I have a scenario that I commonly run into. It's simple to do with a
standard ADO Transaction, but not so much with NH (that I know of).
I have 2 tables to update. The first contains profile information
(Profile) and the other (Work) contains records changes that need to
be made and the status of those changes. For each update to the
Profile table, there will be an update to the status on the Work
table.
If the update to the Profile table fails, I need to update the
status on the Work table.
If the update to the Profile table succeeds, and the update to the
Work table fails, I need to rollback the transaction.
The problem is that I don't know if the update to the Profile table
failed until I commit the transaction. I tried to do a Flush on the
Profile to catch the exception so I could write the status to the Work
table, but then my Commit fails with the exception caused from the
Profile update.
How can I handle this? In a typical ADO Transaction, my first call
will throw, but I can catch and still update the other tables in the
transaction.
Here's sort of what my code looks like - pretty standard. This is not
my actual code, so please focus on the problem, not that I'm not
disposing my transaction or closing my session ;) :
try
{
ITransaction trans = _session.BeginTransaction();
var work = _repo.GetWork();
var profile = _repo.GetProfile(work.ProfileId);
try
{
profile.UpdateWithNewValues(work);
_session.SaveOrUpdate(profile);
_session.Flush();
work.Status = "Success";
}catch{
work.Status = "Failure";
}
_session.SaveOrUpdate(work);
trans.Commit();
}catch{
trans.Rollback();
}
I realize that Flush() is not going to work, but I don't know how else
to do this.
Some clarifications needed on your requirements.
1) >> If the update to the Profile table succeeds, and the update to the Work table fails, I need to rollback the transaction
I would have thought that Work is like an audit trail update and should not have failed if Profile update works. If this is the case, then you should not rollback your transaction. However, having said this, your code already comply to this requirement.
2) >>If the update to the Profile table fails, I need to update the status on the Work table.
If update fails, then you would have rolled back your transaction. You will not be able to update the Work table unless you have two separate transactions (one for both Profile and Work (as current) and then a separate one just for Work). Does this make sense to you?
I don't see a problem with having a trans.Commit prior to the flush. Here's an example (slightly modified too look like yours):
Profile profile;
Work work;
ITransaction tx;
try
{
session.SaveOrUpdate(profile);
work.Status = "Success";
session.SaveOrUpdate(work);
tx.Commit();
}
catch (Exception) // wroh oh...
{
try
{
work.Status = "Failure";
session.SaveOrUpdate(work);
tx.Commit();
}
catch (Exception)
{
if (!tx.WasRolledBack)
{
tx.Rollback();
session.Clear();
}
throw;
}
}
finally
{
if (session.IsOpen)
{
// Whatever happened, Flush/Persist at the end.
session.Flush();
}
}
Related
Execute the following server code, and then check the promotion table and task table in the database. The related fields have been updated correctly, which indicates that the transaction has been successfully committed.
using (ITransaction tx = session.BeginTransaction())
{
try
{
Promotion p = session.Get<Promotion>(request.PromotionId);
p.Status = PromotionStatus.Canceled;
foreach (Task task in p.Tasks)
{
if (task.AnnounceStatus == TaskAnnounceStatus.New)
{
task.AnnounceStatus = TaskAnnounceStatus.PromotionCanceled;
task.CancelTime = DateTime.Now;
//session.Update(task);
}
}
tx.Commit();
}
catch
{
tx.Rollback();
throw;
}
}
Then execute the following query(Query A), the data obtained is also the updated value. It looks like everything is very good.
tasks = session.Query<Task>().Where(p => p.AnnounceStatus == Model.TaskAnnounceStatus.New && p.ProcessStatus == Model.TaskProcessStatus.New).ToList();
However, if I execute a query on the task using the following code before committing the transaction, the result of the above query(Query A) will get the old unmodified value. At the same time, what you see in the database is still the correctly updated value.
Task task = session.Get<Task>(taskId);
So I modified the first piece of code and explicitly called the update method (see the code at the comment), and everything worked fine this time.
My guess is that Nhibernate's cache is causing the above problem. I use syscache2 to manage the second-level cache, the cache was set to ReadWrite, and use sessionFacotry.getCurrentSession to manage Nhibernate's session.
Hope someone can help me explain how this works.
You execute query session.Get<Task>(taskId); first. This loads the entity in first level cache.
Then in your transaction, you Get the Promotion entity. The Task is the IEnumerable property of it. As lazy load may be, your foreach loop iterate through Task entity with ID taskID - Modifies it - Updates it - Transaction is successful. As all this is happening inside the transaction, your initial entity returned by session.Get<Task>(taskId); is not updated. It still hold the old values.
