I am running the command runcluvfy stage -post hwos -n mynodeA,mynodeB -verbose and get the error message “Shared storage check failed on nodes “mynodeA,mynodeB”. I have read other posts in which people have said you can ignore this. However, I need to be sure. What can be done to make absolutely sure our architecture is set up properly and prevent problems from occurring later.
CLUVFY is coded to use some checking mechanism to check sharedness of disks and if these checking mechanism fails we get the errors. This however, doesn’t signify that you really have a problem with shared disks because cluvfy may sometimes not recognise a disk type correctly.
A possible cause why Cluvfy shows disks are not shared could be a incorrect OS/hardware setup.
If you are sure that everything is setup correctly then to cross-verify that disks are shared amongst Nodes, you need to write and read to these shared disk from each of the RAC Nodes.
The below is an example to check this out.
NOTE – You can use any OS utility to check this. I am using DD command here.
For windows there is also a DD utility available through http://uranus.it.swin.edu.au/~jn/linux/rawwrite/dd.htm
To check if the disks are shared correctly between nodes perform the following steps.
on node A –
% hostname > C:\disktest1.a
% dd if=C:\ disktest1.a of=\\.\a count=1
(Replace \\.\a with the name of your shared disk partition)
on node B –
% dd if=\\.\a of=C:\disktest1.b count=1
% more C:\disktest.b
If the result of disktest.b show the nodename from node A, we can confirm that node A can share the disk with node B. This does not proof that node B can share his files with node A. So a second test is needed but now from node B to node A.
on node B –
% hostname > C:\disktest2.a
% dd if=C:\disktest2.a of=\\.\a count=1
on node A —
% dd if=\\.\a of=C:\disktest2.b count=1
% more C:\disktest2.b
And again if the nodename of node B is mentioned in the output of the file on node A you know the disks are shared correctly.
Note: In the steps we assume we have a 2 node cluster environment. If there are more nodes you need to repeat the steps on the other nodes as well