Skip to content

uladz-zubrycki/Reseed

Repository files navigation

NuGet version (Reseed) NuGet stats (Reseed)

Disclaimer: Versions of the library up to 1.0.0 don't follow semantic versioning and public api is a subject of possibly breaking changes.

đź“Ł Check out Discussions if you've got questions or any ideas to suggest.

About

Reseed library allows you to initialize and clean integration tests database in a convenient, reliable and fast way.

It covers cases similar to what a few other libraries intend to do:

Library is written in C# and targets netstandard2.0, therefore could be used by both .NET Framework and .NET/.NET Core applications.

This is how it's supposed to be used:

  • you describe the state of database needed for your tests;
  • you initialize database prior to running any tests, so that Reseed is able to do its magic;
  • you insert, delete or restore data to the previous state in between of the test runs, so that you're sure that each test has the same data to operate on;
  • you clean database after running the tests.

As simple as that.

Problem

Integration tests operating on a real database do modify the data, so some sort of isolation is required to avoid inter-test dependencies, which are usually a reason of random failures and flaky tests.

There are a few ways to address that issue, each of which has its pros and cons. E.g:

  • Make tests responsible for restoring the data they change;
  • Prepare data needed in the tests themselves and assert in some smart way on this data only;
  • Create and initialize fresh database from scratch for every test;
  • Use database backups to restore database to the point you need;
  • Use database snapshots for the restore;
  • Wrap each test in transaction and revert it afterwards;
  • Execute scripts to delete all the data and insert it again.

Reseed implements some of the concepts above and takes care of the database state management, while allowing you to focus on testing the business logic instead.

Design

Main idea is not to insert data directly, as for example NDbUnit does, but to generate scripts for the database initialization and cleanup and then execute those, when needed. This gives more control to the consumer and makes some optimization tricks possible. Scripts tend to be descriptive and well-formed in order to be readable by human and could, if needed, be adjusted manually before the execution.

It's possible to generate scripts for data seeding or cleanup only or both, depends on your needs.

The only entry point to all the functionality is the Reseeder class and this is how the library usage might look in its simplest form:

const string connectionString = "Server=myServerName\myInstanceName;Database=myDataBase;User Id=myUsername;Password=myPassword;";
var reseeder = new Reseeder();
var seedActions = reseeder.Generate(connectionString, SeedMode.Basic(
    BasicInsertDefinition.Script(),
    CleanupDefinition.Script(CleanupMode.PreferTruncate(), CleanupTarget.Excluding())),
    DataProviders.Xml(".\Data"));

reseeder.Execute(connectionString, seedActions.PrepareDatabase);
reseeder.Execute(connectionString, seedActions.RestoreData);
reseeder.Execute(connectionString, seedActions.CleanupDatabase);

See the Examples section below to get acquainted with the API and library behavior or check the Samples section to see how it could be integrated into your application testing framework.

Features

  • It's possible to generate scripts for either data seeding or cleanup only or both;
  • Reseed is able to order tables graph (tables are nodes, foreign keys are edges), so that foreign key constraints are respected in the insertion and deletion scripts;
  • Alternatively Reseed could simply disable foreign key constraints to deal with such data dependecies;
  • It detects cyclic foreign key dependencies on both tables and rows levels, so that loops don't break anything. More on this in Constraints resolution;
  • You could specify Custom cleanup scripts for specific tables to ignore rows during data cleanup;
  • Data schema is read from the database and there is no need to describe it manually (e.g NDbUnit requires XSD files);
  • There is an optional data validation step, which allows to detect data inconsistencies (like duplicated primary key values);
  • It's possible to omit identity columns, Reseed will generate them for you;

Limitations

  • Data could be described in xml files or inline as csharp objects, other formats aren't supported for now. If this isn't sufficient for your case, you could provide own implementation of IDataProvider type. See Data Providers for details and examples;
  • MS SQL Server is the only database supported for now.

