如何创建通用SQL Server存储过程以基于插入和删除的触发器执行插入审计表

Question

I have implemented an audit trail framework based on the information provided by the first answer to the following post:

SQL Server history table - populate through SP or Trigger?

Ultimately, the framework that I have implemented uses three triggers per table that insert audit information based on changes to the tables.

My insert and delete audit triggers are fairly simple. However, the update triggers are far more complex because the trigger has to check to determine whether or not each column is under audit control and then perform an insert based on whether or not the column values in the Inserted and Deleted columns are equal or not since I don't want to write unnecessary audit records. Ultimately, I want to know if there is a way to write a stored procedure that will reduce the amount of code in my trigger by allowing me to dynamically perform the insert statement below. Basically, I envision the trigger firing the sproc with each column name that is under audit control and then the stored procedure will used the column name to perform the code snippet below. Currently, I have the code below for every column under audit control which unfortunately results in lots of redundant code.

Revised Trigger After Suggested Changes

CREATE TRIGGER [dbo].[Audit_Customers_Update] ON [dbo].[Customers]
FOR UPDATE AS

select FirstName,LastName into #deleted from deleted;

declare /*const*/ @TABLE_NAME sysname = '[table name]';

declare f cursor
local
forward_only
read_only
for
  select c.name, quotename(c.name, '[')
  from
    sys.columns c
    inner join sys.types t on c.system_type_id = t.system_type_id
  where
    c.object_id = object_id(@TABLE_NAME)
    and c.is_computed = 0
    and c.is_identity = 0
    and t.name not in ('text', 'image', 'timestamp', 'xml')
    and (substring(COLUMNS_UPDATED(), ((c.column_id - 1) / 8) + 1, 1) & power(2, (c.column_id - 1) % 8)) > 0
  ;

declare @field_name sysname, @field_name_sanitised sysname;
create table #results (row_id int not null,
                       field_name sysname not null,
                       oldval nvarchar(150) null,
                       newval nvarchar(150) null);

-- For each changed field, insert what exactly changed into #results

open f;

fetch next from f into @field_name, @field_name_sanitised;
while @@fetch_status = 0
begin
  declare @query nvarchar(4000);

  set @query =  N'insert into #results(row_id, field_name, oldval, newval)
                  select d.row_id, @field_name, d.' + @field_name_sanitised + N', i.' + @field_name_sanitised + N'
                  from
                    #deleted d inner join ' + @TABLE_NAME + N' i on d.row_id = i.row_id
                  where
                    (d.' + @field_name_sanitised + N' <> i.' + @field_name_sanitised + N')
                    or
                    (case when d.' + @field_name_sanitised + N' is null then 1 else 0 end <> case when i.' + @field_name_sanitised + N' is null then 1 else 0 end);'
                ;    

  exec sp_executesql
    @stmt = @query,
    @params = N'@field_name sysname',
    @field_name = @field_name
  ;

  fetch next from f into @field_name, @field_name_sanitised;
end;

close f;
deallocate f;

-- Do something meaningful to #results here

How do I access #results? Do I have to use a cursor?

Answer 1

We've solved that problem in the following way.

select <list of tracked columns here> into #deleted from deleted;

declare /*const*/ @TABLE_NAME sysname = '[table name]';

declare f cursor
local
forward_only
read_only
for
  select c.name, quotename(c.name, '[')
  from
    sys.columns c
    inner join sys.types t on c.system_type_id = t.system_type_id
  where
    c.object_id = object_id(@TABLE_NAME)
    and c.is_computed = 0
    and c.is_identity = 0
    and t.name not in ('text', 'image', 'timestamp', 'xml')
    and (substring(COLUMNS_UPDATED(), ((c.column_id - 1) / 8) + 1, 1) & power(2, (c.column_id - 1) % 8)) > 0
  ;

declare @field_name sysname, @field_name_sanitised sysname;
create table #results (row_id int not null, field_name sysname not null, oldval nvarchar(150) null, newval nvarchar(150) null);

