Thursday, March 16, 2006

Veritas Volume Manager Cheat Sheet


Veritas Volume Manager Cheat Sheet:

The Setup: an Oracle database supporting an application called ABC,
and given the following:

Assuming that we need to create filesystems intended for Oracle and:

we are using a SID of 'ABC'

we have nine disks, c0t0d1 through c0t0d9

we will be creating six volumes: /u01/oradata/ABC ( 4gb ),
/u02/oradata/ABC (4gb ), /u03/oradata/ABC (8gb ), /u04/oradata/ABC (
4gb ), /u01/app/oracle/product/8.0.5_ABC ( 4gb ), and /u01/admin/ABC (
4gb ).

We added these disks to Volume Manager Control during installation
with vxinstall, or later with vxdiskadd(1m). Both vxinstall and
vxdiskadd can add either initialize a previously unused disk, or
'encapsulate' a disk already containing data, but not under Volume
Manager control.

we could take the following steps to set up our filesystems.
Adding an EMC disk to Veritas

Use the inq utility to see if you have SCSI visibility to the disk

Use vxdisk -o alldgs list to see if Veritas has seen the disk

if not, try the following, and then repeat the previous step:


vxdctl initdmp

vxdctl enable

Creating a Disk Group in Volume Manager:

Create a vxvm disk group for the Oracle SID using vxdg(1m). It's
important that all of the disks/filesystems necessary for the ABC
database to run are included in the disk group, otherwise the database
could not be moved to another system.

The easiest way to do this is to use vxdiskadd
vxdiskadd c0t0d0 c0t0d1 c0t0d2 . . .

Doing it the hard way:
vxdg init ABCdg ABCdg01=c0t0d1

This would create a disk group called ABCdg containing one physical
disk, c0t0d1, which will be referred to by it's name within vxvm,
ABCdg01. You cannot initiate a disk group without specifying at least
one disk as a member of the group. You should not include any disks
destined to be part of an Oracle Database in the default vxvm group,
rootdg. This allows you to use the vxdg 'deport' and 'import' commands
to migrate an entire vxvm disk group to another host.

Now that we've create the disk group for our database, we can add the
rest of our disks to it:

vxdg -g ABCdg adddisk ABCdg02=c0t0d2

vxdg -g ABCdg adddisk ABCdg09=c0t0d9

Creating Volumes With vxassist

vxassist(1m) is a vxvm command that acts as a front-end to other vxvm
commands, much as newfs(1m) acts as a simpler-to-use front-end to
mkfs(1m). Using vxassist with it's -v flag will display the actual
vxvm commands used during volume construction. vxassist also
front-ends vxfs commands and normal UNIX utilities during volume
construction. An example of this is using a single vxassist command to
set up a volume. Using vxassist simplifies many vxvm tasks, but you
should pay careful attention to the defaults you will inherit when
setting up volumes using this tool. Understanding the contents of the
/etc/default/vxassist file is important, especially when striping

To make our 8gb /u03/oradata/ABC volume, we could use the following
command line:

vxassist -g ABCdg make ABC01 8g

Breaking it down into it's component parts:


-g ABCdg # this volume will belong to the ABCdg disk group

make # tells vxassist that we are creating a new volume

ABC01 # name the volume

8g # specify the size of the volume

Here we create the other five volumes:

vxassist -g ABCdg make ABC02 8g

vxassist -g ABCdg make ABC03 8g

vxassist -g ABCdg make ABC04 8g

vxassist -g ABCdg make ABC05 8g

vxassist -g ABCdg make ABC06 8g

Creating a File System:

We now lay filesystems down on our newly created volumes. We assume
here that you've agreed with the Oracle DBA's to use a block size of
8192 for our filesystems. Not specifying a block size allow Veritas to
dynamically assign one for you based upon the size of the volume. Our
volumes would default to 1024 for a block size on our 4 GB volumes,
and 2048 for our 8GB volume. We will specify the largefiles option,
because the mkfs_vxfs(1m) command uses nolargefiles as a default,
unlike mkfs_ufs(1m) on Solaris 2.6 or greater, which uses largefiles
as a default. If we forgot to do so, we could use fsadm (
fsadm_vxfs(1m) ) to set the largefiles bit on. fsadm can also query a
mounted filesystem for the current setting of this flag. Be certain
that whatever flags you set during file system creation or
modification are reflect in vfstab, or the resulting file system may
be un-mount-able.

mkfs -F vxfs -o bsize=8192,largefiles /dev/vx/rdsk/ABCdg/ABC01

Mounting a File System:

Creating and mounting filesystems go hand in hand. You must be certain
that any special flags set at creation time are reflected in
/etc/vfstab, or on the command line if there is no entry in vfstab for
your filesystem.

mount -F vxfs -o largefiles /dev/vx/dsk/ABCdg/ABC01 /u03/oradata/ABC

Resizing a mounted volume and file system:

/etc/vx/bin/vxresize -g rootdg -F vxfs archive 16g

Recovering disks under VM control:

(if you want to live dangerously, try looking at


* Do a vxprint -g mydg first to see what disks your volumes lie on!
* Do a vxdisk list and save it
* Do an inq -et and save it!
* You may wish to review moving a volume off of a particluar disk

1) comment the volume(s) you're destroying out of vfstab

2) umount the volume

3) stop it:
vxvol stop volume01

4) remove it:
vxedit -rf rm volume01

5) remove the disk(s) the volume was on from the disk group ( up to
the last disk )
vxdg -g mydg rmdisk dmdiskname01

6) Remove the the disk(s) from VXVM control
vxdisk rm c1t1d51s2

7) If you wish do get rid of the disk group ( in order to recover the
last disk ), you must destroy the disk group:
vxdg destroy mydg

8) remove the last disk
vxdisk rm c2t7d9s2

9) If you're removing the disks from the EMC port, be sure to clean up:
vxdctl enable
vxdctl initdmp

10) Freak out and realize you whacked the wrong thing, use /root/eotw
to recover.

