SQL Server Nonclustered Index on Primary Key – Speed Up Deletes and Prevent Deadlocks

deadlocknonclustered-indexsql serversql server 2014

I have a very high traffic database. In the application they will issue deletes, and frequently these deletes will deadlock with other deletes on the same table. I am researching on ways to remedy this, and one answer I saw was to ensure that the deletes have the fastest pathway to the record.

Currently, all of the deletes follow this form:

(@0 int) delete [dbo].[table] where (table_id = @0)

Each of these tables has a primary key and clustered index on table_id.

My question is, could adding a non-clustered index on table_id help speed these deletes up and prevent deadlocks from occurring?

Deadlock graph:

<deadlock victim="process2b53a408c8"><process-list><process id="process2b53a408c8" taskpriority="0" logused="284" waitresource="KEY: 19:72057597368270848 (0e2c6d2527ac)" waittime="2034" ownerId="14899585071" transactionname="user_transaction" lasttranstarted="2018-03-07T09:47:18.833" XDES="0x50746b1c0" lockMode="S" schedulerid="7" kpid="967168" status="suspended" spid="185" sbid="2" ecid="0" priority="0" trancount="2" lastbatchstarted="2018-03-07T09:47:18.850" lastbatchcompleted="2018-03-07T09:47:18.850" lastattention="1900-01-01T00:00:00.850" clientapp="redacted_service" hostname="redacted_host" hostpid="164656" loginname="redacted_domain\_SQL_redacted_database_P" isolationlevel="read committed (2)" xactid="14899585071" currentdb="19" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056" databaseName="redacted_database"><executionStack><frame procname="adhoc" line="1" stmtstart="16" stmtend="120" sqlhandle="0x020000001335f1027371c186c8d7405191cdef00ddd0ebe70000000000000000000000000000000000000000">
unknown     </frame><frame procname="unknown" line="1" sqlhandle="0x0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000">
unknown     </frame></executionStack><inputbuf>
(@0 int)DELETE [dbo].[redacted_table]
WHERE ([id] = @0)    </inputbuf></process><process id="process218b0a5468" taskpriority="0" logused="284" waitresource="KEY: 19:72057597368270848 (c94c2713ec0f)" waittime="2036" ownerId="14899585063" transactionname="user_transaction" lasttranstarted="2018-03-07T09:47:18.833" XDES="0x1fcd0c4d10" lockMode="S" schedulerid="4" kpid="962372" status="suspended" spid="384" sbid="2" ecid="0" priority="0" trancount="2" lastbatchstarted="2018-03-07T09:47:18.850" lastbatchcompleted="2018-03-07T09:47:18.847" lastattention="1900-01-01T00:00:00.847" clientapp="redacted_service" hostname="redacted_host" hostpid="164656" loginname="redacted_domain\_SQL_redacted_database_P" isolationlevel="read committed (2)" xactid="14899585063" currentdb="19" lockTimeout="4294967295" clientoption1="671088672" clientoption2="128056" databaseName="redacted_database"><executionStack><frame procname="adhoc" line="1" stmtstart="16" stmtend="120" sqlhandle="0x020000001335f1027371c186c8d7405191cdef00ddd0ebe70000000000000000000000000000000000000000">
unknown     </frame><frame procname="unknown" line="1" sqlhandle="0x0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000">
unknown     </frame></executionStack><inputbuf>
(@0 int)DELETE [dbo].[redacted_table]
WHERE ([id] = @0)    </inputbuf></process></process-list><resource-list><keylock hobtid="72057597368270848" dbid="19" objectname="redacted_database.dbo.redacted_table_2" indexname="PK_redacted_table_2" id="lock9d21dab80" mode="X" associatedObjectId="72057597368270848"><owner-list><owner id="process218b0a5468" mode="X" /></owner-list><waiter-list><waiter id="process2b53a408c8" mode="S" requestType="wait" /></waiter-list></keylock><keylock hobtid="72057597368270848" dbid="19" objectname="redacted_database.dbo.redacted_table_2" indexname="PK_redacted_table_2" id="lock4de4b1800" mode="X" associatedObjectId="72057597368270848"><owner-list><owner id="process2b53a408c8" mode="X" /></owner-list><waiter-list><waiter id="process218b0a5468" mode="S" requestType="wait" /></waiter-list></keylock></resource-list></deadlock>

table definition:

