There is a .NET 4.7 WebAPI application working with SQL Server using Entity Framework and hosting NServiceBus endpoint with MSMQ transport.
Simplified workflow can be described by a controller action:
[HttpPost]
public async Task<IHttpActionResult> SendDebugCommand()
{
var sample = new Sample
{
State = SampleState.Initial,
};
_dataContext.Set<Sample>().Add(sample);
await _dataContext.SaveChangesAsync();
sample.State = SampleState.Queueing;
var options = new TransactionOptions
{
IsolationLevel = IsolationLevel.ReadCommitted,
};
using (var scope = new TransactionScope(TransactionScopeOption.Required, options, TransactionScopeAsyncFlowOption.Enabled))
{
await _dataContext.SaveChangesAsync();
await _messageSession.Send(new DebugCommand {SampleId = sample.Id});
scope.Complete();
}
_logger.OnCreated(sample);
return Ok();
}
And DebugCommand handler, that is sent to the same NServiceBus endpoint:
public async Task Handle(DebugCommand message, IMessageHandlerContext context)
{
var sample = await _dataContext.Set<Sample>().FindAsync(message.SampleId);
if (sample == null)
{
_logger.OnNotFound(message.SampleId);
return;
}
if (sample.State != SampleState.Queueing)
{
_logger.OnUnexpectedState(sample, SampleState.Queueing);
return;
}
// Some work being done
sample.State = SampleState.Processed;
await _dataContext.SaveChangesAsync();
_logger.OnHandled(sample);
}
Sometimes, message handler retrieves the Sample from the DB and its state is still Initial, not Queueing as expected. That means that distributed transaction initiated in the controller action is not yet fully complete. That is also confirmed by time-stamps in the log file.
The 'sometimes' happens quite rarely, under heavier load and network latency probably affects. Couldn't reproduce the problem with local DB, but easily with a remote DB.
I checked DTC configurations. I checked there is escalation to a distributed transaction for sure. Also if scope.Complete() is not called then there will be no DB update neither message sending happening.
When the transaction scope is completed and disposed, intuitively I expect both DB and MSMQ to be settled before a single further instruction is executed.
I couldn't find definite answers to questions:
Is this the way DTC work? Is this normal for both transaction parties to do commits, while completion is not reported back to the coordinator?
If yes, does it mean I should overcome such events by altering logic of the program?
Am I misusing transactions somehow? What would be the right way?
In addition to the comments mentioned by Evk in Distributed transaction with MSMQ and SQL Server but sometimes getting dirty reads here's also an excerpt from the particular documentation page about transactions:
A distributed transaction between the queueing system and the persistent storage guarantees atomic commits but guarantees only eventual consistency.
Two additional notes:
NServiceBus uses IsolationLevel.ReadCommitted by default for the transaction used to consume messages. This can be configured although I'm not sure whether setting it to serialized on the consumer would really solve the issue here.
In general, it's not advised to use a shared database between services as this highly increases coupling and opens the door for issues like you're experiencing here. Try to pass relevant data as part of the message and keep the database an internal storage for one service. Especially when using web servers, a common pattern is to add all the relevant data to a message and fire it while confirming success to the user (as the message won't be lost) while the receiving endpoint can store the data to it's database if necessary. To give more specific recommendations, this requires more knowledge about your domain and use case. I can recommend the particular discussion community to discuss design/architectural question like this.
I'm using the source code like this:
Database db1 = new Database(); //init 1 db connection
db1.BeginTransaction();
//this function used to check exist customer
//in this function, I also use Database db2 = new Database(); db2.Close();
CheckExistCustomer();
InsertCustomer(db1, strInsert); //this function worked correct, use db1
if(iErrorCode == ErrorStatus.SUCCESSED)
db1.CommitTransaction();
else db1.RollbackTransaction();
As you see, I have 2 db connections. Can I use them? When db2.Close(), it won't affect to current db1, right? Should I use only 1 db connection (db1)?
When I run CheckExistCustomer(), the program will hang on. I don't know why. Any clue?
Please advise.
I'm very appriciated for your help.