Seed actions generation

Scripts that Reseed generates are named "seed actions" and are represented by the SeedActions type. It contains all the actions grouped in a few stages with a property per stage. Every stage is basically an ordered collection of seed actions, while each action in its turn is an instance of ISeedAction type.

class SeedActions 
{
    IReadOnlyCollection<OrderedItem<ISeedAction>> PrepareDatabase;
    IReadOnlyCollection<OrderedItem<ISeedAction>> RestoreData;
    IReadOnlyCollection<OrderedItem<ISeedAction>> CleanupDatabase;
}

The stages are:

  • PrepareDatabase

    At this stage Reseed executes some infrastructure related actions like creation of temporary database objects. Should be executed the first and once per test fixtures.

  • RestoreData

    Actions on these stage do actually restore the data to its initial state. You might think of it as a combination of data cleanup followed by data insertion. Should be executed before every test run.

  • CleanupDatabase

    At this stage objects created on the PrepareDatabase stage are deleted if there were any. Should be executed the last and once per test fixtures.

To generate actions you use Reseeder.Generate instance method by configuring generation behavior the way you like, see Operation modes documentation section for additional details.

Reseeder.Generate(AnySeedMode);

Seed actions execution

To execute the actions you simply pass a collection representing a specific stage to the Reseeder.Execute instance method.

Reseeder.Execute(string, IReadOnlyCollection<OrderedItem<ISeedAction>>, TimeSpan?);

If you have Reseeder reseeder, string connectionString and SeedActions seedActions variables, then it's:

reseeder.Execute(connectionString, seedActions.PrepareDatabase);
reseeder.Execute(connectionString, seedActions.RestoreData);
reseeder.Execute(connectionString, seedActions.CleanupDatabase);

Optionally you could provide a TimeSpan to specify an action execution timeout; otherwise defaults of the underlying system are used (e.g 30 seconds for SqlCommand).

reseeder.Execute(connectionString, seedActions.RestoreData, TimeSpan.FromSeconds(5));

Operation modes

Reseed is able to operate in a few modes, so that actions it outputs could differ depending on the configuration. See the Performance section for some starting data and do some tests for your case to choose the mode, which suits your case the best.

Static SeedMode type is an entry point to the Reseed behaviour configuration.

Basic mode

SeedMode.Basic(BasicInsertDefinition, AnyCleanupDefinition, IDataProvider[]);

This is the most basic and the most robust mode of operation. In this mode Reseed generates single insert script for all the entities as well as the only delete script to cleanup database. Data restore is a combination of delete and insert scripts executed one after another.

You might have insertion logic generated either as a script or stored procedure. The latter could be beneficial for large insert scripts.

BasicInsertDefinition.Script();
BasicInsertDefinition.Procedure(ObjectName);

Temporary tables mode

SeedMode.TemporaryTables(string, TemporaryTablesInsertDefinition, AnyCleanupDefinition, IDataProvider[]);

This mode is more tricky, thus less reliable, but at the same time it is able to provide a great speed boost.

Reseed firstly creates copy of the target tables in some temporary schema, then fills those with data with use of insert script and copies data to the target tables in the end. Therefore insertion, which is slower than the internal database data transfer, is executed just once.

There are a few ways to transfer the data, which are exposed at TemporaryTablesInsertDefinition type:

  1. Script:

    TemporaryTablesInsertDefinition.Script();

    Data is copied with use of insert statements like INSERT columns INTO TABLE (SELECT columns FROM temp.TABLE).

  2. Procedure:

    TemporaryTablesInsertDefinition.Procedure(ObjectName);

    Similarly to the Script mode it uses insert statements, but the insert logic is saved as stored procedure.

  3. Sql bulk copy:

    TemporaryTablesInsertDefinition.SqlBulkCopy(Func<SqlBulkCopyOptions, SqlBulkCopyOptions>)

    SqlBulkCopy type is used to transfer data to the target tables.

  4. BCP (Work in progress): Uses BCP utility to copy the data.

CleanupOnly mode

Sometimes you want to manage data insertion on your own and need just a database cleanup, this is the mode to be used then. Note that you don't need to specify IDataProvider for that case, which is a reasonable behavior as you don't need any data to be inserted.