Then, you again session.Query<Task>() outside the transaction. This time, NHibernate see that the entity with same identifier is already loaded in session cache (with session.Get<Task>(taskId); query), it does not load that entity again, it simply returns the entity already in session cache. As that entity hold the old values, you see the problem.
To confirm this, put all these queries inside the transaction block and check the result.
Alternatively, manage so scope of session properly.
Understand that your ISession is your Unit Of Work; scope it carefully.
i am trying to retrieve data from database via nhibernate's CreateSQLQuery on a stored proceduure. Something like the following code.
then I am basically doing a session transaction commit, however the commit throws an "cannot update" exception. It is trying to execute a update statement on CustomEntityDao.
const string selectSQL = "EXEC GetDataSP #Id = :Id";
var query = Session.CreateSQLQuery(selectSQL);
query.SetString("Id", "10");
query.AddEntity(typeof (CustomEntityDao));
var entityList = query.List<CustomEntityDao>();
try
{
Session.Transaction.Commit();
}
catch (Exception ex)
{
throw ex;
}
My question is why are the entities been treated as modified, as you can see in the code I am only doing a query.
May be there is something else going on your code, what I can suggest is use NHibernate Profiler a trial version of which can be downloaded from www.nhprof.com and monitor the SQL commands that are being fired and notice what objects are being retrieved.
Also I dont understand why are you committing the transaction in the first place.
To solve this particular problem you could always use NHibernates StatelessSession which doesnt track entities or you can also use Session.Evict and ask Nhibernate to stop tracking particular objects.
In NHibernate 3.0, FlushMode.Auto does not work when running under an ambient transaction only (that is, without starting an NHibernate transaction). Should it?
using (TransactionScope scope = new TransactionScope())
{
ISession session = sessionFactory.OpenSession();
MappedEntity entity = new MappedEntity() { Name = "Entity", Value = 20 };
session.Save(entity);
entity.Value = 30;
session.SaveOrUpdate(entity);
// This returns one entity, when it should return none
var list = session.
CreateQuery("from MappedEntity where Value = 20").
List<MappedEntity>();
}
(Example shamelessly stolen from this related question)
In the NHibernate source I can see that's it's checking whether there's a transaction in progress (in SessionImpl.AutoFlushIfRequired), but the relevant method ( SessionImpl.TransactionInProgress) does not consider ambient transactions - unlike its cousin ConnectionManager.IsInActiveTransaction, which does consider ambient transactions.
Good news. Thanks to Jeff Sternal (who nicely identified the problem) I updated https://nhibernate.jira.com/browse/NH-3583 and thanks to the NH staff, there's already a fix and a pull request so in the upcoming release 4.1.x.x this ISSUE will be fixed.
You should use an explicit NHibernate transaction always.
using (TransactionScope scope = new TransactionScope())
using (ISession session = sessionFactory.OpenSession())
using (ITransaction transaction = session.BeginTransaction())
{
//Do work here
transaction.Commit();
scope.Complete();
}
I see you also wrote in the NH dev list - while this can change in the future, that's how it works now.
The answer provided by Diego does not work in the case where you have an oracle database.
(releated question). The session.BeginTransaction will fail because the connection is already part of a transaction.
Looks like we have to write some code around this problem in our application (WCF,NHibernate, Oracle), but it just feels like something that NHibernate should provide out of the box.
So if anyone has a good answer, it would be really appreciated.
For me, I don't know the reason behind, but by forcing session flush before session is disposed seemed to worked for me. e.g.
using(session)
{
//do your work
session.Flush();
}
I verified that this works with my distributed transaction, since without doing this, I always get "transaction has aborted" when TransactionScope is disposed.
Even set session.FlushMode to Commit did NOT work for me.
I have a suite of integration tests that run inside transactions.
Sometimes it seems that NHibernate transactions are not being correctly rolled back. I can't work out what causes this.
Here is a slightly simplified overview of the base class that these integration test fixtures run with:
public class IntegrationTestFixture
{
private TransactionScope _transactionScope;
private ConnectionScope _connectionScope;
[TestFixtureSetUp]
public virtual void TestFixtureSetUp()
{
var session = NHibernateSessionManager.SessionFactory.OpenSession();
CallSessionContext.Bind(session);
_connectionScope = new ConnectionScope();
_transactionScope = new TransactionScope();
}
[TestFixtureTearDown]
public virtual void TestFixtureTearDown()
{
_transactionScope.Dispose();
_connectionScope.Dispose();
var session = CurrentSessionContext.Unbind(SessionFactory);
session.Close();
session.Dispose();
}
}
A call to the TransactionScope's commit method is never made, therefore how is it possible that data still ends up in the database?
