Press "Enter" to skip to content

Category: Locks, Blocks, and Deadlocks

Intent Shared Locks in RCSI

Published 2021-11-23 by Kevin Feasel

Tomas Zika troubleshoots a deadlocking problem:

I was analyzing a deadlock graph and there was a mystery lock of type IS (Intent Shared). That was weird by itself because the database has Read Committed Snapshot Isolation (RCSI) enabled, which is the Optimistic Concurrency model that shouldn’t take shared locks. All the statements were contained in this database. Also, the locked table was seemingly unrelated to anything that has been going on in the deadlock report.

Click through for an image, a repro script, and an explanation as to what exactly is going on.

Comments closed

Finding Processes Waiting on Spinlocks

Published 2021-11-08 by Kevin Feasel

David Fowler helps us find processes waiting on spinlocks:

Spinlock contention is always a real headache to deal with. I recently saw an issue when spinlock contention on SOS_CACHESTORE was making the server virtually unresponsive. The issue was very intermittent with no obvious pattern but the assumption is that it was caused by a particular process in the application. Finding what that process was the tricky part, they don’t show up as waiting tasks so your usual scripts for looking for waiting processes may not work here.

It’s almost never spinlocks, but when it is, this is how you figure it out.

Comments closed

BULK_OPERATION Locks during a NOLOCK Scan

Published 2021-11-02 by Kevin Feasel

Paul Randal does some explaining:

I had an email question over the weekend where someone noticed that while executing a scan of a heap using NOLOCK, there was a BULK_OPERATION lock held on the heap for the duration of the scan. The question was why is the BULK_OPERATION lock needed, as surely there’s no way for the NOLOCK scan to read a problematic page?

Well, the answer is that the extra lock is needed *precisely* because the NOLOCK scan *can* read a problematic page if there’s a bulk operation happening on the heap at the same time.

And don’t call me Shirley.

Click through for a demonstration of this answer.

Comments closed

When APPEND_ONLY_STORAGE_INSERT_POINT is a Bottleneck

Published 2021-08-30 by Kevin Feasel

Paul Randal explains the APPEND_ONLY_STORAGE_INSERT_POINT latch and describes when and why this might be a problem:

Continuing my series of articles on latches, this time I’m going to discuss the APPEND_ONLY_STORAGE_INSERT_POINT latch and show how it can be a major bottleneck for heavy update workloads where either form of snapshot isolation is being used.

I strongly recommend you read the initial post in the series before this one, so you have all the general background knowledge about latches.

As always, read the whole thing.

Comments closed

Finding Bad (Worse?) NOLOCK Statements across Instances

Published 2021-08-04 by Kevin Feasel

Aaron Bertrand powers up for about six episodes straight, but the results are amazing:

In Part 1 of this series, I showed how to use a Visitor pattern to walk through one or more T-SQL statements to identify a problematic pattern where NOLOCK hints are applied to the target of an update or delete. The method in my initial examples was very manual, though, and won’t scale if this problem might be widespread. We need to be able to automate collecting a potentially large number of statements across an entire environment, and then try to eliminate false positives without manual intervention.

Read on to see how you can take what Aaron wrote last time and make it scalable.

Comments closed

The Costs and Benefits of Dirty Reads

Published 2021-08-03 by Kevin Feasel

Chad Callihan explains what a dirty read is and does a cost-benefit analysis on it:

When you are not careful with your transaction isolation levels or you get sneaky with the NOLOCK hint, one problem you can encounter is a dirty read. Let’s look at a short example to demonstrate a dirty read.

In a vacuum, I’m not necessarily opposed to the idea of dirty reads because you can find legitimate cases in which they can be useful. In practice, I’m generally very much in opposition because of two reasons: first, Read Committed Snapshot Isolation eliminates the majority of those reasons; and second, because the misuse is almost always in the direction of over-use of NOLOCK hints.

Comments closed

When DBCC_OBJECT_METADATA becomes a Bottleneck

Published 2021-07-28 by Kevin Feasel

Paul Randal takes us through another latch:

Continuing my series of articles on latches, this time I’m going to discuss the DBCC_OBJECT_METADATA latch and show how it can be a major bottleneck for consistency checks prior to SQL Server 2016 under certain circumstances. The issue affects DBCC CHECKDB, DBCC CHECKTABLE, and DBCC CHECKFILEGROUP, but for clarity I’ll just reference DBCC CHECKDB for the rest of this post.

You might wonder why I’m writing about an issue that affects older versions, but there are still a huge number of SQL Server 2014 and older instances out there, so it’s a valid topic for my series.

Read on to understand what DBCC_OBJECT_METADATA does and how it can become a bottleneck on those older versions of SQL Server.

Comments closed

FGCB_ADD_REMOVE Latch

Published 2021-06-22 by Kevin Feasel

Paul Randal looks at a particular latch:

Most latch class names are tied directly to the data structure that they protect. The FGCB_ADD_REMOVE latch protects a data structure called an FGCB, or File Group Control Block, and there will be one of these latches for each online filegroup of each online database in a SQL Server instance. Whenever a file in a filegroup is added, dropped, grown, or shrunk, the latch must be acquired in EX mode, and when figuring out the next file to allocate from, the latch must be acquired in SH mode to prevent any filegroup changes. (Remember that extent allocations for a filegroup are performed on a round-robin basis through the files in the filegroup, and also take into account proportional fill, which I explain here.)

Read on to understand what can cause this particular latch to become a bottleneck in your system.

Comments closed

Foreign Key Constraints and Blocking

Published 2021-06-18 by Kevin Feasel

Paul White takes a look at blocking due to foreign key checks:

This article covers one such consideration that does not receive much publicity: To minimize blocking, you should think carefully about the indexes used to enforce uniqueness on the parent side of those foreign key relationships.

This applies whether you are using locking read committed or the versioning-based read committed snapshot isolation (RCSI). Both can experience blocking when foreign key relationships are checked by the SQL Server engine.

Under snapshot isolation (SI), there is an extra caveat. The same essential issue can lead to unexpected (and arguably illogical) transaction failures due to apparent update conflicts.

This article is in two parts. The first part looks at foreign key blocking under locking read committed and read committed snapshot isolation. The second part covers related update conflicts under snapshot isolation.

Definitely worth reading the whole thing.

Comments closed