This article describes the tempdb system database, a global resource available to all users connected to a Database Engine instance in SQL Server, Azure SQL Database, or Azure SQL Managed Instance.
Overview
The tempdb system database is a global resource that holds:
User objects that are explicitly created. They include:
Global or local temporary tables and indexes on these tables
Temporary stored procedures
Table variables
Tables returned in table-valued functions
Cursors
User objects that can be created in a user database can also be created in tempdb, however they are created without a durability guarantee, and are dropped when the Database Engine instance restarts.
Internal objects that the database engine creates. They include:
Work tables to store intermediate results for spools, cursors, sorts, and temporary large object (LOB) storage.
Work files for hash join or hash aggregate operations.
Intermediate sort results for operations such as creating or rebuilding indexes (if SORT_IN_TEMPDB is specified), or certain GROUP BY, ORDER BY, or UNION queries.
Each internal object uses a minimum of nine pages: an IAM page and an eight-page extent. For more information about pages and extents, see Pages and extents.
Version stores, which are collections of data pages that hold the data rows that support row versioning. There are two types: a common version store and an online index build version store. The version stores contain:
Row versions that are generated by data modification transactions in a database that uses row versioning-based READ COMMITTED or SNAPSHOT isolation transactions.
Row versions that are generated by data modification transactions for features, such as online index operations, Multiple Active Result Sets (MARS), and AFTER triggers.
Operations within tempdb are minimally logged. tempdb is re-created every time the Database Engine is started so that the system always starts with an empty tempdb database. Temporary stored procedures and local temporary tables are dropped automatically when the session that created them disconnects.
tempdb never has anything to be saved from one uptime period of the Database Engine to another. Backup and restore operations are not allowed on tempdb.
Physical properties of tempdb in SQL Server
The following table lists the initial configuration values of the tempdb data and log files in SQL Server. The values are based on the defaults for the model database. The sizes of these files might vary slightly for different editions of SQL Server.
File
Logical name
Physical name
Initial size
File growth
Primary data
tempdev
tempdb.mdf
8 megabytes
Autogrow by 64 MB until the disk is full
Secondary data files
temp#
tempdb_mssql_#.ndf
8 megabytes
Autogrow by 64 MB until the disk is full
Log
templog
templog.ldf
8 megabytes
Autogrow by 64 megabytes to a maximum of 2 terabytes
All tempdb data files should always have the same initial size and growth parameters.
Number of tempdb data files
Depending on the version of the Database Engine, its configuration, and the workload, tempdb might require multiple data files to mitigate allocation contention.
The recommended total number of data files depends on the number of logical processors on the machine. As general guidance:
If the number of logical processors is less than or equal to eight, use the same number of data files.
If the number of logical processors is greater than eight, use eight data files.
If tempdb allocation contention is still observed, increase the number of data files by multiples of four until the contention decreases to acceptable levels, or make changes to the workload.
The following table lists the default value for each database option in the tempdb database and whether the option can be modified. To view the current settings for these options, use the sys.databases catalog view.
In Azure SQL Database, some aspects of tempdb behavior and configuration are different from SQL Server.
For single databases, each database on a logical server has its own tempdb. In an elastic pool, tempdb is a shared resource for all databases in the same pool but temporary objects created by one database are not visible to other databases in the same elastic pool.
Objects in tempdb, including catalog views and dynamic management views (DMVs), are accessible via a cross-database reference to the tempdb database. For example, you can query the sys.database_files view:
SQL
SELECT file_id,
type_desc,
name,
size,
max_size,
growth
FROM tempdb.sys.database_files;
In Azure SQL Managed Instance, some aspects of tempdb behavior and default configuration are different from SQL Server.
You can configure the number of tempdb files, their growth increments, and their maximum size. For more information on configuring tempdb settings in Azure SQL Managed Instance, see Configure tempdb settings for Azure SQL Managed Instance.
