SLES10 and iSCSI - numerous documentation errors and system incons 2007-03-07 - By Alexei_Roudnev
Back
SLES10 uses new open-iscsi. Aside of the fact, that it is new and should not
be used in servers before we used it on desktops for 2 - 3 years,
it caused a numerous troubles during setting up things:
1) There are 2 documents, 1 here
http://en.opensuse.org/Open-iSCSI_and_SUSE_Linux
and other here
http://www.novell.com/documentation/sles10/index.html?page=/documentation
/sles10/sles_admin/data/cha_planning.html
No one is 100% correct (but first have more relation to the reality vs
second - if you follow it, without reinstalling iscsi,
you have a chance to win).
Novell document is wrong in many places:
- says nothing about replacing _netdev by hotplug (!)
- send us to /var/lib/open-iscsi while it is /var/lib/open-iscsi
- don't say anything _how to mount_ or _how to use multipath_.
2) Open-iscsi is !@#$. Comparing with iSCSI in SLES9, it lost ability to use
iSCSI names of devices, don't support multi port
configuration, don't allow resizing lun's on the fly, don't have iscsi
mounting (so you must play with hotplug option) and so on.
iSCSI management in SLES10 is few times more difficult vs SLES9 and SuSe9.
3) Ok, now we proceed. yast2 -> iscsi client is tricky - you add discoveries
targets, then they disappear,
then they do not want to connect and so on. I had 2 netApp filers , 2 ports
on each, and 4 LUN's; in SLES9,
configuration took seconds; on SLES10, it require many steps and 3 rounds
with yast, or require about 10 - 20 iscsiadm commands. It is not a surprise
that most documents describe manual configuration.
4) Now you have not a choice except using multipath. And what if I want to
use eth1 by default and eth0 as a backup (only)?
multipath daemon don't understand such configurations (comparing to old
iscsi system). Microsoft added multiport access to the new iscsi, SuSe
removed it... what the humor!...
5) Good news is that multipath understands iscsi disks by default (if
started). Unfortunately LVM can't see this devices when they comes, so
you must start lvm manually (add it) and then play with numerous
dependencies and (again) with hotplug option in fstab.
And you never wil be able to determine , which disk is which - because
multipath eliminate human readable names (from disk/by-path)
and now you have only auto-generated disk id-s.
6) After all, yast2 -> LVM don't recognize pvdisks coming from multipath,
and disks themselves comes from the wrong place by default.
At the end, we are ending up with supercomplicated configuration, without
yast for iSCSI and LVM (you can't use them
in real life, too many bugs), but with _so - so_ working iSCSI . Back to
SLES8 again ?
When you manage to get it all together, it works. At least it don't fail in
first few hours and was able to run Oracle tests. Is it now a good news _new
production grade system don't fail, after you hacked scripts and
configurations to make it work?_
--
To unsubscribe, email: suse-oracle-unsubscribe@(protected)
For additional commands, email: suse-oracle-help@(protected)
Please see http://www.suse.com/oracle/ before posting
|
|
