[DRBD-user] Persistent Device Names

Ted Young mail at tedyoung.me
Wed Jan 25 01:30:07 CET 2012


> I happen to do DRBD over LVM which I like however there is documentation
on doing both...  A
> LVM -> DRBD -> LVM (however little I may understand of why you would want
this)

Thank you as well, Jake.  That would work (since LVM solves this problem by
tracking physical drives by metadata rather than by name); however, I would
prefer to eliminate the extra layer of LVM that is only offering me drive
management, which I can do using other tools.  For example, here is what I
am considering:

I have 26 slots labeled a - z on each server.  To simplify management, I
want to replicate the drive in slot a on server 1 to the one in slot a on
server 2, and so forth.  If the drive failed in slot a I would simply like
to replace it and have everything magically work.  Theoretically, this could
be accomplished; but, apparently, mapping SAS ports (each slot would be a
different SAS port) in linux has never worked reliably.  The only way to
reliably identify a drive is by its ID number or serial number.  

So, I would use udev to map a drive by serial number (which conveniently is
printed on the label) to /dev/sda (representing slot a).

I would configure DRBD using /dev/sda.

In the event of the failure of sda, I would pull the drive, replace it, and
update udev with the serial number of the replacement drive.

While this strategy does require the manual mapping of serial numbers to
slot identifiers, it does eliminate the overhead of having another layer of
LVM while still eliminating random device naming.  Furthermore, since most
drives (except Toshiba drives apparently) report their serial numbers or
WWNs to the OS is very easy to get the serial number from the label during
installation.

Can anyone poke any holes in this?  I will be happy to post complete set of
instructions as well as the results of testing when I get my hardware next
month.

Ted Young






More information about the drbd-user mailing list