SeedMode.CleanupOnly(CleanupDefinition);

As a result you'll get SeedActions instance, which does nothing but database cleanup during its RestoreData phase. See Data cleanup section for cleanup configuration options.

Data cleanup

There is a posibility to configure the way Reseed generates data cleanup scripts, which are needed for some of the seed modes. Configuration is done with use of CleanupDefinition type.

If you don't want Reseed to execute cleanup for you at all, then you could simply use NoCleanup cleanup definition. Just make sure to take care of cleaning the database on your own or it will most likely fail on attempt to insert duplicated rows once.

Otherwise data cleanup logic could be represented either as a script or a stored procedure:

CleanupDefinition.NoCleanup();
CleanupDefinition.Script(CleanupMode, CleanupTarget, bool = false);
CleanupDefinition.Procedure(ObjectName, CleanupMode, CleanupTarget, bool = false);

Data cleanup configuration

If you want cleanup to be executed, then you need to choose, which database objects should be cleaned and how. Use CleanupTarget type for that aim.

There are two ways to choose cleanup targets:

  • Either include every schema/table and specify the ones to ignore

    CleanupTarget.Excluding(
        Func<ExcludingCleanupFilter, ExcludingCleanupFilter>,
        IReadOnlyCollection<(ObjectName table, string script)>);
    
  • Or on contrary start with an empty tables set and explicitly include the ones to clean

    CleanupTarget.Including(
        Func<IncludingCleanupFilter, IncludingCleanupFilter>,
        IReadOnlyCollection<(ObjectName table, string script)>);
    

It's possible to include/exclude the whole schema or a single table.

Data cleanup modes

Also you should specify how each table will be cleaned. There are a few cleanup modes available:

  1. Delete

    CleanupMode.Delete(ConstraintResolutionBehavior);

    In this cleanup mode library generates DELETE FROM statement to clean the table. Foreign key constraints are respected and either tables are ordered to prevent deletion failures or constraints are disabled, the former is used by default.

  2. Prefer truncate

    CleanupMode.PreferTruncate(ObjectName[], ConstraintResolutionBehavior); 

    Pretty much as the previous one, but if Reseed finds that table has no relations, then TRUNCATE TABLE statement is used, which should be a lot faster; DELETE FROM is used otherwise. It's possible to explicitly specify tables to use DELETE FROM for and to choose constraints resolution behavior.

  3. Truncate

     CleanupMode.Truncate(ObjectName[], ConstraintResolutionBehavior);

    Reseed uses TRUNCATE TABLE for every table in spite of the foreign keys presence. It drops foreign keys and recreates them as it's not possible to use TRUNCATE TABLE statement otherwise. Similarly to the PreferTruncate mode you might force usage DELETE FROM for some of the tables and choose constraints resolution behavior for those.

  4. Switch tables (Work in progress)

    TBD

Here is how full CleanupDefinition setup might look like:

CleanupDefinition.Script(
    CleanupMode.PreferTruncate(),
    CleanupTarget.Including(c => c.IncludeSchemas("dbo"))))

Custom cleanup scripts

Sometimes you don't need to clean all the rows from the table, so you might want to provide custom deletion scripts for specific tables.

E.g you have a superadmin user with Id=1, which you want to be never deleted. Here is how the setup could look like for that case:

CleanupDefinition.Script(
    CleanupMode.PreferTruncate(),
    CleanupTarget.Including(
        c => c.IncludeSchemas("dbo"),
        new[]
        {
            (new ObjectName("User"), "DELETE FROM [dbo].[User] WHERE Id <> 1")
        })))

Identity columns reset

It's also possible to reset identity columns current seed values to the initial ones. Simply pass true value for an optional bool reseedIdentityColumns parameter, which is available for most of CleanupDefinition factory methods.

As a result additional script will be added into the RestoreData SeedActions collection.

Here is a setup example:

CleanupDefinition.Script(
    CleanupMode.PreferTruncate(),
    CleanupTarget.Including(c => c.IncludeSchemas("dbo")),
    reseedIdentityColumns: true)

And this is how generated reset script looks like (if we had the only dbo.User table):

DBCC CHECKIDENT
(
	'[dbo].[User]', RESEED, 1
);

Data providers

If you want Reseed to insert data for you, you need to specify how and from where this data should be read. This is encapsulated in the IDataProvider abstraction. All the built-in providers are exposed through the DataProviders static type. Currently you could describe your data as xml or csharp objects.

