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

Re: files not in sync on nfs mounted directories



hi ya karsten

On Tue, 20 Sep 2005, Karsten Bolding wrote:

> > how long is "quite some time" ( seconds, minutes, hours )
> > 	1-2 minutes of delay before data new show up on a bz network is
> > 	possible
> > 
> 
> Some minutes is not un-common.

:-)

> If the file is copied the problem seems to be solved - moving it is not
> enough.

there is a biggg difference to the filesystem when you
	cp xxx soem-place-else/yy
vs
	mv xxx some-place-else/yy  

in either case, if you do use mv, do make sure you run sync on it
 	mv xxx yyy ; sync ; sync

> > what is your nfs options
> 
> none except:
> /home   n02(rw) n03(rw) n04(rw)

add sync at a minimum in addition to rw options

to see the other default options

	mount   ( no other options to see everything )
 
> hard-mounted 

hard mounted  means in the above case
	/home	no2(rw,hard)  no3(rw,hard)

which means if the machine goes down, everybody waits till it comes up

if its soft mounted, you can ^C the hung jobs and recover and restart
all the jobs and use some other duplicate files from other working
machines

things like ls, df and all other commands wont hang and use
up the network and cpu and memory and everything is waaiting
for everybody elses nfs timeouts

- i do NOT use hard mounts ... i dont like to wait ..

	but that assumes i have multiple data that if it happens
	to get corrupted due to system crashes, i have good data
	elsewhere

- I use the automounter for data directories but have
> sometimes experienced time outs on those and didn't want that during a
> simulation that might take several days. 

sounds like you have a bad network ...
	- use intel nic cards ...
	- use the latest kernels ...
	- on and on to tweek the network
 
> > do you run sync to flush the data to the disk
> > 
> no - normally I don't  - actually I never do.

time to start since your data is not showing up on the other end

> > different apps does different things to data it thinks is current
> > on itself vs what's on the master ( it's trying to be too smart )
> 
> the app in this case is just vi (or another editor).

in that case .. simple sync after mv and in the mount table should
clean things up a bit

c ya
alvin



Reply to: