开发者

A shared List/Map among several threads in F#?

I'm doing a program to handle many blocking I/O operations at a time by spawning an Agent/MailboxProcessor per operation. I've got a bunch of files I've cached in memory in a Map which I want to share among these agents. However, I've al开发者_Go百科so got a FileSystemWatcher to callback whenever changes are made to the files, so that I can update the cache.

How do I make this happen without risking the cache being corrupted by multi-threaded read and write ?

It seems to me that the Map is already based on pointers to objects, so would that automatically solve my problem as I'm simply changing the pointers to the new objects as they are loaded, or is this a broken understanding of it?

Thanks


It seems to me that the Map is already based on pointers to objects, so would that automatically solve my problem as I'm simply changing the pointers to the new objects as they are loaded, or is this a broken understanding of it?

I think your understanding is correct. You can just have a single mutable reference to an immutable Map. Writing a new map to the reference is atomic so there is no need to synchronize that.


When I've seen similar Erlang programs, the systems are set up like this:

  • You can wrap up the FileSystemWatcher with a MailboxProcessor, that way you're handling incoming updates as messages and not windows events. Your FileSystemWatcherProcess can hold a list of children who are listening, and push out updates as needed. This is basically the same thing event-based programming, only with messages and actors instead.

    Your FileSystemWatcherProcess should not need to maintain a your cache of files, it just blindly pushes out messages.

  • OR You have a master process which holds the state of the map. File SystemWatcher sends updates to the master. Each child thread holds a reference to the master, so that each time they finish processing an item or batch of items, they send a message to the master process requesting the latest Map.

Neither system requires any locking.


Following up to Jon's answer. if you end up having multiple writers, then instead of locks you can always do CAS:

 let updateMap value =
    let mutable success = false
    while not success do
      let v = !x
      let result = Interlocked.CompareExchange(x, v.Add(value), v)
      success <- Object.ReferenceEqual(v, result)

And, if targeting only .Net 4.0 is an option for you, you shouldn't be inventing all that stuff yourself: there is a System.Collections.Concurrent.ConcurrentDictionary class which implements concurrently readable and writable dictionary already.

0

上一篇:

下一篇:

精彩评论

暂无评论...
验证码 换一张
取 消

最新问答

问答排行榜