Monitoring database size

One thing every DBA ought to have (and many do!) is a way to monitor and routinely report on the size of databases and their growth.

This is important when it comes time to discuss the purchase or expansion of a SAN, NAS or other storage subsystem.

Frequently management will make decisions that significantly accelerate database growth, yet the nature of the change makes it almost impossible to predict with any accuracy how fast the growth will be.

Having a good means of tracking, measuring and reporting on this growth is vital to our success when wearing our DBA hat.

The script below is one I use.  Every SQL Server install I do, the first thing I do after getting the server up on its legs is I create the database 'DBA'.  This holds all my management and measurement 'stuff'.  This script creates the table tbl_dbsize (in the comment).  Then, whenever I run it, the script gets information on the database files for each database on the server and the size of the file in MB.  The "inserted on" timestamp column gives me a way to measure the growth over time.

We recently made the decision to start scanning copies of important documents at the company where I am employed; these document scans can be added to a 'file room' tab in our ERP system.  This is great for our ability to attach document copies to transactions in the system.  However, when all our locations start doing this and we have no way of knowing with certainty the size of the scanned files, close monitoring is prudent.

So, I'm ratcheting up the frequency of this job from monthly to weekly right now in order to see what impact these scans will have on growth of the database.

If you don't have a script or job like this, give this one a try.


USE [DBA]
/*
Report: Database Size
Contents:   size of database files

CREATE TABLE tbl_dbsize (
[inserted on] DATETIME,
[DB Name] NVARCHAR(255),
[Logical file] NVARCHAR(255),
[Physical file]    NVARCHAR(1000),
[MB] INTEGER
)
;
*/

INSERT INTO tbl_dbsize
SELECT GETDATE()[inserted on],
DB_NAME(database_id) AS [DB Name],
Name AS [Logical file],
Physical_Name [Physical file], (size*8)/1024 SizeMB
FROM sys.master_files




SELECT * FROM dbsize; -- <-- right now I don't have the need for an SSRS report, I dump the answer set into an Excel spreadsheet that I can use to track the growth of files over time.

As I indicated in my previous post, I'm not enamored of the geospatial mapping in SQL Server reports. But what if you don't have a brilliant person who can write code to interface to Google maps or Bing? What if your executives would rather use Google Earth?  What can we do with geospatial data under those circumstances?

Well this is what we can do. We can create a script that will write our kml file data file for use in the Google Earth application.

Here is the exercise. Suppose the President of your company comes to you and asks if it is possible to show the total sales a customer did with your company for the year, and then to plot that data in Google Earth.

Suppose further that the request is that when you hover over the marker, particular data is displayed such as the total number of orders, the type of order, the sale amount and some address information identifying the customer (number, name, city, etc.).

Of course this exercise pre-supposes that you: a) have a version of SQL Server that supports geospatial data, and b) that you have coordinates for the customer so that their location can be plotted in Google Earth.

From that point forward it is merely a straightforward matter of:

a) understanding Google Earth's kml file structure

b) a bit about manipulating markers

c) stringing kml syntax and customer data together into strings that can be inserted between the 'header' and 'footer' sections of your kml file.

So here is the first part of the script:


USE [<database>]  -- set the location to the database holding your customer info.

/*  -- here is a google red dot marker URL
GOOGLE MARKERS
http://maps.google.com/mapfiles/ms/icons/red-dot.png
*/

DECLARE @kmlhdr  NVARCHAR(MAX) -- your kml header text
DECLARE @kmlftr  NVARCHAR(MAX) -- your kml footer text
DECLARE @red  NVARCHAR(255) -- your marker text
DECLARE @divcode NVARCHAR(4)   -- an operating division to report sales on
DECLARE @yr  INT           -- the year to report on

SET @divcode =   '03'
SET @yr =        2012
SET @red =  'http://maps.google.com/mapfiles/ms/icons/red-dot.png'

-- first the easy part; set the footer
SET @kmlftr = '</Document></kml>'

-- more involved, but once you have it down, it is pretty simply and static
SET @kmlhdr = 
(
-- use a select statement to string kml 'snippets' together with descriptive info
SELECT 
'<?xml version="1.0"?><kml xmlns="http://earth.google.com/kml/2.0">
<Document>

-- below is description data that shows up in the Places section of Google Earth.
-- careful consideration of how you want to group and report sales can pay off
-- by using the tags below against data that is grouped by the tags.
-- then when multiple kml files are loaded for different sale types for 
-- different divisions, Places can be used to toggle each kml file on/off to 
-- view sales concentrations.

