Entity Framework Code First Migrations in a Team Environment

The managing of a database schema has seen several advances over the years in the .NET stack. Circa 2010, Visual Studio began shipping with a new Database Project, and some SKUs included data compare tools, which helped compare data schema and the data itself. Around the same time, Entity Framework was launched, at first only offering support in  Model and Database First approaches, it was soon offered with the now, super popular, Code First approach.

The introduction of Code First was in reaction to people who wanted to keep their time in writing code and out of a database designer. So if we’re going to stay out of the database designer, how do we manage our database’s schema? Enter Migrations.  Code First Migrations can analyze your model and automatically create schema update files to get your database updated with your latest model changes. When I started using Migrations, I was in love. It was seriously time saving and super easy to use.

Soon enough, after touting, and bragging about how I found the new hotness, I introduced migrations on a project that I was working with a few other people on. The first week went…well…not good. I really thought this whole thing was gearing up for me to end up with some serious egg on my face.

The Pain, Oh The Pain

So, in order to understand why there was pain, we first need to understand how migrations work. Before doing any deep dives, I made assumptions. My main assumption that when I started with code first was that with  the ‘Add-Migration’ command, the magic going on was that the little minions inside were going out to my database and making notes about what’s different between the database and my code first model. I mean, this is logical, right, this is how we track schema changes using the Schema Compare tool in the past. Well, wrong-o. Code first doesn’t care about your database at all.

In reality, migrations are doing their comparisons against the last known state of the model. It does that this by making use of a snapshot file. You see, whenever you issue an ‘Add-Migration <some migration>’ command, migrations creates a snapshot of your entire model. So the next time you issue an ‘Add Migration <some other migration>’, migrations compares the current state of the model to the snap shot that was created in ‘some migration’. Seeing the problem yet?

So this all works fine and dandy for you. So now lets imagine that we have two developers Sarah and Roger. Sarah does some work, creates a migration, SarahMigration1, and checks in. Roger, gets latest, and sees Sarah’s migration, SarahMigration1. He begins doing some work and creates a migration of his own, RogerMigration1. So behind the scenes, migrations compares Roger’s changes against the snap shot and spits out a migration file. Sweet. But Roger isn’t done working and doesn’t check in his changes yet. Meanwhile, Sarah needs to create another migration, so she issues an ‘Add Migration SarahMigration2’ command.  So what happens? Migrations doesn’t know about RogerMigrationOne, or the snapshot that was created along with it, so migrations compares Sarah’s current model against the only snapshot it knows about, which is SarahMigration1. So a migration file is created that looks good, since Sarah knows nothing about Roger’s changes, her new migration looks and seems right. What what happened behind the scenes is what’s going to cause some problems. The snapshot that was created in Sarah’s last command, doesn’t include Rogers’s changes. So Sarah checks in and Roger gets latest. Since migrations lists migration by date, he will see SarahMigration1, RogerMigration1, and SarachMigration2.  Ok…seems ok. Oh, but it’s so not. When Roger goes to create his second migration, migrations is going to compare his current model against the latest snap shot, SarahMigration2, but the snapshot in SarahMigration2 knows nothing about RogerMigrationOne changes, so migrations is going to generate all of Roger’s changes, including the changes that are already defined in RogerMigrationOne.  Whoops.

The Solution

I like to call this the, ‘Oww, My Toes’ scenario. So the obvious solution might be, just delete RogerMigration1. While that might work in some scenarios, if you’ve updated your database, you’re going to be in a world of hurt if you do that…so don’t do that.

To fix this scenario, you have 2 options:

Option One: Generate A Blank Migration
  1. Write all of your pending model changes to a migration and update your database.
  2. Get latest from source control
  3. Issue an ‘Add-Migration <some name> -IgnoreChanges’ command

This method will result in a blank migration file being added to your codebase, but this solution is quick and easy, so the tradeoff of having blank files may be worth it. Note that this option will not work for you if you have already gotten latest before writing your pending model changes to a migration, so you may want to get into the habit of writing your changes to a migration before getting latest.

Option Two: Regenerate the Snapshot
  1. Find the migration before the one that matches your current database status.
  2. Issue command ‘Update-Database – TargetMigration [migrationbeforecurrent]’
  3. This will revert your database ‘back to good’ if you will.
  4. At this point, you can delete your migration that is in between the other two as well as the latest migration that had the duplicate migration code.
  5. Issue Command ‘Add-Migration MyNewMigration’
  6. This will create a new migration that
  7. Issue Command ‘Update-Database

This should bring the database base back to good and inline with the migrations files, as well as creating a new snapshot of the database that matches the current code first model. Note that this option only works if the latest migration exists in your local workspace only. It cannot have been committed to source control.

Whew.

Best Practices For Teams Using Migrations

  1. Turn off Automatic Migrations – This isn’t just for teams, this is for pretty much anyone using migrations. Automatic migrations takes all control away from the developer on how the database is migrated. You should be in control of the migration files to avoid, ‘It just deleted all my data scenarios’ (Yes, it will do this. Be extra sure you have automatic migration off for anything in Production)
  2. Get latest often – Always having the latest helps to reduce the number of times this scenario occurs.
  3. Check in often – On the contrary, be a good team member by getting new migration checked in as soon as possible.
  4. Designate a migration master – This certainly isn’t going to work for all teams, and it comes with some risk that the migration master becomes a bottle neck for the whole team. But basically, the other developers make changes to the context model, but the actual migration files are always created by one person, that person is always guaranteed to have the correct snapshot of the model if they are the only one creating migration files. This is the only option to completely eliminate the ‘Oww, My Toes’ scenario.

