2 node cluster configuration

[Dinesh_15]

Hi Folks

First time came to know about such a great form.So i hope I will get the sollution. Almost new HP-UX.

Prob: I have to Configure a two node cluster. Not sure about the procedures.
But I got the procedure some how. Can anybody have a look on this and have some corrections if needed.

The procedure




1)umount all the file systems and remove all the entries from /etc/fstab, which need to be included in cluster( first time it mounted for checking purpose after creating the VG and LVs)

2)Deactivate all the volumes which need to be included in cluster.( vgchange –a n /dev/vg01)

3)Vgexport –p –v –m /hp/vg01.map /dev/vg01. Use this command to create map files for all the VGs.
Copy all the exported map files to other nodes.

4)Make VG directories and group control files for all VGs on second node and then import them using the map filles.The minor no of group control file should be same as the primary node.. Is this compulsory.Will other minor no for this group control file will work?
vgimport –v -m /hp/vg01.map /dev/vg01 /dev/dsk/c7t2d0 /dev/dsk/c7t3do ( These are the alternate paths as seen in the other node for the same disks as of primary node)

5)Create Directory structure of file system which was created on node 1

6)vgchange –a y /dev/vg01 Activate VG on node 1 and node 2.
vgchange –a n /dev/vg01 etc etc on node 2
I am little bit confused about this step.
I think it should be vgchange -c a

7)C
reate cmclnodelist file in /etc/cmcluster on both node and add entries for both the nodes.
Eg.. It will have entries
Node1
Node2

8)cmquerycl –v –C /etc/cmcluster/ori.config –n –n .
This will create cluster config file named ori.config
This will have entries of Cluster name, Node Name and their corresponding Lan cards including primary lan , heartbeat lan, and standby lan if configured.,VG name.etc
edit the config file as required

9)cmcheckconf –k –v –C /etc/cmcluster/ori.config
This will check the cluster configuration file

10)vgchange –a n /dev/vg01
Is this should be vgchange -c n or -a n???
(Please clarify)
Test the cluster using the cmruncl –v and cnviewcl –v

11)For cluster Autostart change the /etc/rc.config.d/cmcluster


Package creation
12)mkdir /etc/cmcluster/PKG
cmmakepkg –p /etc/cmcluster/PKG/pkg.conf---
This will create package template
edit and save this file

13)cmmakepkg –s /etc/cmcluster/PKG/dbci.cntl
This will create controle file to control the behaviour of start and stop of package
--------- edit and save this file

Copy dbci.cntl to all other node

For changing in cluster such as segment change do the appropriate changes , delete the pkg and app cmcheck and cmapply config on both the systems
cmcheckconf –v –C /etc/clcluster/ori.config –k –p /etc/cmcluster/PKG/pkg.conf

cmapplyconf –k –v –C /etc/clcluster/ori.config –p /etc/cmcluster/ PKG/pkg.conf


Disable the vgactivation while system boots in /etc/lvmrc . Make it 0

14) start the cluster on primary node
cmruncl


Sorry for posti

[Shaikh Imran]

Hi Dinesh,

I think this will help you a lot.

http://docs.hp.com/hpux/onlinedocs/B3936-90079/B3936-90079.html

Regards,

[bhavin asokan · Accepted solution]

hi dinesh ,


1. Make sure that service guard product is installed.
2. If possible take a ignite backup of vg00
3. create backups of all important files (like password , fstab, etc)
4. create /etc/hosts.equiv file and put entries of servers in that file.
5. umount all the file systems and remove all the entries from /etc/fstab, which need to be included in cluster( first time it mounted for checking purpose after creating the VG and LVs).
6. Deactivate all the volumes which need to be included in cluster.( vgchange -a n /dev/vg01).
7. Vgexport â s â p â v â m /hp/vg01.map /dev/vg01 (-s is for shared mode .you are using the same diskarray for both nodes in the cluster. when you are uding â s option with vgexport it puts a vgid in the map file .At the time of importing you donâ t have to specify the PVâ s for that vg. Other wise you should have to specify the PVâ s included on that vg.
8. Copy all the exported map files to other nodes.
9. Make VG directories and group control files for all VGs on second node and then import them using the map filles. vgimport -s â v â m /hp/vg01.map /dev/vg01.

10. Create Directory structure of file system which was created on node 1
11. Check by activation of vg01 in node2.deactivate it after checking.
12. Activate the vgâ s in cluster aware mode. vgchange â c â y vg01
13. Create cmclnodelist file in /etc/cmcluster on both node and add entries for both the nodes. Eg.. It will have entries
Node1
Node2
14. Move to /etc/cmcluster.
15. cmquerycl â C cmclconfig.ascii â n node1 â n node2
16. make a folder for package inside /etc/cmcluster. eg: pkgora.
17. Move to /etc/cmcluster/pkgora
18. Create the package template by cmmakepkg -p pkg.conf
19. Create the package control script file by cmmakepkg -s pkg.cntl
20. Edit the cmclconfig.ascii ,pkg.conf,pkg.cntl files as necessary
21. Copy the configuration files to other nodes.
22. Check the configuration
cmcheckconf -C cmclconfig.ascii -p pkgora/pkg.conf

23. after successful checking apply cmapplyconf command.
cmapplyconf -C cmclconfig.ascii -p pkgora/pkg.conf

24. Disable the vgactivation while system boots in /etc/lvmrc . Make it 0
25. start the cluster on primary node
26. if you want to start the cluster at the time of starting put the entry in /etc/rc.config.d/cmcluster to 1

27. check the cluster by stopping in primary node and starting in secondary node .