<name>DISTRIBUTION Sales SC Division</name>      
<description>Division Sales | My Company, Inc.</description>   

-- next comes 3 style sections; 
-- A1 is the normal size of the marker

<Style id="A1"><IconStyle><scale>1</scale><Icon><href>'+@red+'</href></Icon><hotSpot x=".3" y=".8" xunits="fraction" yunits="fraction" /></IconStyle><LabelStyle><scale>0</scale></LabelStyle><BalloonStyle><text><![CDATA[ <b>$[name]</b><br />$[address]<br />$[description]<br/><br/>$[geDirections] ]]></text></BalloonStyle></Style>\n

-- ok a bit about the above kml code; CDATA is going to format some info in 

-- each <Placemark> row we create in the next T-SQL script section.

-- A2 is going to show the marker 'inflated' when the President mouses over it
<Style id="A2"><IconStyle><scale>1.5</scale><Icon><href>'+@red+'</href></Icon><hotSpot x=".3" y=".8" xunits="fraction" yunits="fraction" /></IconStyle><LabelStyle><scale>1.0</scale></LabelStyle><BalloonStyle><text><![CDATA[ <b>$[name]</b><br />$[address]<br />$[description]<br/><br/>$[geDirections] ]]></text></BalloonStyle></Style>\n

-- A0 is the style map
<StyleMap id="A0"><Pair><key>normal</key><styleUrl>#A1</styleUrl></Pair><Pair><key>highlight</key><styleUrl>#A2</styleUrl></Pair></StyleMap>'

)

-- I won't go into the details of the above style sections, you can read up on 

-- them with the Google Earth documents.  Just trust me, a lot of time when into

-- figuring it out, and it works!!

-- Again, the point of this exercise is to show how data can be pulled from 

-- invoicing information and customer information, and mapped in Google Earth.

-- so this part of the code is simple mechanics to create some sales data... 

-- I'll assume you, the reader can pull your own data.

-- Gets basic sales information...

SELECT ISNULL(wo.[serv-code],'')[serv-code],
i.[disp-no], i.[order-no],
CASE WHEN i.[disp-no]<> '' THEN 'service' ELSE 'distribution' END [ordtype], 
i.[Document], i.[Invoice-date], 
i.[Charge-cust], i.[Cust-no], i.[Ship-name], i.[Ship-address__1], i.[Ship-city], i.[Ship-st], i.[Ship-zip],
i.[Gross-amt]+i.[Extra-Charge__9]+i.[Extra-Charge__10][Gross-amt] 
INTO #stg_invoicedata
FROM [dbo].[invoice] AS i
LEFT OUTER JOIN [dbo].[wo] AS wo ON i.[disp-no]=wo.[wo-no]
WHERE DATEPART(yyyy,i.[Invoice-date]) = @yr
AND i.[div-code]=@divcode
;

-- cubes the sales data to get a sales total ytd for each customer
SELECT
CASE WHEN GROUPING([serv-code])= 1 THEN 'All' ELSE [serv-code] END [serv-code],
CASE WHEN GROUPING([ordtype])= 1 THEN 'All' ELSE [ordtype] END [ordtype],
-- CASE WHEN GROUPING([Document])= 1 THEN 'All' ELSE [Document] END [Document], 
CASE WHEN GROUPING([Charge-cust])= 1 THEN 'All' ELSE [Charge-cust] END [Charge-cust], 
CASE WHEN GROUPING([Cust-no])= 1 THEN 'All' ELSE [Cust-no] END [Cust-no], 
CASE WHEN GROUPING([Ship-name])= 1 THEN 'All' ELSE [Ship-name] END [Ship-name], 
CASE WHEN GROUPING([Ship-address__1])= 1 THEN 'All' ELSE [Ship-address__1] END [Ship-address__1], 
CASE WHEN GROUPING([Ship-city])= 1 THEN 'All' ELSE [Ship-city] END [Ship-city], 
CASE WHEN GROUPING([Ship-st])= 1 THEN 'All' ELSE [Ship-st] END [Ship-st], 
SUM([Document])[Orders],
SUM([Gross-amt])[Dollars]
INTO #stg_invoicecube2
FROM #stg_invoicecube
GROUP BY [serv-code],[ordtype],[Document],[Charge-cust],[Cust-no],[Ship-name],[Ship-address__1],[Ship-city],[Ship-st]
WITH cube
HAVING  [serv-code]<> 'All'
AND [ordtype]<> 'All'
AND [Charge-cust]<> 'All'
AND [Cust-no]<> 'All'
AND [Ship-name]<> 'All'
AND [Ship-address__1]<> 'All'
AND [Ship-city]<> 'All'
AND [Ship-st]<> 'All'
;

-- joins the sales data with the customer data to get the coordinates
SELECT c.*, c2.[Zip-code], c2.[num-lat],  c2.[num-lng], 
CASE WHEN c2.[num-lat]='0' THEN 'geocode'  END [num-latflag], 
CASE WHEN c2.[num-lng]='0' THEN 'geocode'  END [num-lngflag] 
INTO #tmp_invoicecube 
FROM #stg_invoicecube2 c
INNER JOIN [dbo].[customer] AS c2 ON c.[Cust-no]=[c2].[Cust-no]
;

So below is a row from the table just to make clear what we want...