Update I never really got my head around the way NHibernate treats transactions but I found that calling Session.Flush() within a transaction would sometimes result in the data remaining in the database, even if the transaction is then rolled back. I am not sure why you can't call Flush, but then roll back. This is a pity because during integration testing you want to be able to hit the database and flush is the only way to do this sometimes.
I had an issue with using the Identity generator that sounds similar. In a nutshell, when saving an object using an identity generator, it has to write to the database in order to get the identity value and can do this outside a transaction.
Ayende has a blog post about this.
What's the best practice for handling exceptions in NHibernate?
I've got a SubjectRepository with the following:
public void Add(Subject subject)
{
using (ISession session = HibernateUtil.CurrentSession)
using (ITransaction transaction = session.BeginTransaction())
{
session.Save(subject);
transaction.Commit();
}
}
And a Unit Test as follows:
[Test]
public void TestSaveDuplicate()
{
var subject = new Subject
{
Code = "En",
Name = "English"
};
_subjectRepository.Add(subject);
var duplicateSubject = new Subject
{
Code = "En",
Name = "English1"
};
_subjectRepository.Add(duplicateSubject);
}
I got to the point of handling the error generated by the unit test and got a bit stuck. This fails as expected, though with a GenericADOException, I was expecting a ConstraintViolationException or something similar (there is a uniqueness constraint on the subject code at database level).
The ADOException wraps a MySQL Exception that has a sensible error message but I don't want to start breaking encapsulation by just throwing the inner exception. Particularly as MySQL isn't finalised as the back end for this project.
Ideally I'd like to be able to catch the exception and return a sensible error to the user at this point. Are there any documented best practice approaches to handling NHibernate Exceptions and reporting back up to the user what went wrong and why?
Thanks,
Matt
I would handle it in the Add method as such:
public void Add(Subject subject)
{
using (ISession session = HibernateUtil.CurrentSession)
using (ITransaction transaction = session.BeginTransaction())
{
try
{
session.Save(subject);
transaction.Commit();
}
catch (Exception ex)
{
transaction.Rollback();
// log exception
throw;
}
}
}
In the catch block, you should first rollback the transaction and log the exception. Then your options are:
Rethrow the same exception, which is what my version does
Wrap it in your own exception and throw that
Swallow the exception by doing nothing, which is very rarely a good idea
You don't have any real options for handling the exception in this method. Assuming that the UI calls this method, it should call it in its own try..catch and handle it by displaying a meaningful error message to the user. You can make your unit test pass by using the ExpectedException(type) attribute.
To answer your question directly, you should create your own "sensible error" by extending Exception and throw that with the original exception as its InnerException. That's the exception wrapping technique I listed in (2).
All the Nhibernate exceptions are non recoverable, you could revisit the design of the app/data layer if you are trying to recover from nhibernate exceptions .
You can also Take a look at spring.net 's exception translation implementaion
Also you manually handling transactions on exceptions is tedious and error prone, take a look at nhibernate's contextual sessions .
Spring.net also has some nice helpers around nhibernate .
The general question is going to be, what do you want to tell the user, and who is the user?
If the user will sometimes be another computer (i.e., this is a web service), then you would want to use the appropriate mechanism to return a SOAP Fault or HTTP error.
If the user will sometimes be a UI of some sort, then you may want to display a message to the user, but what would you tell the user so he can do something about it? For instance, most web sites will say, "sorry, we had an unexpected error", no matter what the reason. That's because there's usually nothing the user could do about the error.
But in either case, the choice of how to tell "the user" is a matter for the Presentation layer (UI tier), not for the DAL. You should possibly wrap exceptions from the DAL in another exception type, but only if you're going to change the message. You don't need your own exception class, unless your callers would do something different if it's a data access exception rather than some other kind.
I'd probably validate the input before saving the object; that way you can implement whatever validation you like (e.g. check the length of the Subject Code as well as the fact that there aren't any duplicates), and pass back meaningful validation errors back to the user.
The logic is as follows; exceptions are used to indicate exceptional circumstances that your program doesn't cater for. A user entering a duplicate Subject Code in your example above is something your program should be catering for; so, rather than handling an exception because a DB constraint gets violated (which is an exceptional event and shouldn't be occurring), you'd want to handle that scenario first and only attempt to save the data when you know that the data you're saving is correct.
The advantage with implementing all validation rules in your DAL is that you can then ensure that the data going into your DB is valid as per your business processes in a consistent manner, rather than relying on the constraints in your DB to catch those for you.