CMES-IT
asked on
Advanced T-SQL JOIN statment / stored procedure advice
Hi,
I'm trying to create a stored procedure for use on our website. It will be used a lot. I can write the stored procedure, but it's a non-trivial JOIN, and I'm worried about the performance. I'd like for someone who (hopefully) knows a lot more about SQL than I do to provide a very quick, lightweight solution for SQL Server 2005.
Here's what I'm dealing with:
I've got a people table that contains, among other things, a PeopleID, FirstName, and LastName field.
I've got a 'widget' table that contains, among other things, a WidgetID and WidgetName. A Widget can also have up to five owners. So, right now, the widget table contains five fields, WidgetOwner1, WidgetOwner2, etc. that are related to the PeopleID field in the People table.
The application that is to sit on the website will be a Widget Search. I'd like to encapsulate the Widget Search into a single stored procedure. Users will be able to search by WidgetID, WidgetName, and WidgetOwner. Let's say someone searches for all widgets where John Doe is the owner. The results will be (roughly):
ID: Name: Owners:
001 | BlueWidget | John Doe, Molly Johnson
004 | RedWidget | Jane Doe, John Doe, George Williams
009 | GreenWidget | Jack Doe, Molly Johnson, John Doe
I'm concerned because John Doe can be WidgetOwner1, WidgetOwner2, etc. but once he's found, this will have to join back to the People table to get his name. Every owner will go through this. Also, the list could be quite long, and run often.
So... I know how to do this, but probably not the "best way" or "fastest way" to do it. So, I'm looking for some advice.
How would you do it? Why is your method fast? Can it be encapsulated in a single Stored Procedure?
If necessary I can change the table structure so that the WidgetOwners are not in the same table as the WidgetID, WidgetName, etc. The solution can have as many tables as you like.
Here is the current table structure:
TABLE People:
[PersonID] [smallint] IDENTITY(1,1) NOT NULL
[FirstName] [varchar](24) NOT NULL
[LastName] [varchar](24) NOT NULL
TABLE Widget:
[WidgetID] [varchar](14) NOT NULL
[WidgetName] [varchar](64) NOT NULL
[WidgetOwner1] [smallint] NOT NULL
[WidgetOwner2] [smallint] NULL
[WidgetOwner3] [smallint] NULL
[WidgetOwner4] [smallint] NULL
[WidgetOwner5] [smallint] NULL
Thanks!
I'm trying to create a stored procedure for use on our website. It will be used a lot. I can write the stored procedure, but it's a non-trivial JOIN, and I'm worried about the performance. I'd like for someone who (hopefully) knows a lot more about SQL than I do to provide a very quick, lightweight solution for SQL Server 2005.
Here's what I'm dealing with:
I've got a people table that contains, among other things, a PeopleID, FirstName, and LastName field.
I've got a 'widget' table that contains, among other things, a WidgetID and WidgetName. A Widget can also have up to five owners. So, right now, the widget table contains five fields, WidgetOwner1, WidgetOwner2, etc. that are related to the PeopleID field in the People table.
The application that is to sit on the website will be a Widget Search. I'd like to encapsulate the Widget Search into a single stored procedure. Users will be able to search by WidgetID, WidgetName, and WidgetOwner. Let's say someone searches for all widgets where John Doe is the owner. The results will be (roughly):
ID: Name: Owners:
001 | BlueWidget | John Doe, Molly Johnson
004 | RedWidget | Jane Doe, John Doe, George Williams
009 | GreenWidget | Jack Doe, Molly Johnson, John Doe
I'm concerned because John Doe can be WidgetOwner1, WidgetOwner2, etc. but once he's found, this will have to join back to the People table to get his name. Every owner will go through this. Also, the list could be quite long, and run often.
So... I know how to do this, but probably not the "best way" or "fastest way" to do it. So, I'm looking for some advice.
How would you do it? Why is your method fast? Can it be encapsulated in a single Stored Procedure?
If necessary I can change the table structure so that the WidgetOwners are not in the same table as the WidgetID, WidgetName, etc. The solution can have as many tables as you like.
Here is the current table structure:
TABLE People:
[PersonID] [smallint] IDENTITY(1,1) NOT NULL
[FirstName] [varchar](24) NOT NULL
[LastName] [varchar](24) NOT NULL
TABLE Widget:
[WidgetID] [varchar](14) NOT NULL
[WidgetName] [varchar](64) NOT NULL
[WidgetOwner1] [smallint] NOT NULL
[WidgetOwner2] [smallint] NULL
[WidgetOwner3] [smallint] NULL
[WidgetOwner4] [smallint] NULL
[WidgetOwner5] [smallint] NULL
Thanks!
ASKER
Hi regbes,
Yep, I know. I inherited this mess from someone else (as usual, right?!) But that's why I said: "If necessary I can change the table structure so that the WidgetOwners are not in the same table as the WidgetID, WidgetName, etc."
For the purposes of my application, the WidgetOwners table would also have to have an OwnerNumber, because it's important for me to know if John Doe is Owner1, or Owner2. But anyhow, yes, the tables can be in 3NF if we need them to be.
Still, though, that would result in a join of three tables. And I'd still like some advice on how to make that join as fast and lightweight as possible.
Thanks!
Yep, I know. I inherited this mess from someone else (as usual, right?!) But that's why I said: "If necessary I can change the table structure so that the WidgetOwners are not in the same table as the WidgetID, WidgetName, etc."
For the purposes of my application, the WidgetOwners table would also have to have an OwnerNumber, because it's important for me to know if John Doe is Owner1, or Owner2. But anyhow, yes, the tables can be in 3NF if we need them to be.
Still, though, that would result in a join of three tables. And I'd still like some advice on how to make that join as fast and lightweight as possible.
Thanks!
the issue with your current design is that your query's where caluse will have to look something like
where owner1 = 123
or owner2 = 123
or owner3 = 123
or owner4 = 123
or owner5 = 123
this is realy 5 seperate queries in one
assuming you already did some sort of lookup on person to get the ID if not the joins you would have to have are just plain ugly
you would also have to index all 5 cols for speed
on the other hand one or two well place indexes on each of your tables (clusterd on the id's non clusterd on people and widget names) will keep evrything simple and quick
where owner1 = 123
or owner2 = 123
or owner3 = 123
or owner4 = 123
or owner5 = 123
this is realy 5 seperate queries in one
assuming you already did some sort of lookup on person to get the ID if not the joins you would have to have are just plain ugly
you would also have to index all 5 cols for speed
on the other hand one or two well place indexes on each of your tables (clusterd on the id's non clusterd on people and widget names) will keep evrything simple and quick
ASKER
Hi regbes,
I agree. Assume the tables are in 3NF as you proposed above. How would you write the SQL for your solution?
Thanks.
I agree. Assume the tables are in 3NF as you proposed above. How would you write the SQL for your solution?
Thanks.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
Thanks regbes,
My joins looked like yours, so we're on the same wavelength. That's what I was looking for - another opinion, or sanity check ;) I appreciate your thoughts.
Thanks again.
My joins looked like yours, so we're on the same wavelength. That's what I was looking for - another opinion, or sanity check ;) I appreciate your thoughts.
Thanks again.
how far are you in the devlopment process? i ask becuase your tables are not in 3rd normal form.
if it is still possible suggest you change your table design to the following
TABLE Widget:
[WidgetID] [varchar](14) NOT NULL
[WidgetName] [varchar](64) NOT NULL
TABLE WidgetOwners:
[WidgetID] [varchar](14) NOT NULL
[PersonID] [smallint] NOT NULL
im sure you can see how this will simplify the task you are describing