使用C＃无法从SQL Server中选择超过700000行

Question

I couldn't fetch more than 700000 rows from SQL Server using C# - I get a "out-of-memory" exception. Please help me out.

This is my code:

using (SqlConnection sourceConnection = new SqlConnection(constr))
{
    sourceConnection.Open();

    SqlCommand commandSourceData = new SqlCommand("select * from XXXX ", sourceConnection);

    reader = commandSourceData.ExecuteReader();
}

using (SqlBulkCopy bulkCopy = new SqlBulkCopy(constr2))
{
    bulkCopy.DestinationTableName = "destinationTable";

    try
    {
        // Write from the source to the destination.
        bulkCopy.WriteToServer(reader);
    }
    catch (Exception ex)
    {
        Console.WriteLine(ex.Message);
    }
    finally
    {
        reader.Close();
    }
}

I have made up small console App based on the given solution 1 but ends up with same exception also i have posted my Memory process Before and After Before Processing :

After adding the command timeout at the read code side, Ram Peaks up,

Answer 1

That code should not cause an OOM exception. When you pass a DataReader to SqlBulkCopy.WriteToServer you are streaming the rows from the source to the destination. Somewhere else you are retaining stuff in memory.

SqlBulkCopy.BatchSize controls how often SQL Server commits the rows loaded at the destination, limiting the lock duration and the log file growth (if not minimally logged and in simple recovery mode). Whether you use one batch or not should have no impact on the amount of memory used either in SQL Server or in the client.

Here's a sample that copies 10M rows without growing memory:

using System;
using System.Collections.Generic;
using System.Data.SqlClient;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

namespace SqlBulkCopyTest
{
    class Program
    {
        static void Main(string[] args)
        {
            var src = "server=localhost;database=tempdb;integrated security=true";
            var dest = src;

            var sql = "select top (1000*1000*10) m.* from sys.messages m, sys.messages m2";

            var destTable = "dest";

            using (var con = new SqlConnection(dest))
            {
                con.Open();
                var cmd = con.CreateCommand();
                cmd.CommandText = $"drop table if exists {destTable}; with q as ({sql}) select * into {destTable} from q where 1=2";
                cmd.ExecuteNonQuery();
            }

            Copy(src, dest, sql, destTable);
            Console.WriteLine("Complete.  Hit any key to exit.");
            Console.ReadKey();
        }

        static void Copy(string sourceConnectionString, string destinationConnectionString, string query, string destinationTable)
        {
            using (SqlConnection sourceConnection = new SqlConnection(sourceConnectionString))
            {
                sourceConnection.Open();

                SqlCommand commandSourceData = new SqlCommand(query, sourceConnection);

                var reader = commandSourceData.ExecuteReader();

                using (SqlBulkCopy bulkCopy = new SqlBulkCopy(destinationConnectionString))
                {
                    bulkCopy.BulkCopyTimeout = 60 * 10;
                    bulkCopy.DestinationTableName = destinationTable;
                    bulkCopy.NotifyAfter = 10000;
                    bulkCopy.SqlRowsCopied += (s, a) =>
                    {
                        var mem = GC.GetTotalMemory(false);
                        Console.WriteLine($"{a.RowsCopied:N0} rows copied.  Memory {mem:N0}");
                    };
                     // Write from the source to the destination.
                     bulkCopy.WriteToServer(reader);

                }
            }
        }


    }
}

Which outputs:

. . .
9,830,000 rows copied.  Memory 1,756,828
9,840,000 rows copied.  Memory 798,364
9,850,000 rows copied.  Memory 4,042,396
9,860,000 rows copied.  Memory 3,092,124
9,870,000 rows copied.  Memory 2,133,660
9,880,000 rows copied.  Memory 1,183,388
9,890,000 rows copied.  Memory 3,673,756
9,900,000 rows copied.  Memory 1,601,044
9,910,000 rows copied.  Memory 3,722,772
9,920,000 rows copied.  Memory 1,642,052
9,930,000 rows copied.  Memory 3,763,780
9,940,000 rows copied.  Memory 1,691,204
9,950,000 rows copied.  Memory 3,812,932
9,960,000 rows copied.  Memory 1,740,356
9,970,000 rows copied.  Memory 3,862,084
9,980,000 rows copied.  Memory 1,789,508
9,990,000 rows copied.  Memory 3,903,044
10,000,000 rows copied.  Memory 1,830,468
Complete.  Hit any key to exit.

Answer 2

Something went horribly wrong in your design if you even try to process 700k Rows in C#. That you fail at this is to be expected.

If this is data retrieval for display: There is no way the user will be able to process that amount of data. And filtering down from 700k Rows in the GUI is just a waste of time and Bandwidth. 25-100 fields at once is about the limit. Do filtering or pagination on the Query side so you do not end up retrieving orders of magnitude more then you can actually process.

If this is some form of Bulk insert or Bulk modification: Do that kind of operation in the SQL Server, not in your code. Retrieving, processing in C# and then posting back just adds layers of Overhead. If you add the 2 way Network transfer, you will easily triple the time this will take.

