C# Discussion Boards - CodeProject

2-Mar-10 6:23

Hello,

I am developing a windows service that will process
xml files as they come in.
I have a FileSystemWatcher that is watching a folder
and processing the files when the fileSystemWatcher1_Created
event is raised.
I created a FileProcessor class that handles processing a file,
it splits the incoming xml file into many different files and
then pushes data to a database using sqlBulkCopy, etc.

So right now my code looks like this...

private void fileSystemWatcher1_Created(object sender, FileSystemEventArgs e)
{
    Logger.AddInformation(string.Format("File {0} {1}.", e.Name, e.ChangeType));

    FileProcessor fileProcessor = new FileProcessor(e.Name);
    bool retValue = fileProcessor .ProcessFile();

    Logger.AddInformation(string.Format("Processed File {0} {1}.", e.Name, retValue ? "Successfully" : "Unsuccessfully"));
}

But I have read that the FileSystemWatcher can be unreliable in some
situations so I wanted to build in a backup mechanism that polls
and checks the folder periodically for files that may have been missed,
or were already in the folder when the service was started, etc.

So my goal was to have some kind of object that maintains a queue
of files that need to be processed.
Then when the fileSystemWatcher1_Created event is raised instead of
processing the file at that time I would just add the file to the queue.
And also when the polling if I encountered a new file I would add it to the
queue as well (making sure that file was not already processed or already
on the queue).

Problem is I'm not really sure how to implement this.
I'm imagining I may need another Thread or Threads.
I have some code that creates another thread and uses a timer so that
could potentially be my backup polling mechanism running every x seconds
scanning folder and adding files to the queue.

But I still don't fully understand how to make it work.
I'm feeling like something would need to manage the queue and keep
processing the files until it's empty, and not sure if that should be
happening on the main windows service thread or a separate thread, etc.

I'm trying to keep the design as simple as possible and hopefully limit it
to just 1 windows service.
I just want to process the files 1 by 1 anyway, so I'm not looking to spawn
a new thread for every file or anything like that.

Can anyone point me in the right direction?
Maybe slap together some skeleton code that would help me understand.
Or are there any good articles or examples of doing something like this or
a design pattern I should look into?

Thanks!

modified on Tuesday, March 2, 2010 1:29 PM

Paulo Zemek2-Mar-10 6:36

2-Mar-10 6:36

If I understood your real intention, I can help.

You will need a HashSet<string>;
A ManualResetEvent;
And two threads.

One thread can look like this:

while(!stopRequested)
{
  GetsAllFilesInTheDirectory
  lock(hashset)
    AddAllFilePathsToTheHashSet.

  if at least one file was added, set the manualResetEvent.

  Wait for some time (a minute?)
}

The other thread will look like this:

while(!stopRequested)
{
  manualResetEvent.WaitOne();

  if (stopRequested)
    break;

  manualResetEvent.Reset();

  lock(hashset)
  {
    foreach(var path in hashset)
    {
      ... do what you need...
    }

    hashset.Clear();
  }
}

And, the event of the file created, you must:

lock(hashset)
  add the path of the file to the hashset.

manualResetEvent.Set();

And that's all.

When finishing your application, you must set stopRequested (which must be volatile) and also must set your manualResetEvent.

PIEBALDconsult2-Mar-10 8:50

PIEBALDconsult

2-Mar-10 8:50

That appears to be rather memory-intensive. Dead | X|

And it doesn't appear to address existent files being updated or replaced (or still being written?).

Paulo Zemek2-Mar-10 9:48

2-Mar-10 9:48

I am only saying how it is possible to do it the way the original poster asked.

When I do something "like" that I always consider that processed files will be removed from that directory (in general, go to a OK or Error folder), and that files are never changed. I really don't know what's the intent of the original poster, but I think I answered it.

Also, it does not uses much memory, as a single database connection (as you suggest) needs to keep an active TCP/IP connection (and depending on the driver will have a secondary thread of it's own).

krazykoder2-Mar-10 10:55

2-Mar-10 10:55

Yes, after file is processed it gets moved to a different Processed or Error folder.

PIEBALDconsult2-Mar-10 12:38

PIEBALDconsult

2-Mar-10 12:38

(I don't know who down-voted that, I guess I'll go up-vote it.)

Paulo Zemek wrote:
processed files will be removed

Not in my experience, other than archiving or purging at some later time, perhaps daily or weekly. I mostly work with third-party products which may produce files I can read or copy, but I dare not alter them in any way.

Paulo Zemek wrote:
keep an active TCP/IP connection

I don't leave my database connections open. And I often have my Services running on the same server as the database, so maybe it's a named pipe rather than a network connection anyway.

krazykoder2-Mar-10 9:23

Re: Windows Service, FileSystemWatcher + Polling [modified]

2-Mar-10 9:23

Thanks for your reply.

Still digesting it but have a couple questions.
When you say I will need 2 threads, are you counting the services default thread as 1 of the 2? The one that has the OnStart() & OnStop() events and my file Created event.

What exactly is stopRequested?
I'm guessing this is just a flag bool variable just not sure where it is (or is supposed to be) defined.
Do I set this to true when in service OnStop() event, and then additional threads finish up?

Thanks

Paulo Zemek2-Mar-10 9:39

2-Mar-10 9:39

You will need two more threads.

The original thread deals with the OnStart, OnStop and the FileSystemWatcher event.
One thread deals with the "time by time lookup in the directory"
And one thread deals with the real processing.

The stopRequested is a volatile bool. In OnStop you must set it to true and also call manualResetEvent.Set(), so it can check that stop was requested (instead of being kept in wait mode) and return.

It will be good to have another manualResetEvent. So, in the thread that runs from time to time, instead of calling Thread.Sleep(60000) you will call
otherManualResetEvent.WaitOne(60000);

So, in OnStop you set it and it returns immediatelly and, when it is running, it will have the time-out after 60 seconds and will execute again.

The stopRequested must be set to true before setting the events, or else you can end-up in a situation where you set the event, it tries to run, does nothing, resets the event and then you set the variable... so your thread will not exit.

modified on Tuesday, March 2, 2010 3:49 PM

krazykoder3-Mar-10 9:32

3-Mar-10 9:32

Thanks Paulo,

Been reading up on threading a bit and still looking into using your solution.

Still have a few more questions...

while(!stopRequested)
{  
	manualResetEvent.WaitOne();  
	if (stopRequested)    
		break;  
	manualResetEvent.Reset();  
	lock(hashset)  
	{    
		foreach(var path in hashset)    
		{
			... do what you need...    
		}    
		hashset.Clear();  
	}
}

... do what you need...

1) if it locks that hashset and then loops thru and processes
the files 1 by 1 (which could potentially take a couple minutes a file)
then wont the other threads need to sit and wait until they can obtain the lock on the hashset? For some reason to me that seems like it is defeating the purpose of using the extra thread that processes the files.

If the thread that processes the files is running and locks the hashset for 5 minutes say, and 5 more files get dropped into the IN directory causing 5 more filesystemwatcher.fileCreated events to happen, the 1st one will be waiting to obtain a lock on the hashset and the other 4 events will be waiting in line (and filling up the file system watcher buffer) won't they?

I was trying to avoid doing much work in the filesystemwatcher.fileCreated event (as many seem to suggests) to stop it from bottlenecking and filling it's buffer and cause events to get lost.

Maybe I'm missing something, but seems like if it sits there and waits for lock to become available, it could have just done the work itself.

Maybe I can alter the logic to lock hashset, get 1st file, unlock hashset, process the file, lock hashset, remove the file just processed from hashset , unlock hashset, loop again.

Does that make any sense or is this not a legit concern?

Paulo Zemek3-Mar-10 13:33