SQLServerCentral / Article Discussions / Article Discussions by Author / Discuss content posted by VM / Persisted vs Non-persisted Computed columns / Latest Posts

RE: Persisted vs Non-persisted Computed columns

timothy bates — Wed, 20 Oct 2010 08:50:14 GMT

Nice question, I appreciate the intent of the question as I see it, 'Persisted does not mean permanent.'

RE: Persisted vs Non-persisted Computed columns

rtelgenhoff — Thu, 07 Oct 2010 11:25:33 GMT

http://msdn.microsoft.com/en-us/library/ms189292.aspx#BKMK_persisted'You can create an index on a computed column that is defined with a deterministic, but imprecise, expression if the column is marked PERSISTED'--If I wanted to find all those who received a work permit on a give day, this query works but shows a table scan.SELECT [DOWorkPermit] FROM [Employee_2] WHERE [DOWorkPermit] = '12/31/1993';--Add an index on the persisted, computed column:CREATE NONCLUSTERED INDEX IX_WorkPermit ON [Employee_2] ([DOWorkPermit]);GO--The query still shows a table scan because DOBirth is in the output list and is needed to find the computed DOWorkPermit day:SELECT [DOWorkPermit] FROM [Employee_2] WHERE [DOWorkPermit] = '12/31/1993';-- drop and recreate the index including the DOBirth columnDROP INDEX IX_WorkPermit ON [Employee_2];GOCREATE NONCLUSTERED INDEX IX_WorkPermit ON [Employee_2] ([DOWorkPermit]) INCLUDE ([DOBirth]);GO--an index scan is now usedSELECT [DOWorkPermit] FROM [Employee_2] WHERE [DOWorkPermit] = '12/31/1993'My question is, couldn't an index seek be done on the persisted, indexed value rather than having to compute it while doing a scan? Is it because there's only 1 record in the table or am I missing or misinterpreting something...

RE: Persisted vs Non-persisted Computed columns

Sean Lange — Thu, 07 Oct 2010 07:47:45 GMT

[quote][b]Nils Gustav Stråbø (10/7/2010)[/b][hr]You won't believe how many people that actually use NOLOCK or READ_UNCOMMITTED to "fix" concurrency and locking problems. One of our vendors use NOLOCK in every(!) SELECT statement in a system we have purchased from them. They have never bothered to explain the reason for doing it, but my guess, after doing some investigation, is that it is caused by two things:- Lots of missing indexes, which results in excessive scans- Bad programming, because when I save data in the front-end, SQL Profiler showed that before the transaction for the update statement was committed, another connection from my pc was opened trying to read the updated row(s).[/quote]I worked as a consultant on a system for a major health insurance company and they had the nolock hint on every single query in their system. Keeping in mind that this is the system that pays disbursements both from "paper" claims and real time debit cards our balances were always out of whack because we would authorize payment when there was not enough money due to dirty reads. I can't tell you how many times I suggested that they should find better ways of dealing of locked tables. :w00t:

RE: Persisted vs Non-persisted Computed columns

Nils Gustav Stråbø — Thu, 07 Oct 2010 07:29:22 GMT

Ok, I can see that my comment about using NOLOCK to fix concurrency issues could be explained a bit better, and that there are legitimate reasons to use it, for example in reporting scenarios.But I cannot see where I completely condemn the use of NOLOCK. I mentioned that one of our vendors use in in ALL SELECT statements, and most of them has nothing to do with reporting functionality. They are plain SELECTS that returns just a few rows, but because the client app is poorly programmed (as I tried to explain) the have "fixed" it with NOLOCK.

RE: Persisted vs Non-persisted Computed columns

sjimmo — Thu, 07 Oct 2010 07:14:47 GMT

[quote]HugoMostly queries that report aggregated data for an audience that doesn't care about the exact number, but only wants an indication. If the numbers my query produces will be rounded to the nearest million dollars anyway, I can live with a result that might be a couple of thousand dollars off. [/quote]Completely agree, which is why we unfortunately use it everywhere. Our environment is very large and growing. We literally have several thousand users hitting against the OLTP databases making hourly decisions. Unfortunately we are unable to replicate all of the necessary data to other locations for them to review though it is moved to a warehouse daily which is used primarily for trending as well as market basket and many other things. Reports are happening at the same time as OLTP. I appologize if you felt that I was picking, but I just wanted to get you to expand upon your answer. Thank you.[quote]NilsDid I ever say that this hint was never to be used? For reporting purposes where the possibility of reading dirty data does not affect the quality of the report then go ahead and use it. I've done it myself many times. What I mentioned in my previous post was misuse due to poor programming. [/quote]Unfortunately, your initial posting read very definite and you posted the 2 reasons that you felt were the the use of locking hints. If I misread, I am sorry - but how many others?