You are doing transaction management in this code. in Transaction you are performing operation on two DB, This is called Distributed Transactions, I guess BeginTransaction() method does not support Distributed Transaction, Please use TransactionScope class
TransactionScope: Avoiding Distributed Transactions
I had thought I was clever. But in light of recent discoveries, I'm not so sure any more. During the page life cycle, there could by any number of database interactions. Some back to back, others spread out. So I invented an object that keeps an instance of an SQL connection alive in the HttpContext.Items dictionary. Every db request then uses this connection and when the http request has ended, I properly dispose of the connection. We're looking at a couple hundred milliseconds the connection would be open, and with some heavy http caching, running out of available connections isn't a concern.
The point was to prevent additional round trips due to the establishing of new connections. But when I stumbled upon the knowledge of connection pooling, I think it pretty invalidates the usefulness of preserving the SqlConnection. Or does it?
Is scenario A the same as Scenario B, performance wise? Which would you recommend? Does scenario B provide no performance gains, and even possibly hinders it because of some edge cases where the connection might not disposed of properly? Forgive the pseudo-ness in the examples, I don't want to clutter them with barf.
A
using (var connection = new SqlConnection(connectionString))
{
using (var command = new SqlCommand("...", connection))
{
... doing database stuff ...
}
}
... traversing the stack ...
using (var connection = new SqlConnection(connectionString))
{
using (var command = new SqlCommand("...", connection))
{
... doing database stuff ...
}
}
B
var connectionKeeper = new ConnectionKeeper();
// Add to the context items so it can be used anywhere
Context.Items.Add("Connection", connectionKeeper);
... traversing the stack ...
using (var command = new SqlCommand("...", connectionKeeper.Connection))
{
... doing database stuff
}
... traversing the stack ...
using (var command = new SqlCommand("...", connectionKeeper.Connection))
{
... doing database stuff
}
... traversing the stack ...
// The end of the request
sqlKeeper.Dispose();
Use your code from section A. Please let the connection pool do it's job. Avoid keeping a static SqlConnection around at all costs. The connection pool was designed for this.
Here's an MSDN article for your reference.
SQL Server Connection Pooling (ADO.NET)
No point in doing that in code unless you turn connection pooling off.
And you should have a real serious think before doing that, that's extreme circumstances.
Connection pooling was invented to address the situation you are trying to address with this "permanent" connection, so it will actually interfere with the built in optimisations and increase the volume, complexity, and fragility of your code.
I have a ASP.net based application.
The CPU on the SQL Server box is constantly ~90 - 100%
There are a lot of inneficient queries, which I am currently working on, however, looking at the code from a previous coder, he never seemed to close (or dispose) the SqlConnection
When I run the folloing query, I get around 450 connections that are "Awaiting Command"
SELECT Count(*) FROM
MASTER.DBO.SYSPROCESSES WHERE
DB_NAME(DBID) = 'CroCMS' AND DBID != 0
AND cmd = 'AWAITING COMMAND'
Is this likely to be causing a problem?
I read this and it seems to relate:
http://www.pythian.com/news/1270/sql-server-understanding-and-controlling-connection-pooling-fragmentation/
We are also getting a lot of timeouts, specifically when replication is enabled..
I'm not sure if this is related.. Have disabled replication (transactional) for now and it seems ok..
(This server is a subscriber to our in office Database server)
Would disposing of the SQL connection object help?
Yes, dispose them. Otherwise ignore them for now. Possibly the pool is as large because the statements are slow. I would more suggest:
Fixing the statements.
Check the applicaion that it only uses one connection PER REQUEST (i.e. not open multiple at the same time).
If the problem does not get better after optiomizing SQL - you can revisit the pool.
You should always dispose the command object when your done with it. that way the connection pooling can be used better.
easist is to use the using statment.
using (
var sqlCommand = new SqlCommand(
"storedprocname",
new SqlConnection("connectionstring"))
{ CommandType = CommandType.StoredProcedure })
{
// do what you should.. setting params executing etc etc.
}
I'm performing a large number of INSERTS to a SQLite database. I'm using just one thread. I batch the writes to improve performance and have a bit of security in case of a crash. Basically I cache up a bunch of data in memory and then when I deem appropriate, I loop over all of that data and perform the INSERTS. The code for this is shown below:
public void Commit()
{
using (SQLiteConnection conn = new SQLiteConnection(this.connString))
{
conn.Open();
using (SQLiteTransaction trans = conn.BeginTransaction())
{
using (SQLiteCommand command = conn.CreateCommand())
{
command.CommandText = "INSERT OR IGNORE INTO [MY_TABLE] (col1, col2) VALUES (?,?)";
command.Parameters.Add(this.col1Param);
command.Parameters.Add(this.col2Param);
foreach (Data o in this.dataTemp)
{
this.col1Param.Value = o.Col1Prop;
this. col2Param.Value = o.Col2Prop;
command.ExecuteNonQuery();
}
}
this.TryHandleCommit(trans);
}
conn.Close();
}
}
I now employ the following gimmick to get the thing to eventually work:
private void TryHandleCommit(SQLiteTransaction trans)
{
try
{
trans.Commit();
}
catch (Exception e)
{
Console.WriteLine("Trying again...");
this.TryHandleCommit(trans);
}
}
I create my DB like so:
public DataBase(String path)
{
//build connection string
SQLiteConnectionStringBuilder connString = new SQLiteConnectionStringBuilder();
connString.DataSource = path;
connString.Version = 3;
connString.DefaultTimeout = 5;
connString.JournalMode = SQLiteJournalModeEnum.Persist;
connString.UseUTF16Encoding = true;
using (connection = new SQLiteConnection(connString.ToString()))
{
//check for existence of db
FileInfo f = new FileInfo(path);
if (!f.Exists) //build new blank db
{
SQLiteConnection.CreateFile(path);
connection.Open();
using (SQLiteTransaction trans = connection.BeginTransaction())
{
using (SQLiteCommand command = connection.CreateCommand())
{
command.CommandText = DataBase.CREATE_MATCHES;
command.ExecuteNonQuery();
command.CommandText = DataBase.CREATE_STRING_DATA;
command.ExecuteNonQuery();
//TODO add logging
}
trans.Commit();
}
connection.Close();
}
}
}
I then export the connection string and use it to obtain new connections in different parts of the program.
At seemingly random intervals, though at far too great a rate to ignore or otherwise workaround this problem, I get unhandled SQLiteException: Database file is locked. This occurs when I attempt to commit the transaction. No errors seem to occur prior to then. This does not always happen. Sometimes the whole thing runs without a hitch.
No reads are being performed on these files before the commits finish.
I have the very latest SQLite binary.
I'm compiling for .NET 2.0.
I'm using VS 2008.
The db is a local file.
All of this activity is encapsulated within one thread / process.
Virus protection is off (though I think that was only relevant if you were connecting over a network?).
As per Scotsman's post I have implemented the following changes:
Journal Mode set to Persist
DB files stored in C:\Docs + Settings\ApplicationData via System.Windows.Forms.Application.AppData windows call
No inner exception
Witnessed on two distinct machines (albeit very similar hardware and software)
Have been running Process Monitor - no extraneous processes are attaching themselves to the DB files - the problem is definitely in my code...
Does anyone have any idea whats going on here?
I know I just dropped a whole mess of code, but I've been trying to figure this out for way too long. My thanks to anyone who makes it to the end of this question!
brian
UPDATES:
Thanks for the suggestions so far! I've implemented many of the suggested changes. I feel that we are getting closer to the answer...however...
The code above technically works however it is non-deterministic! It is not guaranteed to do anything aside from spin in neutral forever. In practice it seems to work somewhere between the 1st and 10th iteration. If i batch my commits at a reasonable interval damage will be mitigated but I really do not want to leave things in this state...
More suggestions welcome!
It looks like you failed to link the command with the transaction you've created.
Instead of:
using (SQLiteCommand command = conn.CreateCommand())
You should use:
using (SQLiteCommand command = new SQLiteCommand("<INSERT statement here>", conn, trans))
Or you can set its Transaction property after its construction.
While we are at it - your handling of failures is incorrect:
The command's ExecuteNonQuery method can also fail and you are not really protected. You should change the code to something like:
public void Commit()
{
using (SQLiteConnection conn = new SQLiteConnection(this.connString))
{
conn.Open();
SQLiteTransaction trans = conn.BeginTransaction();
try
{
using (SQLiteCommand command = conn.CreateCommand())
{
command.Transaction = trans; // Now the command is linked to the transaction and don't try to create a new one (which is probably why your database gets locked)
command.CommandText = "INSERT OR IGNORE INTO [MY_TABLE] (col1, col2) VALUES (?,?)";
command.Parameters.Add(this.col1Param);
command.Parameters.Add(this.col2Param);
foreach (Data o in this.dataTemp)
{
this.col1Param.Value = o.Col1Prop;
this. col2Param.Value = o.Col2Prop;
command.ExecuteNonQuery();
}
}
trans.Commit();
}
catch (SQLiteException ex)
{
// You need to rollback in case something wrong happened in command.ExecuteNonQuery() ...
trans.Rollback();
throw;
}
}
}
Another thing is that you don't need to cache anything in memory. You can depend on SQLite journaling mechanism for storing incomplete transaction state.
Run Sysinternals Process Monitor and filter on filename while running your program to rule out if any other process does anything to it and to see what exacly your program is doing to the file. Long shot, but might give a clue.
We had a very similar problem using nested Transactions with the TransactionScope class. We thought all database actions occurred on the same thread...however we were caught out by the Transaction mechanism...more specifically the Ambient transaction.
Basically there was a transaction higher up the chain which, by the magic of ado, the connection automatically enlisted in. The result was that, even though we thought we were writing to the database on a single thread, the write didn't really happen until the topmost transaction was committed. At this 'indeterminate' point the database was written to causing it to be locked outside of our control.
The solution was to ensure that the sqlite database did not directly take part in the ambient transaction by ensuring we used something like:
using(TransactionScope scope = new TransactionScope(TransactionScopeOptions.RequiresNew))
{
...
scope.Complete()
}
Things to watch for:
don't use connections across multiple threads/processes.
I've seen it happen when a virus scanner would detect changes to the file and try to scan it. It would lock the file for a short interval and cause havoc.
I started facing this same problem today: I'm studying asp.net mvc, building my first application completely from scratch. Sometimes, when I'd write to the database, I'd get the same exception, saying the database file was locked.
I found it really strange, since I was completely sure that there was just one connection open at that time (based on process explorer's listing of active file handles).
I've also built the whole data access layer from scratch, using System.Data.SQLite .Net provider, and, when I planned it, I took special care with connections and transactions, in order to ensure no connection or transaction was left hanging around.
The tricky part was that setting a breakpoint on ExecuteNonQuery() command and running the application in debug mode would make the error disappear!
Googling, I found something interesting on this site: http://www.softperfect.com/board/read.php?8,5775. There, someone replied the thread suggesting the author to put the database path on the anti-virus ignore list.
I added the database file to the ignore list of my anti-virus (Microsoft Security Essentials) and it solved my problem. No more database locked errors!
Is your database file on the same machine as the app or is it stored on a server?
You should create a new connection in every thread. I would simplefy the creation of a connection, use everywhere: connection = new SQLiteConnection(connString.ToString());
and use a database file on the same machine as the app and test again.
Why the two different ways of creating a connection?
These guys were having similiar problems (mostly, it appears, with the journaling file being locked, maybe TortoiseSVN interactions ... check the referenced articles).
They came up with a set of recommendations (correct directories, changing journaling types from delete to persist, etc). http://sqlite.phxsoftware.com/forums/p/689/5445.aspx#5445
The journal mode options are discussed here: http://www.sqlite.org/pragma.html . You could try TRUNCATE.
Is there a stack trace during the exception into SQL Lite?
You indicate you "batch my commits at a reasonable interval". What is the interval?
I would always use a Connection, Transaction and Command in a using clause. In your first code listing you did, but your third (creating the tables) you didn't. I suggest you do that too, because (who knows?) maybe the commands that create the table somehow continue to lock the file. Long shot... but worth a shot?
Do you have Google Desktop Search (or another file indexer) running? As previously mentioned, Sysinternals Process Monitor can help you track it down.
Also, what is the filename of the database? From PerformanceTuningWindows:
Be VERY, VERY careful what you name your database, especially the extension
For example, if you give all your databases the extension .sdb (SQLite Database, nice name hey? I thought so when I choose it anyway...) you discover that the SDB extension is already associated with APPFIX PACKAGES.
Now, here is the cute part, APPFIX is an executable/package that Windows XP recognizes, and it will, (emphasis mine) ADD THE DATABASE TO THE SYSTEM RESTORE FUNCTIONALITY
This means, stay with me here, every time you write ANYTHING to the database, the Windows XP system thinks a bloody executable has changed and copies your ENTIRE 800 meg database to the system restore directory....
I recommend something like DB or DAT.
While the lock is reported on the COMMIT, the lock is on the INSERT/UPDATE command. Check for record locks not being released earlier in your code.