Patton, Matthew F, CTR, OSD-PA&E wrote:
Classification: UNCLASSIFIED
> A better approach is to export not an GNBD but an iSCSI
> device from DRBD.
how about exporting each single HD as iSCSI instead of GNBD in the first
place then? Is the iSCSI subsystem more robust in that sense? Then on
the softwareRaid master instead of exporting a filesystem, export a
raid'ed iSCSI device. If the RAID meta-data were spread out onto
multiple disks (which actually would mean multiple machines) then if the
master went down any of the others should be able to rebuild the RAID
set or at least run in degraded mode.
In theory, yes. This is what we have "working" in the lab using GNBD.
Under "normal" fail-over conditions, the remaining node does re-assemble
a degraded RAID device. The problems mainly stem from GNBD under
certain fail-over conditions. Both in importing/exporting the component
disks, and then importing/exporting the assembled raid device.
I'm working out of the office today but will post our init scripts for
the GNBD equivalent, on the understanding that you post back your iSCSI
stuff and mods to our scripts.
Regards,
James
--
Linux-cluster@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/linux-cluster