( configure the pkg ip in the pkg.cntl file you can configure cluster lock disks also for )

you can also see th

[bhavin asokan]

sorry,

i forgot to attach the file.

regds,

[bhavin asokan]

i think some problem for reading the above
attaching the same procedure

regds,

[Geoff Wild]

What version of Service Guard? as of 11.16, cmclnodelist no longer used.

What are you setting Autostart to in /etc/rc.config.d/cmcluster ?

My 2 cents - it should be 0.
If a cluster fails, and both/or 1 server reboots, you may not want to have the cluster start automatically - until after you determine what went wrong - for example, you could corrupt your database if there are disk issues or something to that effect...instead, start manually - modify the control script to only mount file systems - not start applications, etc...

Rgds...Geoff

[Dinesh_15]

Hi All

Imran Tnaks for the link . I am already having the same.

Bavin.
Thanks for nice procedure given.
Please also clarify the following.
1)Is it must that System IP ( which is assigned to NIC) and Package IP ( which is floating ) should be in same Subnet.
Above that is it must that hearbet IP should be in different subnet than above two IPs.
Eg. Package IP 10.2.0.250
Node 1 IP 10.2.0.251
Node 2 IP 10.2.0.252
( Here Subnet is 10.2.0.0)

Hearbeat for 1st Node 192.168.1.2
Hearbeat for 1st Node 192.168.1.1
( Here Subnet is 192.168.1.0)

If yes/no . why it is so?

2)In the cluster configuration file i.e cmclconfig.ascii there should be entries for the IPs of Network and Hearbeat like

NODE_NAME NODE1
NETWORK_INTERFACE lan0
HEARTBEAT_IP 10.2.0.252
NETWORK_INTERFACE lan4
NETWORK_INTERFACE lan5
HEARTBEAT_IP 192.168.1.2
FIRST_CLUSTER_LOCK_PV /dev/dsk/c7t0d0

NODE_NAME KNODE2
NETWORK_INTERFACE lan0
HEARTBEAT_IP 10.2.0.251
NETWORK_INTERFACE lan1
NETWORK_INTERFACE lan2
HEARTBEAT_IP 192.168.1.1
FIRST_CLUSTER_LOCK_PV /dev/dsk/c5t0d0

Is these entries are right. i think these are wrong..
For both the IPs it is showing as HEARTBEAT_IP only.. Actually 10.2.0.252 and 10.2.0.251 are my Server IP and Package is 10.2.0.250

Please clarify.

With Regards
Dinu

[Bharat Katkar]

Hi Dinesh,

System IP and PAckage IP should be in same subnet.
Then It is recommended and not compulsory to have different subnet for Heartbits.
When you use different subnet your public network is reduced with heartbit traffic. This adds to your network performance.

Hope that helps.
Regards,

[Sridhar Bhaskarla]

Hi Dinesh,

1. Yes. System IP and the package IP should be on the same subnet. ServiceGuard runs the command

cmmodnet -a -i

This means you should already have the subnet defined on the system corresponding to the package IP. Package IP attaches as a secondary interface to the interface that is in the same subnet.

2. Your IP assignments are looking good. I am not seeing a second heartbeat. You should have a second heartbeat otherwise it will become a single point of failure against the purpose having ServiceGuard.

3. When you ran cmquerycl to capture the information to build the cluster initially, serviceguard will probe the network and the disks. The network interfaces that are found with less latency will be assigned as HEART_BEAT interfaces. Moreover, in your case, there are supposed to be two HEART BEAT networks. Hence, the system picked both the interfaces as Heartbeat. There is no way system can differentiate between a heartbeat and public interfaces *initially* so it takes the decision based on the response times.

-Sri

[Dinesh_15]

Hi Sri

Thnakd for your Nice Help.
As you mentioned , if I don't have second Lan for how to add the same.
Whether the standby lan is not shared by the lan which is standby for the system ( i.e 10.2.0.251 0r 251).
I mean to say that incase of the sytem lan fails then the network traffic will be shifted to the standby lan..Right
Like this if heart beat Lan fails whether it is not shifted to standby Lan which was meant for System Standby. Is it not shared by System and Hearbeat Lan

Please calrify

Rgds
Dinu

[Sridhar Bhaskarla]

Hi Dinesh,

A cluster's integrity is based on the heartbeast messages exchanged betweeen the nodes in that cluster. If a system in the cluster doesn't receive heartbeat messages, then cluster reformation will occur and the packages will be moved to the node that acquires the cluster lock (disk|quorum). And other node may crash depending on the cluster configuration. With a single heartbeat network, there is always a risk of losing heartbeats. So, it's recommended to use second heartbeat.

The above is entirely different from the standby configuration you have for the system IP. The standby interface will only cover your system lan. Not heartbeat lan.

I would suggest you identify another interface on each of the systems and connect them through a seperate switch and assign them IP addresses in a different subnet. Since these are going to be private, you can use any IP addresses like 192.168.2.1 etc.,

It's not advisable to have hearbeats through data network as there is a good chance for congestion.

-Sri

[Dinesh_15]

Hi Sri

First time I am going thru this forum and what a great experts are there ..
really fantastic.

i really appreciate your help to understand the cluster concept better.. Thanks

I will add the second Lan to hearbeat and let you know.To add the second lan to Hearbeat, I think I have to modify the cluster configuration file and again apply the change using the cmapplyconf command..

Thnaks a lot

Rgds
Dinu

[Dinesh_15]

Hi All

Thnaks a lot your sincere help.

I got the clustering working perfectly.

Once again thanks a lot especially for Shridhar and Bavin

Let me to close this thread.

Rgds
Dinu