3

When testing some queries, I set the recovery model to full on a database and ran two identical 1M row inserts, with and without TABLOCK.

On the one with TABLOCK, I got around 9295 log records on a sql server 2008 instance, and around 8714 log records on a SQL Server 2017 instance.

When running the insert without tablock, I get around 1035659 records for the 2008 instance and around 1068599 records for the 2017 instance.

The reason for testing on sql server 2008 is to match the data loading performance guide's statement on the recovery models for ML operations :

Minimally logged operations are available only if your database is in bulk-logged or simple recovery mode.

So, what am Is seeing here if it is not minimal logging?

Use DatabaseName

ALTER DATABASE DatabaseName SET RECOVERY FULL;
GO
BACKUP DATABASE DatabaseName TO DISK = '\\location\DatabaseName.bak';

BACKUP LOG DatabaseName TO DISK = '\\location\DatabaseName_log.trn';
GO

IF OBJECT_ID(N'dbo.Accounts', N'U') IS NOT NULL
BEGIN
    DROP TABLE dbo.Accounts;
END;
GO

CREATE TABLE dbo.Accounts( AccountID INT PRIMARY KEY NOT NULL,
                            AccountName varchar(255),
                            DateCreated DATETIME2);

-- Insert 1M Rows into dbo.Account without TABLOCK
GO
SET STATISTICS IO, TIME ON;
INSERT INTO dbo.Accounts  (AccountID,AccountName,DateCreated)
SELECT TOP(1000000)
        ROW_NUMBER() OVER (ORDER BY (SELECT NULL)),
        'Name N ' + CAST(ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) AS varchar(255)),
        DATEADD(MINUTE,-ROW_NUMBER() OVER (ORDER BY (SELECT NULL)),GETDATE())
FROM MASTER..SPT_VALUES SPT1
CROSS APPLY MASTER..SPT_VALUES SPT2;
GO
-- check the amount of records in the log file
SELECT count(*)
FROM
fn_dbLog(NULL,NULL);
--1035659 rows
GO

-- clear the log
BACKUP LOG DatabaseName to disk = '\\location\DatabaseName_log2.trn';

GO

--drop the table

IF OBJECT_ID(N'dbo.Accounts', N'U') IS NOT NULL
BEGIN
    DROP TABLE dbo.Accounts;
END;
GO
-- create the table
CREATE TABLE dbo.Accounts( AccountID INT PRIMARY KEY NOT NULL,
                            AccountName varchar(255),
                            DateCreated DATETIME2);

-- Insert 1M Rows into dbo.Account WITH TABLOCK
GO
SET STATISTICS IO, TIME ON;
INSERT INTO dbo.Accounts WITH(TABLOCK) (AccountID,AccountName,DateCreated)
SELECT TOP(1000000)
        ROW_NUMBER() OVER (ORDER BY (SELECT NULL)),
        'Name N ' + CAST(ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) AS varchar(255)),
        DATEADD(MINUTE,-ROW_NUMBER() OVER (ORDER BY (SELECT NULL)),GETDATE())
FROM MASTER..SPT_VALUES SPT1
CROSS APPLY MASTER..SPT_VALUES SPT2;
GO
-- check the amount of records in the log file
SELECT count(*)
FROM
fn_dbLog(NULL,NULL);

--9295 rows
6

When using TABLOCK your logging even in full model is called "efficient logging", it's when instead of logging every insert row-by-row the whole pages are logged.

It would be minimally logged in simple and bulk logged modes with only page allocations in the log, but in full recovery model you have fully formatted pages of data.

So the number of records that go to the log in simple and full models when inserting with tablock will be nearly the same but the content is different: in simple/bulk logged there will be only page numbers, in full there will be complete pages.

And it is really full logging since the insert operation can be fully reconstituted using these log records while in simple/bulk logged you have only information sufficient to rollback the insert.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.