Xml data provider

As the name assumes data should be represented in xml format. Data might be described in the only data file or split into a few, the way that suits you more. Use DataProviders.Xml to use this approach.

DataProviders.Xml(string);
DataProviders.Xml(string, Func<string, bool>);
DataProviders.Xml(string, string, Func<string, bool>);

It's possible to control from what location data should be read and what files to read:

  • you need to specify a root folder, which is to be scanned recursively;
  • you could specify file name pattern to use only files, which match it(see Directory.EnumerateFiles for supported pattern features);
  • you could filter your files additionally by providing a Func<string, bool> predicate to check each file's name.

Rules to describe the data are simple:

  • Each row should be represented as xml element with nested elements for each column.
  • Element name should match the target table name and additionally it's possible to specify a database schema name explicitly by using notation [schema].[table] for your row elements (dbo is used by default);
  • Nested element names should match column names;
  • Nullable columns could be skipped;
  • Columns with default value could be skipped;
  • Identity columns could be skipped (more on this in Data extension).

E.g we have a table User (Id identity not null, FirstName not null, LastName not null), it could be represented this way:

<Users>
    <User>
        <Id>1</Id>
        <FirstName>John</FirstName>
        <LastName>Doe</LastName>
    </User>
    <dbo.User>
        <FirstName>Alice</FirstName>
        <LastName>Kane</LastName>
    </dbo.User>
</Users>

A few things to note here:

  • We used Users as root element, but its name could be any;
  • We'll get two rows inserted into the dbo.User table;
  • We omitted schema name for the first row and explicitly specified it for the other. dbo schema is assumed by default, so there is no difference;
  • We omitted Id value for the second row, but it's an identity column, so value will be generated automatically.

Inline data provider

Another approach to provide your data is to define it inlined, in your csharp code. Use DataProviders.Inline for that aim.

DataProviders.Inline(Func<InlineDataProviderBuilder, IDataProvider>);

InlineDataProviderBuilder has a few methods to add entities in a fluent code-style and a Build method, which should be called the last to create a IDataProvider instance.

It supports a few ways to represent each entity:

  • Entity type, which has a name and a collection of Property objects for entity properties;
  • Collection of Property objects per entity, while entity name is specified separately as an ObjectName instance;
  • IDictionary<string, string> object per entity, which maps property name to its value, while entity name is specified separately as an ObjectName instance.

Configuration might look this way, if we had dbo.User table with three columns: Id, FirstName and LastName:

DataProviders.Inline(builder =>
    builder
        .AddEntities(
            new Entity("User", new[]
            {
                new Property("Id", "1"),
                new Property("FirstName", "Alice"),
                new Property("LastName", "Freeman"),
            }))
        .AddEntities(
            new ObjectName("User", "dbo"),
            new[]
            {
	    	new Property("Id", "2"), 
                new Property("FirstName", "Bob"),
                new Property("LastName", "Spencer"),
            },
	    new[]
            {
	    	new Property("Id", "3"), 
                new Property("FirstName", "Kelly"),
                new Property("LastName", "Space"),
            })
        .AddEntities(
            new ObjectName("User"),
            new Dictionary<string, string>
            {
	    	["Id"] = 4,
                ["FirstName"] = "Jenny",
                ["LastName"] = "Lee"
            },
	    new Dictionary<string, string>
            {
	    	["Id"] = 5,
                ["FirstName"] = "Anthony",
                ["LastName"] = "Turton"
            })
        .Build())

Constraints resolution

TBD

Data Extension

Reseed is able to extend the provided data for any IDataProvider instance before generating the insertion scripts.

