I had an interesting project at work today. We are looking to add 9 more Oracle instances to our current development environment. I had a 400GB volume for archive logs which was more than enough space when we first spec'ed the system. My (somewhat) dilemma was figuring out how to grow this space. We use VxVM on HP-UX 11.31 and I have a large number of 25GB RAID5 LUNS presented to the server. I then have them striped via VxVM. which essentially gives us RAID 5+0. The performance is great. But it took a some fishing around to solve. Anyway here is what you need to know.
1. You need to present extra luns from your array to the host. (There is a gotcha coming)
2. Run ioscan -C disk.
3. Run insf -C disk
4. Run vxdctl enable (rescans disks)
5. Run vxdisksetup on the disks.
6. Use vxdg with adddisk to add disks to the diskgroup.
7. Use vxassist with the relayout options to add the disks to your stripe. (Bang, your dead).
This is where I ran into issues. I got this error:
I read that a relayout would need some temporary space to do the relayout, but nothing I read stated just how much space I would need. In my case I added 4 x 25GB disks and thought that was enough (my mistake). My current stripe was accross 16 disk and I was going to stripe across 20. My command was as follows:
vxassist -g mydg relayout datavol layout=stripe ncol=20 stwidth=64k
It all seemed straight forward. So I went to my old friend Google to look for answers. It took a little while to find the correct search string. But I came across this page.
http://seer.entsupport.symantec.com/docs/248062.htm
Here is the key info
There is a nice chart below this paragraph. I found I needed to add 20% of my current space to the disk group in order for the relayout to work. It's funny, when I took the VxVM class from HP, they never mentioned this. So, I added the space and was able to do the relayout. It took about an hour to perform the relayout.
After you relayout, you need to grow your volume and extend the filesystem. And that's it. You can do all this while the system is up and running.
1. You need to present extra luns from your array to the host. (There is a gotcha coming)
2. Run ioscan -C disk.
3. Run insf -C disk
4. Run vxdctl enable (rescans disks)
5. Run vxdisksetup on the disks.
6. Use vxdg with adddisk to add disks to the diskgroup.
7. Use vxassist with the relayout options to add the disks to your stripe. (Bang, your dead).
This is where I ran into issues. I got this error:
VxVM vxassist ERROR V-5-1-6345 Cannot allocate 15728640 blocks of disk space \
required by the relayout operation for column addition
VxVM vxassist ERROR V-5-1-4037 Relayout operation aborted. (7)
I read that a relayout would need some temporary space to do the relayout, but nothing I read stated just how much space I would need. In my case I added 4 x 25GB disks and thought that was enough (my mistake). My current stripe was accross 16 disk and I was going to stripe across 20. My command was as follows:
vxassist -g mydg relayout datavol layout=stripe ncol=20 stwidth=64k
It all seemed straight forward. So I went to my old friend Google to look for answers. It took a little while to find the correct search string. But I came across this page.
http://seer.entsupport.symantec.com/docs/248062.htm
Here is the key info
Relayout operations require temporary disk space. This temporary space is allocated from the free space available in the disk group that contains the volume being relayed out ...
There is a nice chart below this paragraph. I found I needed to add 20% of my current space to the disk group in order for the relayout to work. It's funny, when I took the VxVM class from HP, they never mentioned this. So, I added the space and was able to do the relayout. It took about an hour to perform the relayout.
After you relayout, you need to grow your volume and extend the filesystem. And that's it. You can do all this while the system is up and running.
Well explained, and very comprehensive. Thanks!
ReplyDeleteWas looking for relayout explained today and this is
ReplyDeleteVERY much appreciated by a "tech head" like me.