Renaming a VM Disk:

Since names like "ABCdg10" are not as descriptive as they could be,
you can rename it to something more useful.

vxedit rename ABCdg10 ABCdg_hot_spare

Using long disk names can be more descriptive, but will make using
vxva trickier, as it will truncate your disk name somewhat if it is
too long.

Moving a Disk Group to Another System:

1) umount the disks:

umount /u01/app/oracle/product/8.0.5_ABC

umount /u01/admin/ABC

umount /u01/oradata/ABC

umount /u02/oradata/ABC

umount /u03/oradata/ABC

umount /u04/oradata/ABC

2) Stop the volumes:

vxvol -g ABCdg stopall

3) deport the disk group from one system:

vxdg deport ABCdg

4) import them on another system:

vxdg import ABCdg

5) Start all of the volumes on the new system and resync mirrors in
the background:

vxrecover -g ABCdg -sb

6) mount the filesystems

mount /u01/app/oracle/product/8.0.5_ABC

mount /u01/admin/ABC

mount /u01/oradata/ABC

mount /u02/oradata/ABC

mount /u03/oradata/ABC

mount /u04/oradata/ABC

This procedure assumes that the second system can see the disks, and
also that you have commented the file systems out of vfstab on the old

Rename a Disk Group

dfwns19 # df -k | grep reports
/dev/vx/dsk/reportsdg/ncr01 /u01/app/oracle/admin/REPORTS
/dev/vx/dsk/reportsdg/ncr02 /u01/app/oracle/product/8.0.5REPORTS
/dev/vx/dsk/reportsdg/ncr03 /u01/oradata/REPORTS
/dev/vx/dsk/reportsdg/ncr04 /u02/oradata/REPORTS
/dev/vx/dsk/reportsdg/ncr05 /u03/oradata/REPORTS
/dev/vx/dsk/reportsdg/ncr06 /u04/oradata/REPORTS
/dev/vx/dsk/reportsdg/ncr07 /u11/oraarch/REPORTS

dfwns19 # umount /u01/app/oracle/admin/REPORTS
dfwns19 # umount /u01/app/oracle/product/8.0.5REPORTS
dfwns19 # umount /u01/oradata/REPORTS
dfwns19 # umount /u02/oradata/REPORTS
dfwns19 # umount /u03/oradata/REPORTS
dfwns19 # umount /u04/oradata/REPORTS
dfwns19 # umount /u11/oraarch/REPORTS

dfwns19 # vxvol -g reportsdg stopall

dfwns19 # vxdg deport reportsdg

dfwns19 # vxdg -n nclproddg import reportsdg

dfwns19 # vxdg list
Mirroring and then breaking mirrors

# make an sd from your dm ( use the -g, or it won't work. The book is wrong )
# sdname dmname,start_offset,sd_length
vxmake -g dwproddg sd dwprodd15-01 dwprodd15,0,8389440

# create a plex from your sd
vxmake -g dwproddg plex dw01-02 sd=dwprodd15-01

# attach the plex to your volume
vxplex -g dwproddg att dw01 dw01-02

# watch it silver ( check out the pause and slow options. badass )
dfwora03 # vxtask -l list
Task: 162 RUNNING
Operation: PLXATT Vol dw01 0.0
Started: Tue Dec 05 03:47:29 2000
Throttle: 0
Progress: 18.33% 1537352 of 8388608 Blocks
Work time: 1 minute, 26 seconds (06:23 remaining)

# more than one sd per plex for plexen > one dm
vxmake -g dwproddg sd dwprodd17-01 dwprodd17,0,8833920
vxmake -g dwproddg sd dwprodd18-01 dwprodd18,0,7944000
vxmake -g dwproddg plex dw03-02 sd=dwprodd17-01,dwprodd18-01
vxplex -g dwproddg att dw03 dw03-02

# now for the scary part. dis-associating the plexes and sd's
vxplex -o rm dis dw01-01
# zap the dm from the dg
vxdg -g dwproddg rmdisk dwprodd01
# nuke the da
vxdisk rm c2t2d13s2

nclproddg enabled 949356971.2501.dfwns19

Renaming a disk group and volume

bansecdbdg -> secproddg used as an example

# Unmount all volumes in the bansecdbdg disk group.
# Stop all volumes
vxvol -g bansecdbdg stopall

# Deport bansecdbdg
vxdg deport bansecdbdg

# Import bansecdbdg using the -n option
vxdg -n secproddg import bansecdbdg

# Recover the disk group by the new dg name
vxrecover -g secproddg -sb

# Rename each volume
vxedit -g secproddg rename bansecdb01 sec01
vxedit -g secproddg rename bansecdb02 sec02
vxedit -g secproddg rename bansecdb08 sec08

# Update vfstab, your mount points and remount the volumes

Resizing a disk, while specifying the disk

# Resize volume u11, and only use disk dev20dg23
# x requires the volume to grow (i.e. does not let you specify a
# smaller than existing volume size), and b backgrounds the
# task (according to the tech)
./vxresize -F vxfs -bx -g dev20dg u11 16g dev20dg23

Moving a volume off of a particluar disk

# Move u11 off of dm dev20dg20
vxassist -g dev20dg move u11 !dev20dg20

No comments: