| author | ms.author | ms.reviewer | ms.date | ms.service | ms.topic |
|---|---|---|---|---|---|
rwestMSFT |
randolphwest |
dfurman |
02/17/2025 |
sql |
include |
Changes an existing resource governor workload group configuration, and optionally assigns it to a different resource governor resource pool.
Note
To modify resource governor configuration in [!INCLUDEssazuremi-md.md], you must be in the context of the master database on the primary replica.
:::image type="icon" source="../../includes/media/topic-link-icon.svg" border="false"::: Transact-SQL syntax conventions.
ALTER WORKLOAD GROUP { group_name | [default] }
[ WITH
([ IMPORTANCE = { LOW | MEDIUM | HIGH } ]
[ [ , ] REQUEST_MAX_MEMORY_GRANT_PERCENT = value ]
[ [ , ] REQUEST_MAX_CPU_TIME_SEC = value ]
[ [ , ] REQUEST_MEMORY_GRANT_TIMEOUT_SEC = value ]
[ [ , ] MAX_DOP = value ]
[ [ , ] GROUP_MAX_REQUESTS = value ] )
]
[ USING { pool_name | [default] } ]
[ ; ]
The name of an existing user-defined workload group or the resource governor built-in default workload group.
default must be in brackets ([]) or quotation marks ("") when used with ALTER WORKLOAD GROUP to avoid a conflict with DEFAULT, which is a system reserved word. For more information, see Database identifiers.
Built-in resource pools and workload groups use all lowercase names, such as default. Use the lower case default on servers that use a case-sensitive collation. Servers with case-insensitive collation treat default, Default, and DEFAULT as the same value.
Specifies the relative importance of a request in the workload group. The default value is MEDIUM.
IMPORTANCE is local to the resource pool that contains the workload group. Workload groups of different importance inside the same resource pool affect each other, but don't affect workload groups in other resource pools.
Specifies the maximum amount of query workspace memory that a single request can take from the pool. value is a percentage of the resource pool size defined by MAX_MEMORY_PERCENT. Default value is 25.
In [!INCLUDE sssql17-md] and older, value is an integer and the allowed range is from 1 through 100.
Starting with [!INCLUDE sssql19-md], the value can be fractional using the float data type. The allowed range is from 0 through 100.
Important
The amount specified only refers to query workspace memory obtained via query memory grants.
It is not recommended to set value too large (for example, greater than 70) because the server may be unable to set aside enough free memory for other concurrent queries. This can lead to a memory grant time out error 8645.
Setting value to 0 or a small value might prevent queries with operators that require workspace memory, such as sort and hash, from running in user-defined workload groups. If the query memory requirements exceed the limit defined by this parameter, the following behavior occurs:
- For user-defined workload groups, the server tries to reduce the degree of parallelism (DOP) of the request (query) until the memory requirement falls under the limit, or until DOP equals 1. If the query memory requirement is still greater than the limit, error 8657 occurs and the query fails.
- For the
internalanddefaultworkload groups, the server permits the query to obtain the required memory.
In either case, error 8645 might occur if the server has insufficient physical memory.
Specifies the maximum amount of CPU time, in seconds, that a batch request can use. value must be 0 or a positive integer. The default setting for value is 0, which means unlimited.
When the maximum CPU time is exceeded, the cpu_threshold_exceeded extended event and a trace event are generated. For more information, see CPU Threshold Exceeded Event Class.
In Azure SQL Managed Instance, when the maximum CPU time is exceeded, resource governor aborts the request with error 10961.
In [!INCLUDE ssnoversion-md], resource governor doesn't abort the request by default. However, starting with [!INCLUDE sssql16-md] SP2 and [!INCLUDE sssql17-md] CU3, resource governor aborts a request with error 10961 when trace flag 2422 is enabled and the maximum CPU time is exceeded.
Specifies the maximum time, in seconds, that a query can wait for a memory grant from the query workspace memory to become available. value must be 0 or a positive integer. The default setting for value, 0, uses an internal calculation based on query cost to determine the maximum time.