Answer 3

NB: Per DavidBrowne's answer , it seems I'd misunderstood how the batching of the SqlBulkCopy class works. The refactored code may still be useful to you, so I've not deleted this answer (as the code is still valid), but the answer is not to set the BatchSize as I'd believed. Please see David's answer for an explanation.

---

Try something like this; the key being setting the BatchSize property to limit how many rows you deal with at once:

using (SqlConnection sourceConnection = new SqlConnection(constr))
{
    sourceConnection.Open();
    SqlCommand commandSourceData = new SqlCommand("select * from XXXX ", sourceConnection);
    using (reader = commandSourceData.ExecuteReader() { //add a using statement for your reader so you don't need to worry about close/dispose

        //keep the connection open or we'll be trying to read from a closed connection

        using (SqlBulkCopy bulkCopy = new SqlBulkCopy(constr2))
        {
            bulkCopy.BatchSize = 1000; //Write a few pages at a time rather than all at once; thus lowering memory impact.  See https://docs.microsoft.com/en-us/dotnet/api/system.data.sqlclient.sqlbulkcopy.batchsize?view=netframework-4.7.2
            bulkCopy.DestinationTableName = "destinationTable";

            try
            {
                // Write from the source to the destination.
                bulkCopy.WriteToServer(reader);
            }
            catch (Exception ex)
            {
                Console.WriteLine(ex.Message);
                throw; //we've caught the top level Exception rather than somethign specific; so once we've logged it, rethrow it for a proper handler to deal with up the call stack
            }
        }
    }

}

Note that because the SqlBulkCopy class takes an IDataReader as an argument we don't need to download the full data set. Instead, the reader gives us a way to pull back records as required (hence us leaving the connection open after creating the reader). When we call the SqlBulkCopy 's WriteToServer method, internally it has logic to loop multiple times, selecting BatchSize new records from the reader, then pushing those to the destination table before repeating / completing once the reader has sent all pending records. This works differently to, say, a DataTable , where we'd have to populate the data table with the full set of records, rather than being able to read more back as required.

One potential risk of this approach is, because we have to keep the connection open, any locks on our source are kept in place until we close our reader. Depending on the isolation level and whether other queries are trying to access the same records, this may cause blocking; whilst the data table approach would have taken a one-off copy of the data into memory and then closed the connection, avoiding any blocks. If this blocking is a concern you should look at changing the isolation level of your query, or applying hints... Exactly how you approach that would depend on the requirements though.

NB: In reality, instead of running the above code as is, you'd want to refactor things a bit, so the scope of each method is contained. That way you can reuse this logic to copy other queries to other tables. You'd also want to make the batch size configurable rather than hard-coded so you can adjust to a value that gives a good balance of resource usage vs performance (which will vary based on the host's resources).
You may also want to use async methods, to allow other parts of your program to progress whilst you're waiting on data to flow from/to your databases.

Here's a slightly amended version:

public Task<SqlDataReader> async ExecuteReaderAsync(string connectionString, string query) 
{
    SqlConnection connection;
    SqlCommand command; 
    try 
    {
        connection = new SqlConnection(connectionString); //not in a using as we want to keep the connection open until our reader's finished with it.
        connection.Open();
        command = new SqlCommand(query, connection);
        return await command.ExecuteReaderAsync(CommandBehavior.CloseConnection);  //tell our reader to close the connection when done.
    } 
    catch 
    {
        //if we have an issue before we've returned our reader, dispose of our objects here
        command?.Dispose();
        connection?.Dispose();
        //then rethrow the exception
        throw;
    }
}
public async Task CopySqlDataAsync(string sourceConnectionString, string sourceQuery, string destinationConnectionString, string destinationTableName, int batchSize)
{
    using (var reader = await ExecuteReaderAsync(sourceConnectionString, sourceQuery))
        await CopySqlDataAsync(reader, destinationConnectionString, destinationTableName, batchSize);
}
public async Task CopySqlDataAsync(IDataReader sourceReader, string destinationConnectionString, string destinationTableName, int batchSize)
{
    using (SqlBulkCopy bulkCopy = new SqlBulkCopy(destinationConnectionString))
    {
        bulkCopy.BatchSize = batchSize; 
        bulkCopy.DestinationTableName = destinationTableName;
        await bulkCopy.WriteToServerAsync(sourceReader);
    }
}
public void CopySqlDataExample()
{
    try 
    {
        var constr = ""; //todo: define connection string; ideally pulling from config 
        var constr2 = ""; //todo: define connection string #2; ideally pulling from config 
        var batchSize = 1000; //todo: replace hardcoded batch size with value from config
        var task = CopySqlDataAsync(constr, "select * from XXXX", constr2, "destinationTable", batchSize); 
        task.Wait(); //waits for the current task to complete / if any exceptions will throw an aggregate exception
    } 
    catch (AggregateException es)
    {
        var e = es.InnerExceptions[0]; //get the wrapped exception 
        Console.WriteLine(e.Message);
        //throw; //to rethrow AggregateException 
        ExceptionDispatchInfo.Capture(e).Throw(); //to rethrow the wrapped exception
    }
}