[Tip] FS-Manipulation in 7.1.3-Clusters

UNIX AIX
1

In principle it is possible to create FSes on shared volume groups with cluster methods (also known as C-SPOC), either from the command line or using the respective SMIT menus. In practice, however, many options are missing in the C-SPOC variants. For instance, it is not possible to add a mountgroup to a newly created FS.

Many (me included) rely therefore on non-cluster methods (the normal crfs -command, et. al.) even in cluster environments. One common method, using the non-documented command imfs , is like this (we are supposing a two-node active/passive-cluster here):

root@activenode # crfs ....
root@passivenode # varyonvg -ncA <sharedvg>
root@passivenode # imfs <sharedvg>
root@passivenode # varyonvg -ncP <sharedvg>

This will not work any more with HACMP 7.1.3 upwards because IBM introduced a new (and very badly documented) fencing mechanism. If you try the above on a newer cluster you will immediately find the disks in <sharedvg> in the errpt as missing. The only method to get clear from this is to cycle the cluster completely.

Instead use the following method now:

root@activenode # crfs ....
root@passivenode # /usr/es/sbin/cluster/events/utils/cl_set_vg_fence_height -c <sharedvg> rw
root@passivenode # varyonvg -ncA <sharedvg>
root@passivenode # imfs <sharedvg>
root@passivenode # varyonvg -ncP <sharedvg>
root@passivenode #  /usr/es/sbin/cluster/events/utils/cl_set_vg_fence_height -c <sharedvg> ro

If you want to try the - old and new - method use a cluster you can afford to powercycle. There, you will be able to watch the problem unfold in hacmp.out - or the C-SPOC log, if you have used it.

I hope this helps.

bakunin
________________________
Sources:

1 Like
2

http://www.redbooks.ibm.com/redbooks/pdfs/sg245433.pdf p. 354 :wink:

3

@agent.kgb, when providing a link please also describe what we can expect to find there, preferably by quoting the relevant part.

After reading that page, I gather what you mean to say is that imfs is in fact documented.

4

To be honest, i haven't seen that part of the LVM redbook (or maybe i have but couldn't remember). Fact is that there is no man page for the command, it is not mentioned in IBMs online documentation and - save for the passing quote agent.kgb found - nowhere else in IBMs documents.

Even the redbook is very terse about it. A half-sentence hardly qualifies for "documentation".

Alas, this is similar to some other parts of the OS, which are also very poorly (or not at all) documented: mkinstallp , pre-7.1 HACMP cluster-commands and so on.

bakunin

5

bakunin is right. The command imfs is not documented. The redbook is the only documentation about a lot of intermediate-level LVM commands and it is very outdated. There are some other undocumented and very useful commands in the redbook, such as ipl_varyon , getlvodm or lvlstmajor . IBM doesn't like if you know too much about AIX internals :wink:

6

This is absolutely true!

It might be a worthwile community project to write documentation for these undocumented commands. getlvodm , for instance, i have used literally for decades now (it must have been introduced in AIX 4.1.x or so). Anybody interested?

bakunin

@Scrutinizer: i took agent.kgbs link as a rather witty comeback regarding my use of the word "undocumented". No harm done. ;-))

7

do you have a spare domain? :wink: I have some spare servers.