at the perf level. Other scenarios use the time function perform collection and are not affected.
• timer: The time function used in all scenarios to collect the operator execution time. In cluster 8.2.0 and earlier versions, only this method is used.
• opt: The database prioritizes selecting TSC for operator self-timing collection based on node conditions. If TSC is not available, the default time function is used for time collection. This policy is supported only by 8.2.1.230 clusters.

Default value: timer

• The TSC has two methods of converting the time, including the TSC frequency and TSC conversion factors. By default, only the TSC frequency can be used on the x86 platform, and the TSC conversion factor is prioritized on the ARM platform. You can view the TSC conversion information for the current or all nodes through TSC-related views or functions.

• In a cluster installation scenario, the default value of this parameter is tsc. In an upgrade scenario, the default value of this parameter is timer to ensure forward compatibility.

resource_track_cost

Parameter description: Specifies the minimum execution cost for resource monitoring on statements in the current session. This parameter is valid only when enable_resource_track is set to on.

Type: USERSET

Value range: an integer ranging from –1 to INT_MAX

• –1 indicates that resource monitoring is disabled.
• A value greater than or equal to 0 indicates that statements whose execution cost exceeds this value will be monitored.

Default value: 0

The default value of this parameter is 0 for a new cluster. In upgrade scenarios, the default value of this parameter is the same as that of the source version.

resource_track_duration

Parameter description: Specifies the minimum statement execution time that determines whether information about jobs of a statement recorded in the real-time view (see Table 1) will be dumped to a historical view after the statement is executed. Job information will be dumped from the real-time view (with the suffix statistics) to a historical view (with the suffix history) if the statement execution time is no less than this value. This parameter is valid only when enable_resource_track is set to on.

Type: USERSET

Value range: an integer ranging from 0 to INT_MAX. The unit is second (s).

• 0 indicates that information about all statements recorded in the real-time resource monitoring view (see Table 1) will be archived into historical views.
• If the value is greater than 0, information about statements recorded in the real-time resource monitoring view (see Table 1), whose execution time exceeds this value will be archived into historical views.

Default value: 60s

resource_track_subsql_duration

Parameter description: Filters the minimum execution time of substatements in a stored procedure. This parameter is supported by version 8.2.1 or later clusters.

If the execution time of a sub-statement in a stored procedure is greater than the value of this parameter, the job information is archived to the Top SQL table. This parameter takes effect only when enable_track_record_subsql is set to on.

Type: USERSET

Value range: an integer ranging from 0 to INT_MAX. The unit is second (s).

• If the value is 0, historical information about all substatements in the stored procedure is archived.
• If the value is greater than 0, historical information is archived when the execution time of a substatement in a stored procedure exceeds the value of this parameter.

Default value: 180s

query_exception_count_limit

Parameter description: Specifies the maximum number of times that a job triggers an exception rule. If the number of times that a job triggers an exception rule reaches the upper limit, the job will be automatically added to the blocklist and cannot be executed anymore. The job can be resumed only after it is removed from the blocklist.

Type: USERSET

Value range: an integer ranging from –1 to INT_MAX

• If the value is -1, the number of times that a job triggers an exception rule is not limited. That is, the job will not be automatically added to blocklist even if it triggers an exception rule for many times.
• If the value is greater than or equal to 0, the job will be added to the blocklist immediately when the number of times it triggers an exception rule reaches the threshold. The values 0 and 1 both indicate that a job is added to blocklist once the job triggers an exception rule.

Default value: –1

dynamic_memory_quota

Parameter description: Specifies the memory quota in adaptive load scenarios, that is, the proportion of maximum available memory to total system memory.

Type: SIGHUP

Value range: an integer ranging from 1 to 100

Default value: 80

disable_memory_protect

Parameter description: Stops memory protection. To query system views when system memory is insufficient, set this parameter to on to stop memory protection. This parameter is used only to diagnose and debug the system when system memory is insufficient. Set it to off in other scenarios.

Type: USERSET

Value range: Boolean

• on indicates that memory protection stops.
• off indicates that memory is protected.

Default value: off

query_band

Parameter description: Specifies the job type of the current session.

Type: USERSET

Value range: a string

Default value: empty

enable_dynamic_workload

Parameter description: Specifies whether to enable the dynamic workload management function.

Type: POSTMASTER

Value range: Boolean

• on indicates the dynamic workload management function is enabled.
• off indicates the dynamic workload management function is disabled.

Default value: on

• If memory adaptation is enabled, you do not need to use work_mem to optimize the operator memory usage after collecting statistics. The system will generate a plan for each statement based on the current load, estimating the memory used by each operator and by the entire statement. In a concurrency scenario, statements are queued based on the system load and their memory usage.
• The optimizer may not accurately estimate the number of rows and will probably underestimate or overestimate memory usage. If the memory usage is underestimated, the allocated memory will be automatically increased during statement running. If the memory usage is overestimated, system resources will not be fully used, and the number of statements waiting in a queue will increase, which probably results in low performance. To improve performance, identify the statements whose estimated memory usage is much greater than the DN peak memory and adjust the value of query_max_mem. For details, see Adjusting Key Parameters During SQL Tuning.
• As the memory estimated by the optimizer may be inaccurate, in cluster versions earlier than 8.2.1, the enable_dynamic_workload parameter needs to be disabled to prevent CCN global queuing. This will cause the dynamic load management function to be unavailable. Therefore, enable_global_memctl is introduced in 8.2.1. When a CCN exception occurs, you can disable the enable_global_memctl parameter so that jobs can be delivered to and run in the resource pool.

enable_global_memctl

Parameter description: Specifies whether to enable the global memory management function. This parameter is supported by version 8.2.1 or later clusters.

