Improving zpool import time
Due to many tailed disks on our cluster environment, the zpool import can consume a lot of time. Since we created a seperate directory for each zone and linked the concerned block devices from /dev/vx/dmp to this directory, we were able to start a zpool with "zpool import -d /dev/. That decreases the import time enormously as shown below.
14:57:04:root:gf01sxdb102t:/opt/VRTSvcs/bin/Zpool > time zpool import bais2t_msc real 4m10.956s user 0m0.311s sys 0m12.495s 15:04:56:root:gf01sxdb102t:/opt/VRTSvcs/bin/Zpool > zpool export bais2t_msc
15:05:04:root:gf01sxdb102t:/opt/VRTSvcs/bin/Zpool > time zpool import -d /dev/gf0zsxdb110t/ bais2t_msc real 0m0.811s user 0m0.031s sys 0m0.469s
Could you please add a "import -d" - like Option to the zpool - ressource - type and the start script of the VCS Agent? Resource type definition: type Zpool ( static str ArgList[] = { PoolName, AltRootPath, ChkZFSMounts, ZoneResName, "ZoneResName:State" } str PoolName str AltRootPath boolean ChkZFSMounts = 1 str ZoneResName ) Thanks!
Comments 3 Comments • Jump to latest comment
This feature will be available in upcoming SF/HA VCS 6.0 release (CQ4 2011)
Please think about the alternative directory and how the administrator should manage the links in the directory. There should be a mechanism to be sure, that the directoy and the links to the dmp devices are correct on all nodes.
Anyhow, when providing pools to zones, it sould be done like
use zonecfg :
add fs
set type=zfs
set special=<zpool>/<mountpoint>
set dir=<mountpoint within the zone>
end
Other wiese problerms might occur, for example when trying to use liveupgrade...
Would you like to reply?
Login or Register to post your comment.