09-12-2013 04:34 AM
Hello,
We have a dg which should be site consistent but it is not. Setting up the dg as site consistent (disk tagging, sites added and siteconsistent flag enabled), how to make a volume site consistent?
In the storage foundation cfs and high avalaibility the page 539 says:
8 Turn on the allsites flag for the volume which requires data replication to each site:
But I think it is not correct as you first should mirror the volume and create the dco log:
# vxvol -g test_dg set allsites=on no_mirror1_lv
Solved! Go to Solution.
09-12-2013 06:45 AM
I find there is a lot of confusion about what site consistency provides, so this is my understanding:
Suppose you have 2 volumes, with a single Oracle Database on and both volumes are mirrored across sites , Oracle is running at site1 and site1 looses connection to just ONE of the plexes at site2.
WITHOUT site consistency - this is what would happen:
The plex at site2 is detached and Oracle on site1 continues to write data to both plexes at site1, but only one plex at site2. Site1 fails, so you then switch to site2, and site2 can see the local plex for each volume, but the data is useless as one volume has up-to-date data and the other does not.
WITH site consistency - this is what would happen:
ALL plexes at site2 which have siteconsistent=on are detached and Oracle on site1 continues to write data to both plexes at site1, but the one visable plex at site2 IS not written to as all plexes were detached. Site1 fails, so you then switch to site2, and site2 can see the local plex for each volume, and although the data is old, it is consistent so it can be used.
So if you are configuring site consistency, the assumption is that your volumes are already mirrored and configuring site consistency just ensures a site remains consistent so that some plexes don't have newer data then others on the same site.
Setting allsites=on for a volume means if you try to add a new site to the diskgroup, this will only work if the volumes with allsites=on have a plex on site that is being added.
Mike
09-12-2013 06:45 AM
I find there is a lot of confusion about what site consistency provides, so this is my understanding:
Suppose you have 2 volumes, with a single Oracle Database on and both volumes are mirrored across sites , Oracle is running at site1 and site1 looses connection to just ONE of the plexes at site2.
WITHOUT site consistency - this is what would happen:
The plex at site2 is detached and Oracle on site1 continues to write data to both plexes at site1, but only one plex at site2. Site1 fails, so you then switch to site2, and site2 can see the local plex for each volume, but the data is useless as one volume has up-to-date data and the other does not.
WITH site consistency - this is what would happen:
ALL plexes at site2 which have siteconsistent=on are detached and Oracle on site1 continues to write data to both plexes at site1, but the one visable plex at site2 IS not written to as all plexes were detached. Site1 fails, so you then switch to site2, and site2 can see the local plex for each volume, and although the data is old, it is consistent so it can be used.
So if you are configuring site consistency, the assumption is that your volumes are already mirrored and configuring site consistency just ensures a site remains consistent so that some plexes don't have newer data then others on the same site.
Setting allsites=on for a volume means if you try to add a new site to the diskgroup, this will only work if the volumes with allsites=on have a plex on site that is being added.
Mike
09-12-2013 09:23 AM
Hi,
The guide for setting up site aware storage is in the Cluter File System Administrator's Guide for your OS (Linux is below).
Please refer to page 532 for the setup and configuration.
Doug Snyder, CISSP
All documents - all products, all OS types:
09-13-2013 03:12 AM
OK, I see now the meaning of site consistent and allsites. So, the steps I follow are correct in order to convert a volume into a mirror volume with site consistency, right??
Would you change any step???
Regards,
joagmv
09-13-2013 03:44 AM
There are some steps missing - you mentioned steps 8 and 9 in your opening post, but you don't have all of the other 7 on page 539 (like step 6 for instance):
1 Ensure that the disk group is updated to at least version 140, by running thevxdg upgrade command on it:# vxdg upgrade diskgroup2 On each host that can access the disk group, define the site name:# vxdctl set site=sitename3 Tag all the disks in the disk group with the appropriate site name:# vxdisk [-g diskgroup] settag site=sitename disk1 disk2Or, to tag all the disks in a specified enclosure, use the following command:# vxdisk [-g diskgroup] settag site=sitename encl:encl_name4 Use the vxdg move command to move any unsupported RAID-5 volumes toanother disk group. Alternatively, use the vxassist convert commands toconvert the volumes to a supported layout such as mirror or mirror-stripe.You can use the site and mirror=site storage allocation attribute to ensurethat the plexes are created on the correct storage.5 Use the vxevac command to ensure that the volumes have at least one plexat each site. You can use the site and mirror=site storage allocation attributeto ensure that the plexes are created on the correct storage.6 Register a site record for each site with the disk group:# vxdg -g diskgroup addsite sitename7 Turn on site consistency for the disk group:# vxdg -g diskgroup set siteconsistent=on
09-13-2013 04:04 AM
I already implemented these steps before starting to create the mirror of the volume although I didn't pasted here.
Doubt resolved! Thanks!
Regards,
joagmv