Type: SIGHUP

Value range: Boolean

• on indicates that the global memory management is enabled.
• off indicates that global memory management is disabled.

Default value: on

The dynamic load function consists of two layers of memory management: global memory management and resource pool management. Global memory management determines whether a job can be delivered based on its estimated memory. Resource pool management determines whether a job can be delivered based on resource pool parameters. In versions earlier than 8.2.1, the global memory management function is enabled by default after the dynamic load management function is enabled. The statement memory usage may be underestimated or overestimated by the optimizer. As a result, jobs are queued in the global memory management queue on the CCN. In GaussDB 8.2.1, this parameter is used to control whether to enable the global memory management to improve job efficiency and reduce CCN queue exceptions.

enable_wlm_internal_memory_limit

Parameter description: Specifies whether to enable the built-in limit on estimated statement memory usage in load management. (This parameter is supported by version 8.2.0 or later clusters.)

In the memory management module of load management, some built-in restrictions are imposed on the estimated memory of statements. For example:

• The estimated memory of statements cannot exceed 80% of the memory upper limit of the associated resource pool.
• If the concurrency control parameter active_statements of the resource pool is not set to 1, the estimated memory of the statement cannot exceed 40% of the memory upper limit of the associated resource pool.
• During the estimation of statement memory usage, a range is provided first. The maximum value indicates the memory required for optimal statement running performance. The minimum value indicates the memory required for statement running when data spilling is allowed. The final estimation will be within this range. The maximum estimated memory cannot exceed 90% of the memory upper limit of the associated resource pool.

These built-in restrictions can prevent overestimation of statement memory. If memory usage is overestimated, statements will preoccupy too much memory, causing subsequent jobs to queue and affecting resource utilization. To avoid such problems, the kernel limits the estimated memory usage of a single statement. Execution plans under the built-in restrictions may not be optimal, and may affect the performance of a statement. The memory negative feedback mechanism is provided in 8.2.0 and later cluster versions to alleviate this problem. The enable_wlm_internal_memory_limit parameter is added in 8.2.0 and later versions. You can determine whether to enable the built-in restrictions.

Type: SIGHUP

Value range: Boolean

• on indicates that the built-in restrictions on statement memory estimation are enabled.
• off indicates that the built-in restrictions on statement memory estimation are disabled.

Default value: on

enable_strict_memory_expansion

Parameter description: Specifies whether to enable strict control over the increase of statement memory usage. (This parameter is supported by version 8.2.0 or later clusters.)

The CN calculates the estimated memory for a statement and preempts memory accordingly. If there is sufficient memory, the DN can increase the memory used for a statement to facilitate its execution. If this parameter is enabled, the increase of memory usage for a statement will be strictly controlled. The memory usage of a statement will not be allowed to exceed its estimated maximum usage. The memory usage of an operator is increased proportionally each time, so the memory usage after an increase may exceed the allowed maximum, but to a limited extent.

Type: SIGHUP

Value range: Boolean

• on indicates that strict control over statement memory usage is enabled.
• off indicates that strict control over statement memory usage is disabled.

Default value: off

allow_zero_estimate_memory

Parameter description: Specifies whether the estimated memory usage of a statement can be 0. (This parameter is supported by version 8.2.0 or later clusters.)

If the table queried by a statement does not contain statistics, the estimated memory of the statement on the CN may be 0. In this case, the memory usage of operators in the statement is limited by work_mem. (work_mem is not recommended for operator memory usage control). If work_mem is large and there are many operators in a statement, the actual memory of the statement will be large. If this parameter is set to off, the estimated memory usage cannot be 0 for queries that have not been analyzed. This setting can help reduce unexpected problems.

Type: SIGHUP

Value range: Boolean

• on indicates that the estimated memory usage of a statement can be 0.
• off indicates that the estimated memory usage of a statement cannot be 0.

Default value: on

wlm_memory_feedback_adjust

Parameter description: Specifies whether to enable memory negative feedback in dynamic load management. (This parameter is supported by version 8.2.0 or later clusters.)

Memory is preempted based on the estimated statement memory usage calculated on the CN. If the estimated memory usage of a statement is too high, it will preempt too much memory, causing subsequent jobs to be queued. With the negative memory feedback mechanism, if the cluster memory usage has been overestimated for a period of time, the CCN node will dynamically release some memory for subsequent jobs, improving resource utilization.

