What's new in columnstore indexes

Applies to: SQL Server Azure SQL Database Azure SQL Managed Instance Azure Synapse Analytics Analytics Platform System (PDW)

Learn about which columnstore features available for each version of SQL Server, and the latest releases of SQL Database, Azure Synapse Analytics, and Analytics Platform System (PDW).

Feature summary for product releases

This table summarizes key features for columnstore indexes and the products in which they are available.

Columnstore Index Feature SQL Server 2012 (11.x) SQL Server 2014 (12.x) SQL Server 2016 (13.x)1 SQL Server 2017 (14.x) SQL Server 2019 (15.x) SQL Server 2022 (16.x) SQL Database1 Azure Synapse Analytics dedicated SQL pool
Batch mode execution for multi-threaded queries2 yes yes yes yes yes yes yes yes
Batch mode execution for single-threaded queries yes yes yes yes yes yes
Archival compression option yes yes yes yes yes yes yes
Snapshot isolation and read-committed snapshot isolation yes yes yes yes yes yes
Specify columnstore index when creating a table yes yes yes yes yes yes
Always On supports columnstore indexes yes yes yes yes yes yes yes yes
Always On readable secondary supports read-only nonclustered columnstore index yes yes yes yes yes yes yes yes
Always On readable secondary supports updateable columnstore indexes yes yes yes yes
Read-only nonclustered columnstore index on heap or B-tree yes yes yes 3 yes 3 yes 3 yes 3 yes 3 yes 3
Updateable nonclustered columnstore index on heap or B-tree yes yes yes yes yes yes
Additional B-tree indexes allowed on a heap or B-tree that has a nonclustered columnstore index yes yes yes yes yes yes yes yes
Updateable clustered columnstore index yes yes yes yes yes yes yes
B-tree index on a clustered columnstore index yes yes yes yes yes yes
Columnstore index on a memory-optimized table yes yes yes yes yes yes
Nonclustered columnstore index definition supports using a filtered condition yes yes yes yes yes yes
Compression delay option for columnstore indexes in CREATE TABLE and ALTER TABLE yes yes yes yes yes yes
Support for nvarchar(max) type yes yes yes yes no 4
Columnstore index can have a non-persisted computed column yes yes yes
Tuple mover background merge support yes yes yes yes
Ordered clustered columnstore indexes yes yes yes

1 For SQL Database, columnstore indexes are available in Azure SQL Database DTU Premium tiers, DTU Standard tiers - S3 and above, and all vCore tiers. For SQL Server 2016 (13.x) SP1 and later versions, columnstore indexes are available in all editions. For SQL Server 2016 (13.x) (before SP1) and earlier versions, columnstore indexes are only available in Enterprise Edition.

2 The degree of parallelism (DOP) for batch mode operations is limited to 2 for SQL Server Standard Edition and 1 for SQL Server Web and Express Editions. This limitation refers to columnstore indexes created over disk-based tables and memory-optimized tables.

3 To create a read-only nonclustered columnstore index, store the index on a read-only filegroup.

4 Not supported in dedicated SQL pools but is supported in serverless SQL pool.

SQL Server 2022 (16.x)

SQL Server 2022 (16.x) adds these new features. For more information, see What's new in SQL Server 2022 (16.x).

  • Ordered clustered columnstore indexes improve performance for queries based on ordered column predicates. Ordered columnstore indexes can improve performance by skipping segments of data altogether. This can drastically reduce IO needed to complete queries on columnstore data. For more information, see segment elimination. Ordered cluster columnstore indexes are available in SQL Server 2022 (16.x). For more information, see CREATE COLUMNSTORE INDEX and Performance tuning with ordered clustered columnstore index.

