Lab Currently Closed Hours: 9am–6pm on Friday more »
Setting up a new server involves putting in all its new drives, turning off MegaRAID, setting up mdraid (Linux software RAID) on them, and then installing the operating system. It requires quite a few tricky steps.
The below steps were written for
hal which have LSI RAID
cards that need to be put into JBOD mode. The intructions will vary without LSI
hardware RAID (which we don't use).
Also, MegaCLI isn't very consistent between versions, and in general it's extremely buggy and poorly-written. So you might have to modify the instructions slightly to get something that works.
Because the software that comes with LSI's RAID controllers is terrible. It's called MegaCLI and you will never read anything good about it.
Examples of problems we've had in the past with MegaCLI:
Plus, it's proprietary, which makes getting it installed automatically difficult.
We are sacrificing a bit of measurable performance (mostly because we can't use the battery-backed write cache), but we find it to be a small amount (and worth the operational benefits of never having to touch MegaCLI again).
These assume you're willing to destroy all the data on the server and rebuild
it. They also assume you're currently using MegaRAID. If you're not using
MegaRAID or MegaRAID was already set up and you are booting off the first drive
/dev/sda), then you probably don't have to change anything and can skip the
MegaRAID setup steps.
These commands work pretty reliably but you should still think before pressing enter, because they might need some adjustment.
On boot, enter the LSI pre-boot CLI (press
Ctrl-Y at the right time). The
syntax in the pre-boot CLI seems to be the same as MegaCLI.
Remove all logical drives and put the physical drives in JBOD mode:
$ -CfgLdDel -LALL -aALL $ -PDMakeGood -PhysDrv[252:0,252:1,252:2,252:3] -force -a0 $ -AdpSetProp EnableJBOD 1 -aALL $ -PDMakeJBOD -PhysDrv[252:0,252:1,252:2,252:3] -a0
note: I got an error on jaws on the
PDMakeJBOD, but it worked anyway
Boot into finnix and figure out which drives you want in the RAID.
Make new partition tables on each drive and one large partition to hold the data.
You should make the data partition take almost all of the space on the drive, but not all the way to the end (leave a GB or two). The reason is so that you can replace the drive when it fails with another drive which isn't quite the same size (it might be a few bytes smaller).
for device in /dev/sda /dev/sdb /dev/sdc /dev/sdd; do parted "$device" mklabel gpt parted "$device" mkpart primary 10MB 510GB done
Pick one disk to hold GRUB (I usually do
/dev/sda) and do:
parted /dev/sda mkpart primary ext4 1 5 parted /dev/sda
Figure out the new partition number (typically 2 since it was the second partition created), then run:
parted /dev/sda set 2 bios_grub on
Set up RAID 10, and make sure to use the data partitions (like
and not the entire drive).
mdadm --create -v /dev/md0 --level=raid10 --raid-devices=4 \ /dev/sda1 /dev/sdb1 /dev/sdc1 /dev/sdd1
Set up a GPT partition table on the new RAID volume. Don't forget this or you'll be sorry when you have to abandon the Debian install.
parted /dev/md0 mklabel gpt
Back in Finnix, run
sync to write any changes in memory to disk.
Reboot and launch the Debian installer.
Make sure not to do the "OCF Automated Install" at PXE, do an "expert install" instead. sorry.
When you get to partitioning, use common sense. I recommend switching to
another TTY (
Ctrl+Alt+F2) and using fdisk to create ~40GB root, ~8GB swap,
and the rest as one partition (for LVM). These should be created on the RAID
/dev/md0 if you only have one RAID array).
When asked, install GRUB on the same disk as in step 5 (I recommend
This probably only has to be done if this server is booting into a different drive slot than it has used before. Try booting and see what happens, and if you can't, then try messing with this.
Boot into WebCLI (
Ctrl-H on boot at the LSI screen)
In WebCLI, figure out which disk you added your boot to, and set it as bootable. If you can't find the "Make Bootable" option on the physical drive page, it's probably already bootable. Maybe just restart and see if it works.
I can't find a way to match drive letters inside WebCLI, so you might just need to try all of them in your new array until it works, sorry.
Undo everything from above on
pestilence (enable puppet again, just run
puppet again to revert any changes, and start the DHCP server again)