The only extension available for now is identity columns value generation. If Reseed finds an identity column without a value specified, it will generate a value for it automatically. If such behavior isn't desired, use ReseedOptions.DataExtensionOptions to disable it.

class DataExtensionOptions 
{
    bool GenerateIdentityValues;
}

Data validation

There is an optional data validation step executed the last in the insertion scripts generation flow. It's enabled by default and could be controlled with use of ReseedOptions type.

class ReseedOptions 
{
    bool ValidateData;
}

There is the only validation scenario available for now:

  • Validate primary keys uniqueness.

If any data issues are detected during the validation, the error will be thrown from the Reseed.Generate method.

Additionally Reseed will fail if there are missing values for required columns or unknown column names, but you can't disable the validation of such cases currently, as those are essential for valid scripts generation.

API

TBD

Performance

TBD

Examples

Let's take a look on a simple seeding example. E.g there is the only table with user data and we'd like to test our application behavior upon a database with a few user entities.

Script to create our database might look like this.

CREATE TABLE [dbo].[User] (
    Id int NOT NULL IDENTITY(1, 1) PRIMARY KEY,
    FirstName nvarchar(100) NOT NULL,
    LastName nvarchar(100) NOT NULL,
    ManagerId int NULL,
    Age int NULL
    
    CONSTRAINT [FK_User_ManagerId] FOREIGN KEY (ManagerId) REFERENCES [dbo].[User](Id)
)

And we have the only data file to represent the database state, which is Users.xml (name could actually be any).

<Users>
    <User>
        <FirstName>John</FirstName>
        <LastName>Doe</LastName>
        <Age>23</Age>
        <ManagerId>2</ManagerId>
    </User>
    <User>
        <Id>2</Id>
        <FirstName>Alice</FirstName>
        <LastName>Bart</LastName>
    </User>
    <User>
        <FirstName>Joe</FirstName>
        <LastName>Henessy</LastName>
        <Age>56</Age>
        <ManagerId>2</ManagerId>
    </User>
</Users>

We're going to use Basic SeedMode in this example. This configuration is the simplest and is often the most reasonable.

SeedMode.Basic(
    BasicInsertDefinition.Script(),
    CleanupDefinition.Script(
        CleanupMode.PreferTruncate(), 
        CleanupTarget.Including(f => f.IncludeSchemas("dbo"))))

This is what we get generated. PrepareDatabase and CleanupDatabase are empty collections as no internal database objects are needed in this mode of operation, while RestoreData stage contains two scripts: one to clean the data and another one to insert.

Delete script looks this way:

ALTER TABLE [dbo].[User] NOCHECK CONSTRAINT [FK_User_ManagerId]
DELETE FROM [dbo].[User];
ALTER TABLE [dbo].[User] CHECK CONSTRAINT [FK_User_ManagerId]

A few things to note here:

  • User table has foreign key constraint to itself and to address that dependency cycle Reseed disables the constraint;
  • Even though we used PreferTruncate mode, DELETE FROM statement is used, that's also due to the foreign key constraint presence.

And here is the insert script:

SET IDENTITY_INSERT [dbo].[User] ON
INSERT INTO [dbo].[User] WITH (TABLOCKX) (
	[Id], [FirstName], [LastName]
)
VALUES 
	(2, 'Alice', 'Bart')
INSERT INTO [dbo].[User] WITH (TABLOCKX) (
	[Id], [FirstName], [LastName], [ManagerId], [Age]
)
VALUES 
	(1, 'John', 'Doe', 2, 23), (3, 'Joe', 'Henessy', 2, 56)
SET IDENTITY_INSERT [dbo].[User] OFF

A few notes in regard to the insert script:

  • We specified Id column value equal 2 for the only record, but as it's an identity column, Reseed has taken care of that and provided values 1 and 3 for the other rows automatically;
  • Order of the rows in the script doesn't match the order of the entities in the data file, this is due to the foreign key constraint, which needs rows ordering to respect it. Row with Id=2 should be inserted the first as the other rows has ManagerId=2; insertion of rows in another order will either fail or require FK disabling, which is slower than ordering;
  • We have two rows with same columns specified (Id=1 and Id=3 in the script), those were combined to the only INSERT INTO clause;
  • Optional Age column was present for one row and omitted for the rest.

Samples