Azure SQL Managed Instance supports temporary objects in the same way as SQL Server, where all global temporary tables and global temporary stored procedures are accessible by all user sessions within the same SQL managed instance.
To learn more about tempdb sizes in Azure SQL Managed Instance, review resource limits.
The following operations can't be performed on the tempdb database:
Adding filegroups.
Backing up or restoring the database.
Changing collation. The default collation is the server collation.
Changing the database owner. tempdb is owned by sa.
Creating a database snapshot.
Dropping the database.
Dropping the guest user from the database.
Enabling Change Data Capture.
Participating in database mirroring.
Removing the primary filegroup, primary data file, or log file.
Renaming the database or primary filegroup.
Running DBCC CHECKALLOC.
Running DBCC CHECKCATALOG.
Setting the database to OFFLINE.
Setting the database or primary filegroup to READ_ONLY.
Permissions
Any user can create temporary objects in tempdb.
Users can access only their own non-temporary objects in tempdb, unless they receive additional permissions.
It's possible to revoke the CONNECT permission on tempdb to prevent a database user or role from using tempdb. This isn't recommended because many operations require the use of tempdb.
Optimize tempdb performance in SQL Server
The size and physical placement of tempdb files can affect performance. For example, if the initial size of tempdb is too small, time and resources might be taken up to autogrow tempdb to the size required to support the workload every time the Database Engine instance is restarted.
If possible, use instant file initialization to improve performance of the growth operations for data files.
Starting with SQL Server 2022 (16.x), transaction log file growth events up to 64 MB can also benefit from instant file initialization. For more information, see Instant file initialization and the transaction log.
Preallocate space for all tempdb files by setting the file size to a value large enough to accommodate the typical workload in the environment. Preallocation prevents tempdb from autogrowing too often, which can negatively affect performance.
The files in the tempdb database should be set to autogrow to provide space during unplanned growth events.
Dividing tempdb into multiple data files of equal size can improve efficiency of operations that use tempdb.
To avoid data allocation imbalance, data files should have the same initial size and growth parameters because the Database Engine uses a proportional-fill algorithm that favors allocations in files with more free space.
Set the file growth increment to a reasonable size, for example 64 MB, and make the growth increment the same for all data files to prevent growth imbalance.
To check current size and growth parameters for tempdb, use the following query:
Put the tempdb database on a fast I/O subsystem. Individual data files or groups of tempdb data files don't necessarily need to be on different disks unless you're encountering disk-level I/O bottlenecks.
If there is I/O contention between tempdb and user databases, put tempdb files on disks that differ from the disks that user databases use.
Note
To improve performance, delayed durability is always enabled on tempdb even if the database option DELAYED_DURABILITY is set to DISABLED. Because tempdb is recreated at startup, it doesn't go through a recovery process and doesn't provide a durability guarantee.
Performance improvements in tempdb for SQL Server
Introduced in SQL Server 2016 (13.x)
Temporary tables and table variables are cached. Caching allows operations that drop and create the temporary objects to run very quickly. Caching also reduces page allocation and metadata contention.
The allocation page latching protocol is improved to reduce the number of UP (update) latches that are used.
Logging overhead for tempdb is reduced to reduce disk I/O bandwidth consumption on the tempdb log file.
SQL Setup adds multiple tempdb data files during a new instance installation. Review the recommendations and configure your tempdb in the Database Engine Configuration page of SQL Setup, or use the command-line parameter /SQLTEMPDBFILECOUNT. By default, SQL Setup adds as many tempdb data files as the number of logical processors or eight, whichever is lower.
The AUTOGROW_ALL_FILES property is always turned on for the PRIMARY filegroup.
Introduced in SQL Server 2017 (14.x)
The SQL Setup experience improves guidance for initial tempdb file allocation. SQL Setup warns customers if the initial file size is set to a value greater than 1 GB and if instant file initialization is not enabled, preventing instance startup delays.
The sys.dm_tran_version_store_space_usage dynamic management view tracks version store usage per database. This DMV is useful for DBAs who want to proactively plan tempdb sizing based on the version store usage requirement per database.