Cust-no Ship-name   Ship-address__1     Ship-city Ship-st

10170 MY SAMPLE CUSTOMER 123 Anywhere Street Anytown SC

Orders Dollars Zip-code num-lat  num-lng

1 3066.58 29071 35.43802 -80.522617

-- Now we want to create <Placemark> rows for each row in the table.  

-- In the example above I plot this row from this data.

-- Here is the script to do that...

SELECT '<Placemark><name>ORDERS: '+cast([orders] as nchar(1))+' DOLLARS: '+cast(cast([dollars] as decimal(10,2))as nvarchar(255))+' TYPE: '+Upper([ordtype])+' - BILL TO: '+[cust-no]+' '+[ship-name]+' - SHIP TO: '+
[cust-no]+' '+[ship-name]+' '+[ship-address__1]+'</name><description>'+
[serv-code]+'</description><Point><coordinates>'+cast([num-lng]as nvarchar(255))+','+cast([num-lat] as nvarchar(255))+',0</coordinates></Point><address>'+[ship-address__1]+' '+[ship-city]+', '+cast([ship-st] as nvarchar(255))+' '+cast([zip-code] AS nvarchar(255))+'</address><styleUrl>#A0</styleUrl></Placemark>'

FROM #tmp_invoicecube

If you know a bit about strings, you should be able to follow the SELECT statement above. Simply stated, the name element is going to hold descriptive data about the total number of orders 'ORDERS: ' and the [orders] data value; total sales ' DOLLARS: ' and the [dollars] data value; plus various other data you and your President decide to have shown when a marker is moused over.  Next we provide the geospatial plot between the <Point><coordinates> tags. Lastly we provide the full customer address data between the <address> tags.

Substitute your own column names from your own invoiced data table.  To help you I have encased my column names in square brackets in the SELECT statement.

The resulting text string(s) your script generates should be in the form of the one below.

<Placemark><name>ORDERS: 1 DOLLARS: 3066.58 TYPE: DISTRIBUTION - BILL TO: 10170 MY SAMPLE CUSTOMER - SHIP TO: 10170 MY SAMPLE CUSTOMER 123 Anywhere Street</name><description>DIST</description><Point><coordinates>-80.522617,35.438020,0</coordinates></Point><address>123 Anywhere Street Anytown, SC 29071</address><styleUrl>#A0</styleUrl></Placemark>

Now take the resulting answer sets and paste into a simple text editor. Put the output for your @kmlhdr variable into the text editor first.  Next place the data set for the script that generates the Placemark strings into the text editor. Last, do not forget to place the text for your @kmlftr variable into the text editor. Then save the file and replace the .txt with .kml.

After you have Google Earth installed, double click your .kml file. When your file is opened, you will see a red dot marker designating the location of sales for a Placemark row. When you hover over it with your mouse, you'll see a label that describes the order - total orders and sales dollars - and customer information.

Setting coordinates set once you have spatial data

Although I don't find the spatial maps in SSRS all that useful (I'll post another article in the future on generating kml files for Google Earth); one thing that some folks have found tedious is establishing the coordinates once the latitude and longitude have been updated in a table.

This can be easily done with a while loop, as demonstrated below.  Suppose you want to create a bubble map of sales data.

First you need some invoices.  To be pragmatic here, I'll assume you are familiar with your own invoice tables, so instead of creating them here, we'll just show the code I used to extract invoice information from my employer's system:


use DBA

declare @divcode nvarchar(10)  -- divisional code
declare @yr int                           -- year to extract
set @divcode = '00'              
set @yr=2012


if exists(select * from sysobjects where type = 'u' and [name]='tbl_mapdinvoices')
drop table tbl_mapdinvoices
;