RE: Persisted vs Non-persisted Computed columns

Sree Arjun Div — Thu, 07 Oct 2010 07:11:36 GMT

Nice questionI am much afraid by seeing the questionpost this type questions

RE: Persisted vs Non-persisted Computed columns

Daniel Bowlin — Thu, 07 Oct 2010 06:58:10 GMT

Good question.....lots of reading first thing in the AM.

RE: Persisted vs Non-persisted Computed columns

Nils Gustav Stråbø — Thu, 07 Oct 2010 06:42:36 GMT

Did I ever say that this hint was never to be used? For reporting purposes where the possibility of reading dirty data does not affect the quality of the report then go ahead and use it. I've done it myself many times. What I mentioned in my previous post was misuse due to poor programming.

RE: Persisted vs Non-persisted Computed columns

Hugo Kornelis — Thu, 07 Oct 2010 06:39:47 GMT

[quote][b]sjimmo (10/7/2010)[/b][hr][quote]HugoShame about the (nolock) hints, though. They serve no purpose, and they might lead people to believe that these hints are okay to use. They are not, except in a very limited amount of situations. [/quote]What would these limited situations be?[/quote]Mostly queries that report aggregated data for an audience that doesn't care about the exact number, but only wants an indication. If the numbers my query produces will be rounded to the nearest million dollars anyway, I can live with a result that might be a couple of thousand dollars off.

RE: Persisted vs Non-persisted Computed columns

sjimmo — Thu, 07 Oct 2010 06:17:50 GMT

First - this was a very good question. Learned something today which is great.[quote]HugoShame about the (nolock) hints, though. They serve no purpose, and they might lead people to believe that these hints are okay to use. They are not, except in a very limited amount of situations. [/quote]What would these limited situations be?[quote]NilsYou won't believe how many people that actually use NOLOCK or READ_UNCOMMITTED to "fix" concurrency and locking problems. One of our vendors use NOLOCK in every(!) SELECT statement in a system we have purchased from them. They have never bothered to explain the reason for doing it, but my guess, after doing some investigation, is that it is caused by two things:- Lots of missing indexes, which results in excessive scans- Bad programming, because when I save data in the front-end, SQL Profiler showed that before the transaction for the update statement was committed, another connection from my pc was opened trying to read the updated row(s). [/quote]How about thousands of connections hitting against high volume OLTP databases that are also required for up to the minute reporting where replication is not an issue nor are dirty reads?There are times where hints are required. I may not like them as the norm, but I don't think that one should dismiss them all together.

RE: Persisted vs Non-persisted Computed columns

Nils Gustav Stråbø — Thu, 07 Oct 2010 05:31:20 GMT

[quote][b]Hugo Kornelis (10/7/2010)[/b][hr]and they might lead people to believe that these hints are okay to use. They are not, except in a very limited amount of situations.[/quote]You won't believe how many people that actually use NOLOCK or READ_UNCOMMITTED to "fix" concurrency and locking problems. One of our vendors use NOLOCK in every(!) SELECT statement in a system we have purchased from them. They have never bothered to explain the reason for doing it, but my guess, after doing some investigation, is that it is caused by two things:- Lots of missing indexes, which results in excessive scans- Bad programming, because when I save data in the front-end, SQL Profiler showed that before the transaction for the update statement was committed, another connection from my pc was opened trying to read the updated row(s).

RE: Persisted vs Non-persisted Computed columns

Koen Verbeeck — Thu, 07 Oct 2010 04:33:04 GMT

Nice question.I was afraid it was some trick question of some exotic behaviour of SQL Server (with the no lock hints and everything), but luckily common sense was enough to answer the question.

RE: Persisted vs Non-persisted Computed columns

Hugo Kornelis — Thu, 07 Oct 2010 04:27:39 GMT

Nice question, though maybe a bit too easy.Shame about the (nolock) hints, though. They serve no purpose, and they might lead people to believe that these hints are okay to use. They are not, except in a very limited amount of situations.

RE: Persisted vs Non-persisted Computed columns

Ian Elliott — Thu, 07 Oct 2010 03:56:16 GMT

Argh didn't get it wrong because I didn't know the answer but rather because I can't seem to count this morining.:blush:

RE: Persisted vs Non-persisted Computed columns

GPO — Thu, 07 Oct 2010 01:50:28 GMT

Just one question. What's the significance of the (nolock) hint? Could you argue that none of the answers could be guaranteed in the event of a dirty read? Cheers and many thanks for the question. GPO

RE: Persisted vs Non-persisted Computed columns

SQLRNNR — Thu, 07 Oct 2010 00:34:01 GMT

Thanks for the question.

RE: Persisted vs Non-persisted Computed columns

UMG Developer — Wed, 06 Oct 2010 21:03:12 GMT

Nice question, thanks.Though I think formatting the code into little sections, and including the commented select statements, makes it a little more confusing and harder to read than it could have been.

Persisted vs Non-persisted Computed columns

VM-723206 — Wed, 06 Oct 2010 21:01:55 GMT

Comments posted to this topic are about the item [B]Persisted vs Non-persisted Computed columns[/B]