Video Screencast Help
Search Video Help Close Back
to help
New in the Rewards Catalog: Vouchers for "Symantec Technical Specialist" and "Symantec Certified Specialist" exams.

CFS, NFS and locking oddities

Updated: 21 May 2010 | 1 comment
Jeremy Robst's picture
0 0 Votes
Login to vote
This issue has been solved. See solution.

Hi,
I have a 2 node Solaris 10 cluster running SF5.0 MP3, with cluster mounted, cluster rw volumes (CFS). The volumes are exported via NFS to a mixture of Linux and Solaris (9/10) clients. Since converting the volumes to cluster mounts (previously were just vxfs mounted on a single node at once) I've had reports of various programs failing, which I've traced to problems acquiring locks over NFS.

At first I thought this was just the linux clients using flock (which I now can have problems over NFS) instead of fcntl; but after experimenting with this I now have the following :

Solaris 10 clients - can obtain locks without problems, using the fcntl F_SETLKW or F_SETLK commands
Solaris 9 or Linux clients - can option locks with fcntl F_SETLK ok, but *not* with F_SETLKW (ENOLCK is returned) which confuses me - I don't know what the difference is in acquiring locks between F_SETLK and F_SETLKW (other than waiting / returning immediately of course). Especially when trying to acquire a lock I know will succeed why does F_SETLKW consistently fail and F_SETLK consistently succeed ?

With NFS exported UFS or non cluster VXFS volumes the Solaris 9 / Linux clients can acquire locks with F_SETLKW (and also flock) quite happily, 

Has anybody encountered anything like this or got any ideas ?

Any suggestions gratefully received,

Jeremy

Comments

davidnoy's picture
21
Oct
2009
0 Votes 0
Login to vote

We plan to release full

We plan to release full NFS locking support when shared out of multiple CFS nodes as part of the 5.1 release end of this year.
Until then, you cannot gaurantee that during failure you will not be exposed to potential locking issues.
Therefore I recommend waiting 5 wks or so and picking up our release which resolves these issues.

We had planned  to support NFS lock coordination in a CFS cluster for Sol 10 and onwards. Sol 9 was deemed
unsupportable but I dont know exactly why except maybe that Sol 10 has the right hooks in it so that we are aware
cluster-wide of lock status.

If you need NFS lock coordination earlier than Dec 2009, please contact me.

If you need NFS lock coordination on Sol 9, please contact me.

Thanks.