SG HA Cluster Auto and Maual Failover

[Binubaby]

Hello
Please advice on what could be the complications in terms of redudancy of
having auto failover configuration if two - two packages are running in each cluster nodes in a two node cluster configuration?
Auto failoer or Manual (packages) suggested?

Also advise configuration of cluster haerbeat with public lan -- redudany point of view
(one port primary and two ports standby configurations

Thanks,
BB

[Manix]

I don`t think there wound be any complication
apart from the fact that the packages may increase system & network usage when all are running on the single node.

Depends which application you running with MCSG.

A simple package configuration is as follows

PACKAGE_NAME pkg1
PACKAGE_TYPE FAILOVER
FAILOVER_POLICY CONFIGURED_NODE
FAILBACK_POLICY MANUAL
NODE_NAME ftsys9
NODE_NAME ftsys10
AUTO_RUN YES
LOCAL_LAN_FAILOVER_ENABLED
NODE_FAIL_FAST_ENABLED NO

As far heart beat is concerned it`s always recommended to have at least one dedicated heart beat ( non -public ) and then you may have public /stationary IPs for both heqart beat & data.

NETWORK_INTERFACE lan0
HEARTBEAT_IP 192.168.114.119
NETWORK_INTERFACE lan1 # ( redundant )

NETWORK_INTERFACE lan3
STATIONARY_IP 16.118.114.119
NETWORK_INTERFACE lan4 # ( redundant )


You may use the above configuration if you have four lan cards.

Otherwise remove the redundant lan1.

Hope this will help.

Thanks

Manix

[Viktor Balogh]

Hello,

>two packages are running in each cluster
>nodes in a two node cluster configuration?

Try to spread the load over the two nodes. If there are also test environments among the packages, you can set the second node as a primary node to those.

>Auto failoer or Manual (packages) suggested?

If one node cannot deal with all the packages, set auto failover only for the important packages and take care that all the auto-failover packages can run on a single node in one time. In this case also set a dependency between the packages, in the start script you can incorporate a check like this:

########################################################################
# HEADER FOR ***PRODUCTION*** PACKAGE WITH CONFLICTING TEST/INTEGRATIONS
# PKGS IN THE SAME CLUSTER (e.g. Conflicting mountpoints/deamons etc..)
# This should be insterted directly below "# UNCOMMENT the variables "
# in the PRODUCTION PKG CNTL File.
#
# Check if the TEST or INTEGRATIONS PACKAGE IS RUNNING HERE
# -> If True, Stop the TEST/INTEGRATIONS Package First
# Only do this if the CNTL File has been called with "start" (not halt)
########################################################################

THIS_RUN_ARG=$1
THISHOST=`/usr/bin/hostname`

# ACHTUNG: if the PKG Name ist UPPERCASE und das CNTL File lowercase
# THIS MUST BE CHANGED e.g. with "typeset -u" !
SCRIPT_NAME=${0##*/}
THIS_PKG=${SCRIPT_NAME%.*}

# List all CONFLICTING TEST or INTEGRATIONS PKGs which may not run together with
# this PKG in ALL_CONFLICT_PKGS
ALL_CONFLICT_PKGS="ivwi00"

if [ "$THIS_RUN_ARG" = start ] ; then
for CONFLICT_PKG in "$ALL_CONFLICT_PKGS"
do
if `/usr/sbin/cmviewcl -n ${THISHOST} | /usr/bin/grep ${CONFLICT_PKG} >/dev/null 2>&1` ; then
echo "
###########################################################################
ATTENTION: `/bin/date '+%b %e %X'`
ATTENTION: CONFLICTING DEVELOPMENT PACKAGE ${CONFLICT_PKG}
RUNNIG ON ${THISHOST}!!!
PACKAGE ${CONFLICT_PKG} WILL BE STOPPED IMMEDIATELY ON ${THISHOST}
###########################################################################
"
/usr/sbin/cmhaltpkg -n ${THISHOST} ${CONFLICT_PKG}
fi
done
fi

[Rita C Workman]

You can have the packages failover over to each other with the understanding:

1. You have sufficent resources and/or you understand that will performace may be affected in the event of a failover
2. You are NOT using the same mountpoint names.
To explain:
Dev box/dev pkg has same mountpoint name as product. You could not run the production package and the dev package on the same box at the same time. One must come down before the other can start.

Note: Automatic failover, properly set up, is the point of SG clustering to begin with. It is NOT, however, recommended to have automatic failback!

Regards,
Rita

[Rita C Workman]

...typo...

Dev box/dev pkg has same mountpoint name as product.

Should read:

Dev box/dev pkg has same mountpoint names as production box/pkg.

/rcw

[Binubaby]

Thanks to Everyone

BB