Linux SAN Multipathing

There are a lot of SAN multipathing solutions on Linux at the moment. Two of them are discussesed in this blog. The first one is device mapper multipathing that is a failover and load balancing solution with a lot of configuration options. The second one (mdadm multipathing) is just a failover solution with manuel re-anable of a failed path. The advantage of mdadm multiphating is that it is very easy to configure.

Before using a multipathing solution for a production environment on Linux it is also important to determine if the used solution is supportet with the used Hardware. For example HP doesn’t support the Device Mapper Multipathing solution on their servers yet.

Device Mapper Multipathing

Procedure for configuring the system with DM-Multipath:

1. Install device-mapper-multipath rpm
2. Edit the multipath.conf configuration file:
   • comment out the default blacklist
   • change any of the existing defaults as needed
3. Start the multipath daemons
4. Create the multipath device with the multipath

Install Device Mapper Multipath

# rpm -ivh device-mapper-multipath-0.4.7-8.el5.i386.rpm
warning: device-mapper-multipath-0.4.7-8.el5.i386.rpm: Header V3 DSA signature:
Preparing...                ########################################### [100%]
1:device-mapper-multipath########################################### [100%]

Initial Configuration

Set user_friendly_name. The devices will be created as /dev/mapper/mpath[n]. Uncomment the blacklist.

# vim /etc/multipath.conf

#blacklist {
#        devnode "*"
#}

defaults {
user_friendly_names yes
path_grouping_policy multibus

}

Load the needed modul and the startup service.

# modprobe dm-multipath
# /etc/init.d/multipathd start
# chkconfig multipathd on

Print out the multipathed device.

# multipath -v2
or
# multipath -v3

Configuration

Configure device type in config file.

# cat /sys/block/sda/device/vendor
HP

# cat /sys/block/sda/device/model
HSV200

# vim /etc/multipath.conf
devices {

device {
vendor                  "HP"
product                 "HSV200"
path_grouping_policy    multibus
no_path_retry           "5"
}
}

Configure multipath device in config file.

# cat /var/lib/multipath/bindings

# Format:
# alias wwid
#
mpath0 3600508b400070aac0000900000080000

# vim /etc/multipath.conf

multipaths {

multipath {
wwid                    3600508b400070aac0000900000080000
alias                   mpath0
path_grouping_policy    multibus
path_checker            readsector0
path_selector           "round-robin 0"
failback                "5"
rr_weight               priorities
no_path_retry           "5"
}
}

Set not mutipathed devices on the blacklist. (f.e. local Raid-Devices, Volume Groups)

# vim /etc/multipath.conf

devnode_blacklist {

devnode "^cciss!c[0-9]d[0-9]*"
devnode "^vg*"
}

Show Configured Multipaths.

# dmsetup ls --target=multipath
mpath0  (253, 1)

# multipath -ll

mpath0 (3600508b400070aac0000900000080000) dm-1 HP,HSV200
[size=10G][features=1 queue_if_no_path][hwhandler=0]
\_ round-robin 0 [prio=4][active]
\_ 0:0:0:1 sda 8:0   [active][ready]
\_ 0:0:1:1 sdb 8:16  [active][ready]
\_ 1:0:0:1 sdc 8:32  [active][ready]
\_ 1:0:1:1 sdd 8:48  [active][ready]

Format and mount Device

Fdisk cannot be used with /dev/mapper/[dev_name] devices. Use fdisk on the underlying disks and execute the following command when device-mapper multipath maps the device to create a /dev/mapper/mpath[n] device for the partition.

# fdisk /dev/sda

# kpartx -a /dev/mapper/mpath0

# ls /dev/mapper/*
mpath0  mpath0p1

# mkfs.ext3 /dev/mapper/mpath0p1

# mount /dev/mapper/mpath0p1 /mnt/san

After that /dev/mapper/mpath0p1 is the first partition on the multipathed device.

Multipathing with mdadm on Linux

The md multipathing solution is only a failover solution what means that only one path is used at one time and no load balancing is made.
Start the MD Multipathing Service

# chkconfig mdmpd on

# /etc/init.d/mdmpd start

On the first Node (if it is a shared device)
Make Label on Disk

# fdisk /dev/sda
Disk /dev/sdt: 42.9 GB, 42949672960 bytes
64 heads, 32 sectors/track, 40960 cylinders
Units = cylinders of 2048 * 512 = 1048576 bytes

Device Boot      Start         End      Blocks   Id  System
/dev/sdt1               1       40960    41943024   fd  Linux raid autodetect

# partprobe

Bind multiple paths together

# mdadm --create /dev/md4 --level=multipath --raid-devices=4 /dev/sdq1 /dev/sdr1 /dev/sds1 /dev/sdt1

Get UUID

# mdadm --detail /dev/md4
UUID : b13031b5:64c5868f:1e68b273:cb36724e

Set md configuration in config file

# vim /etc/mdadm.conf

# Multiple Paths to RAC SAN
DEVICE /dev/sd[qrst]1
ARRAY /dev/md4 uuid=b13031b5:64c5868f:1e68b273:cb36724e

# cat /proc/mdstat

On the second Node (Copy the /etc/mdadm.conf from the first node)

# mdadm -As

# cat /proc/mdstat

Restore a failed path

# mdadm /dev/md1 -f /dev/sdt1 -r /dev/sdt1 -a /dev/sdt1

ASM Disk not shown in Oracle Universal Installer (OUI) or DBCA

Terms:

Operating System: Enterprise Linux 4 U5 (RHEL4 U5)

Oracle: 10.2.0.1

Problem:

While installing the ASM Instance with Oracle Universal Installer the ASM Disk, created with oracleasm createdisk, is not shown.

