Hello SHOUJIN,
Sunday, February 8, 2009, 5:36:13 AM, you wrote:
SW> I created a 2 nodes NFS cluster with ZFS+NFS. I used resource
SW> type SUNW.HAStoragePlus for Zpool. During failover, the zpool
SW> needs to be imported in 2nd node.
SW> Unfortunately, the "zpool import" is very slow (need about 2.5
SW> minutes in my system). This caused the whole NFS service failover
SW> very slowly. I searched the ZFS document, there is a work acround
SW> can make the "zpool import" very fast: Set cache file for the
SW> pool, then use "zpool import -d FILE_NAME" to import from cache
SW> file. But I don't know how to add this kind of specail command
SW> line to HA cluster configuration so that the agent can call it
SW> when failover happens. Could anybody help on this?
I don't believe that SC supports cachefile property currently - it
would be a good RFE. You should be able to implemente id yourself via
GDS if really needed - but first I would check what happens if you manually
import a pool with and without cachefile and compare timing. Then last
time I checked HAS+ does import actually twice (looks like first time
it is checking if pool is ok to import and if it is then it imports
again by going thru all the scanning, etc.). Then in recent nevada
build imports are happaning in a more parallel way - so instead of
reading labels one by one it happens in parallel - don't know if it
has been back ported to S10.
--
Best regards,
Robert Milkowski
http://milek.blogspot.com
