A similar argument can be made with multipath (device mapper) although use cases for this spread far beyond just Oracle. Many vendors tout the use of their own multipath software, and while they can have some extra features, I am of the opinion that 99% of the what is required today is available in this free and integrated solution.
Multipath Setup
First off, make sure your storage array is supported by multipath. From my knowledge, all of the major vendors are, but you may have one that isn't on the list. My setup is RedHat 5.4 with an EMC CX array which is convenient as multipath has integrated support for the CLARiiON line. The key section to note from your vendor is any special device entry under /etc/multipath.conf they may need.
Next is to ensure multipath is installed. For RedHat this is in the package device-mapper-multipath, for SuSE it is multipath-tools. Do a package search for multipath and you should find it.
Configuration File
defaults {
user_friendly_names no
}
multipaths {
multipath {
wwid wwid (e.g. 36006016015a0190038585d690b53df11)
alias ora_data1
}
multipath {
wwid wwid (e.g. 36006016015a0190038585d690b53df12)
alias ora_data2
}
}
If you leave user_friendly_name to yes, Linux will create a lovely mpathThe multipath entries take this WWID and turns it into any name specified by the alias entry. Because we want to change the device permissions, a specific name or prefix is used. In this case I have chosen “ora_” but you could use whatever you like, or multiple prefixes for different functions. For example you could have a vote_ or an ocr_ prefix using different permissions although those disks aren't really required anymore with 11gR2.
You can get WWIDs from the storage array or simply let Linux tell you when it scans them, we'll cover that in a later post.
Optional Configuration
Now if you have a different storage array you may also have to add the vendor specified devices entry, perhaps something like this:
devices {
device {
vendor “NETAPP”
product “LUN”
path_group_policy multibus
getuid_callout “/sbin/scsi_id -g -u -s /block/%n”
prio_callout “/sbin/mpath_prio_ontap /dev/%n”
features "1 queue_if_no_path"
path_checker directio
failback immediate
flush_on_last_del yes
}
}
This is only an example, check with your vendor. If they support multipath, they have documentation with the correct entry for the current features and models.
You can also include a blacklist entry for your local devices but I haven't found it to be required. However, if you like an entry such as this would do:
blacklist
{
devnode "^(ram|raw|loop|fd|md|dm-|sr|scd|st)[0-9]*"
devnode "^hd[a-z]"
devnode "^cciss!c[0-9]d[0-9]*[p[0-9]*]"
}
Multipath Daemon
The last step is to get the daemon running. Its job is to reconfigure paths when something breaks or when a link is put back in service.
Under RedHat / SuSE this is relatively simple to do:
chkconfig multipathd on
You can validate this with:
chkconfig --list multipathd
And finally startup the daemon immediately with:
service multipathd start
- or -
/etc/init.d/multipathd start
Udev Configuration
Unfortunately distributions can vary greatly in how their udev rules are laid out, so here is the configuration for RedHat, your mileage may vary. I am going to store my new configuration as part of the multipath rules under /etc/udev/rules.d/40-multipath.rules. The changes from the default are highlighted:
# multipath wants the devmaps presented as meaninglful device names
# so name them after their devmap name
SUBSYSTEM!="block", GOTO="end_mpath"
KERNEL!="dm-[0-9]*", ACTION=="add", PROGRAM=="/bin/bash -c '/sbin/lsmod | /bin/grep ^dm_multipath'", RUN+="/sbin/multipath -v0 %M:%m"
KERNEL!="dm-[0-9]*", GOTO="end_mpath"
PROGRAM!="/sbin/mpath_wait %M %m", GOTO="end_mpath"
ACTION=="add", RUN+="/sbin/dmsetup ls --target multipath --exec '/sbin/kpartx -a -p p' -j %M -m %m"
PROGRAM=="/sbin/dmsetup ls --target multipath --exec /bin/basename -j %M -m %m", RESULT=="?*", NAME="%k", SYMLINK="ora_rdsk/%c", GOTO="update_oracle_devs"
PROGRAM!="/bin/bash -c '/sbin/dmsetup info -c --noheadings -j %M -m %m | /bin/grep -q .*:.*:.*:.*:.*:.*:.*:part[0-9]*-mpath-'", GOTO="end_mpath"
PROGRAM=="/sbin/dmsetup ls --target linear --exec /bin/basename -j %M -m %m", NAME="%k", RESULT=="?*", SYMLINK="ora_rdsk/%c", GOTO="update_oracle_devs"
GOTO="end_mpath"
LABEL="update_oracle_devs"
RESULT=="vote*",OWNER="oracle",GROUP="oinstall",MODE="644"
RESULT=="ocr*",GROUP="oinstall",MODE="640"
RESULT=="ora*",OWNER="oracle",GROUP="dba",MODE="660"
OPTIONS="last_rule"
LABEL="end_mpath"
The ora_rdsk entry is a way of keeping the oracle disk (multipath disk) in a unique location. The default is mpath, its up to you. This means that all of my multipath devices will show up under /dev/ora_rdsk.
GOTO=”end_mpath” is an entry which acts as a catch for devices that aren't specifically sent to update_oracle_devs which of course follows. This is where the magic happens, it basically does a match based on a certain prefix assigns an owner, a group, and permissions. You can get quite creative here if you like with different regular expressions or device entries for different functions.
Strictly speaking you don't require partitions to use ASM under Oracle RAC, however, if you would like to, they will show up as your designated alias plus a p
Putting it All Together
Just allocate your disk, put the entries into /dev/multipath.conf and you are off to the races. Your devices should all magically show up in /dev/ora_rdsk/. Next time I'll cover off how to add and remove devices dynamically so you don't have to know your WWIDs up front or do any nasty rebooting.
Hi,
ReplyDeleteDo you you get the right perms for /dev/mapper/ora_data1
too? Or will the symlink be owned as root?
99-asm.conf
ReplyDeletePROGRAM="/sbin/dmsetup info -c --noheadings -o name casm-09B6-OCR02p1" , RESULT=="casm-09B6-OCR02p1", NAME="test/casm-09B6-test1" , RUN+="/bin/chown oracle:dba /dev/mapper/casm-09B6-OCR02p1"