Migrating a Project from Database First to Code First

Overview

So you just pushed you application to production and used Microsoft’s new shiny ORM. It’s 2008 and you’re on the bleeding edge of .NET technology by implementing Entity Framework. Your EDMX paired with your database project keeps your project nice and organized in source control. Great Job. But fast forward to today, and Entity Framework Code First is all the rage. What do you do with that aging database first design along with that EDMX in all it’s glory ? Nuke it. You don’t need it anymore.

I sure hope you didn’t just blindly nuke it and check in. We still do need that EDMX for a bit, but not for long. We’re going to walk through the process of converting your old busted to the new hotness of Entity Framework Code first.

Migrations are you friend, but not like the kind you leave alone home with your significant other. Be sure to use them, but I highly recommend turning off automatic migrations. Anything that has that much blind control of your app should be something that you should VERY carefully consider before turning on.

Note: This process assumes you are using the Database First approach, and not the Model First Approach. If you use the model-first approach, you will have some leg work to do in order to determine what your EDMX might be doing that cannot be reverse engineered from the database.

Disclaimer: This is a fairly significant change you will be making to your project, so make sure that you plan for the regression testing of everything.

Now, let’s get on with it:

Step 1 : Generate your Context, Entities, and Mapping Files

Microsoft has released a visual studio plugin () that will generate POCOs and a context based on an EDMX. This will save you a whole lot of time. Head on over here, and install this plugin.

Once installed, move over to you project, and right click your Project File. There should now be a context menu item, Entity Framework, Select That, and then Reverse Engineer Code First.

generate_views

Select the database you would like to use to base the reverse engineer process to be based on.

generate_diag1

Once you click OK, a folder will be created in your project called Models that contains your new Context, Entities, and Fluent mapping configurations.

generated_files

Step 2: Remove old context, and update the project to use the new context.

Now that you have created all of your new entities and context, the old one can be removed. Delete the EDMX and all associated files (context.tt, etc).

Step 3: Enable Migrations

As I mentioned before, we are NOT enabling automatic migrations. We are only enabling migrations. This means that we will manually create migrations by using the add-migration syntax in the Package Manager Console.

In the Package Manager Console, Make sure that you set the Default Project to the project that contains your context. Then enter the command Enable-Migrations

enable_migrations

You will notice that a Migrations folder has been created with a Configuration.cs file.  In the Configuration.cs file, make sure Automatic Migrations is set to false.

Step 4: Create and Set Database Initializer

Create a new class called MyDbInitialzier

using System.Data.Entity;
using MyProject.Data.DataAccess.Migrations;

namespace MyProject.Data.DataAccess.EntityFramework
{
    internal sealed class MyDbInitializer : MigrateDatabaseToLatestVersion<MyDbContext, Configuration>
    {
    }
}

You will notice that I the initializer calss inherits from the MigrateDatabaseToLatestVersion class.  It is likely that this is the Initializer behavior that you will want to use if you have an existing database already in production.  If you have special circumstances, be sure to review all of the default initializers and/or look into building a custom initializer.

Step 5: Implement the new Context

You will want to crack open you web config and replace the old connection string (The one with all of the metadata stuff, with a new connection string.  The new connection string should look like any old ADO.NET connection string.

You will now want to replace the references to the old context with the new one. (Shortcut: you could just rename the new one to match the old one’s name).

Note: You may encounter a bit of a gotcha here.  Since the new context is of type DbContext and the old one was of type ObjectContext, you may find that some the compiler is complaining about some things.  The DbContext is kind of a wrapper for the object context that is meant to be lighter weight, there are things you may be using that are not supported by the db context.  You will want to research any of these issues that come up to see if the DbContext can support them. If all else fails, the DbContext can be cast to the ObjectContext if you absolutely need it.  (This will result in a performance hit, so use wisely).  The syntax for getting the ObjectContext from a DbContext is:

public class MyContext: DbContext
{
    public ObjectContext ObjectContext()
    {
        return (this as IObjectContextAdapter).ObjectContext;
    }
}
Step 6: Create Your Initial Migration

If we tried to run the project right now, the application would encounter an error letting you know that there are pending changes that need to be included in a migration before the application can proceed.  We are going to create our initial migration. In the Package Manager Console, enter the command Add-Migration initial

add_migration_initial

In your Migrations folder, a file should have been created: YYYYYYDDHHMMSSS_initial.cs. This should be a total representation of your entire existing database.

EF keeps track of changes to the data model updating a table in your database called __MigrationHistory (in SystemTables) Since your database is existing already, you do not have this table in your database, so when this migration goes to run, it will attempt to re-create all of the objects in your database. This is bad, and we dont want that.  We can use this trick to tell EF to not re-create all of the objects when it attempts to run this migration.  In your initial migration class, comment out all of the code in the Up method.  That’s it, that’s the whole trick.

public partial class initial : DbMigration
 {
    public override void Up()
    {
        // Commented Code Here
    }
}
Step 7: Update the Database

Now is the time to update your database.  In the package manager console, enter the command, Update-Database.

udpate-database

This will create the __MigrationHistory table and will record that it ran this initial migration, so moving forward it will view your database as ‘up to date’ with your data model. (If you want to create the database from scratch using code first from now on, you will need to uncomment this migration.  It can safely be uncommented after it updates the existing database).

That’s it.  You should now be able to run your project. Now you need to regression test everything really well.

Conclusion

By following these steps you should now be fully running on Code First with Migrations.  Happy Coding!