Intelligent query processing features such as adaptive joins and memory grant feedback reduce memory spills on consecutive executions of a query, reducing tempdb utilization.
Introduced in SQL Server 2019 (15.x)
Database Engine doesn't use the FILE_FLAG_WRITE_THROUGH option when opening tempdb files to allow for maximum disk throughput. Since tempdb is recreated on startup, this option isn't needed to provide data durability. For more information on FILE_FLAG_WRITE_THROUGH, see Logging and data storage algorithms that extend data reliability in SQL Server.
Concurrent Page Free Space (PFS) page updates reduce page latch contention in all databases, an issue most commonly seen in tempdb. This improvement changes the concurrency management of PFS page updates so that they can be updated under a shared latch, rather than an exclusive latch. This behavior is on by default in all databases (including tempdb) starting with SQL Server 2019 (15.x). For more information on PFS pages, read Under the covers: GAM, SGAM, and PFS pages.
By default, a new installation of SQL Server on Linux creates multiple tempdb data files, based on the number of logical cores (with up to eight data files). This doesn't apply to in-place minor or major version upgrades. Each tempdb data file is 8 MB, with an auto growth of 64 MB. This behavior is similar to the default SQL Server installation on Windows.
Temporary object metadata contention has historically been a bottleneck to scalability for many SQL Server workloads. To address that, SQL Server 2019 (15.x) introduced a feature that's part of the in-memory database feature family: Memory-optimized TempDB metadata.
Enabling the Memory-optimized TempDB metadata feature removes this bottleneck for workloads previously limited by temporary object metadata contention inside tempdb. Starting with SQL Server 2019 (15.x), the system tables involved in managing temporary object metadata can become latch-free, non-durable, memory-optimized tables.
Tip
Because of current limitations, we recommend enabling Memory-optimized TempDB metadata only when object metadata contention occurs and significantly impacts your workloads.
The following diagnostic query returns one or more rows if temporary object metadata contention is occurring. Each row represents a system table, and returns the number of sessions contending for access to that table at the time when this diagnostic query is executed.
SQL
SELECT OBJECT_NAME(dpi.object_id, dpi.database_id) AS system_table_name,
COUNT(DISTINCT(r.session_id)) AS session_count
FROM sys.dm_exec_requests AS r
CROSSAPPLY sys.fn_PageResCracker(r.page_resource) AS prc
CROSSAPPLY sys.dm_db_page_info(prc.db_id, prc.file_id, prc.page_id, 'LIMITED') AS dpi
WHERE dpi.database_id = 2AND dpi.object_id IN (3, 9, 34, 40, 41, 54, 55, 60, 74, 75)
ANDUPPER(r.wait_type) LIKE N'PAGELATCH[_]%'GROUPBY dpi.object_id, dpi.database_id;
Watch this seven-minute video for an overview of how and when to use Memory-optimized TempDB metadata feature:
Note
Currently, the Memory-optimized TempDB metadata feature is not available in Azure SQL Database, SQL database in Microsoft Fabric, and Azure SQL Managed Instance.
Configure and use Memory-optimized TempDB metadata
The following sections include steps to enable, configure, verify, and disable the Memory-optimized TempDB metadata feature.
Enable
To enable this feature, use the following script:
SQL
ALTERSERVER CONFIGURATION SET MEMORY_OPTIMIZED TEMPDB_METADATA = ON;
For more information, see ALTER SERVER. This configuration change requires a restart of the service to take effect.
You can verify whether or not tempdb is memory-optimized by using the following T-SQL command:
If the returned value is 1 and a restart has occurred after enabling the feature, then the feature is enabled.
If the server fails to start for any reason after you enable Memory-optimized TempDB metadata, you can bypass the feature by starting the Database Engine instance with minimal configuration using the -f startup option. You can then disable the feature and remove the -f option to restart the Database Engine in normal mode.
Bind to resource pool to limit memory usage
To protect the server from potential out-of-memory conditions, we recommend that you bind tempdb to a resource governor resource pool that limits the memory consumed by Memory-optimized TempDB metadata. The following sample script creates a resource pool and sets its maximum memory to 20%, enables resource governor, and binds tempdb to the resource pool.
This example uses 20% as the memory limit for demonstration purposes. The optimal value in your environment might be larger or smaller depending on your workload, and can change over time if the workload changes.
SQL
CREATERESOURCE POOL tempdb_resource_pool
WITH (MAX_MEMORY_PERCENT = 20);
ALTERRESOURCE GOVERNOR RECONFIGURE;
ALTERSERVER CONFIGURATION
SET MEMORY_OPTIMIZED TEMPDB_METADATA = ON
(RESOURCE_POOL = 'tempdb_resource_pool');
This change also requires a service restart to take effect, even if Memory-optimized TempDB metadata is already enabled.
Verify resource pool binding and monitor memory usage
To verify that tempdb is bound to a resource pool and to monitor memory usage statistics for the pool, use the following query:
SQL
WITH resource_pool AS
(
SELECT p.pool_id,
p.name,
p.max_memory_percent,
dp.max_memory_kb,
dp.target_memory_kb,
dp.used_memory_kb,
dp.out_of_memory_count
FROM sys.resource_governor_resource_pools AS p
INNERJOIN sys.dm_resource_governor_resource_pools AS dp
ON p.pool_id = dp.pool_id
)
SELECT SERVERPROPERTY('IsTempdbMetadataMemoryOptimized') AS is_tempdb_memory_optimized_metadata_enabled,
rp.name AS resource_pool_name,
rp.max_memory_percent,
rp.max_memory_kb,
rp.target_memory_kb,
rp.used_memory_kb,
rp.out_of_memory_count
FROM sys.databases AS d
LEFTJOIN resource_pool AS rp
ON d.resource_pool_id = rp.pool_id
WHERE d.name = 'tempdb';
Remove resource pool binding
To remove the resource pool binding while keeping Memory-optimized TempDB metadata enabled, execute the following command and restart the service:
SQL
ALTERSERVER CONFIGURATION
SET MEMORY_OPTIMIZED TEMPDB_METADATA = ON;
Disable
To disable Memory-optimized TempDB metadata, execute the following command and restart the service:
SQL
ALTERSERVER CONFIGURATION
SET MEMORY_OPTIMIZED TEMPDB_METADATA = OFF;
Limitations of Memory-optimized TempDB metadata
Enabling or disabling the Memory-optimized TempDB metadata feature requires a restart.
In certain cases, you might observe high memory usage by the MEMORYCLERK_XTP memory clerk causing out-of-memory errors in your workload.
To see memory usage by the MEMORYCLERK_XTP clerk relative to all other memory clerks and relative to the target server memory, execute the following query:
A system stored procedure can be periodically executed to release MEMORYCLERK_XTP memory that is no longer needed. For more information, see sys.sp_xtp_force_gc (Transact-SQL).
When you use In-Memory OLTP, a single transaction is not allowed to access memory-optimized tables in more than one database. Because of this, any read or write transaction that involves a memory-optimized table in a user database can't also access tempdb system views in the same transaction. If this occurs, you receive error 41317:
Output
A user transaction that accesses memory optimized tables or natively compiled modules cannot access more than one user database or databases model and msdb, and it cannot write to master.
This limitation also applies to other scenarios where a single transaction attempts to access memory-optimized tables in more than one database.
For example, you might get error 41317 if you query the sys.stats catalog view in a user database that contains memory-optimized tables. This happens because the query attempts to access statistics data on a memory-optimized table in the user database and the memory-optimized metadata in tempdb.