-- For each changed field, insert what exactly changed into #results

open f;

fetch next from f into @field_name, @field_name_sanitised;
while @@fetch_status = 0
begin
  declare @query nvarchar(4000);

  set @query =  N'insert into #results(row_id, field_name, oldval, newval)
                  select d.row_id, @field_name, d.' + @field_name_sanitised + N', i.' + @field_name_sanitised + N'
                  from
                    #deleted d inner join ' + @TABLE_NAME + N' i on d.row_id = i.row_id
                  where
                    (d.' + @field_name_sanitised + N' <> i.' + @field_name_sanitised + N')
                    or
                    (case when d.' + @field_name_sanitised + N' is null then 1 else 0 end <> case when i.' + @field_name_sanitised + N' is null then 1 else 0 end);'
                ;    

  exec sp_executesql
    @stmt = @query,
    @params = N'@field_name sysname',
    @field_name = @field_name
  ;

  fetch next from f into @field_name, @field_name_sanitised;
end;

close f;
deallocate f;

-- Do something meaningful to #results here

Related reading:

• COLUMNS_UPDATED
• sys.columns

Answer 2

Ran into a similar problem... figured it out this way... may not be the most elegant solution but works for the compliance guys... So here goes...

Based on the solution given here

The xml is extracted with FOR XML from the trigger that updated the table... The "OldValues" come from the DELETED table and the "NewValues" from the INSERTED table... so the final xml looks like this...

            DECLARE @x XML= '<FieldData>
              <UpdatedColumns>
                <trType>OldValues</trType>
                <ID>5</ID>
                <def_label>TEST_TIE</def_label>
                <def_code />
              </UpdatedColumns>
              <UpdatedColumns>
                <trType>OldValues</trType>
                <ID>4</ID>
                <def_label>RP_TIE</def_label>
                <def_code />
              </UpdatedColumns>
              <UpdatedColumns>
                <trType>OldValues</trType>
                <ID>3</ID>
                <def_label>ERR_TIE</def_label>
                <def_code />
              </UpdatedColumns><UpdatedColumns>
                <trType>NewValues</trType>
                <ID>5</ID>
                <def_label>TEST_TIE</def_label>
                <def_code>A</def_code>
              </UpdatedColumns>
              <UpdatedColumns>
                <trType>NewValues</trType>
                <ID>4</ID>
                <def_label>RP_TIE</def_label>
                <def_code>A</def_code>
              </UpdatedColumns>
              <UpdatedColumns>
                <trType>NewValues</trType>
                <ID>3</ID>
                <def_label>ERR_TIE</def_label>
                <def_code>A</def_code>
              </UpdatedColumns>
            </FieldData>'

            declare @timestamp datetime2= SYSDATETIME()

            select 
                     ID = identity(int,1,1), 
                     T.N.value('local-name(.)', 'nvarchar(100)') as NodeName,
                     T.N.value('../ID[1]','nvarchar(100)') AS table_ID,
                     T.N.value('.', 'nvarchar(100)') as OldValue
            INTO #old
            from @x.nodes('//UpdatedColumns/*') as T(N)
            WHERE T.N.value('../trType[1]', 'nvarchar(100)') ='OldValues'


            select 
                     ID = identity(int,1,1),
                     T.N.value('local-name(.)', 'nvarchar(100)') as NodeName,
                     T.N.value('../ID[1]','nvarchar(100)') AS Table_ID,
                     T.N.value('.', 'nvarchar(100)') as NewValue
            into #new
            from @x.nodes('//UpdatedColumns/*') as T(N)
            WHERE T.N.value('../trType[1]', 'nvarchar(100)') ='NewValues'



            SELECT n.table_ID, n.NodeName, o.OldValue, n.NewValue,@timestamp as transation_time FROM #new n
            left outer JOIN #old o ON n.NodeName = o.NodeName AND n.ID = o.ID 
            WHERE isnull(o.[OldValue],'') <> isnull(n.[newValue],'') AND n.NodeName <> 'trType'



            DROP TABLE #new,#old 
            GO