The ETL process worked fine for the last 3 days. Today it started failing. The reason, Log_Backup. The database was in Full recovery mode and transaction log got full.
As a matter of fact, the staging database has to be in Simple recovery mode.
First to view disk space occupied by database, run this;
sp_helpdb SampleDb
To change database recovery model, run this;
USE SampleDb
GO
SELECT * FROM sys.database_files
--Truncate the log by chaning the database recovery model to SIMPLE
ALTER DATABASE SampleDb
SET RECOVERY SIMPLE
GO
--Shrink the truncated log file to 1MB
DBCC SHRINKFILE (SampleDb, 1)
GO
--Reset the database recovery model, if required
/*
ALTER DATABASE SampleDb
SET RECOVERY FULL
GO
*/
If DBCC SHRINKFILE takes longer, we can use following command to see the progress;
select * from sys.dm_exec_requests
There is a Percentage_Completed and Estimated_Completion_time columns. These columns are not populate for every operation, but they are for shrink. You can find the row for your connection during the shrinking, and inspect the values to get an estimate of completion time. If the values are not changing, you’ll need to investigate whether the process is blocking something.
Upon checking the database’s Log file growth setting, the log file was limited growth of 1GB. So what happened is when the job ran and it asked SQL server to allocate more log space, but the growth limit of the log declined caused the job to failed. I modified the log growth and set it to grow by 50MB and Unlimited Growth and the error went away.
Recently I hit by server memory issue while running SSIS package. I was getting binary data from remote server but the production server was unable to process it because of limited memory.
There are two choices to resolve this. Increase server memory. This wouldn’t solve the problem because the data will grow day by day.
The second is to split the batch into multiple batches say, 300 records per batch. if there are 2000 records then it would be 7 round trips to the remote server to load the data. 6 full 300 and seventh one would be 200.
Here is the design;
I will be using SQL OFFSET FETCH Feature and SSIS Script component for this. For an OLEDB example you can click on the link under Resources.
Declare four variables as follows:
1) vRowCount (Int32): Stores the total number of rows in the source table
2) vRC_IncrementValue (Int32): Stores the number of rows we need to specify in the OFFSET
3) vRC_Increment: (Int32): Stores the number of rows we need to specify in this operation
4) vRCChunkValue (Int32): Specifies the number of rows in each chunk of data in this operation
5) vRCBatchValue (Int32): Specifies the number of rows in each chunk of data
After declaring the variables, we assign a default value for the vRC_ChunkValue variable; in this example, we will set it to 100.
Our select query inside Script source component is like this;
string vSqlStatement = $@"
SELECT *
FROM [dbo].[tblSample]
WHERE 1=1
-- get chunks
ORDER BY SampleID
OFFSET {Variables.vRCIncrementValue} ROWS
FETCH NEXT {Variables.vRCBatchValue} ROWS ONLY";
Next, Add an Execute SQL Task to get the total number of rows from source table and change result set property to Single Row;
Assign the return value to vRowCount variable.
Next, add two expression task to copy values from operating variables to query variables;
Next, Add a For Loop Container, with the following configuration;
Add a data flow task inside For Loop Container. Add a script component and configure it as source;
Configure Data output on Inputs and Outputs tab;
Configure Connection on Connection Manager tab;
Click “Edit Script” and make these changes;
IDTSConnectionManager100 connMgr;
SqlConnection sqlConn;
SqlDataReader sqlReader;
public override void AcquireConnections(object Transaction)
{
//base.AcquireConnections(Transaction);
connMgr = this.Connections.GoldenConn;
sqlConn = (SqlConnection)connMgr.AcquireConnection(null);
}
public override void ReleaseConnections()
{
//base.ReleaseConnections();
connMgr.ReleaseConnection(sqlConn);
}
public override void PreExecute()
{
base.PreExecute();
//create sql statement
string vSqlStatement = $@"
SELECT *
FROM [dbo].[tblSample]
WHERE 1=1
-- get chunks
ORDER BY SampleID
OFFSET {Variables.vRCIncrementValue} ROWS
FETCH NEXT {Variables.vRCBatchValue} ROWS ONLY";
//MessageBox.Show(vSqlStatement);
SqlCommand cmd = new SqlCommand(vSqlStatement, sqlConn);
/*
7200 sec = 120 min = 2 hours. This can be set to 0 for non timeout at all
Will this work? It also depends on server timeout settings. In most SQL install, SQL default timeout for remote queries is 600 seconds, 10 minutes.
