Hi all, First of all I''m just using the plain vanilla CentOS included Xen 3.1 to do my work. I have 3 machines hooked up to a private network which provides access to 1 Gbit networking and access to an Equallogic SAN providing iSCSI. I''ve succesfully setup Xen to use this iSCSI device with LVM and can happily run and migrate between machines. However when setting up cLVM I just get bogged down with Redhat''s clustering so I''ve taken a look at what cLVM provides and decided not to bother with it. Especially since all it does is keep metadata consistent and allow changes to be made from each node (it does not lock the LV''s) This mirrors the setups I''ve seen in books and some articles. Basically what I intend to do: -Have all the *.cfg''s on a Rysnched folder -Designate a master node from where all LVM actions are done -One each LVM change, call a script on each slave to refresh the LVM metadata. My only concern is that I''m wondering if LVM actually pre-allocates the PE (Physical Extents) or allocates them as required. The last thing would be bad (tm) as it would mean that data could get randomly overwritten. I''ve been running a bunch of machines in test and have noticed no problems. Could somebody else share their experiences? Thank you for any information you can share! PS: I have considered the commercial version of Xen many times, but I''d rather not start retooling at this point as everything is working nicely -- Barry van Someren --------------------------------------- LinkedIn: http://www.linkedin.com/in/barryvansomeren Skype: BvsomerenSprout Blog: http://blog.bvansomeren.com KvK: 27317624 irc: BarryNL @ FreeNode _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Sun, May 3, 2009 at 5:32 PM, Barry van Someren <barry@coffeesprout.com> wrote:> However when setting up cLVM I just get bogged down with Redhat''s > clustering so I''ve taken a look at what cLVM provides and decided not > to bother with it.If you use iscsi anyway, why not give each domU its own LUN and dump LVM altogether? I believe there''s an unofficial block-iscsi script that makes setting up domU easier http://lists.xensource.com/archives/html/xen-devel/2007-11/msg00782.html> My only concern is that I''m wondering if LVM actually pre-allocates > the PE (Physical Extents) or allocates them as required. > The last thing would be bad (tm) as it would mean that data could get > randomly overwritten.The first one. Regards, Fajar _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi, Unfortunately I don''t have that level of control over the iSCSI partitioning. (I buy the storage per LUN and I don''t think the provider will like it if I buy 100''s of 10-100 GB LUNS) Otherwise it would definitely be best as you would just devote a whole "hd" to a VM. On Sun, May 3, 2009 at 3:39 PM, Fajar A. Nugraha <fajar@fajar.net> wrote:> On Sun, May 3, 2009 at 5:32 PM, Barry van Someren > <barry@coffeesprout.com> wrote: >> However when setting up cLVM I just get bogged down with Redhat''s >> clustering so I''ve taken a look at what cLVM provides and decided not >> to bother with it. > > If you use iscsi anyway, why not give each domU its own LUN and dump > LVM altogether? > I believe there''s an unofficial block-iscsi script that makes setting > up domU easier > > http://lists.xensource.com/archives/html/xen-devel/2007-11/msg00782.html > >> My only concern is that I''m wondering if LVM actually pre-allocates >> the PE (Physical Extents) or allocates them as required. >> The last thing would be bad (tm) as it would mean that data could get >> randomly overwritten. > > The first one. > > Regards, > > Fajar > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >-- Barry van Someren --------------------------------------- LinkedIn: http://www.linkedin.com/in/barryvansomeren Skype: BvsomerenSprout Blog: http://blog.bvansomeren.com KvK: 27317624 irc: BarryNL @ FreeNode _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Sun, May 3, 2009 at 5:32 AM, Barry van Someren <barry@coffeesprout.com> wrote:> My only concern is that I''m wondering if LVM actually pre-allocates > the PE (Physical Extents) or allocates them as required. > The last thing would be bad (tm) as it would mean that data could get > randomly overwritten.PEs are statically allocated at lvcreate/lvextend time. -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Sun, May 3, 2009 at 5:32 AM, Barry van Someren <barry@coffeesprout.com> wrote:> Basically what I intend to do: > -Have all the *.cfg''s on a Rysnched folder > -Designate a master node from where all LVM actions are done > -One each LVM change, call a script on each slave to refresh the LVM metadata.to be a bit more on the safe side, check the dmsetup tool, especially do a `dmsetup suspend/resume` around the ''change/reload'' process. even better, use some form of locks to make sure non-master nodes are io-suspended before doing the change... soon you''ll have a full cLVM replacement. -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hey, Thanks for that. Actually clvm is less thorough as it only locks the LVM metadata and not the actual access to the device. My beef is not with clvm, it''s good for what it does. The problem is that I can''t get the entire boatload of cluster extra''s that come with CentOS to work reliably (probably operator error) I''m running tests on this right now and it''s all looking good especially since all the changes are done in 1 place and I''ve implemented a queue to make sure that only 1 lvm command can be active at any time. Thank you for the suggestion.. I''m going to roll with it and see how it works. Regards, Barry On Mon, May 4, 2009 at 4:30 PM, Javier Guerra <javier@guerrag.com> wrote:> On Sun, May 3, 2009 at 5:32 AM, Barry van Someren > <barry@coffeesprout.com> wrote: >> Basically what I intend to do: >> -Have all the *.cfg''s on a Rysnched folder >> -Designate a master node from where all LVM actions are done >> -One each LVM change, call a script on each slave to refresh the LVM metadata. > > to be a bit more on the safe side, check the dmsetup tool, especially > do a `dmsetup suspend/resume` around the ''change/reload'' process. > even better, use some form of locks to make sure non-master nodes are > io-suspended before doing the change... soon you''ll have a full cLVM > replacement. > > > -- > Javier >-- Barry van Someren --------------------------------------- LinkedIn: http://www.linkedin.com/in/barryvansomeren Skype: BvsomerenSprout Blog: http://blog.bvansomeren.com KvK: 27317624 irc: BarryNL @ FreeNode _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users