Home/ Questions/Q 184389
In Process

The Archive Base Latest Questions

Editorial Team
  • 0
Asked: 2026-05-11T15:17:00+00:00

I have what looked to me at first glance a very simple problem. I

  • 0

I have what looked to me at first glance a very simple problem. I want to be able to obtain a unique key value with a prefix. I have a table which contains ‘Prefix’ and ‘Next_Value’ columns.

So you’d think you just start a transaction, get the next value from this table, increment the next value in the table and commit, concatenate the prefix to the value, and you are guaranteed a series of unique alphanumeric keys.

However under load, with various servers hitting this stored proc via ADO.NET, I’ve discovered that from time to time it will return the same key to different clients. This subsequently causes an error of course when the key is used as a primary key!

I had naively assumed BEGIN TRAN…COMMIT TRAN ensured the atomicity of data accesses within the scope. In looking into this I discovered about transaction isolation levels and added SERIALIZABLE as the most restrictive – with no joy.

  Create proc [dbo].[sp_get_key]    @prefix nvarchar(3)   as   set tran isolation level SERIALIZABLE   declare       @result nvarchar(32)      BEGIN TRY       begin tran        if (select count(*) from key_generation_table where prefix = @prefix) = 0 begin          insert into key_generation_table (prefix, next_value) values (@prefix,1)       end        declare @next_value int        select @next_value = next_value       from key_generation_table       where prefix = @prefix        update key_generation_table       set next_value = next_value + 1        where prefix = @prefix        declare @string_next_value nvarchar(32)       select @string_next_value = convert(nvarchar(32),@next_value)        commit tran        select @result = @prefix + substring('000000000000000000000000000000',1,10-len(@string_next_value)) + @string_next_value        select @result    END TRY    BEGIN CATCH         IF @@TRANCOUNT > 0 ROLLBACK TRAN          DECLARE @ErrorMessage NVARCHAR(400);         DECLARE @ErrorNumber INT;         DECLARE @ErrorSeverity INT;         DECLARE @ErrorState INT;         DECLARE @ErrorLine INT;          SELECT @ErrorMessage = N'{' + convert(nvarchar(32),ERROR_NUMBER()) + N'} ' + N'%d, Line %d, Text: ' + ERROR_MESSAGE();         SELECT @ErrorNumber = ERROR_NUMBER();         SELECT @ErrorSeverity = ERROR_SEVERITY();         SELECT @ErrorState = ERROR_STATE();         SELECT @ErrorLine = ERROR_LINE();         RAISERROR (@ErrorMessage, @ErrorSeverity, @ErrorState, @ErrorNumber,@ErrorLine)       END CATCH

Here’s the key generation table…

CREATE TABLE [dbo].[Key_Generation_Table](         [prefix] [nvarchar](3) NOT NULL,        [next_value] [int] NULL,   CONSTRAINT [PK__Key_Generation_T__236943A5] PRIMARY KEY CLUSTERED    (     [prefix] ASC   )WITH (PAD_INDEX  = OFF, STATISTICS_NORECOMPUTE  = OFF, IGNORE_DUP_KEY = OFF,             ALLOW_ROW_LOCKS  = ON, ALLOW_PAGE_LOCKS  = ON) ON [PRIMARY]   ) ON [PRIMARY]
Share

Leave an answer

You must login to add an answer.

Need An Account,

1 Answer

  1. Couple of things you have a race condition on your if block. Two requests come in for a new prefix at the same time, both could pass the if block. You should change this around to always insert into your table but in your where clause for the insert do the check to make sure it doesn’t exist. Also I’d recommend using Exists instead of count(*)=0. With Exists once sql finds a row it can stop looking.

    This same thing can happen with your select, you could have two threads both select the same value, then one gets blocked waiting on the update, but then when it returns it will return the old id.

    Modify your logic to update the row first, then get the value you updated it too

       update key_generation_table                set next_value = next_value + 1      where   prefix = @prefix         select @next_value = next_value -1                 from key_generation_table                  where prefix = @prefix

    I’d also look at using the ouput statement instead of doing the second select.

    EDIT

    I’d probally change this to use output since yoru on SQL2005:

    declare @keyTable as table  (next_value int)  UPDATE key_generation_Table set next_value=next_value+1 OUTPUT DELETED.next_value into @keyTable(next_value) WHERE prefix=@prefix  /* Update the following to use your formating */ select next_value from @keyTable
    • 0