The following example script produces this error when Memory-optimized TempDB metadata is enabled:
SQL
BEGIN TRAN;
-- Create an In-memory OLTP transaction that accesses a system view in tempdbSELECTnameFROM tempdb.sys.tables;
-- An attempt to create an In-memory OLTP transaction in the user database failsINSERTINTO <userdatabase>.<schema>.<memory-optimized table>
VALUES (1);
COMMIT TRAN;
Note
This limitation does not apply to temporary tables. You can create a temporary table in the same transaction that accesses a memory-optimized table in a user database.
Queries against system catalog views always use the READ COMMITTED isolation level. When the Memory-optimized TempDB metadata is enabled, queries against system catalog views in tempdb use the SNAPSHOT isolation level. In either case, locking hints are not honored.
Columnstore indexes can't be created on temporary tables when Memory-optimized TempDB metadata is enabled.
As a consequence, the use of the sp_estimate_data_compression_savings system stored procedure with the COLUMNSTORE or COLUMNSTORE_ARCHIVE data compression parameter is not supported when Memory-optimized TempDB metadata is enabled.
Capacity planning for tempdb in SQL Server
Determining the appropriate size for tempdb depends on many factors. These factors include the workload and the Database Engine features that are used.
We recommend that you analyze tempdb space consumption by performing the following tasks in a test environment where you can reproduce your typical workload:
Enable autogrow for tempdb files. All tempdb data files should have the same initial size and autogrow configuration.
Reproduce the workload and monitor tempdb space use.
If you use periodic index maintenance, execute your maintenance jobs and monitor tempdb space.
Use the maximum space used values from the previous steps to predict your total workload usage. Adjust this value for projected concurrent activity, and then set the size of tempdb accordingly.
Monitor tempdb use
Running out of disk space in tempdb can cause significant disruptions and application downtime. You can use the sys.dm_db_file_space_usage dynamic management view to monitor the space used in the tempdb files.
For example, the following example script finds:
Free space in tempdb (not considering free disk space that might be available for tempdb growth)
To monitor page allocation or deallocation activity in tempdb at the session or task level, you can use the sys.dm_db_session_space_usage and sys.dm_db_task_space_usage dynamic management views. These views can help you identify queries, temporary tables, or table variables that are using large amounts of tempdb space.
For example, use the following example script to obtain the tempdb space allocated and deallocated by internal objects in all currently running tasks in each session:
SQL
SELECT session_id,
SUM(internal_objects_alloc_page_count) AS task_internal_objects_alloc_page_count,
SUM(internal_objects_dealloc_page_count) AS task_internal_objects_dealloc_page_count
FROM sys.dm_db_task_space_usage
GROUPBY session_id;
Use the following example script to find the tempdb allocated and currently consumed space by internal and user objects for each session and request, for both running and completed tasks:
SQL
WITH tempdb_space_usage AS
(
SELECT session_id,
request_id,
user_objects_alloc_page_count + internal_objects_alloc_page_count AS tempdb_allocations_page_count,
user_objects_alloc_page_count + internal_objects_alloc_page_count - user_objects_dealloc_page_count - internal_objects_dealloc_page_count AS tempdb_current_page_count
FROM sys.dm_db_task_space_usage
UNIONALLSELECT session_id,
NULLAS request_id,
user_objects_alloc_page_count + internal_objects_alloc_page_count AS tempdb_allocations_page_count,
user_objects_alloc_page_count + internal_objects_alloc_page_count - user_objects_dealloc_page_count - user_objects_deferred_dealloc_page_count - internal_objects_dealloc_page_count AS tempdb_current_page_count
FROM sys.dm_db_session_space_usage
)
SELECT session_id,
COALESCE(request_id, 0) AS request_id,
SUM(tempdb_allocations_page_count * 8) AS tempdb_allocations_kb,
SUM(IIF (tempdb_current_page_count >= 0, tempdb_current_page_count, 0) * 8) AS tempdb_current_kb
FROM tempdb_space_usage
GROUPBY session_id, COALESCE (request_id, 0)
ORDERBY session_id, request_id;
This article is an in-depth look at diagnosing and resolving latch contention in SQL Server. This article was originally published by the SQLCAT team at Microsoft.