A query doesn't always fail when a memory grant time out is reached. A query only fails if there are too many concurrent queries running. Otherwise, the query might only get the minimum memory grant, resulting in reduced query performance.
Specifies the maximum degree of parallelism (MAXDOP) for parallel query execution. The allowed range for value is from 0 through 64. The default setting for value, 0, uses the global setting.
For more information, see MAXDOP.
Specifies the maximum number of simultaneous requests that are allowed to execute in the workload group. value must be 0 or a positive integer. The default setting for value is 0, and allows unlimited requests. When the maximum concurrent requests are reached, a session in that group can be created, but is placed in a wait state until the number of concurrent requests drops below the value specified.
Associates the workload group with the user-defined resource pool identified by pool_name, or with the default resource pool. If pool_name isn't provided, or if the USING argument isn't specified, the workload group is associated with the built-in default pool.
default is a reserved word and when specified in USING, must be enclosed in brackets ([]) or quotation marks ("").
Built-in resource pools and workload groups use all lowercase names, such as default. Use the lower case default on servers that use a case-sensitive collation. Servers with case-insensitive collation treat default, Default, and DEFAULT as the same value.
ALTER WORKLOAD GROUP is allowed on the default workload group, but not on the internal group.
Changes to the workload group configuration don't take effect until after ALTER RESOURCE GOVERNOR RECONFIGURE is executed.
For more information, see Resource governor and Resource governor workload group.
For a given query, effective MAXDOP is determined as follows:
MAXDOPas a query hint is honored as long as it doesn't exceed the workload groupMAX_DOPsetting.MAXDOPas a query hint always overrides themax degree of parallelismserver configuration. For more information, see Server configuration: max degree of parallelism.- Workload group
MAX_DOPoverrides themax degree of parallelismserver configuration and theMAXDOPdatabase scoped configuration.
The MAXDOP limit is set per task. It isn't a per request or per query limit. During an execution of a parallel query, a single request can spawn multiple tasks that are assigned to a scheduler. For more information, see the Thread and task architecture guide.
When a query is marked as serial at compile time (MAXDOP = 1), it can't execute with parallelism at run time regardless of the workload group or server configuration setting. After MAXDOP is determined for a query, it can only be lowered due to memory pressure. Workload group reconfiguration does not affect queries waiting in the memory grant queue.
When you change a plan affecting setting such as MAX_DOP, the new setting takes effect in previously cached plans only after executing DBCC FREEPROCCACHE (<pool_name>), where <pool_name> is the name of a resource governor resource pool used by the current workload group.
- If changing
MAX_DOPto 1, executingDBCC FREEPROCCACHEisn't required because parallel plans can run in serial mode. However, such a plan might be less efficient than a plan compiled as a serial plan. - If changing
MAX_DOPfrom 1 to 0 or a value greater than 1, executingDBCC FREEPROCCACHEisn't required. However, serial plans can't run in parallel, so clearing the respective cache allows new plans to potentially be compiled using parallelism.
Warning
Clearing cached plans from a resource pool that is associated with more than one workload group affects all workload groups using the user-defined resource pool identified by <pool_name>.
For performance reasons, index creation is allowed to use more memory workspace than initially granted. Resource governor supports this special handling. However, the initial grant and any additional memory grants are limited by the workload group and resource pool settings.
The memory consumed to create a nonaligned index on a partitioned table is proportional to the number of partitions involved. If the total required memory exceeds the per-query limit enforced by the REQUEST_MAX_MEMORY_GRANT_PERCENT workload group setting, index creation might fail. Because the default workload group allows a query to exceed the per-query limit with the minimum required memory to start for backward compatibility, you might be able to create the same index using the default workload group if the default resource pool has enough total memory.
Requires the CONTROL SERVER permission.
The following example shows how to change the importance of requests in the default group from MEDIUM to LOW.
ALTER WORKLOAD GROUP [default]
WITH (IMPORTANCE = LOW);
ALTER RESOURCE GOVERNOR RECONFIGURE;The following example shows how to move a workload group from the pool that it's currently into the default pool.
ALTER WORKLOAD GROUP adHoc
USING [default];
ALTER RESOURCE GOVERNOR RECONFIGURE;