select IDENTITY(int, 1,1) rowid,
cb.[div-code], cb.[Cust-no], c.[address],c.[city],c.[st],c.[zip-code],c.[map-grid],c.[num-lat],c.[num-lng], cb.[Total]
into tbl_mapdinvoices
from
(
SELECT
case when GROUPING([div-code])=1 then 'All' else [div-code] end [div-code],
case when grouping([Charge-cust])=1 then 'All' else [Charge-cust] end [Charge-cust],
case when grouping([Cust-no])=1 then 'All' else [Cust-no] end [Cust-no],
sum([Gross-amt])+sum([Tax-amt__1])+sum([Tax-amt__2])+sum([Tax-amt__3])+sum([Extra-Charge__1])+sum([Extra-Charge__2])+sum([Extra-Charge__3])
+sum([Extra-Charge__4])+sum([Extra-Charge__5])+sum([Extra-Charge__6])+sum([Extra-Charge__7])+sum([Extra-Charge__8])+sum([Extra-Charge__9])
+sum([Extra-Charge__10])[Total]
FROM [dbo].[invoice]
where DATEPART(yyyy,[invoice-date]) IN (@yr)
group by [div-code], [Charge-cust], [Cust-no]
with CUBE
having [div-code] IN ( @divcode )
and [Charge-cust] <> 'All'
and [Cust-no] <> 'All'
) cb
inner join customer c on cb.[Cust-no]=c.[Cust-no]
and c.[num-lat]+c.[num-lng]<>0
;

Notice the IDENTITY column above?  The column rowid will serve as a means to 'loop through' the entire table.  If one has say, 10 or 20 invoices in the table, manually updating the table to get coordinates isn't a big deal.  But if you have a table of 1,000 or 10,000 or 100,000 invoices to map, well, you need a better way.

Here's a redacted listing of the first two rows of my invoice table to give you an idea of what we have so far:
rowid    div-code  city                       st      num-lat              num-lng           total
1          00            MARYVILLE TN 35.754724        -83.971210 524.10
2          00            IRVING               TX   32.917917        -97.002038 511.31


Now we're going to add a column to our table for coordinates, we will call it 'spatialdata'

alter table tbl_mapdinvoices add spatialdata geography;

Next, we set up the WHILE LOOP to create the coordinates for us.  We're going to do a number of things here, so let's look at the code:


declare @counter int
set @counter = 1
declare @sql nvarchar(max)
declare @lat decimal(10,6)
declare @lng decimal(10,6)

while @counter < (select MAX(rowid)+1 from tbl_mapdinvoices)
begin
set @lat = (select ([num-lat]) from tbl_mapdinvoices where rowid = @counter)
set @lng = (select ([num-lng]) from tbl_mapdinvoices where rowid = @counter)
set @sql = 'update tbl_mapdinvoices set spatialdata ='+''''+'POINT('+(select cast(@lng as varchar(16)) )+' '+(select cast(@lat as varchar(16)) )+')'+''''
set @sql = @sql+' where rowid ='+(select cast(@counter as nvarchar(10)) )+';'
--print @sql
exec sp_executesql @sql
set @counter = @counter+1
end

First we declare our counter as an integer and set it's value to '1'.
While declaring "stuff", we're going to create a string for our UPDATE sql, as well as declaring our latitude and longitude variables.

We initiate our loop with a WHILE statement that essentially says, while the counter variable is less than the value of the maximum row id plus 1 from the table of invoices continue looping.
(You may be asking 'Why not just use a cursor'?  The answer is that there are multiple posts and myriad books and other analyses that demonstrate cursors are rarely more efficient than using other T-SQL tactics. I have therefore abandoned using cursors completely as I have never found a need for them.)

Next we set our latitude variable to the value of the num-lat column in our invoice table where the rowid in the table is equal to the counter.  Likewise we set the value of longitude variable to our num-lng column value.

Now we employ the DYNAMIC SQL technique to build our UPDATE string.
If you are not familiar with DYNAMIC SQL, once you have an invoice table built with a rowid as described in the code, comment out the exec sp_executesql @sql statement and uncomment out the print @sql statement and run the WHILE LOOP.  You should see a series of statements in your answer set that look something like this:

   update tbl_mapdinvoices set spatialdata ='POINT(-83.971210 35.754724)' where rowid =1;
   update tbl_mapdinvoices set spatialdata ='POINT(-97.002038 32.917917)' where rowid =2;
   update tbl_mapdinvoices set spatialdata ='POINT(-80.949954 34.071876)' where rowid =3;
   update tbl_mapdinvoices set spatialdata ='POINT(-79.221837 35.431453)' where rowid =4;

(DYNAMIC SQL is a good way to programmatically build statements for subsequent execution.)
Don't forget to uncomment the exec sp_executesql @sql statement and comment out the print @sql statement before you run the code though.
The exec sp_executesql @sql statement is the last thing in the WHILE LOOP before incrementing the counter and returning to the top of the loop.