Type: SIGHUP

Value range: string

• on indicates that memory negative feedback is enabled.
• off indicates that memory negative feedback is disabled.
• on(Time_required_for_triggering_negative_feedback, Estimated_memory_percentage_required_for_triggering_negative_feedback) indicates that memory negative feedback is enabled. For example, on(60,50) indicates that to trigger the negative feedback mechanism, the memory must be overestimated for 60 consecutive seconds, and the preempted memory needs must exceed 50% of the available memory. By default, the wait time before the negative feedback mechanism takes effect is 50 seconds. The minimum estimated total memory usage for triggering the mechanism is over 40% of the available system memory.

Default value: on

bbox_dump_count

Parameter description: Specifies the maximum number of core files that are generated by GaussDB(DWS) and can be stored in the path specified by bbox_dump_path. If the number of core files exceeds this value, old core files will be deleted. This parameter is valid only if enable_bbox_dump is set to on.

Type: USERSET

Value range: an integer ranging from 1 to 20

Default value: 8

When core files are generated during concurrent SQL statement execution, the number of files may be larger than the value of bbox_dump_count.

io_limits

Parameter description: This parameter has been discarded in version 8.1.2 and is reserved for compatibility with earlier versions. This parameter is invalid in the current version.

Type: USERSET

Value range: an integer ranging from 0 to 1073741823

Default value: 0

io_priority

Parameter description: This parameter has been discarded in version 8.1.2 and is reserved for compatibility with earlier versions. This parameter is invalid in the current version.

Type: USERSET

Value range: enumerated values

• None
• Low
• Medium
• High

Default value: None

session_respool

Parameter description: Specifies the resource pool associated with the current session.

Type: USERSET

If you set cgroup_name and then session_respool, the Cgroups associated with session_respool take effect. If you reverse the order, Cgroups associated with cgroup_name take effect.

If the Workload Cgroup level is specified during the cgroup_name change, the database does not check the Cgroup level. The level ranges from 1 to 10.

You are not advised to set cgroup_name and session_respool at the same time.

Value range: a string. This parameter can be set to the resource pool configured through create resource pool.

Default value: invalid_pool

enable_transaction_parctl

Parameter description: whether to control transaction block statements and stored procedure statements.

Type: USERSET

Value range: Boolean

• on: Transaction block statements and stored procedure statements are controlled.
• off: Transaction block statements and stored procedure statements are not controlled.

Default value: on

session_history_memory

Parameter description: Specifies the memory size of a historical query view.

Type: SIGHUP

Value range: an integer ranging from 10 MB to 50% of max_process_memory

Default value: 100MB

topsql_retention_time

Parameter description: Specifies the retention period of historical Top SQL data in the gs_wlm_session_info and gs_wlm_operator_info tables.

Type: SIGHUP

Value range: an integer ranging from 0 to 3650. The unit is day.

• If it is set to 0, the data is stored permanently.
• If the value is greater than 0, the data is stored for the specified number of days.

Default value: 30

• Before setting this GUC parameter to enable the data retention function, delete data from the gs_wlm_session_info and gs_wlm_operator_info tables.
• The default value of this parameter is 30 for a new cluster. In upgrade scenarios, the default value of this parameter is the same as that of the source version.

transaction_pending_time

Parameter description: maximum queuing time of transaction block statements and stored procedure statements if enable_transaction_parctl is set to on.

Type: USERSET

Value range: an integer ranging from –1 to INT_MAX. The unit is second (s).

• –1 or 0: No queuing timeout is specified for transaction block statements and stored procedure statements. The statements can be executed when resources are available.
• Value greater than 0: If transaction block statements and stored procedure statements have been queued for a time longer than the specified value, they are forcibly executed regardless of the current resource situation.

Default value: 0

This parameter is valid only for internal statements of stored procedures and transaction blocks. That is, this parameter takes effect only for the statements whose enqueue value (for details, see PG_SESSION_WLMSTAT) is Transaction or StoredProc.

wlm_sql_allow_list

Parameter description: Specifies whitelisted SQL statements for resource management. Whitelisted SQL statements are not monitored by resource management.

Type: SIGHUP

Value range: a string, which contains a maximum of 1,024 characters

Default value: empty

• One or more whitelisted SQL statements can be specified in wlm_sql_allow_list. If multiple SQL statements are to be whitelisted, use semicolons (;) to separate them.
• The system determines whether SQL statements are monitored based on the prefix match. The SQL statements are case insensitive. For example, if wlm_sql_allow_list is set to 'SELECT', all SELECT statements are not monitored by the resource management module.
• The system identifies spaces at the beginning of the parameter value. For example, 'SELECT' and ' SELECT' have different representations. ' SELECT' filters only the SELECT statements with spaces at the beginning.
• The system has some whitelisted SQL statements by default, which cannot be modified. You can query the default whitelisted SQL statements and the SQL statements that have been successfully added to the whitelist by GUC through the system view gs_wlm_sql_allow.
• New SQL statements cannot be appended to the whitelisted SQL statements specified by wlm_sql_allow_list but can be set only through overwriting. To add an SQL statement, query the original GUC value, add the new statement to the end of the original value, separate the statements with a semicolon (;), and set the GUC value again.