  • Predicate pushdown with clustered columnstore rowgroup elimination of strings uses boundary values to optimize string searches. All columnstore indexes benefit from enhanced segment elimination by data type. Starting with SQL Server 2022 (16.x), these segment elimination capabilities extend to string, binary, and guid data types, and the datetimeoffset data type for scale greater than two. Previously, columnstore segment elimination applied only to numeric, date, and time data types, and the datetimeoffset data type with scale less than or equal to two. After upgrading to a version of SQL Server that supports string min/max segment elimination (SQL Server 2022 (16.x) and later versions), the columnstore index will not benefit this feature until it is rebuilt using a REBUILD or DROP/CREATE.

SQL Server 2019 (15.x)

SQL Server 2019 (15.x) adds these new features:

Functional

Starting with SQL Server 2019 (15.x), the tuple mover is helped by a background merge task that automatically compresses smaller OPEN delta rowgroups that have existed for some time as determined by an internal threshold, or merges COMPRESSED rowgroups from where a large number of rows has been deleted. Previously, an index reorganize operation was needed to merge rowgroups with partially deleted data. This improves the columnstore index quality over time.

SQL Server 2017 (14.x)

SQL Server 2017 (14.x) adds these new features.

Functional

  • SQL Server 2017 (14.x) supports nonpersisted computed columns in clustered columnstore indexes. Persisted computed columns are not supported in clustered columnstore indexes. You cannot create a nonclustered columnstore index on a computed column.

SQL Server 2016 (13.x)

SQL Server 2016 (13.x) adds key enhancements to improve the performance and flexibility of columnstore indexes. These improvements enhance data warehousing scenarios and enable real-time operational analytics.

Functional

  • A rowstore table can have one updateable nonclustered columnstore index. Previously, the nonclustered columnstore index was read-only.

  • The nonclustered columnstore index definition supports using a filtered condition. To minimize the performance impact of adding a columnstore index on an OLTP table, use a filtered condition to create a nonclustered columnstore index on only the cold data of your operational workload.

  • An in-memory table can have one columnstore index. You can create it when the table is created or add it later with ALTER TABLE (Transact-SQL). Previously, only a disk-based table could have a columnstore index.

  • A clustered columnstore index can have one or more nonclustered rowstore indexes. Previously, the columnstore index did not support nonclustered indexes. SQL Server automatically maintains the nonclustered indexes for DML operations.

  • Support for primary keys and foreign keys by using a B-tree index to enforce these constraints on a clustered columnstore index.

  • Columnstore indexes have a compression delay option that minimizes the impact of the transactional workload on real-time operational analytics. This option allows for frequently changing rows to stabilize before compressing them into the columnstore. For details, see CREATE COLUMNSTORE INDEX (Transact-SQL) and Get started with Columnstore for real-time operational analytics.

Performance for database compatibility level 120 or 130

  • Columnstore indexes support read committed snapshot isolation level (RCSI) and snapshot isolation (SI). This enables transactional consistent analytics queries with no locks.

  • Columnstore supports index defragmentation by removing deleted rows without the need to explicitly rebuild the index. The ALTER INDEX ... REORGANIZE statement removes deleted rows, based on an internally defined policy, from the columnstore as an online operation

  • Columnstore indexes can be access on an Always On readable secondary replica. You can improve performance for operational analytics by offloading analytics queries to an Always On secondary replica.

  • Aggregate Pushdown computes the aggregate functions MIN, MAX, SUM, COUNT, and AVG during table scans when the data type uses no more than 8 bytes, and is not a string data type. Aggregate pushdown is supported with or without GROUP BY clause for both clustered columnstore indexes and nonclustered columnstore indexes. On SQL Server, this enhancement is reserved for Enterprise edition.

  • String Predicate pushdown speeds up queries that compare strings of type VARCHAR/CHAR or NVARCHAR/NCHAR. This applies to the common comparison operators and includes operators such as LIKE that use bitmap filters. This works with all supported collations. On SQL Server, this enhancement is reserved for Enterprise edition.

