[ovirt-users] small cluster

[Jeremiah Jahn]

you probably shouldn't snapshot your data directory though, as
postgres pretty much does it's own thing as far as journalling and
what not. It also has it's own snapshotting capabilities.  again just
an FYI and chances are bringing  back a db from a snapshot could cause
all sorts of problems with your PG-XC cluster, esp in a mutli-master
configuration. Generally we just drop the db, and resync from the
beginning, but we use slony, and not XC, so your milage may vary.
I've got nothing on the remote start though :-)

On Thu, Jul 17, 2014 at 10:23 AM, Jeremiah Jahn
<jeremiah at goodinassociates.com> wrote:
> you can use puppet to remotely reboot the machine.. FYI
>
> http://stackoverflow.com/questions/8528392/remotely-shutdown-reboot-linux-boxes-without-ssh
>
> On Thu, Jul 17, 2014 at 10:17 AM, urthmover <urthmover at gmail.com> wrote:
>> Thank you for the insight and advice Jeremiah.
>>
>> I think that creating a base image on one machine and using puppet/scripting
>> to create the others is a great idea (I’ve been using clonezilla and
>> manually editing IP/hostname afterwards).
>>
>> As for not using overt, the aspects that I’d be missing are remotely
>> rebooting a hung machine.  Also, taking a snapshot of the entire guest is
>> relatively easy/without reboot/started remotely.
>>
>> As for “Why Mini’s” we have a bunch of extra machines on hand and off lease.
>>
>> If anyone else has other thoughts and suggestions, please share.
>>
>> Thanks,
>>
>> --
>> urthmover
>>
>> On July 17, 2014 at 9:04:12 AM, Jeremiah Jahn
>> (jeremiah at goodinassociates.com) wrote:
>>
>> Just a couple of thoughts for you. You are correct. Gluster would not
>> be a happy thing for for a DB. But for that matter, no network file
>> system would be good for postgres or any DB. Your SSD's probably max
>> out at 6Gb/s while your nics on a mini only go up to 1Gb/s. The whole
>> point of postgres-xc is that it takes care of all of the replication
>> and redundancy. Depending on your usage, your probably going to want
>> all of the 16 GB of ram for your indexes as well. I'd be very tempted
>> to make an install image from one mini and use it to add/create the
>> other nodes with puppet or just a custom script to configure the image
>> for addition to the pg-xc cluster. You're not going to gain a whole
>> lot of anything by running ovirt on your mini's except some slowdown.
>> if your servers that PG will be running on are linux, then you don't
>> really need more than 10GB for a linux install. If you're going to
>> use your mini's for other guests careful of the memory you use so you
>> don't make your dba unhappy. and using gluster on the exported 100G
>> partition would only net you about 500G for a storage domain if you've
>> got gluster replication going, which is not a bad idea. And finally,
>> why mac mini's? pretty pricey for server hardware unless your planning
>> on using them to host osx guests, which I'm not sure can actually be
>> done with anything but vmware, which is even a hack, at the moment.
>>
>> just my 2 cents as a person who runs gluster, ovirt, and a postgres cluster.
>>
>> On Wed, Jul 16, 2014 at 2:50 PM, urthmover <urthmover at gmail.com> wrote:
>>> After further investigation and reading. Glusterfs is not really designed
>>> for database operations. So I am retracting one question and curious about
>>> anyone’s thoughts regrading the new set of questions.
>>>
>>> Should I partition the mirrored pair into two slices (100GB and 800GB).
>>> Then present ovirt with 10 storage domains each being a 100G partition for
>>> the OS of each guest. Then use nfs for the 800G /data partitions (not
>>> using
>>> /data as a storage domain within overt, just as plain old nfs mounts hard
>>> coded to each guest machine)
>>>
>>> Should I present each mac mini’s mirrored pair as an nfs share to
>>> ovirt-engine? This would create 10 1TB storage domains. Then create 10
>>> large 800GB /data partitions (a /data for each guest).
>>>
>>> Should I NOT use ovirt and just run each mac mini as a mirrored pair of
>>> disks and a standalone server?
>>>
>>> --
>>> urthmover
>>>
>>> On July 16, 2014 at 12:12:16 PM, urthmover (urthmover at gmail.com) wrote:
>>>
>>> I have 10 mac minis at my disposal. (Currently, I’m booting each device
>>> using a centos6.5 usbstick leaving the 2 disks free for use)
>>>
>>> GOAL:
>>> To build a cluster of 10 servers running postgres-xc
>>>
>>> EQUIPMENT:
>>> 10 mac mini: i7-3720QM at 2.60GHz/16G RAM/1x1gbit NIC/2x1TB SSDs (zfs
>>> mirrored)
>>>
>>> REQUEST:
>>> Please run the software application postgres-xc (a multi-master version of
>>> postgres). I'm told by the DBA that disk IO is the most important factor
>>> for the tasks that he’ll be running. The DBA wants 10 servers each with a
>>> 50G OS partition and a 800GB /data.
>>>
>>> THOUGHTS:
>>> I have a few ideas for how to accomplish but I'm unsure which is the best
>>> balance between disk IO and overall IT management of the environment.
>>>
>>> QUESTIONS FOR THE LIST:
>>> Should I present each of the 10 mac mini’s mirrored disks to glusterfs
>>> thus
>>> creating a large 10TB storage area. Then connect the storage area to
>>> ovirt-engine creating on 10TB storage domain, and use it as the storage
>>> domain for 10 large 800GB disks (a /data for each guest) ?
>>>
>>> Should I present each mac mini’s mirrored pair as an nfs share to
>>> ovirt-engine? This would create 10 1TB storage domains. Then create 10
>>> large 800GB /data partitions (a /data for each guest).
>>>
>>> Should I NOT use ovirt and just run each mac mini as a mirrored pair of
>>> disks and a sandalone server?
>>>
>>> LASTLY:
>>> I’m open to any other thoughts or ideas for how to best accomplish this
>>> task.
>>>
>>> Thanks in advance,
>>>
>>> --
>>> urthmover
>>>
>>>
>>> _______________________________________________
>>> Users mailing list
>>> Users at ovirt.org
>>> http://lists.ovirt.org/mailman/listinfo/users
>>>