iSCSI troubles

[Matthew Crocker]

On Jan 12, 2008, at 12:39 AM, Matthew Geier wrote:

> iscsiadm --mode discovery --type sendtargets --portal 192.168.130.1
> (Assuming 192.168.130.1 is one of the iSCSI target addresses. This
> program contacts the storage on that ip address and asks for the rest
> and squirrels them away in /var/lib/iscsi/
> g
> I had to go in there and remove the two iSCSI ports that I don't
> actually have connected. iscsiadm found and mapped all 4 iSCSI ports,
> not just the two I'm using. I've got the iSCSI running over a pair of
> dedicated interfaces with the Ethernet directly connected back-to- 
> back.

Ok, I did that and it showed all 4,  I have connectivity to all 4 so  
I'll leave them in there.

> The restart the iSCSI transport daemons. I can't remember if it iscsid
> or iSCSI or both scripts.
>
> The smagent should find a SCSI device it likes after this.

It didn't, I was getting that error message.

I had to do iscsiadm -mode node --target [...]  --login first

Do I have to log in to all 4 of the targets returned by discovery in  
order to get full speed/redundancy?

Do I need to put the login stuff in /etc/rc.d/rc.local in order for it  
to survive a reboot?

I have MPP installed, how do I make sure that is configured correctly

> I still have one remaining issue - I can't get the iSCSI luns auto
> mounted at boot. I've added '_netdev' to the fstab, trying to delay  
> the
> mount of the file systems till after network ports are up, but this
> hasn't helped. So far all that's worked is putting a mount -a' in
> /etc/rc.local which is the last thing to run as the system boots. This
> doesn't seem right some how.

I'm going to be running Xen on the blades and using the iSCSI LUNs for  
the xen guest machines.  I don't want the host machine mounting the  
filesystems.  I do want to have all 20 LUNs available to all 10 blades  
so that any one blade can run any xen instance.  What is the method to  
insure that a LUN ends up on the same /dev/sdX device on every boot?

> I've got 5Tb of disk, I did successfully use LVM to join 3 LUNs
> together and make a single 5Tb file system, although in the end I will
> stick with the 2TB MD3000i limit and run multiple 2TB file systems on
> the raw LUNs. (For backup reasons I can't make the file systems too
> large...)

I have 15  300Gb 15k SAS drives in my MD3000i.  I don't need anywhere  
near that amount of space.  My plan is to setup 2 RAID 5s (0-6, 7-13)  
and use drive 14 as a hot-spare.  The first RAID will be my working  
data set, the second RAID will hold replication data & backups.  At  
least until I can afford a separate drive array for backups.

-Matt