*/
cmd.CommandTimeout = 7200;
sqlReader = cmd.ExecuteReader();
}
public override void PostExecute()
{
base.PostExecute();
sqlReader.Close();
}
public override void CreateNewOutputRows()
{
try
{
while (sqlReader.Read())
{
{
SampleDataBuffer.AddRow();
SampleDataBuffer.SampleID = sqlReader.GetString(0);
SampleDataBuffer.AddDate = sqlReader.IsDBNull(7) ? null : sqlReader.GetString(7);
}
}
}
catch (Exception ex)
{
//set to true to cause execution to abort
bool cancel = false;
//raise the error event to SSIS
ComponentMetaData.FireError(-1, "CreateNewOutputRows()", ex.Message, "", -1, out cancel);
}
}
Next, Add Script task inside For Loop container to calculate remaining rows;
Edit Script and add this;
// make sure all rows are accounted for
int rowCount = (int)Dts.Variables["User::vRowCount"].Value;
int rowIncrement = (int)Dts.Variables["User::vRC_Increment"].Value;
int rowChunkValue = (int)Dts.Variables["User::vRC_ChunkValue"].Value;
//this is our new offset value
rowIncrement = rowIncrement + rowChunkValue;
Dts.Variables["User::vRC_IncrementValue"].Value = rowIncrement;
//calculate remaining rows
int remainingRows = rowCount - rowIncrement;
//MessageBox.Show($"RowCount: {rowCount}\nRowIncrmenet: {rowIncrement}\nRowChunkValue{rowChunkValue}\nRemainingRows{remainingRows}");
if ((remainingRows <= rowChunkValue))
{
//short circuit
Dts.Variables["User::vRC_BatchValue"].Value = remainingRows;
//for loop assign expression is [@vRC_Increment = @vRC_Increment + @vRC_ChunkValue], let's reverse this for last loop iteration
Dts.Variables["User::vRC_Increment"].Value = rowIncrement - rowChunkValue;
}
OFFSET FETCH is a feature added to the ORDER BY clause beginning with the SQL Server 2012 edition. It can be used to extract a specific number of rows starting from a specific index.
Here is a simple example;
SELECT * FROM [dbo].[tblSample]
order by SampleId
OFFSET 0 ROWS
FETCH NEXT 10 ROWS ONLY
This instruct SQL Server to start from first record and return 10 records only.
Let’s write a dynamic sql to get data in batches; For example we have 7 records from database and we would like to get 3 records on each round trip;
DECLARE @RowCount INT, @RowIncrement INT, @RowsInChunk INT, @RemainingRows INT, @SourceQuery NVARCHAR(255)
DECLARE @RowIncrementValue INT, @BatchValue INT
DECLARE @SQL_Select NVARCHAR(4000);
DECLARE @RowIncrementCheck INT;
SET @RowCount = 7;
SET @RowIncrement = 0;
SET @RowsInChunk = 3;
--initial value
SET @RowIncrementValue = @RowIncrement
SET @BatchValue = @RowsInChunk
--loop through all records
WHILE (@RowIncrement <= @RowCount)
BEGIN
--prepare SQL statement
SET @SQL_Select =
'SELECT *
FROM [dbo].[tblSample]
order by SampleId
OFFSET ' + CAST(@RowIncrementValue AS NVARCHAR) + ' ROWS ' + CHAR(13) +
'FETCH NEXT ' + CAST(@BatchValue AS NVARCHAR) + ' ROWS ONLY'
--EXECUTE sp_executesql @SQL_Select
PRINT @SQL_SELECT
SET @RowIncrement = @RowIncrement + @RowsInChunk
SET @RowIncrementValue = @RowIncrement
--calculate remaining rows
SET @RemainingRows = @RowCount - @RowIncrement
PRINT 'Remaining Rows ' + CAST(@RemainingRows AS NVARCHAR)
IF (@RemainingRows <= @RowsInChunk) AND (@RowIncrement <= @RowCount)
BEGIN
--short circuit this loop
SET @BatchValue = @RemainingRows
SET @RowIncrement = @RowCount
END
END
Here is the result of 3 dynamic sql for 7 records;
SELECT *
FROM [dbo].[tblSample]
order by SampleId
OFFSET 0 ROWS
FETCH NEXT 3 ROWS ONLY
SELECT *
FROM [dbo].[tblSample]
order by SampleId
OFFSET 3 ROWS
FETCH NEXT 3 ROWS ONLY
SELECT *
FROM [dbo].[tblSample]
order by SampleId
OFFSET 6 ROWS
FETCH NEXT 1 ROWS ONLY
In the query above, OFFSET 0 is used to skip 0 rows and FETCH 3 ROWS ONLY is used to extract only 3 rows.
This is suggested code block for creating stored procedures in SQL Server;
BEGIN TRY
BEGIN TRANSACTION
--All your insert/update/delete/merge goes in here
COMMIT TRANSACTION
END TRY
BEGIN CATCH
SELECT ERROR_NUMBER() AS ErrorNumber
,ERROR_MESSAGE() AS ErrorMessage
ROLLBACK TRANSACTION
END CATCH
While manipulating multiple tables, the transaction would be roll back if there is any error.