create table [dbo].[redacted_table](
    [id] [int] identity(1,1) not for replication not null,
    [loan_id] [int] not null,
    [user_role_id] [int] not null,
    [assigned_by_user_id] [int] not null,
    [out_for_assignment] [bit] not null,
    [assignment_date] [datetime] not null,
    [recognize_date] [datetime] not null,
    [routing_source] [varchar](50) null,
    [request_guid] [uniqueidentifier] null,
 constraint [PK_redacted_table] primary key clustered 
(
    [id] asc
)with (pad_index = off, statistics_norecompute = off, ignore_dup_key = off, allow_row_locks = on, allow_page_locks = on, fillfactor = 90) on [PRIMARY]
) on [PRIMARY]

go

alter table [dbo].[redacted_table] add  constraint [DF_redacted_table_assignment_date]  default (getdate()) for [assignment_date]
go

alter table [dbo].[redacted_table] add  constraint [DF_redacted_table_recognize_date]  default (getdate()) for [recognize_date]
go

alter table [dbo].[redacted_table]  with check add  constraint [FK_redacted_table_redacted_table3] foreign key([loan_id])
references [dbo].[redacted_table3] ([id])
go

alter table [dbo].[redacted_table] check constraint [FK_redacted_table_redacted_table3]
go

alter table [dbo].[redacted_table]  with check add  constraint [FK_redacted_table_user_redacted_table4] foreign key([user_role_id])
references [dbo].[user_redacted_table4] ([id])
go

alter table [dbo].[redacted_table] check constraint [FK_redacted_table_user_redacted_table4]
go

alter table [dbo].[redacted_table]  with check add  constraint [FK_redacted_table_redacted_table5] foreign key([assigned_by_user_id])
references [dbo].[redacted_table5] ([id])
go

alter table [dbo].[redacted_table] check constraint [FK_redacted_table_redacted_table5]
go

Best Answer

Since your statement is deleting rows via a WHERE clause on the primary key, table_id, adding a non-clustered index on table_id is unlikely to help, and may well increase the number of deadlocks occurring.

DELETE FROM ... deletes the row from the table (or clustered index), as well as every non-clustered index defined on the table where the row is present. Adding an extra index necessitates and extra delete operation.

Take this very simple example:

IF OBJECT_ID(N'dbo.DeleteTest', N'U') IS NOT NULL
DROP TABLE dbo.DeleteTest;
GO
CREATE TABLE dbo.DeleteTest
(
    table_id int NOT NULL
        CONSTRAINT PK_DeleteTest
        PRIMARY KEY CLUSTERED
        IDENTITY(1,1)
    , someVal varchar(2000) NOT NULL
        CONSTRAINT DF_DeleteTest_someVal
        DEFAULT ((CRYPT_GEN_RANDOM(2000)))
);
GO

--insert 500 rows
INSERT INTO dbo.DeleteTest DEFAULT VALUES;
GO 500

--turn on "Actual" execution plans
DELETE 
FROM dbo.DeleteTest 
WHERE table_id = 90;

The actual execution plan for the above DELETE:

enter image description here You should probably add the CREATE TABLE statements to your question, along with details about the deadlock via a deadlock graph.

Now, if we add an additional non-clustered index, and run another delete:

CREATE INDEX IX_DeleteTest
ON dbo.DeleteTest (table_id);

DELETE 
FROM dbo.DeleteTest 
WHERE table_id = 91;

We see the following plan:

enter image description here

As you can see in the two images, the second delete costs nearly twice as much; 0.0132841 for the first delete, and 0.0232851 for the second delete.

If you use SentryOne Plan Explorer (free!) to look at execution plans, you can see the number of additional non-clustered index-delete operations taking place:

enter image description here

Looking at your deadlock graph, in combination with the list of non-clustered indexes in your question, it looks like the deadlock is likely caused by simultaneous deletes of rows contained on the same page. Adding a ROWLOCK hint to your delete may prevent this particular deadlock. You could also test using SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; prior to the delete operation. Be aware that changes the isolation level for all statements occuring afterwards within the same batch. Ensure you reset the transaction isolation level if you execute other statements after the delete operation.

The deadlock shows the wait resource is the same for both statements, except the row number: waitresource="KEY: 19:72057597368270848 (0e2c6d2527ac)" The row number is the number inside brackets. @Kin shows a great way to look at the details of the waitresource in his answer here

ROWLOCK
Specifies that row locks are taken when page or table locks are ordinarily taken. When specified in transactions operating at the SNAPSHOT isolation level, row locks are not taken unless ROWLOCK is combined with other table hints that require locks, such as UPDLOCK and HOLDLOCK.

Related Question