Columnstore Index Delete

Introduction

The Columnstore Index Delete is not really an actual operator. You can encounter it in graphical execution plans in SSMS (and other tools), but if you look at the underlying XML of the execution plan, you will see that it is actually either an Index Delete or a Clustered Index Delete operator.

SQL Server currently supports three types of index storage: rowstore, columnstore, and memory-optimized. Indexes of each of those types can be the target of an Index Delete or Clustered Index Delete, as indicated by the Storage subproperty of the Object property. When the Storage property is RowStore or MemoryOptimized, then the normal Index Delete or Clustered Index Delete icon is used, but when Storage is ColumnStore, and only a single index is listed in the Object property, then Management Studio and Plan Explorer show a different icon instead (Azure Data Studio does show the regular icon for Index Delete or Clustered Index Delete in these cases).

While clustered and nonclustered indexes each have their own operator for rowstore storage, these types are lumped together as a single icon and operator name in SSMS. Whether the targeted columnstore index is a clustered or a nonclustered index is only visible in parentheses behind the operator name (Clustered vs Nonclustered), or in the Logical Operation property (Clustered Index Delete vs Index Delete).

Visual appearance in execution plans

Depending on the tool being used, an Index Delete or Clustered Index Delete operator with just a single delete target that has its Storage property equal to ColumnStore is displayed in a graphical execution plan as shown below:

SSMS and ADS (current versions)	Legacy SSMS	Plan Explorer	Paste The Plan
			(shown as Index Delete or Clustered Index Delete)

Algorithm

The Columnstore Index Delete is not a real operator but merely a visual indicator that an Index Delete or Clustered Index Delete operator is deleting data from a columnstore index. Please check the appropriate pages for a full description of the algorithms of those operators.

Input or Seek Predicates

The generic description of the Clustered Index Delete operator shows that the rows to be deleted can be either provided by a child operator, or be defined in a Seek Predicate property. For the (nonclustered) Index Delete, only the former method is supported.

A Columnstore Index Delete also can only read rows from a child operator; a Seek Predicate property is not supported. This is true even when the target columnstore index is the clustered index. This makes sense when you realize that, due to their storage structure, seek operations are not possible on columnstore indexes.

The input rows will always contain a “ColStoreLocnnnn” input, which is a single internal value that represents the location of the row to be deleted. This “columnstore locator” is actually a combination of two values: the rowgroup number, and the ordinal position of the row within that rowgroup.

Operator properties

The properties below are specific to the Columnstore Index Delete operator, or have a specific meaning when appearing on it. For all other properties, see Common properties. Properties that are included on the Common properties page but are also included below for their specific meaning for the Index Insert operator are marked with a *.

(Note that most of these properties are exactly the same as for the Index Delete operator; they are repeated here for ease of use).

Property name	Description
DMLRequestSort	When set to true, the delete operation might qualify for minimal logging if additional conditions apply. Details of those additional conditions and the requirements for DMLRequestSort to be true can be found here.
Object	The index that the Colummnstore Index Delete operator will delete rows from, using four part naming (database, schema, table, index). The subproperties of the Object property represent the four name parts separately, but also include these additional properties: Index Kind: Represents what kind of index data is deleted from. For a Columnstore Index Delete, this can be ViewNonClustered (for nonclustered columnstore indexes on an indexed view), NonClustered (for all other nonclustered columnstore indexes), or Clustered (for all clustered columnstore indexes. OnlineInbuildIndex: This property is present and set to 1 when a concurrent online index operation is running. The actual target is the temporary new index structure that is being built by an Online Index Insert operator. Storage: Determines the storage type of the index. For a Columnstore Index Delete, this is always ColumnStore.
Partitioned	This property is present and set to True when the target of the Columnstore Index Delete is a partitioned index.
WithUnorderedPrefetch	As far as currently known, prefetching is never used (and hence this property never included) for a Columnstore Index Delete operator.

Implicit properties

This table below lists the behavior of the implicit properties for the Columnstore Index Delete operator.

Property name	Description
Batch Mode enabled	The Columnstore Index Delete operator supports row mode execution only.
Blocking	The Columnstore Index Delete operator is non-blocking.
Memory requirement	The Columnstore Index Delete operator does not have any special memory requirement.
Order-preserving	The Columnstore Index Delete operator is fully order-preserving.
Parallelism aware	The Columnstore Index Delete operator does not support parallelism. It can only be used in a serial plan, or in a serial section of a parallel plan.
Segment aware	The Columnstore Index Delete operator is not segment aware.