NFSv4 : open blocked
J. Bruce Fields
bfields at fieldses.org
Wed Mar 21 12:06:51 EDT 2007
On Wed, Mar 21, 2007 at 01:08:10PM +0100, Vincent ROQUETA wrote:
> Here is a simple test that blocks NFSv4 on kernel 2.6.21-rc3+CITI, using 16K
> pages.
>
>
> 1/ Run fsstress during few time (3 minutes) with this parameters
> fsstress -l 0 -d /mnt/nfsv4/ -n 1000 -p 50 -r
> 2/ cat an existing file on NFS
> For example :
...
> Open remains blocked, until I kill the "cat" process. In similar conditions I
> have caused a kernel bug (umount process). Anyone have more ideas to
> investigate those bugs ?
The obvious first question is whether they're likely to be client- or
server- side. In this case we could probably determine that by looking
at the traffic on the wire, and seeing whether the client is waiting for
the server to respond to some request (maybe the open). or we could
keep the client constant while varying the server configuration (kernel
version or maybe page size), or the reverse--keeping the server constant
while varying the client.
In cases where the client is blocked, a sysrq-t dump can be helpful too
sometimes.
--b.
More information about the NFSv4
mailing list