ASM bug?
- From: Maureen English <maureen.english@xxxxxxxxxx>
- To: oracle-l@xxxxxxxxxxxxx
- Date: Wed, 01 Oct 2014 15:44:15 -0800
We're building a new system and the sysadmin and dba working on it discovered a
problem.
This is RHEL5 and Oracle 11.2.0.4 RAC.
At boot, or when instructed to do so, Oracle ASM scans disks and creates an index of which ASM labels are on which
disks, however it does not do anything to tell the kernel that it intends to use the disks it has identified. It is
not until ASM mounts the disks in a diskgroup that it marks them as "in use".
Until a disk is actually in use (opened), the kernel will happily unmap or re-map dm devices that Oracle ASM has
already scanned. Oracle ASM will try to modify the wrong disk if the kernel remaps an unused dm device to a different
disk before ASM mounts it.
When multipath is run with the -F option, it will flush all multipath devices that are not in use. When multipath
re-maps the devices, it may assign different device numbers to the disks it flushed. There is no way to tell Oracle
ASM to forget about a label it has scanned without deleting the ASM labels off the disk or rebooting the system. It
is very important not to flush a disk that ASM has already identified. Do not use multipath -F if there are any
unmounted ASM disks presented to the system.
Is this a known bug, or are there some rules regarding ASM and multipath that
we missed?
I'm not actually working on this system, but thought it sounds like something
others would have encountered. I'm still
searching through Metalink docs....
- Maureen
--
//www.freelists.org/webpage/oracle-l
Other related posts:
