DragonFly BSD
DragonFly users List (threaded) for 2006-09
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: NFS locking issues? (where to go for info?)


From: nospam@xxxxxxxxxxxxx (Jamie)
Date: 14 Sep 2006 19:17:05 GMT

In <200609141800.k8EI0DSf043244@xxxxxxxxxxxxxxxxxxxx>,
Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx> mentions:
>:Has anyone had much success with NFS locking? (where would one
>:go if they wanted to understand the details?)
>:
>:I've read that it's been fixed for some time, that I need to run
>:rpc.statd and rpc.lockd (and in fact, mutt complains unless I run
>:these) I'm assuming 'statd' locks the local file and lockd talks
>:to statd when a remote wants a lock?
>
>    It isn't going to work.  The locks will be honored locally but we
>    do not have a working lockd, and even FreeBSD which has spent a lot
>    of time trying to get a working lockd doesn't have one that is
>    reliable.   NFS locking in general is not reliable and I would not
>    recommend that anyone try to do it.
>
>    Ultimately we will be able to support locking across DragonFly boxes
>    as part of our clustering work, but I really doubt we will ever have
>    reliable NFS locking across disparate operating systems.

Hmm.. Thanks! Glad to know it's not going to work, so I can stop chasing
after it. :-)

Anyone recommend an alternate (fast) network filesystem? I'm not exactly 
crazy about NFS, it's just that it's the default. Locks aren't a huge deal
I guess. (glad to know it's still a problem though)

One thing I used to do (before hearing I shouldn't) is:

	if( symlink(pid,"file.lock")){
		# I have a lock, readlink() gives others the PID.
	}

Are symlinks atomic across NFS?

Funny, seems like such a basic, simple thing yet it remains so illusive.

As far as the clustering, this should really be interesting to see, I have
this picture in my mind of a unified virtual filesystem that tries to
allocate physical space as close to the process that is generating/requesting 
data as possible. Sort of transparently, with a "file" being possibly split
between several machines. 

That seems like it'd really be tricky to do. Taking a machine down might involve
copying terrabytes of data, then freaking out when there isn't enough space
to on the other machines. Files might suddenly disappear and then reappear later.

Very interesting problem. My head hurts just thinking about it!

Jamie
-- 
http://www.geniegate.com                    Custom web programming
guhzo_42@xxxxxxxxx (rot13)                User Management Solutions



[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]