[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-API] redundant xcp server with drbd



Well, We've choose iscsi fiiling to make it 'shareable' (I means allows XCP to think it shared).

Are you using XCP or just xen (xend)? How you made storage manager to think about /dev/drbd as shared storage?

On 06.03.2012 23:15, Denis Cardon wrote:
Hi Georges,

We've tried that for internal laboratory. Problem: sometime drbd do
additional IO and making storage out of sync. We've spend few weeks on
this set up (primary/primary drbd, and iscsi export with LVMoISCSI SR),
but it getting unstable after some time, so we gave up.

Actually I have not encountered instability in my test setup (the SR was define directly over /dev/drbd0, no iscsi in between). During the test, I was running iozone in a loop in VMs that were switch from host every few minutes in a loop (linux and windows VMs) for a whole 48 hours.

I have'nt encountered any issues during that stress test (set aside the global warming). But when I went ahead with disaster recovery scenario, it was a no go : severing the primary-primary drbd link left me with a brain split to solve by hand.

I was thinking to use the more conservative approach with the following scenario. It adds the iscsi latency to the io path, but :

  XCP1 in xenpool                  XCP2 in xenpool
==================              ====================
drbd-sr1 (primary) <----->   drbd-sr1 (secondary)
 exported to xenpool
 through iscsi target

drbd-sr2 (secondary) <----->   drbd-sr2 (primary)
                                 exported to xenpool
                                 through iscsi target

I have a few year of practice of drbd with primary/secondary setup on iscsi san, and I am pretty confident the data should be safe with this kind of setup, but I'm wondering of the behavior of a dom0 xen in this situation.

I'll give it a shot in the next few days.

Thanks for your input.

Denis








On 06.03.2012 13:47, Denis Cardon wrote:
Hello everyone,

I wanted to know if any of you had some experience with building a
redundant xcp/xenserver cluster based on drbd. In this setup I would
like to avoid a separate iscsi SAN, and have replicated local storage.

There are docs on the internet showing how to set up a dual primary
drbd SR on xenserver/xcp. It works fine with live migration and all,
but when trying disaster scenario, like severing the drbd replication
link, one gets a split brain scenario that cannot be unsolved. It
should be doable if xapi had a smart understanding of drbd, but
currently it is just a SR like others. So it is a no-go for the time
being.

I also tried a drbd primary-secondary scenario, with re-export with
iet iscsi target, but for the moment I have unresolved performance
issues (will try also with scst target).

So I wanted to know if any of you had any experience or hints with
this kind of setup : a dual active/active XCP servers with local SR
replicated with primary-secondary DRBD, re-exported with iscsi or aoe
or whatever.

Cheers,

Denis Cardon



_______________________________________________
xen-api mailing list
xen-api@xxxxxxxxxxxxx
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api



_______________________________________________
xen-api mailing list
xen-api@xxxxxxxxxxxxx
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.