There is a multithreads application, that works with large DB file (>600 Mb). "Database is locked" problem started when I added blob data, and started operate with >30 Kb of BLOB data per request. I think problem related to small HDD speed. It looks like SQLite deletes -journal file, one thread of my application got out of lock (because -journal file was applied and deleted), and other my thread want to do smth with DB, but SQLite still updates DB file... Sure, I can do a minute delays after each DB call, but this is not a solution, because I need more speed.
Now I use session per conversation (per thread) implementation. So there is one ISessionFactory per application object and many ISession objects.
There are my helper classes (as you can see I use IsolationLevel.Serializable and CurrentSessionContext = ThreadStaticSessionContext):
public abstract class nHibernateHelper
{
private static FluentConfiguration _configuration;
private static IPersistenceContext _persistenceContext;
static nHibernateHelper() {}
private static FluentConfiguration ConfigurePersistenceLayer()
{
return Fluently.Configure().Database(FluentNHibernate.Cfg.Db.SQLiteConfiguration.Standard.ShowSql().UsingFile(_fileName).IsolationLevel(IsolationLevel.Serializable).MaxFetchDepth(2)).
Mappings(m => m.FluentMappings.AddFromAssemblyOf<Foo>()).CurrentSessionContext(typeof(ThreadStaticSessionContext).FullName);
}
public static ISession CurrentSession
{
get { return _persistenceContext.CurrentSession; }
}
public static IDisposable OpenConnection()
{
return new DbSession(_persistenceContext);
}
}
public class PersistenceContext : IPersistenceContext, IDisposable
{
private readonly FluentConfiguration _configuration;
private readonly ISessionFactory _sessionFactory;
public PersistenceContext(FluentConfiguration configuration)
{
_configuration = configuration;
_sessionFactory = _configuration.BuildSessionFactory();
}
public FluentConfiguration Configuration { get { return _configuration; } }
public ISessionFactory SessionFactory { get { return _sessionFactory; } }
public ISession CurrentSession
{
get
{
if (!CurrentSessionContext.HasBind(SessionFactory))
{
OnContextualSessionIsNotFound();
}
var contextualSession = SessionFactory.GetCurrentSession();
if (contextualSession == null)
{
OnContextualSessionIsNotFound();
}
return contextualSession;
}
}
public void Dispose()
{
SessionFactory.Dispose();
}
private static void OnContextualSessionIsNotFound()
{
throw new InvalidOperationException("Ambient instance of contextual session is not found. Open the db session before.");
}
}
public class DbSession : IDisposable
{
private readonly ISessionFactory _sessionFactory;
public DbSession(IPersistenceContext persistentContext)
{
_sessionFactory = persistentContext.SessionFactory;
CurrentSessionContext.Bind(_sessionFactory.OpenSession());
}
public void Dispose()
{
var session = CurrentSessionContext.Unbind(_sessionFactory);
if (session != null && session.IsOpen)
{
try
{
if (session.Transaction != null && session.Transaction.IsActive)
{
session.Transaction.Rollback();
}
}
finally
{
session.Dispose();
}
}
}
}
And there is repository helper class. As you can see there are locks by every DB call, so concurrence DB call can't appear, for different threads too, because _locker object is static.
public abstract class BaseEntityRepository<T, TId> : IBaseEntityRepository<T, TId> where T : BaseEntity<TId>
{
private ITransaction _transaction;
protected static readonly object _locker = new object();
public bool Save(T item)
{
bool result = false;
if ((item != null) && (item.IsTransient()))
{
lock (_locker)
{
try
{
_transaction = session.BeginTransaction();
nHibernateHelper.CurrentSession.Save(item);
nHibernateHelper.Flush();
_transaction.Commit();
result = true;
} catch
{
_transaction.Rollback();
throw;
}
//DelayAfterProcess();
}
}
return result;
}
//same for delete and update
public T Get(TId itemId)
{
T result = default(T);
lock (_locker)
{
try
{
result = nHibernateHelper.CurrentSession.Get<T>(itemId);
}
catch
{
throw;
}
}
return result;
}
public IList<T> Find(Expression<Func<T, bool>> predicate)
{
IList<T> result = new List<T>();
lock (_locker)
{
try
{
result = nHibernateHelper.CurrentSession.Query<T>().Where(predicate).ToList();
}
catch
{
throw;
}
}
return result;
}
}
I use previous classes like this (I call nHibernateHelper.OpenConnection() once per thread). Repository is instantiated by singletone:
using (nHibernateHelper.OpenConnection())
{
Foo foo = new Foo();
FooRepository.Instance.Save(foo);
}
I tried to change IsolationLevel to ReadCommited, but this not changes a problem. Also I tried to solve this problem by change SQLite journal mode from journal to WAL:
using (nHibernateHelper.OpenConnection())
{
using (IDbCommand command = nHibernateHelper.CurrentSession.Connection.CreateCommand())
{
command.CommandText = "PRAGMA journal_mode=WAL";
command.ExecuteNonQuery();
}
}
This helped on computers with fast HDD, but on some I got same error. Then I tried to add "DB update file exist" check to repository, and delay after each save/update/delete procedure:
protected static int _delayAfterInSeconds = 1;
protected void DelayAfterProcess()
{
bool dbUpdateInProcess = false;
do
{
string fileMask = "*-wal*";
string[] files = Directory.GetFiles(Directory.GetCurrentDirectory(), fileMask);
if ((files != null) && (files.Length > 0))
{
dbUpdateInProcess = true;
Thread.Sleep(1000);
}
else
{
dbUpdateInProcess = false;
}
} while (dbUpdateInProcess);
if (_delayAfterInSeconds > 0)
{
Thread.Sleep(_delayAfterInSeconds * 1000);
}
}
Same solution (check for DB update file) not worked for -journal file. It reported, that -journal file was deleted, but I still got errors. For -wal file it works (as I think. I need more time to test it). But this solution seriously brake program.
Maybe you can help me?