Link to home
Start Free TrialLog in
Avatar of street9009
street9009Flag for United States of America

asked on

MSSQL - Lock Row from reading by other programs

Hello,

I have an issue where I have a table that stores a unique value in it. Many programs read this unique value and then increment it. You can see the problem already - if two programs read at the same time (or even close to the same time), they get the same unique value and the result is a duplicate. The database structure cannot be changed (to use an auto-number for example).

There's another wrinkle here - I need to give priority to one program (web application) over the rest (since the web app can time out and the others can wait). So this solution needs to handle the instance where one application is doing reads/writes and if the web does a read, the other application needs to wait.

How can this be accomplished?
Avatar of Ray Paseur
Ray Paseur
Flag of United States of America image

The database structure cannot be changed (to use an auto-number for example).
That makes the design a non-starter.  The thing you're describing is called a "race condition" because the outcome is unpredictable until the race is run.  We do not design database tables this way because we do not have any way of predicting how they will work under a load.  They cannot be unit-tested with any confidence.

This might give some good ideas.
https://technet.microsoft.com/en-us/library/jj856598(v=sql.110).aspx

With databases (in general) the terms to search for include "lock" and "transaction."  Those, together with "MSSQL" might be helpful for you as you look for alternate ideas.
Avatar of Dave Baldwin
The only way I can think to do what you are asking is to write an EXE that runs on the server that all of the other programs use to access the database.  It has to arbitrate between the accesses to do what you want.  You're not going to be able to get PHP or the MS SQL server to do that.  Unless you want to entirely rewrite one of them along with the drivers for them.
Keep all logic inside a transaction. This will lock the table until new value is stored. Example:
BEGIN TRAN

-- Get the next value
SELECT @value = myValue+1
FROM MyTable

-- Update the value
UPDATE MyTable
SET myValue = @value

COMMIT

-- Use the new value as you need
INSERT INTO AnotherTable (myValueField, anotherField, ...)
VALUES (@value, 'bla bla bla', ...)

Open in new window

ASKER CERTIFIED SOLUTION
Avatar of Scott Pletcher
Scott Pletcher
Flag of United States of America image

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
Avatar of skullnobrains
skullnobrains

So this solution needs to handle the instance where one application is doing reads/writes and if the web does a read, the other application needs to wait.

that would require to abort the other transaction entirely

i'm pretty sure we'd be more helpful in finding a way to achieve your goal without that manual increment at all. what is the id used for ?

it could seem reasonable to increment the value first, and then let the programs do what they need so there is no race conditions at all and little contention so you don't need to bother with priorities...
is incrementing AFTER doing the work actually necessary ?

would that be a problem if the ids have gaps ? or are not incremental at all ?
if you can use arbitrary values or allow gaps, each app can have it's own reserved set of ids which can help letting them run at the same time.

also sql server has record-based versioning and can allow parallel execution in such cases. depending on your requirements in terms of consistency, that feature might be usable as well but i'd rather look into other possibilities first.
Avatar of street9009

ASKER

i'm pretty sure we'd be more helpful in finding a way to achieve your goal without that manual increment at all. what is the id used for ?

It's a transaction number. Yes the system design is TERRIBLE in this case but I can't change it. The way the system itself works is it looks to that field and gets a value and then puts the next # back.

it could seem reasonable to increment the value first, and then let the programs do what they need so there is no race conditions at all and little contention so you don't need to bother with priorities...
is incrementing AFTER doing the work actually necessary ?

No, I don't suppose that it is necessary to do it afterwards, but I'm not sure what effect it would have. Literally all the "work" that's done is read, increment, write (then use what was read). It amazes me that there is duplication but there is.


The read and increment should happen in the same statement.

This makes a lot of sense to me. I'm going to try this first.
SOLUTION
Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
It will be a while before I know if this has the same flaw, but I don't expect it will. I liked Scott's solution best so gave him the bulk of the points (with an assist from skull). It was very straight-forward and simple. Thanks everyone!
good to see you got something. you can test your current setup by locking the table or row manually for a few minutes ( or more depending on how often your multiple apps are scheduled )  which will force the contention issue to appear.