Can't use "eject" to identify the server

Question

I'm a software engineer and our lab usually has several servers in a rack with a KVM switch. We're developers and we usually have several projects happening at once so the servers are frequently imaged and admittedly we aren't good at keeping them labeled, etc.

The point is, up until recently, as soon as we walked up to a terminal it was our habit to use eject to see which physical server we were actively selecting on the KVM switch, for obvious reasons.

Now we are transitioning into servers without optical drives. So the old eject trick is obsolete. There must be a simple command to flash the "ID" indicator on the front panel of the computer, but after Googling as many variations of the question as I could think of, I got nothing.

The servers are running Scientific Linux 6.4. Thanks for any help you can offer.

In a perfect world, of course. But they change so frequently that it's just not practical. — MerrillFraz, Dec 10 '13 at 14:29
Even if you can't label servers with their NAME, you can label them with the KVM channel on which they are plugged. That shouldn't change that often ! — mveroone, Dec 10 '13 at 14:31
Does `sgpio` work for you? It's originally meant for drive bay enclosures but depending on your hardware it might be useful. — Jenny D, Dec 10 '13 at 14:34
You might also contact your hardware vendor and ask if they've got any programs to do this. — Jenny D, Dec 10 '13 at 14:55
`spgio` did nothing, and there is nothing in `/sys/class/leds/` — MerrillFraz, Dec 10 '13 at 15:09
Audible ping and pull cables until the beeping stops? If they didn't want me unplugging then they would have labeled then :) — Grant, Apr 08 '14 at 21:26

score 7 · Accepted Answer · answered Dec 10 '13 at 15:00

Most modern servers come with an out-of-band management option, eg HP's ILO, Dells DRAC, IBM's RSA, so you don't actually need to go to the rack anymore to manage them. Simply use the correct form of KVM over IP to take over the console from the comfort of your desk!

They come with virtual media which uses an ISO image on your desktop to emulate an CD/DVD drive you can install from. Virtual power switches to power up, down or reset the server and more.

The out-of-band management card has it's own independent IP-address from the server it is managing. So label the actual servers and and give the out-of-band management card the same name and you'll see the current hostname of the OS once you take over the console.

And oftebn the out-of-band-management card allows you to light an identification light for the rare cases you do need physical access.

That seems to be the case here. I've contacted our vendor to see if we have such an option. Thanks everyone for the feedback. — MerrillFraz, Dec 10 '13 at 15:35

score 7 · Answer 2 · answered Dec 10 '13 at 15:08

I also love eject! anyway, you need to check support package from your server vendor.

for example, HP provides a command line utility to check/set/unset UID LED for their proliant servers. and iLO, HP's remote management tool, provide same function.

score 7 · Answer 3 · answered Dec 10 '13 at 15:11

If your out-of-band management does not allow you to flash the indicator, you could try ethtool if you have a spare/empty network interface

ethtool -p eth2

Should flash the LED for eth2:

   -p --identify
          Initiates adapter-specific action intended to enable an operator to easily identify the adapter by sight.  Typically this involves blinking one or more LEDs on the specific network port.

Then you can go search the server with an empty but blinking network interface. This should also work for interfaces that are up and running (if they're all connected), but then you'd have to distinguish between the ethtool regular interval and the normal blinking that shows the interface activity :)

It would work, but not practical. Some of our racks are downright inaccessible from behind, plus when we are identifying a computer installed in on of our systems we can't get to the back without pulling out the server. — MerrillFraz, Dec 10 '13 at 15:30

score 6 · Answer 4 · answered Dec 10 '13 at 15:12

6

If they've still got internal speakers, I used to get mileage out of doing cat /kernel/unix > /dev/audio, and then running around a dark and empty trading floor homing in on the shrieking machine.

answered Dec 10 '13 at 15:12

MadHatter

78,442
20
178
229

1

This is so far what we came up with, too. It works, but our factory is not quiet so it isn't ideal. – MerrillFraz Dec 10 '13 at 15:27

Journeyman Geek · Answer 5 · 2013-12-10T15:15:43.887

4

Well, get creative - there's USB controlled LED device like the blink(1) or the blink stick that may work, though the blink(1) seems out of stock everywhere.

For a more ghetto solution, get a bunch of the cheapest USB keys you can, stick them on a USB port, and when you need to identify a system, create and delete a small file (or just keep writing to it) repeatedly until you find the system. Then you can use the activity light to find the system.

edited Dec 10 '13 at 15:15

answered Dec 10 '13 at 15:10

Journeyman Geek

6,969
3
31
49

1

Although creative, that is a frightening solution. My co-workers are already disorganized, that sounds like fuel to the fire! – MerrillFraz Dec 10 '13 at 15:26

score 2 · Answer 6 · answered Dec 10 '13 at 23:30

2

AFAIK all of the modern servers come with the BMC [Baseboard management card] and provide an IPMI compliant interface both accessible from the OS and network,

Other vendors should provide something similar, in case of Dell servers this is the generic IPMI command which starts flashing the chassis LED:

ipmitool chassis identify 1 and ipmitool chassis identify 0 to stop the flashing

answered Dec 10 '13 at 23:30

Martino Dino

1,145
1
10
17

Confirmed on HP server as well. Also, `ipmitool chassis identify force` turns on the light indefinitely. – 200_success Dec 11 '13 at 01:01

Can't use "eject" to identify the server

6 Answers6