I have an idea to add "missing migration" check to my CI pipeline. This check should validate that DbSnapshot that is present in a branch matches models in that branch (in another words, it should validate that if I will create a migration, Up and Down would be empty).
Straightforward way of doing this would be to execute dotnet ef migrations add Test and validate that resulting files match "empty migration" ones, but this seems like a hack.
Is there an easier way of doing this? Like dotnet ef migrations --verify-snapshot-up-to-date or something?
Related
Context
I'm working a pipeline to manage a dot net core project. In my pipeline, at the build stage I run dotnet ef migrations script --context "MyDbContext" --project "./MySolution/MyDataProject/" --output "./migrationScript.sql" --idempotent to generate scripts which can later be run against the various test, staging, and production environments databases to sync the schema (I use the same approach in non-production environments as in production to ensure those scripts are tested).
It's my understanding that these scripts aren't generated based on the database context, but rather use the migrations which are already defined in the project; with those migrations having been defined via the add command (e.g. dotnet ef migrations add vNext --context "MyDbContext" --project "./MySolution/MyDataProject/"). As such, it's possible that a developer may make changes to the data model, forget to run the migrations add command to add a new migration, and thus create a solution that will fail to update the schema.
Question
Is there a way to test if the migrations are in sync with the code / if a new migration is needed?
Thoughts around Solutions
I'd hoped that by running dotnet ef migrations add when there are no changes to the dbContext there would be no output; so I could test for the presence of a new file and have the pipeline terminate if this had been missed (checking nothing into git; so the new migration isn't persisted / it's left for the developer to manually rerun this).
Looking at the available options, there's only add, remove, list, and script; so nothing obvious for performing this check.
I could run the add command before the script command in my pipeline to ensure it's been run; but then there may be issues with future runs, since if the added migration isn't pushed to git, the next iteration won't be aware of this migration.
If I were to push this new migration to git that resolves that issue; but then creates a new issue that every build creates a new migration, so we'll have a lot of redundant clutter / this approach won't be sustainable.
The best solution I can think of is to run the add command, then inspect the generated scripts to see if the Up or Down methods have anything in the functions' bodies; but that feels hacky; and I'm not certain it's enough (e.g. do the methods in the .Designer.cs file ever change without producing anything in Up/Down in the .cs file?).
I'm sure MS would have created something in the toolset for this; but I'm struggling to find it. Thank-you in advance for any help.
Existing / Similar Answers
In EF Core, how to check whether a migration is needed or not? - This point answers the question about applying migrations, but not about generating them.
Something that appears to work with a few simple test cases I've tried is to add a test migration, and then check if the <ContextName>ModelSnapshot.cs file has changed in the migrations directory.
That file only appears to change when the core migration has been amended.
I have a question about editing migration files in EF Core.
Example: I have created 3 tables using separate migration for each: AddTable1, AddTable2, AddTable3.
After that several days later I decided to change the type of status column in Table1 to be int and not string.
After updating the AddTable1 migration file, deleting the database and executing update-database command the column was still defined with string type.
I noticed that all the .Designer.cs files still were using string type for status column despite my edit of Table1 migration file.
I copied the content of the AppDbContextModelSnapshot.cs file and manually updated all the .Designer.cs files for all migrations (note: the method name in source and .Designer.cs files is different).
And that helped. I'm sure that this is not the recommended way to use migrations.
My question is: is there a command which will do what I did - 'reset' the '.Designer.cs' files?
If not, how to make one? :)
Think of migrations as Git commits. You don't rewrite previous commits once you've shared them with other team members (or in our case, applied a migration to a database). Instead, you create new migrations with additional changes.
If you haven't applied a migration or shared it with teammates, you can remove it and re-scaffold:
Remove-Migration
Add-Migration FixedUpMigration
See the EF Core Migrations docs for basic usage.
With EF Migrations you can add migration scaffoldings and then keep running 'Update-Database' to apply changes. And in order to rollback the following works:
Update-Database -TargetMigration: <xyzMigration>
This is great for updating your dev. DB. However when I automate migrations using MigrateDatabaseToLatestVersion for deploying to other environments like test and prod. adding migration files for any changes and having them reflect works well.
But, in case I wanted to rollback to a specific migration how do I achieve that? 'Update-Database -TargetMigration:' only updates your local dev. DB or whatever your connection string is pointing to locally. And that has to be run via Package Manager Console. Is rollback not an option for actual deployments? Do you have to just 'Add-Migration' and specifiy a new migration that has all the changes you want reflected?
For a production database doing an Add-Migration with the changes you want would probably make the most sense, especially if you have multiple deployments.
Other options include:
Using DbMigrator.Update() with a parameter that indicates the target migration. (I'd probably use a separate build, just for this purpose...seems tricky to handle...especially when you add additional migrations in the future).
Update-Database also has optional arguments for -ConnectionString, but that would require setting up Visual Studio on a machine with access to the database you're targeting.
I am using Entity Framework and MVC5 with Code First Methodology. I also use branching in TFS to work on different modules without impacting a release. I forgot to enable migrations in one of my branches and wanted to see if anyone had a solution to create the database changes.
Thanks in advance.
Here are the steps I performed to make this work (after merging all the branches):
Enable-migration -force (in the main branch - I had to use -force because one of the branches had enabled migrations already).
Next I ran update-database, however this failed because the migration file from the sub-branch was setup to create all the tables, instead of just the changes. I deleted the create table code for all tables, and kept the create table for the new table and run update-database again.