Once this WHILE loop is finished, you will have populated the spatialdata column of your table.
Below is a redacted listing of the first two rows of my invoice table:


rowid city                  st num-lat         num-lng         spatialdata
1         MARYVILLE TN 35.754724 -83.971210 0xE6100000010CCFC0C8CB9AE0414015E3FC4D28FE54C0
2         IRVING         TX 32.917917 -97.002038 0xE6100000010CA2B8E34D7E7540405AD6FD63214058C0


Now we want to convert the spatialdata into coordinates.

alter table tbl_mapdinvoices add coordinates nvarchar(255);


update tbl_mapdinvoices
set coordinates = convert(nvarchar(255),[spatialdata])
;

Below is a redacted listing of the first two rows of my invoice table with coordinates.



rowid city                  st num-lat         num-lng         coordinates
1         MARYVILLE TN 35.754724 -83.971210 POINT (-83.97121 35.754724)
2         IRVING         TX 32.917917 -97.002038 POINT (-97.002038 32.917917)


Now I have the table ready to map the data.
To see an excellent example of how to spatially map data in SSRS, see the posting at this link:
http://www.mssqltips.com/sqlservertip/2174/maps-in-sql-server-2008-r2-reporting-services/

Those pesky aggregate reports - USING CUBE

Over the course of using SQL Server for the years, I confess I have come up with myriad ways of producing those pesky aggregate reports that management so loves; such a report being the YTD sales of customers.

Some of the methods I have created have been to UNION multiple tables holding the data for different groups, building derived tables, doing the equivalent with CTE, and probably many others.  I've tried using stored procedures as well, but invariably each request is an ad-hoc one, and the stored procedure approach led nowhere because of it.

Then came 'CUBE'.  I have settled off on using this because it very easily and readily creates the information I need.  (Now if Microsoft would only improve the PIVOT operation so that multiple aggregates could be handled... but I digress).

Here is an example of how to use a CUBE operation to simplify the summation of sales data for a report of customers' sales YTD.

First we need to build a customer table....

-- create some customers
if exists(select * from sysobjects where type = 'U' and name = 'customer')
drop table customer
;

SELECT '28549'[billing #], '28549'[customer #], cast('ABC Corporation' as nvarchar(255))[name], cast('123 Anywhere Place' as nvarchar(255))[Address],
cast('Irmo' as nvarchar(255))[City], 'SC'[St]
into customer;

insert into customer SELECT '32568'[billing #], '32568'[customer #], 'ABC Corporation'[name], '123 Anywhere Place'[Address], 'Irmo'[City], 'SC'[St];
insert into customer SELECT '46598'[billing #], '46598'[customer #], 'Hannibal Cartons'[name], '1235 Courtney Blvd.'[Address], 'Walterboro'[City], 'SC'[St];
insert into customer SELECT '84956'[billing #], '84956'[customer #], 'Quattle Container'[name], '3857 Cedar Road'[Address], 'Lexington'[City], 'SC'[St];
insert into customer SELECT '51324'[billing #], '51324'[customer #], 'Simmons Freightlines'[name], '456 Anywhere Place'[Address], 'Irmo'[City], 'SC'[St];

select * from customer
;

billing #	customer #	name	Address	City	St
28549	28549	ABC Corporation	123 Anywhere Place	Irmo	SC
32568	32568	ABC Corporation	123 Anywhere Place	Irmo	SC
46598	46598	Hannibal Cartons	1235 Courtney Blvd.	Walterboro	SC
84956	84956	Quattle Container	3857 Cedar Road	Lexington	SC
51324	51324	Simmons Freightlines	456 Anywhere Place	Irmo	SC

Next we need to build some invoice data.  Being the pragmatic sort, since I don't need to provide breakdowns by items sold, I don't need the line data, so I'll be using data from the invoice header table only.


-- create some invoice headers
if exists(select * from sysobjects where type = 'U' and name = 'invoicehdr')
drop table invoicehdr
;

