Database maintenance can be a full -time job when working within a large corporation. Cleanup or deletion of records can especially be time-consuming and error prone when dealing with large data tables with millions of records. Often developers will try to run a basic delete statement without knowing about batching or truncating the activity log. Below is a simple method for performing a batch delete.

Define you DELETE statement

The first step is to define your DELETE statement with the appropriate WHERE clause. If for example you want to delete all records before a set date then your query would look similar to:

DELETE FROM my_table WHERE action_ts < '06/01/2009'

Implementing the basic batch loop

To implement a basic loop we need to define some variables to keep track of a few variables. First we define @err to obtain the error id if an error occours, and @nCount to hold the total number of rows deleted.

DECLARE @err INT, @nCount INT
SELECT @nCount=0

Next we implement or while statement to execute or delete statement while more records are available to be deleted. In this example we will test if any rows exist matching or deletion statement.

WHILE EXISTS (SELECT * FROM my_table WHERE action_ts < '06/01/2009')
BEGIN
    ... 
END

In our while loop we will delete our records in batches by use of the TOP command that limits how many records are effected. We will be using a batch size of 10000.

DELETE TOP 10000 FROM my_table WHERE action_ts < '06/01/2009'

After our delete query we want to check if any errors occurred and if so break out of our batch loop. This is done by obtaining the current error status and if an error occurred then breaking out of the loop by the break command.

SELECT @err=@@ERROR, @nCount=@nCount+@@ROWCOUNT
IF @err=0
BEGIN
    ...
END
ELSE
BEGIN
    RAISERROR 20001 'Failed to delete date from table my_table'
    BREAK
END

Dumping the activity log and output statements

To be notified on how many rows are being deleted we can output the @nCount variable that is udpated after every deletion with the number of rows affected by the last delete statement. We can use the print statement to output the number.

PRINT '%l! rows deleted', @nCount

Also when dealing with a large amount of rows it is possible for the activity log to get full for the database. To clean up the log use the dump tran command with the truncate_only option.

DUMP TRAN my_db WITH TRUNCATE_ONLY

The Full Batch Delete Script

Putting the entire script together we end up with:

DECLARE @err INT, @nCount INT
SELECT @nCount=0
 
WHILE EXISTS (SELECT * FROM my_table WHERE action_ts < '06/01/2009')
BEGIN
 
    DELETE TOP 10000 FROM my_table WHERE action_ts < '06/01/2009'
 
    SELECT @err=@@ERROR, @nCount=@nCount+@@ROWCOUNT
    IF @err=0
        BEGIN
            DUMP TRAN my_db WITH TRUNCATE_ONLY
            PRINT '%l! rows deleted', @nCount
        END
    ELSE
        BEGIN
            RAISERROR 20001 'Failed to delete date from table my_table'
            BREAK 
        END
END

For those that worry about the extra time needed for doing the exists select clause in the while loop you could optionally use a separate variable to hold the latest row count and test for if the row count is greater than 0.

, , , , , ,

It is often the case that developers new to databases will try to perform large data modifications without batching. When this happens in most cases the modification will fail with the error of the transaction log being full after you are presented with the message The transaction log in the database is almost full. Your transaction is suspended until space is made available in the log. What is happening is that the update/insert modification is trying to change too many rows without committing so the transaction log is getting filled up and once it is full the modification fails and no changes get saved to the database.

To correct this it is good practice to perform your database modifications in batches, and upon each iteration of the batch to clear your log transaction file. The dump tran command notifies Sybase that you want to clear the transaction log of a database and with which options. In these cases you simply want to clear it without backing up the log to a file so you would use the with turncate_only option.

DUMP TRAN MY_TABLE WITH truncate_only

Resources

, , , , ,