Node replacement in an active cluster

[Eric SAUBIGNAC]

Hi everybody,

I will have to replace a node in an MCSG cluster with a new one. The new node will have the same hpux 11, host name, IP, and other things than the old one : my purpose is just replacing the hardware.

I thought of several ways to do that :

1 : No break down : nice !
1a) just do fresh install on the new box (OS, applications, patches, ...)
1b) be aware of getting the same instance numbers with the special files of disks and lan cards (ioinit)
1c) make any necessary OS adjustements (hostname, IP, passwd, printers, lan services, ...)
1d) then, just copy /etc/cmcluster from the old box to the new one
1e) shutdown and unplug the old box from the lan then plug and start the new one.

2 : Little break down
2a) just do fresh install on the new box (OS, applications, patches, ...)
2b) DO NOT be aware of getting the same instance numbers with the special files of disks and lan cards
2c) make any necessary OS adjustements (hostname, IP, passwd, printers, lan services, ...)
2d) copy all ascii configuration files from /etc/cmcluster from the old box to the new one
2e) make any necessary adjustements to cmcluster.ascii, package.ascii, control.sh due to the new hardware
2f) prevent the new box from starting the cluster at boot time
2g) stop the existing cluster
2h) shutdown and unplug the old box
2i) plug and start the new box
2j) do cmcheckconf, cmapplyconf from the new box
2k) restart the cluster

3 : No break down, nothing to do : very nice !
3a) make_recovery from the old box
3b) intall the new box with
3c) stop and unplug the old box from the lan then plug and start the new one.

Any comment ? Which way would you choose ? I suppose that the 3rd one is a just a beautiful dream ?

Eric

PBFWME : Please Be Fair With My English

[James R. Ferguson · Accepted solution]

Hi Eric:

I'd opt for a cold install with current patches and follow the guidelines in document #UXSGKBAN00000071 ("Rejoining a Node To A Cluster After Hardware Upgrade").

...JRF...

[melvyn burnard]

you have missed one obvious other option, and that is to rebuild the cluster as a single node, i.e. do an on-line deletion of the node you are removing, replace the hardware after setting up the new system, ensure all connectivity is ok, adjust the cluster/package files as neceesary and add the new node on-line.

[Eric SAUBIGNAC]

Hi James & Melvyn,

your answers are finally similar : i didn't know that it's possible to make some changes on an active cluster, since release 10.10. I was still working on an older basis ! So thanks for this information

But ... as i don't know anything about the client's cluster / system / application configuration i will modify, i think i will try a mix :

a) suppress the old node without stopping the cluster (as you suggested)
b) make some adjustements on the old box around MC/SG (as exporting cluster vg, ...)
c) make_recovery and install the new box with it
d) plug and start the new box
e) configure MC/SG with new hardware
f) add the node to the cluster without stopping the cluster

expected result : no break down but maybe the transfert of the package ... good.

I will tell there later (01/29/2001 ?) if it was a good idea to use make_recovery on an unknown system

Thank's again

Eric (PBFWME)

[Eric SAUBIGNAC]

Late to close this post ... but it is never too late to make well !