select '46598'[billing #], '46598'[customer #], cast('05-11-2012' as datetime)[invoice-date], 1256.55[gross], cast(1256.55*.05 as decimal(10,2))[tax] into invoicehdr
insert into invoicehdr select '28549'[billing #], '28549'[customer #], cast('06-30-2012' as datetime)[invoice-date], 125.36[gross], cast(125.36*.05 as decimal(10,2))[tax]
insert into invoicehdr select '28549'[billing #], '28549'[customer #], cast('07-12-2012' as datetime)[invoice-date], 326.10[gross], cast(326.10*.05 as decimal(10,2))[tax]
insert into invoicehdr select '28549'[billing #], '28549'[customer #], cast('07-28-2012' as datetime)[invoice-date], 42.10[gross], cast(42.10*.05 as decimal(10,2))[tax]
insert into invoicehdr select '32568'[billing #], '32568'[customer #], cast('01-21-2012' as datetime)[invoice-date], 442.30[gross], cast(442.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '32568'[billing #], '32568'[customer #], cast('03-19-2012' as datetime)[invoice-date], 442.30[gross], cast(442.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '32568'[billing #], '32568'[customer #], cast('04-19-2012' as datetime)[invoice-date], 442.30[gross], cast(442.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '32568'[billing #], '32568'[customer #], cast('05-19-2012' as datetime)[invoice-date], 442.30[gross], cast(442.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '32568'[billing #], '32568'[customer #], cast('06-19-2012' as datetime)[invoice-date], 442.30[gross], cast(442.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '32568'[billing #], '32568'[customer #], cast('07-19-2012' as datetime)[invoice-date], 442.30[gross], cast(442.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '46598'[billing #], '46598'[customer #], cast('01-19-2012' as datetime)[invoice-date], 546.30[gross], cast(546.30*.05 as decimal(10,2))[tax]
insert into invoicehdr select '46598'[billing #], '46598'[customer #], cast('03-10-2012' as datetime)[invoice-date], 342.20[gross], cast(342.20*.05 as decimal(10,2))[tax]
insert into invoicehdr select '51324'[billing #], '51324'[customer #], cast('03-14-2012' as datetime)[invoice-date], 156.89[gross], cast(156.89*.05 as decimal(10,2))[tax]
insert into invoicehdr select '51324'[billing #], '51324'[customer #], cast('04-23-2012' as datetime)[invoice-date], 423.65[gross], cast(423.65*.05 as decimal(10,2))[tax]

select * from invoicehdr
;

billing #	customer #	invoice-date	gross	tax
46598	46598	5/11/2012	1256.55	62.83
28549	28549	6/30/2012	125.36	6.27
28549	28549	7/12/2012	326.1	16.31
28549	28549	7/28/2012	42.1	2.11
32568	32568	1/21/2012	442.3	22.12
32568	32568	3/19/2012	442.3	22.12
32568	32568	4/19/2012	442.3	22.12
32568	32568	5/19/2012	442.3	22.12
32568	32568	6/19/2012	442.3	22.12
32568	32568	7/19/2012	442.3	22.12
46598	46598	1/19/2012	546.3	27.32
46598	46598	3/10/2012	342.2	17.11
51324	51324	3/14/2012	156.89	7.84
51324	51324	4/23/2012	423.65	21.18

Now we have a customer table and we have some invoice data for those customers; we are ready to create a sales ytd report.

Using the CUBE operation, we can easily and rapidly get our summary data as shown below:

select
case when grouping(i.[billing #])=1 then 'All' else [billing #] end [billing #],
case when grouping(i.[customer #])=1 then 'All' else [customer #] end [customer #],
sum(i.gross)gross,
sum(i.tax)tax
from invoicehdr as i with (nolock)
group by i.[billing #], i.[customer #]
with cube
order by [billing #], [customer #]
;

To use the 'with cube' operation, you need to use the GROUPING() statement on the columns you want to group by. The GROUPING statement will return a value of 1 for the group and a 0 for each element in the group. However, I like to case my GROUPING() statements to return the word, 'All' for the total and then the data for each element in the group.  It just makes more sense to me. You can use whatever you choose, that's one of the points of being pragmatic.

The query returns the following answer set:

billing #	customer #	gross	tax
28549	28549	493.56	24.69
28549	All	493.56	24.69
32568	32568	2653.8	132.72
32568	All	2653.8	132.72
46598	46598	2145.05	107.26
46598	All	2145.05	107.26
51324	51324	580.54	29.02
51324	All	580.54	29.02
All	28549	493.56	24.69
All	32568	2653.8	132.72
All	46598	2145.05	107.26
All	51324	580.54	29.02
All	All	5872.95	293.69

Notice that for the billing # column I get a row back for each billing # in the invoice table, and I get an "All" row back for each data in the customer # column; plus 1 row where the customer # value is "All". Additionally, I get a row of data for each customer # and an "All" row for each customer #. These rows are basically the group total rows.  I don't find them of much value, because I usually create our company's reports in SSRS. SSRS provides a grouping function in the tablix control to render group and sub-group totals so these rows only get in the way.

