I want to find a way to control EF's underlying db connection & transaction to make sure my application is using only one connection at a time during one transaction (I will need to support both Oracle and SQL Server).
I found this good article which comes with a lot of recommendations, but brings up (like all the other articles I have read) the TransactionScope. Well, I would like to stay away of TransactionScope, if possible...
Could I have a solution for this playing with only pure DbConnection & DbTransaction or this is simply impossible or wrong?
Even more, I found this article here, stating at section :
"Specifying Your Own Transaction"
Just as you can override the default behavior with connections, you
can also control transaction functionality. If you explicitly create
your own transaction, SaveChanges will not create a DbTransaction. You
won't create a System.Common.DbTransaction, though. Instead, when
creating your own transactions, you need to use a
System.Transaction.TransactionScope object.
But there is no explaination...
I am using Entity Framework 5.0. Could you please help me understand in order to choose correct for my application? It would be ideal to show me some good patterns of usage.
Thanks in advance!
Note: I am planning this because of the transactions escalating to DTC for Oracle Data Provider.
Entity Framework 6 has two features which may help with this:
Explicit Transaction Support
Ability to create a DbContext from a DbConnection.
If you did want to use EF5, you'd need to use a TransactionScope:
var context = new MyContext();
using (var transaction = new TransactionScope())
{
MyItem item = new MyItem();
context.Items.Add(item);
context.SaveChanges();
item.Name = "Edited name";
context.SaveChanges();
transaction.Complete();
}
As mentioned in the linked article, you will need to reference System.Transactions to get TransactionScope.
Entity Framework maintains its own transaction which is sufficient. However it gives you flexibility of committing or discarding the changes in transaction. If you do not call SaveChanges() method then it will discard the transaction. Also if you are using the same DbContext for many transactions, then it would be using same connection. If you use two or more DbContext at the same time, then it will use separate connections which is ideal situation. Here is a very important point I would like to make is that it would be a waste of Entity Framework technology to implement own transaction. If you want to do so, then I would suggest to use your own DB implementation in traditional way.
Related
I just want to know if I want to rollback all changes in database if transaction not complete, is there any difference between
using (TransactionScope transaction = new TransactionScope())
and
using (var dbContextTransaction = context.Database.BeginTransaction())
I am confused when read these two:
Connection.BeginTransaction and Entity Framework 4?
and
https://learn.microsoft.com/en-us/ef/ef6/saving/transactions
**note that I use entity framework 4 in my project if its necessary
From Programming Microsoft SQL Server 2012:
There are a number of pain points with explicit transactions. The first difficulty lies in the requirement that every SqlCommand object used to perform updates inside the transaction must have its Transaction property set to the SqlTransaction object returned by BeginTransaction. This means that you must take care to pass along the SqlTransaction object to any place in your code that performs an update, because failing to assign it to the Transaction property of every SqlCommand object in the transaction results in a runtime exception. The issue is compounded when you need to track the SqlTransaction object across multiple method calls that perform updates. It becomes even harder to manage things when these methods need to be flexible enough to work whether or not a transaction is involved or required.
The problem is worse when working with any of the other technologies we'll be covering later that provide abstraction layers over the raw objects. For example, because a SqlDataAdapter actually wraps three distinct SqlCommand objects (for insert, update, and delete), you must dig beneath the data adapter and hook into its three underlying command objects so that you can set their Transaction properties. (We don't generally recommend that you mix and match different data access APIs within your application, but if you must transactionalize updates across a combination of technologies, implicit transactions make it easy.)
The TransactionScope object, introduced as part of a dedicated transaction management API with .NET 2.0 in 2005, lets you code transactions implicitly. This is a superior approach that relieves you from all of the aforementioned burdens associated with explicit transactions. So the guidance here is to always work with implicit transactions whenever possible. You will write less code that is more flexible when you allow the framework to handle transaction management for you. However, it is still also important to understand explicit transactions with the SqlTransaction object, as you might need to integrate with and extend existing code that already uses explicit transactions. Therefore, we will cover both transaction management styles to prepare you for all situations.
The transaction management API offers many more benefits besides implicit transactions. For example, TransactionScope is capable of promoting a lightweight transaction (one associated with a single database) to a distributed transaction automatically, if and when your updates involve changes across multiple databases.
There are two pitfalls with TransactionScope you should be aware of.
First is that it will, by default, create a transaction with SERIALIZABLE isolation level, which, for SQL Server, is a poor choice. So you should always create your TransactionScope like this:
public class TransactionUtils
{
public static TransactionScope CreateTransactionScope()
{
var transactionOptions = new TransactionOptions();
transactionOptions.IsolationLevel = IsolationLevel.ReadCommitted;
transactionOptions.Timeout = TransactionManager.MaximumTimeout;
return new TransactionScope(TransactionScopeOption.Required, transactionOptions);
}
}
See rant here: using new TransactionScope() Considered Harmful
The second is that TransactionScope supports Distributed Transactions. So it will enable you to enlist different connections and even different resource providers in a single transaction. While this is occasionally useful, it's more often a pitfall. If you accidently end up with a distributed transaction you can accidently take a dependency on having a distributed transaction coordinator in your environment. So you should take steps to avoid having a distributed transaction, like shutting down Microsoft Distributed Transaction Coordinator (MSDTC) in your development environment. And ensure that any time you have multiple methods enlisted in a transaction, they don't both have a SqlConnection open at the same time.
While Database. BeginTransaction() is used only for database related operations transaction, System. Transactions. ... TransactionScope for mixing db operations and C# code together in a transaction.
please see Below Links.Hope they help you:
TransactionScope vs Transaction in LINQ to SQL
Database.BeginTransaction vs Transactions.TransactionScope
I am writing unit test cases for testing a framework. The unit test adds and modifies the data in an existing database. I need to rollback all the changes done to database after a test finishes, i.e. delete the rows added and revert the rows modified.
I am using Entity Framework 6 for accessing the database. The underlying database is SQL Server. Is there any support provided by EF6 to achieve this?
Presently I am storing the changes in a list and I refer the this list for cleaning up the database. But using this approach leaves some residue randomly. I am not sure of the reason though, maybe some race condition or something.
Looking for some minimal and smart alternative for it. Thanks in advance :)
you can wrap your test in a transaction and don't commit changes:
using (TransactionScope scope = new TransactionScope()) {
//do your stuff
}
But for unit testing propouses you can use Effort - Entity Framework Unit Testing Tool which provide in-memory database operations.
EDITED to reply last comments
You can use an overloaded TransactionScope contructor to control the IsolationLevel, so you can choose to read uncommited changes or not.
If your proxy isn't inside the Transaction, please, check that the connection string is the same, so ado.net can identify the connection and enlist the connection in the same transaction.
If the connection string is not the same, you probably will need to activate the Distributed Transaction Coordinator. Here you have an explanation how DTC scalation occurs: TransactionScope automatically escalating to MSDTC on some machines?
I've been reading a document from Microsoft's patterns & practices group (Data Access for Highly-Scalable Solutions: Using SQL, NoSQL, and Polyglot Persistence).
In Chapter 3, in the section "Retrieving Data from the SQL Server Database", the authors discuss using Entity Framework to load entities from a database. Here's a bit of their sample code:
using (var context = new PersonContext())
{
Person person = null;
using (var transactionScope = this.GetTransactionScope())
{
person = context.Persons
.Include(p => p.Addresses)
.Include(p => p.CreditCards)
.Include(p => p.EmailAddresses)
.Include(p => p.Password)
.SingleOrDefault(p => p.BusinessEntityId == personId);
transactionScope.Complete();
}
// etc...
}
Note the use of a custom transaction scope via the GetTransactionScope method, implemented in their base context class like so:
public abstract class BaseRepository
{
private static TransactionOptions transactionOptions = new TransactionOptions()
{
IsolationLevel = IsolationLevel.ReadCommitted
};
protected virtual TransactionScope GetTransactionScope()
{
return new TransactionScope(TransactionScopeOption.Required, transactionOptions);
}
}
Working with Transactions (EF6 Onwards) on MSDN states:
In all versions of Entity Framework, whenever you execute SaveChanges() to insert, update or delete on the database the framework will wrap that operation in a transaction [...] the isolation level of the transaction is whatever isolation level the database provider considers its default setting. By default, for instance, on SQL Server this is READ COMMITTED. Entity Framework does not wrap queries in a transaction. This default functionality is suitable for a lot of users
The bold emphasis is mine.
My question: is the use of a TransactionScope as shown above overkill, particularly for data reads, when using Entity Framework?
(I realised after posting that the answer to this question might be somewhat opinion-based, sorry about that.)
The question is a little open ended. But may prove useful for people learning about dirty reads. Isolation and EF.
And you have read EF Transaction Scope ef6 and we have a clear question.
Generally i would say let EF manage the scope.
Beginners dont consider uncommitted reads and it is a good default for EF.
You may have a valid reason to want to control scope.
Or need to use an existing transaction. But remains for specialist use.
So now the real question.
Is it good practice to include such defensive programming around isolation.
My view:
Only if it doesnt make the code harder to maintain, and harder to reuse.
Oracle and SQL server have default Read Committed. I would expect so on other DBs.I would therefore conclude, most likely unecessary protection that adds complexity.
I wouldnt add it to my code.
It depends.
If you want dirty reads, serializable reads,etc anything other than default isolation level, then you need to wrap you queries inside a transaction scope.
To maintain data integrity, it is sometimes useful to use an explicitly defined multi-statement transaction so that if any part of the transaction fails, it all gets rolled back. The decision on whether or not to use a transaction should not be entered into lightly. Explicit transactions, especially distributed ones, come at a major cost to performance and open the system up to deadlocks and to the possibility that a transaction could be left open, locking the affected table and blocking all other transactions. In addition, the code is far more difficult to develop and maintain.
In my experience, I have found that using a transaction scope in C# code is more complicated than using an explicit transaction in SQL. So for operations that would benefit from a transaction, and really any sufficiently complicated query, I would recommend creating a stored procedure to be called from the ORM.
Have you read the following link ?
https://msdn.microsoft.com/en-us/library/vstudio/bb738523%28v=vs.100%29.aspx
If you were to use the normal behavior of saveChanges without it begin part of a transactionscope you would not be able to extend the transaction beyond the boundaries of the database. Using the transaction scope like in the reference above makes the message queue part of the transaction. So if for some reason the sending of the message fails, the transaction fails too.
Of course you might consider this a more complex scenario. In simple applications that don't require such complex logic, you are probably safer to use the plain saveChanges api.
I am using NServiceBus with Distributed Transactions. Normally this works fantastic. Either my message succeeds or fails. All or nothing.
However, I am also trying to record diagnostic data to my database. This is done on a separate ObjectContext. For that, I would like it to save to the database regardless of the success or failure of the Distributed Transaction.
Is there a way to have one (and only one) data context connect to the database outside the distributed transaction? (And keep the rest of them in?)
Try this to suppress your ambient distributed transaction for auditing:
using (new TransactionScope(TransactionScopeOption.Suppress))
{
// Create logging context and audit your data
}
One option (if using3.0) is to create a UoW:
http://andreasohlund.net/2011/11/21/unit-of-work-in-nservicebus-3-0/
And in there suppress the transaction and do the logging.
NSB has a built in way to do auditing. I would recommend turning this on and dealing with the logging/audit trail in another endpoint.
I'm creating and dropping databases on the fly for some integration tests. I'm doing all the database management at the ADO.NET level. For the tests I'm using Entity Framework because the entities is one part of what I am testing. The problem is that after I do this:
using (ProjectEntities db = new ProjectEntities(cs)) {
}
I cannot drop the database anymore. It says it is in use. How do I release it so it can be dropped?
I actually had the same problem at the ADO.NET level and what I did was:
new SqlCommand("USE [master]", DatabaseConnection).ExecuteNonQuery();
but I'm not sure how to perform something with the same effect for the Entity Framework connection. I've trying manually disposing the db object (although the using clause should guarantee that) and I've also tried manually closing the db.Connection. Neither helped. If I could run SQL directly no the Entity Framework connection, I believe I'll be able to do it. Or maybe there's another way?
You might need to explicitly close all connections, I believe that the connection are being pooled and it's one of the pooled connections that is still maintaining an active connection to the DB. Try using SqlConnection.ClearAllPools
I blogged about a similar issue in Entity Framework Object Context - AWAITING COMMAND.