Hi.
We have an issue with some apparent caching going on in nfs clients
supplied with SLES 9.1 and RedHat 5. We use nfs to manage the a queue
of tasks in a distributed environment thus
ServerA ServerB
\ /
\ /
/remote/drive
|_ pending
|_ processing
|_ processed
ServerA and ServerB look for files in the pending directory, the
content of which describe the task to perform. If a file is found, the
server which found it will move it to the processing directory and
start working, once complete, the task will be moved to the processed
directory.
The problem is that ServerA picks up the task and moves it into
processing and sometime after (about 1/10 second in recent testing),
ServerB scans the pending directory and sees the task which has been
processed by ServerB. It tries to move the file to the processing
directory before commencing with the task and this move fails which
makes me think it does not go back to the nfs server to stat the file,
only to perform move (and presumably to read the contents).
I have mounted the file system with the -noac option. I have also set
up the export with no_wdelay option. Is there anything I can do to
ensure I do not get any stale data. We currently try to refresh the
directory by doing a directory scan, but this does not appear to be
completely effective.
Unfortunately this is a much simplified example and it would not be
straightforward to replace this file based queue with a sockets
equivalent.
Many Thanks In Advance,
|
Similar Threads
Linear Mode