Here is how to filter them out easily... A good YTD report is going to show more data about the customer than their billing and customer numbers, so we want to perform a join of the CUBE query answer set and the customer table as shown below:


select cb.[billing #], cb.[customer #], c.[name], c.[Address], c.City, c.St, cb.gross, cb.tax
from
(
select
case when grouping(i.[billing #])=1 then 'All' else [billing #] end [billing #],
case when grouping(i.[customer #])=1 then 'All' else [customer #] end [customer #],
sum(i.gross)gross,
sum(i.tax)tax
from invoicehdr as i with (nolock)
group by i.[billing #], i.[customer #]
with cube
) cb
inner join customer as c with (nolock) on cb.[billing #]=c.[billing #] and cb.[customer #]=c.[customer #]
order by gross desc
;

Since I know there is no billing # or customer # data with the value "All" the inner join is going to filter out the group and subgroup total rows out of my cube.  Since the billing and customer numbers are identical in this data set, and in the interest of having space on my post page for a tidy report, I'll eliminate the billing # from the answer set:

customer #		name	Address	City	St	gross	tax
32568		ABC Corporation	123 Anywhere Place	Irmo	SC	$ 2,653.80	$ 132.72
46598		Hannibal Cartons	1235 Courtney Blvd.	Walterboro	SC	$ 2,145.05	$ 107.26
51324		Simmons Freightlines	456 Anywhere Place	Irmo	SC	$    580.54	$   29.02
28549		ABC Corporation	123 Anywhere Place	Irmo	SC	$    493.56	$   24.69

Since no billing # matches the 'All' value in the cube data set, and likewise for the customer #, as I wrote above, these rows are eliminated from the report.  All we have are the total sales year-to-date for the customers in our customer table.

This is so much simpler than any other method I've ever used.  If you don't know about the CUBE and ROLLUP clauses you need to check them out.  They're tremendous time savers (and head ache savers)!!

Use INFORMATION_SCHEMA.COLUMNS to help build your INSERT, SELECT or other query

Last month I posted on a 'pragmatic' way to use internal tables to help build query statements dealing with large tables.

However, as INFORMATION_SCHEMA is the "new, ANSI standard" way to go, I thought I'd show that it can be used in a similar manner.

Also, I've often read posts where it is alluded that INFORMATION_SCHEMA should be used over sysinternals such as sysobjects and syscolumns.

I recently read this post from SQLServerCentral.com that shows this assertion is false at this time.
http://www.sqlservercentral.com/Forums/Topic1272370-1526-1.aspx#bm1277421

In the case of listing column names, either way is OK.  However, you should be aware of when and why you choose to use either.  There are some risks involved in always relying on INFORMATION_SCHEMA as Gail Shaw points out.

So, in the interest of preventing this site from going stale, even as I try to get on top of my project pile, I offer this short post to show that you can indeed continue to use this technique with the newer INFORMATION_SCHEMA.COLUMNS view to get a list of table column names in order of their position in the table.

Try this comparison to help you get comfortable with using the new "IS" views.


use dba -- < pick your own database containing large tables

-- pick a large table and substitute its name for 'nat'.

-- THIS IS THE OLD WAY
select name from syscolumns where id in (select id from sysobjects where name = 'nat') order by colorder
-- ANSWER SET
/*EventLog, RecordNumber, TimeGenerated, TimeWritten, EventID, EventType, EventTypeName, EventCategory, EventCategoryName, SourceName, Strings, ComputerName, SID, Message, Data*/

-- THIS IS THE NEW WAY
-- ANSWER SET
select column_name from information_schema.columns where table_name = 'nat' order by ordinal_position
/*EventLog, RecordNumber, TimeGenerated, TimeWritten, EventID, EventType, EventTypeName, EventCategory, EventCategoryName, SourceName, Strings, ComputerName, SID, Message, Data*/

Personally, I'm starting to like the new way better as I don't have to reference the INFORMATION_SCHEMA.tables in order to pull the data from INFORMATION_SCHEMA.columns.
The old way I have to use sysobjects to get the table id for the syscolumns to use.

You can use the INFORMATION_SCHEMA views in a number of interesting ways to help you reduce the time it takes to work with larger objects in your databases.


The following views apply ...

INFORMATION_SCHEMA.CHECK_CONSTRAINTS
INFORMATION_SCHEMA.COLUMN_DOMAIN_USAGE
INFORMATION_SCHEMA.COLUMN_PRIVILEGES
INFORMATION_SCHEMA.COLUMNS
INFORMATION_SCHEMA.CONSTRAINT_COLUMN_USAGE
INFORMATION_SCHEMA.CONSTRAINT_TABLE_USAGE
INFORMATION_SCHEMA.DOMAIN_CONSTRAINTS
INFORMATION_SCHEMA.DOMAINS
INFORMATION_SCHEMA.KEY_COLUMN_USAGE
INFORMATION_SCHEMA.PARAMETERS
INFORMATION_SCHEMA.REFERENTIAL_CONSTRAINTS
INFORMATION_SCHEMA.ROUTINE_COLUMNS
INFORMATION_SCHEMA.ROUTINES
INFORMATION_SCHEMA.SCHEMATA
INFORMATION_SCHEMA.TABLE_CONSTRAINTS
INFORMATION_SCHEMA.TABLE_PRIVILEGES
INFORMATION_SCHEMA.TABLES
INFORMATION_SCHEMA.VIEW_COLUMN_USAGE
INFORMATION_SCHEMA.VIEW_TABLE_USAGE
INFORMATION_SCHEMA.VIEWS

Take time to become familiar with the information these views provide.  It can really help you increase your productivity writing T-SQL.

What happened to the posts?

Well, start a new blog, then get buried in projects after only 3 posts... embarrassed!

I hope to be back soon.  Apologies for the delay, but gotta earn a living ;-)

SQL Middle Earth - putting that hobbit id to pragmatic use

What's a hobbit in SQL Server?


A 'hobbit' is a HOBT (heap or btrieve) value.
There, doesn't that clear things up?


For more about what a HOBT is and why it was named thus, refer to 
http://thehobt.blogspot.com/2009/02/what-heck-is-sql-server-hobt-anyway.html.


I will hold forth on the usefulness I found for understanding what a hobbit is as follows, if you use
a commercial monitoring tool for SQL Server, (such as SQL Monitor from Red Gate Software), when your monitoring encounters a deadlock and you start to drill in you may notice a line such as the one below:




SPID Victim Lock details                         Statement type
138                      dbid: 5, object id: 72057594727628800, index id: (0100e420fa5a) UPDATE Line #: 1
169 Victim process dbid: 5, object id: 72057594727628800, index id: (0100e420fa5a) UPDATE Line #: 18


Now that object id, 72057594727628800, is not a real object id, you won't find it in the sysobjects table.  It is a hobbit object id.  When I click on the line for SPID 138, I find that the HOBT id is associated with object id 440335512, and when I click on the other line, I get 1644479766.


But I don't want to take a lot of time poking all the links and buttons in SQL Monitor (Oh, sometimes I do because I need to know the details).  But if your company is like the company I'm currently employed at, the applications running on SQL Server are most likely 3rd party software.  In other words, a) you don't have the source code; b) even if you did the vendor would take a dim view of you rummaging around in and worse, changing their code; c) you are too busy doing what your employer pays you to do, why should you work on the vendor's code for them?


So the most urgent thing for me is simply to find out as much as I can, as quickly as I can for the deadlock so  that I can open an issue on the Vendor's trouble ticket system and then they can go figure out why the code is prone to deadlocking.


Here's a nifty little bit of code that has helped me out in doing this:


USE [SERVICE] -- <-- your database name goes here


DECLARE @hobbit BIGINT
DECLARE @hobbitobj NVARCHAR(255)
DECLARE @hobbitidx INT


DECLARE @victimhobbit BIGINT
DECLARE @victimhobbitobj NVARCHAR(255)
DECLARE @victimhobbitidx INT


DECLARE @victimobject BIGINT
DECLARE @object BIGINT




SET @victimhobbit = 72057594727628800 
SET @victimobject = 1644479766
SET @hobbit = 72057594727628800 
SET @object = 440335512


SET @hobbitobj = (SELECT [object_id] FROM sys.partitions WHERE hobt_id = @hobbit)
SET @hobbitidx = (SELECT index_id FROM sys.[partitions] WHERE  hobt_id = @hobbit)
SET @victimhobbitobj = (SELECT [object_id] FROM sys.partitions WHERE hobt_id = @victimhobbit)
SET @victimhobbitidx = (SELECT index_id FROM sys.[partitions] WHERE  hobt_id = @victimhobbit)




SELECT 'victim'msg, hobt_id, object_name(p.[object_id]) [object], p.[object_id], index_id FROM sys.partitions p WHERE hobt_id = @victimhobbit
SELECT hobt_id, object_name(p.[object_id]) [object], p.[object_id], index_id FROM sys.partitions p WHERE hobt_id = @hobbit
SELECT 'victim'msg,[object_id], [name], [index_id], [type_desc] FROM sys.[indexes] AS i WHERE [object_id] = @hobbitobj AND [index_id]=@victimhobbitidx
SELECT [object_id], [name], [index_id], [type_desc] FROM sys.[indexes] AS i WHERE [object_id] = @hobbitobj AND [index_id]=@hobbitidx
SELECT 'victim'msg, name, id FROM sysobjects WHERE [id] = @victimobject
SELECT name, id FROM sysobjects WHERE [id] = @object


The results tell me about the name of the object associated with both 'hobbits', in this instance 2 SPIDS wanted access to the exact same object at the exact same time; moreover they were attempting to access the same object on the same index, and finally I have the name of the 2 stored procedures that were running.


So I can quickly copy this information into a ticket, the most important and useful part is that I can provide the vendor with the names of the stored procedures that deadlocked.


This has helped me by equipping the vendor with the information they need to start figuring out what they can do to reduce or eliminate deadlocks in their code.


Have fun!