Hi,
I just installed the 1998.12.14 raid driver kernel patch and raidtools
released earlier today, and I'm happy to report that it works like a charm.
In fact, the server I tested it on is going into production later today.
I took it through all of my usual tests, which I've detailed below.
There were three general testing areas: (1) test drive failure and
reconstruction of new drive, (2) check drive failure and reconstruction when
the device names of the drives change, (3) test a power failure and the
reconstruction that follows.
In my raid setup, I've got three raid1 devices (md0, md1 and md2) which are
based off the disks sda and sdb. I don't use the autostart to run my md
devices, so I have not tested the autostart feature.
-- Test one, drive failure and reconstruction:
(1) Got the system running with md0, md1, and m2 up. md2 was in readonly
mode.
(2) Killed power to sdb while system was running. This caused a lockup
because the scsi driver could not handle loosing a drive.
(3) Rebooted the machine (power to sdb still off) and everything came up
fine. The md devices started in degraded mode and I was able to boot linux.
(4) Restored power to sdb and rebooted the machine. The md devices still
came up in degraded mode.
(5) Used the raidhotadd command to add the appropriate partition on sdb back
into each md device. Background reconstruction started and finished.
(6) Rebooted and the md devices started in non-degraded mode like they
should.
-- Test two, drive failure and reconstruction with device name changes:
(1) Got the system running with md0, md1, and m2 up. md2 was in readonly
mode.
(2) Killed power to sda while system was running. System locked up again
because the scsi driver.
(3) Rebooted the machine (power to sda still off) and the md devices came up
in degraded mode. Got the "device name has changed from sdb7 to sda7 since
last import" message from the raid driver. This message is what we expect to
see. Was able to boot linux.
(4) Restored power to sda and rebooted the machine. The md devices came up
in degraded mode. The raid driver informed me that "device name has changed
from sda9 to sdb9 since last import" which is true.
(5) Used the raidhotadd command to add partitions on sda back into each md
device. Background reconstruction did its thing.
(6) Rebooted and the md devices came up in non-degraded mode.
-- Test three, power failure and reconstruction:
(1) Killed power to whole system while md devices were up and running.
(2) Restarted and the md devices background reconstructed themselves without
any user intervention.
-----
Note: I use an initial ram disk to start /dev/md0 (on which root is
mounted), and then start the rest from /etc/rc.d/rc.sysinit. I like this
because I feel it gives me more control.
Only one complaint: Using the "raidhotadd" command on a md device which is
in readonly mode still crashes the system. Something should stop the user
from doing this, either in the kernel ioctl or in the raidtools command.
Great job Molnar!
- David Harris
Principal Engineer, DRH Internet Services