  • Enhancements for batch mode operations by leveraging vector based hardware capabilities. The Database Engine detects the level of CPU support for AVX 2 (Advanced Vector Extensions) and SSE 4 (Streaming SIMD Extensions 4) hardware extensions, and uses them if supported. On SQL Server, this enhancement is reserved for Enterprise edition.

Performance for database compatibility level 130

  • New batch mode execution support for queries using any of these operations:

    • SORT
    • Aggregates with multiple distinct functions. Some examples: COUNT/COUNT, AVG/SUM, CHECKSUM_AGG, STDEV/STDEVP
    • Window aggregate functions: COUNT, COUNT_BIG, SUM, AVG, MIN, MAX, and CLR
    • Window user-defined aggregates: CHECKSUM_AGG, STDEV, STDEVP, VAR, VARP, and GROUPING
    • Window aggregate analytic functions: LAG, LEAD, FIRST_VALUE, LAST_VALUE, PERCENTILE_CONT, PERCENTILE_DISC, CUME_DIST, and PERCENT_RANK
  • Single-threaded queries running under MAXDOP 1 or with a serial query plan execute in batch mode. Previously, only multi-threaded queries ran with batch execution.

  • Memory optimized table queries can have parallel plans in SQL InterOp mode both when accessing data in rowstore or in columnstore index.

Supportability

These system views are new for columnstore:

These in-memory OLTP-based DMVs contain updates for columnstore:

Limitations

  • For in-memory tables, a columnstore index must include all the columns; the columnstore index cannot have a filtered condition.
  • For in-memory tables, queries on columnstore indexes run only in InterOP mode, and not in the in-memory native mode. Parallel execution is supported.

Known issues

Applies to: SQL Server, Azure SQL Database, Azure SQL Managed Instance, Azure Synapse Analytics dedicated SQL pool

  • Currently, LOB columns (varbinary(max), varchar(max), and nvarchar(max)) in compressed columnstore segments are not affected by DBCC SHRINKDATABASE and DBCC SHRINKFILE.

SQL Server 2014 (12.x)

SQL Server 2014 (12.x) introduced the clustered columnstore index as the primary storage format. This allowed regular loads as well as update, delete, and insert operations.

  • The table can use a clustered columnstore index as the primary table storage. No other indexes are allowed on the table, but the clustered columnstore index is updateable so you can perform regular loads and make changes to individual rows.
  • The nonclustered columnstore index continues to have the same functionality as in SQL Server 2012 (11.x) except for additional operators that can now be executed in batch mode. It is still not updateable except by rebuilding, and by using partition switching. The nonclustered columnstore index is supported on disk-based tables only, and not on in-memory tables.
  • The clustered and nonclustered columnstore index has an archival compression option that further compresses the data. The archival option is useful for reducing the data size both in memory and on disk, but does slow query performance. It works well for data that is accessed infrequently.
  • The clustered columnstore index and the nonclustered columnstore index function in a very similar way; they use the same columnar storage format, same query processing engine, and the same set of dynamic management views. The difference is primary versus secondary index types, and the nonclustered columnstore index is read-only.
  • These operators run in batch mode for multi-threaded queries: scan, filter, project, join, group by, and union all.

SQL Server 2012 (11.x)

SQL Server 2012 (11.x) introduced the nonclustered columnstore index as another index type on rowstore tables and batch processing for queries on columnstore data.

  • A rowstore table can have one nonclustered columnstore index.
  • The columnstore index is read-only. After you create the columnstore index, you cannot update the table by INSERT, DELETE, and UPDATE operations; to perform these operations you must drop the index, update the table and rebuild the columnstore index. You can load additional data into the table by using partition switching. The advantage of partition switching is you can load data without dropping and rebuilding the columnstore index.
  • The columnstore index always requires extra storage, typically an additional 10% over rowstore, because it stores a copy of the data.
  • Batch processing provides 2x or better query performance, but it is only available for parallel query execution.