Finally to get the changes from the second sub-branch, I ran the add-migration command, followed by the udpate-database command.
I have been using EF migrations for some time now and have more than 100 migration files in my project. I would like to consolidate these into a single migration before moving forward - ie I want to replace the existing InitialCreate migration with a new version that takes all my subsequent changes into account so I can then delete all the other migration files.
I do this very easily if I am not concerned with losing all the data in the DB, but I am.
How can I achieve this whilst keeping all data intact and also retaining the ability to recreate the database from scratch (without data) by just running Update-Database (which I believe is not possible using the approach outlined by Julie Lerman)?
Consider reading this nice article from Rick Strahl :
https://weblog.west-wind.com/posts/2016/jan/13/resetting-entity-framework-migrations-to-a-clean-slate
Basically the solution is not trivial and needs more than just reseting all the migrations into one
because you have two scenarios that needs to fit in ONE migration class:
Create a new database => the migration class should contain every table creation
My database is already up to date => I need an empty migration class
Solution:
The idea of this process is basically this: The database and the EF schema are up to date and just the way you want it, so we are going to remove the existing migrations and create a new initial migration.
In summary, the steps to do this are:
Remove the _MigrationHistory table from the Database
Remove the individual migration files in your project's Migrations folder
Enable-Migrations in Package Manager Console
Add-migration Initial in PMC
Comment out the code inside of the Up method in the Initial Migration
Update-database in PMC (does nothing but creates Migration
Entry) Remove comments in the Initial method You've now essentially
reset the schema to the latest version.
once the commented out migration has been executed on the desired database, uncomment the migration code
If you're not concerned with keeping this migrations, what I've done is delete everything in your migrations folder, and then target a new database in the connection string (or pass in a new one). After that, you can just run the add-migration command:
add-migration InitialCreate
And it should create the migration for you.
Below procedure has the benefit of working without doing anything with the DBs, __MigrationHistory can stay as-is. Also it will work if you have multiple different environments with different versions of the structure - provided you have the branches to match.
I turn the last migration into an initial migration. The trick is to use the oldest version of the code and DB that is in use, replace its last migration with a new initial migration and delete all previous migrations. Newer branches keep the more recent migrations so those will still work after merging to older branches.
So start in the OLDEST branch - PROD, normally - and do:
Remove all but the last migration
Remove the migration code in both the "Up" and "Down" methods in the last migration
Change build action of the last migration to "None" to let EF ignore it
Change active connection to point to a local DB database.
Make sure this local DB database does not exist
add-migration Initial
Copy Up and Down code from the created "Initial" migration to the last migration
Delete Initial migration
Change build action of the last migration back to "Compile"
Check in
Merge changes up
Test in DEV branch on LocalDB DB - it should do the new initial migration as well as the subsequent ones with no issues
Test in main branch on the latest DB - it shouldn't do anything
Note above only works if you don't add stuff to the migrations that EF doesn't do itself. E.g. if you add DB views etc. than the newly created migration won't get those, it only gets the scripts EF generates based on your code.
Removing all migrations or regenerating them has drawbacks so we took an approach we merged all older migrations.
It require a bit of scripting. You read about the details here https://www.bokio.se/engineering-blog/how-to-squash-ef-core-migrations/ and see the scripts here https://github.com/bokio/EFCoreTools/tree/main/MigrationSquasher
The basic is the following steps though (copied from the blog post):
Overview of our approach
Create a new fresh database from the old migrations (We will use
this for comparison later)
Find a suitable target migration to be
the new initial (We picked one about 3 months old)
Write a script to
merge the Up() methods of all earlier migrations into the Up()
method of a new migration. We ignored Down() because we don't use it
for old migrations.
Generate this migration and add it to the
project. In our case we called it
20200730130157_SquashedMigrations1.cs. We used the snapshot from the
target migration we had picked.
Generate a 2nd prep migration that
inserts into the migrations history that
20200730130157_SquashedMigrations1.cs has already run. We called
this 20200730130156_SquashedMigrations1_prep.cs. Note the slightly
smaller timestamp on that one to make sure it runs before the real
migrations.
Delete the old migrations
Point our config to a new
database and run the migrations.
Compare that the schema we generate
is equal using the Sql Schema Compare in Visual Studio.
Work through
the issues until we have equal schemas. This part is a bit
complicated but I will get back to it.
Merge and 🤞 (Ok, we did run
more tests both on local and staging databases)
I hope this helps someone else. The EF team is looking at improving this story so if you have feedback on your requirements it probably help them to post that now https://github.com/dotnet/efcore/issues/2174.
We had the same problem.
The general solution we found was to
Archive the old migration / context on a Nuget Package (including dependancies and with a different namespace to avoid conflicts)
Delete all migrations and create a new Init (InitV2) Migration from scratch.
Change the startup sequence of our application:
If the database contains the first old Init migration then
Migrate the database using the Nuget Package to be sure it will be up to date
Then erase the content of the __EFMigrationHistory table and Insert the new migration in the table
After that use the standard Migrate method on the new context
That's it !
This solution is relatively simple. It solve all cases
New database (will be using the InitV2)
Old Database (will be upgrading to the last V1 Migration and after that to the last version of the v2 Database)
Be aware, if you used custom scripts/sql(...) in the v1 migrations, you have to check if the v2 Init migrations needs it.
To be sure it was OK, we created an empty database from v1 migrations and another one from v2 init migration and did a schema and data diff (with Visual Studio SQL Server Tools)