As part of planning an Entity Framework migration, in order to debug data movement, I would often use the -Script parameter to generate the script.
I could then take this script to Query Analyzer and wrap it in a transaction in order to test it manually.
I came across a situation where we needed a Go statement to execute the script properly. The following code was added to the migration in order to output a Go in the proper place.
Sql("GO");
This adds a GO statement in the proper position when -Script is used. But when -Script isn't used. I get the exception...
System.Data.SqlClient.SqlException (0x80131904): Could not find stored procedure 'GO'.
Is there a safe way to add a Go command to the script?
I hit the exact same situation recently. My EF code migrations often introduce a new table or column, and then I also put data migrations using Sql(...) into those migrations that sometimes want to reference the new table/column. As you pointed out, when run as an EF code migration, each statement appears to be issued as a discrete batch to the DB, and hence there aren't any issues. However, to satisfy the production deployment constraints, we turn a set of Code migrations from a sprint into a single script (using -Script) to present a single aggregate SQL script migration for the deployment team. This script file sometimes fails, as you pointed out, due to it attempting to process a single T SQL batch from a single code migration where later statements attempt to refer to structure that was only defined earlier in the batch.
I don't particularly like either of the two approaches I've taken for now to mitigate this, but here they are:
a. If I happen to be thinking about it at the time, I split the code migration into two migrations, so that when they are scripted, they are in two (or more) separate batches. I don't like this because there is no feedback during the development of the Code Migration that this is necessary, and hence it seems error prone.
b. When I generate the aggregate scripts, I run them against a scratch DB to prove them out, and I end up manually injecting "GO" statements where necessary in that script. This is an annoying process to have to go back and do, and results in -Script output that isn't a 100% reflection of the Code Migrations.
I haven't spent much time digging into the source code of EF Code Migrations yet to see if I can understand why it interprets "GO" as a stored proc, and whether there is anything in the source code that would point to a way to provide a directive that would avoid that.
internal sealed class Configuration : DbMigrationsConfiguration<Context>
{
public Configuration()
{
AutomaticMigrationsEnabled = false;
const string providerInvariantName = "System.Data.SqlClient";
SetSqlGenerator(providerInvariantName, new BatchingMigrationSqlGenerator(GetSqlGenerator(providerInvariantName)));
}
protected override void Seed(Context context)
{
}
}
internal class BatchingMigrationSqlGenerator : MigrationSqlGenerator
{
private readonly MigrationSqlGenerator migrationSqlGenerator;
public BatchingMigrationSqlGenerator(MigrationSqlGenerator migrationSqlGenerator)
{
this.migrationSqlGenerator = migrationSqlGenerator;
}
public override IEnumerable<MigrationStatement> Generate(IEnumerable<MigrationOperation> migrationOperations, string providerManifestToken)
{
var migrationStatements = migrationSqlGenerator.Generate(migrationOperations, providerManifestToken).ToArray();
foreach (var migrationStatement in migrationStatements)
{
migrationStatement.BatchTerminator = "GO";
}
return migrationStatements;
}
}
I ended up using two different Configuration
classes when I run the migrations with and without the -Script
parameter. In one of my Configuration
classes I wrap its MigrationSqlGenerator
in a custom implementation, which adds the GO
statements.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With