Solution:

Define the Scanorder in /etc/sysconfig/oracleasm config file. For example, if the used multipathing device is /dev/md1, you have to force the ASMlib to scan the /dev/md* paths before the /dev/sd* paths.

# vim /etc/sysconfig/oracleasm
# ORACLEASM_SCANORDER: Matching patterns to order disk scanning
ORACLEASM_SCANORDER="md sd"

Also make sure that the needed packages are installed for using ASM with ASMlib.

Make sure that the needed packages are installed.

• oracleasmlib-2.0 – the ASM libraries
• oracleasm-support-2.0 – utilities needed to administer ASMLib
• oracleasm – a kernel module for the ASM library

More Inofs:

Metalink Note:394956.1

Grid Control Error: Agent unreachable

Terms:

Operating System: Enterprise Linux 4 U5 (RHEL4 U5)

Oracle: 10.2.0.2

Problem:

Communication from the Oracle Management Service host to the Agent host failed. Agent crashes after a few hour of correct working.

$ cd /u01/app/oracle/product/10.2.0/agent10g/xen1.pool/sysman/log/
$ more emagent.trc

2007-11-15 06:06:56 Thread-4112513952 ERROR util.files: ERROR: nmeufos_new: failed in lfiopn on file:

/u01/app/oracle/product/10.2.0/agent10g/xen1.pool/sysman/emd/agntstmp.txt.erro
r = 24 (Too many open files)
2007-11-15 06:06:56 Thread-4112513952 ERROR pingManager: Error in updating the agent time stamp file
2007-11-15 06:06:59 Thread-4112513952 ERROR fetchlets.healthCheck: GIM-00104: file not found
LEM-00031: file not found; arguments: [lempgmh] [lmserr]
LEM-00033: file not found; arguments: [lempgfm] [Couldn't open message file]
LEM-00031: file not found; arguments: [lempgmh] [lmserr]

Solution:

Important is the message about too many open files. This is caused by operating system running out of open file handles limit (default for linux is 1024).

You need to increase the file descriptor soft limit per shell to 4096 from default 1024.

# vim /etc/security/limits.conf

#oracle              soft    nofile  1024
oracle              soft    nofile  4096
oracle              hard    nofile  65536

# sysctl -p

There are also a few infos about this problem from Oracle:

http://download-west.oracle.com/docs/cd/B16240_01/doc/relnotes.102/b31594/toc.htm

Oracle Clusterware Installation – Timed out waiting for the CRS stack to start

Terms:

Operating System: Enterprise Linux 4 U5 (RHEL4 U5)

Oracle: 10.2.0.1

Clusterware: 10.2.0.1

Cluster Interconnect: Ethernet (private connection)

Problem:

While running the root.sh script on the last cluster node, during the CLusterware Installation, the following error message occurs.

Clean remove Oracle Clusterware (CRS) 10GR2 from a RHEL4

The problem with Oracle Clusterware (also known as Cluster Ready Service – CRS) is, that there is no built-in mechanism from Oracle to clean remove the Clusterware and all of it’s files distributed over the OS filesystem. The follwowing example script removes the Oracle Clusterware completely. The operating system is a RHEL4 U5.

This script has to be edited for personal use. $ORA_CRS_HOME is here under /opt/oracle.


########### script ###############

#!/bin/bash

echo

echo "Remove the Oracle Clusterware Service ?"

echo

echo "Enter y[yes] or n[no] to exit"read comit

if [ $comit == "n" ]; then

echo "Exit from Script without any change..."

exit 1

else

echo "Start to Shutdown and Remove Oracle Clusterware ..."

echo

/etc/init.d/init.evmd stop

/etc/init.d/init.evmd disable

/etc/init.d/init.cssd stop

/etc/init.d/init.cssd disable

/etc/init.d/init.crsd stop

/etc/init.d/init.crsd disable

/etc/init.d/init.crs stop

/etc/init.d/init.crs disable

rm -rf /etc/oracle /etc/oraInst.loc /etc/oratab

rm -rf /etc/init.d/init.crsd /etc/init.d/init.crs /etc/init.d/init.cssd /etc/init.d/init.evmd

rm -rf /etc/rc2.d/K96init.crs /etc/rc2.d/S96init.crs etc/rc3.d/K96init.crs \

/etc/rc3.d/S96init.crs /etc/rc4.d/K96init.crs /etc/rc4.d/S96init.crs \

/etc/rc5.d/K96init.crs /etc/rc5.d/S96init.crs /etc/rc.d/rc0.d/K96init.crs \

/etc/rc.d/rc1.d/K96init.crs /etc/rc.d/rc6.d/K96init.crs /etc/rc.d/rc4.d/K96init.crs

cp /etc/inittab.orig /etc/inittab

rm -rf /etc/inittab.crs /etc/inittab.no_crs

rm -rf /tmp/*

rm -rf /tmp/.oracle

rm -rf /usr/local/bin/dbhome /usr/local/bin/oraenv /usr/local/bin/coraenv

rm -rf /var/tmp/.oracle

rm -rf /opt/oracle/*

echo

echo "Remove on one Node the Shared Devices"

echo "rm -rf /u03/oracrs/*"

echo

fi

########### end script ###############

After running that script on a system it should be possible to reinstall Oracle Clusterware without any problems.