About locking with REORGANIZE

[rrg92]

Original text says that REORGANIZE not block concurrent readers and writers. But is not true.

Use following code to check this:

-- create some test table
create table __reorgtest(c int);

-- create sample nonclustered index
create index x on __reorgtest(c);

-- insert some data
insert into __reorgtest values(2020);
go 1000


-- Open another session and run this (we call Session X):
begin tran;;
alter index x on __reorgtest reorganize;

-- Open another session and run this (we call session Y):
begin tran;;
select count(*) from __reorgtest

-- On this session, check tran_locks! 
select * from sys.dm_tran_locks where request_session_id in (  X,Y ) -- replace and y by sessions id opened above.

-- Current behavior on my tests: Session Y gets blocked by X.
-- THe inverse can be checked: Try start first a select (wth holdlock) or update operation on some row, and try start X(reorganize). Reorganize gets blocked!

Behavior checked on

• Microsoft SQL Server 2019 (RTM-GDR) (KB4517790) - 15.0.2070.41 (X64) Oct 28 2019 19:56:59 Copyright (C) 2019 Microsoft Corporation Developer Edition (64-bit) on Windows 10 Home Single Language 10.0 (Build 18363: )
• Microsoft SQL Server 2016 (SP2-CU10) (KB4524334) - 13.0.5492.2 (X64) Oct 4 2019 19:14:08 Copyright (c) Microsoft Corporation Standard Edition (64-bit) on Windows Server 2012 R2 Standard 6.3 (Build 9600: )

[PRMerger9]

@rrg92 : Thanks for your contribution! The author(s) have been notified to review your proposed change.

[rothja]

@rrg92 Thanks for the feedback. @pmasl Could you review this change and sign off on it if you approve? Thank you!

[ktoliver]

@pmasl Hi, could you take a look at this PR? Thank you.

[rrg92]

Hey guys, this PR make some sense?

[ktoliver]

@pmasl Could you review this PR? Thanks.

[pmasl]

Your test has REORG encapsulated in an open tran. That way anything blocks. The use of REORG as an autocommit statement is done online, which means only the singleton record impacted blocks or is blocked. In summary, your test doesn't represent the intended or real-use case of REORG.

[rrg92]

Thanks @pmasl !

I opened an Extended events session to check you says and I confirmed that.

This is lock held using begin tran:

Last is an Exclusive lock.

Bellow are all lock held into this object without begin tran:

No exclusives, just intents and shared!! Nice

I search in SQL docs and not find anything about that fact autocommit mode change this behavior... Do you guess that adding this info into doc is pertinent? I guess most of users don't know this.

[pmasl]

Locking behavior is different between explicit transactions and auto-commit (aka implicit transactions which is the default) for every (or most) DML and DDL.

If any I'd suggest adding to that note something to the effect of:

"If using explicit transactions instead of the default implicit transaction mode, the locking behavior of REORGANIZE becomes more restrictive. For more information on implicit transactions, see SET IMPLICIT_TRANSACTIONS (Transact-SQL)."

[PRMerger6]

@rrg92 : Thanks for your contribution! The author(s) have been notified to review your proposed change.

[rrg92]

I changed, putting original text and adding a Note on about the fact of autocommit.

[PRMerger7]

@rrg92 : Thanks for your contribution! The author(s) have been notified to review your proposed change.

[rrg92]

All done, text will looks like this:

[pmasl]

#sign-off

[PRMerger15]

@ktoliver : Thanks for your contribution! The author(s) have been notified to review your proposed change.