NHibernate version: 1.0.2.0

Hello All,

I'm using the syscache from the contrib along with Nhibernate, and I find that very rarely, the NHibernate.AssertionFailure below fires. It is very difficult to reproduce, and thus has been challenging to debug.

The exception is stating that there is already an entry in syscache's _map for this object, when we're trying to .Add() it (so the _map, which is Hashtable, throws).

The code on top of the stack trace below is just .Save()'ing objects with Guids as keys. The guids are obtained from Guid.NewGuid() so it cannot be the case that 2 different objects have the same key.

This seems similar to a post fom July in the forum, which sadly has no replies: http://forum.hibernate.org/viewtopic.php?t=962045

Any ideas are most appreciated. Many thanks.


Full stack trace of exception that occurs:
************** Exception Text **************
NHibernate.TransactionException: Commit failed with SQL exception ---> NHibernate.AssertionFailure: Exception releasing cache locks ---> System.ArgumentException: Item has already been added. Key in dictionary: 'NHibernate-Cache:Msg.Framework.RUdpTransport.RMOutMsg:c24d7914-0898-47d7-a4a8-40c652c9db39@-1965736198' Key being added: 'NHibernate-Cache:Msg.Framework.RUdpTransport.RMOutMsg:c24d7914-0898-47d7-a4a8-40c652c9db39@-1965736198'
at System.Collections.Hashtable.Insert(Object key, Object nvalue, Boolean add)
at System.Collections.Hashtable.Add(Object key, Object value)
at NHibernate.Caches.SysCache.SysCache.Put(Object key, Object value)
at NHibernate.Cache.ReadWriteCache.AfterInsert(Object key, Object value, Object version)
at NHibernate.Impl.ScheduledInsertion.AfterTransactionCompletion(Boolean success)
at NHibernate.Impl.SessionImpl.AfterTransactionCompletion(Boolean success)
--- End of inner exception stack trace ---
at NHibernate.Impl.SessionImpl.AfterTransactionCompletion(Boolean success)
at NHibernate.Transaction.AdoTransaction.Commit()
--- End of inner exception stack trace ---

Hello All,

FYI, I've taken a bit of time and looked at the syscache code, and it looks like it suffers from a race condition.

NHibernate's use of it is threadsafe (behind a lock) but the syscache listens to callbacks from the WebCache that uses when objects expire. Since this map is not locked, NH's thread and this callback thread can race and collide on this _map. This fix for this is to lock the map when we access it.

Is this syscache contribution in a project somewhere where I can look at a bug list, or is just a random piece of code some sent...somewhere to be included in the NH contrib?

As for the issue I encountered in the original post, I've noticed that when doing Insert()'s, NHibernate seems to call Put() on the cache twice for the same object. Thus, if the object is evicted from the cache but the callback thread has not called into us to update the map, we can, on insertions, end up with this sequence:

1. First Put() from NH - puts the object into the cache and map.
2. cache evicts object, starts callback.
3. Second Put() from NH comes in and, seeing that the object is not in the cache, trys to .Add() it to the map, which throws because the id is still there.

The solution for this is to change the .Add() to a [] = in the SysCache::Put() else clause.

Evan

EDIT: Changed a map to a cache in the #3 point in the list, so it makes sense.

The NHibernate issue tracking system is here.

Have you tried using the latest code?

I'm using the syscache code from the official NHibernateContrib here: http://sourceforge.net/project/showfile ... p_id=73818

Is there other code I should be using, beyond that in this release?

If so, could you point me towards it? Many thanks.

Evan

Hi Evan,

What is your current status with this issue? I am also seeing this occasionally and it is frustratingly difficult to pinpoint the cause of it or figure out how to implement some safety measures/workarounds.

My site is in development at the moment with very low traffic and I am very worried that this error will go banana's when there is a high volume of traffic on the site. The error seems to "stick" once it has occured and I usually have to reset the web process to clear the cache...not nice!

Would appreciate any assistance you can provide on this.

Thanks,

Jason

Have you tried a different cache? It would help to isolate whether the problem is in SysCache, NHibernate, or somewhere else.

This bug should be fixed in 1.0.3 which will be out tomorrow.

That's great news...thanks guys.

I was wondering if my problem might be due to running 2 copies of the application on my computer and the cache being shared in some way...is that possible? However, the problem has occurred when I shutdown all of the web processes and just ran one of the applications so I figured there might be some other cause too.

Anyway, I have added different cache region prefixes for each application just in case.

FYI: I upgraded to 1.0.3 and this problem does seem to have been resolved.

Thanks,

Jason