What is a kernel upgrade reboot ?

[luchtzakje]

Hello,

I don't know much about servers, can anyone explain me what a kernel upgrade reboot is ?

greetings,

Bart

[Ron]

They are upgrading the operating system (In the Unix world, refered to as the kernel), and the machine needs to be rebooted afterwards.

Just like your PC... only better!

[JPC-Les]

A kernel upgrade reboot is a notice we post in the forums to make clients aware that their respective server will be going down for a reboot.

After a new kernel is installed the server must be rebooted for it to take effect hence the reboot.

[Ron]

Beat ya, Les

[JPC-Les]

Doh!

[luchtzakje]

Is it normal that a Kernal Upgrade Reboot takes over 1 hour and 30 minutes ??? Since my website is still not working

[JPC-Les]

As mentioned in the other thread the techs are working on the server.

[Ron]

Anytime a machine is taken down, there is a pretty good possibility that it won't come back up.

Whenever I would schedule server maintenance which would require actually shutting down power to the box(es), I would ensure that there was an adequate supply of spare drives, backups and power supplies.
Disk drives, especially, don't like to spin back up after being shut down.

A few years ago, on one Sun development machine in particular, I would have to physically take out the drive and manually rotate it rapidly back and forth then stick it back in the drive then try and fire up the server.

Why they are continually having to do fsck's after a soft reboot...? Maybe the sync command isn't working... sync usually needs to be called manually several times, as in "sync;sync;sync;reboot"

[tank]

Personally I would rather a scheduled downtime that takes a little longer than a crash that could put you out of business for a while. However, I do find it strange that servers are being rebooted during the workweek in the middle of the day.

[Spathiphyllum]

Originally posted by Ron
Anytime a machine is taken down, there is a pretty good possibility that it won't come back up...
Disk drives, especially, don't like to spin back up after being shut down.

I've never heard of this phenomenon before. Is it restricted to a particular brand or interface? I'm guessing you were working with SCSI all of the time but I've never seen this happen in any server or desktop that I've built. Also, was it SUN and its derivatives only or did you experience this with Linux or Windows species, too?

Granted, I'm not doing industrial strength serving, but what you've experienced sounds like serious trouble. Whatever happened to TCO (total cost of ownership) at a reasonable price? Is it a case of not making things like they used to?

[Ron]

When I said "pretty good" I wasn't implying a huge percentage like 50% or anything. It's just that things happen, and sometimes a machine with a new kernel won't come back up. some slight variation in configuration from one machine to the next can make one machine come back and another refuse to restart back to multiuser modes.

As a not-too-great example, look at the upgrade to mysql. Most sites are fine, others have a problem. That's why development servers testing is a must for pre-production verification prior to a production rollout.

When you shut down 100's of drives, some of them years old, that have been spinning constantly for a year, a couple of the spindles can be expected to not want to spin back up.

When I was in the corporate world, I had our own backup hardware (CPUs, NICs, CPU boards, drives, clocks, power supplies, one or two of each) available for normal daily use (my USAs and their manager were fully capable of diagnosing and replacing failed components much more quickly than the 3 hour response times of the maintenance contract). The maintenance company would then swap the failed component for a new one. Frequently, they wouldn't have the exact replacement part anyway, and would ask if we had one in stock. "Risk Management" was my constant concern in life.

Everything, including all of the drives, were covered by our on-site maintenance contracts, and when performing a large data-center event, I would require our maintenance company have a technician with lots of spare parts available.

So incidents were reduced by planned preventative maintenance, and when an incident occured anyway, downtime was minimized. Can you imagine how little non-events are recognized by senior management? A good infrastructure manager is the least noticed, while an incompetent one is like "Scotty" on Star Trek; always putting out fires and looking like a hero.

[Ron]

PS Not that Scotty was incompetent... he was dealing with unexpected events like battles with doomsday machines.

[Ron]

PPS JAGs USAs are great at putting out fires, and I've seen some signs of incremental improvement. I've had experiences where the same problem recurred many times before they've looked for and found the problem. I've seen them put together a nice proactive monitoring and notification system.

I don't see, however, anything that goes on behind the scenes.... just like senior management.

[Spathiphyllum]

Originally posted by Ron
When I said "pretty good" I wasn't implying a huge percentage like 50% or anything.

Whew! You had me terrified for a moment.

When you shut down 100's of drives, some of them years old, that have been spinning constantly for a year, a couple of the spindles can be expected to not want to spin back up.

OK, I understand that. An odd sticky head or spindle is understandable but not completely forgivable... I love the warranty of "We'll replace your drive if it fails." Heck, that's the cheapest part of a replacement, nevermind the inevitable irreplaceable data that just happened not to get backed-up.

Frequently, they wouldn't have the exact replacement part anyway... I would require our maintenance company have a technician with lots of spare parts available.

I'll bet they LOVED that. So much for just-in-time-inventory on product lines with a shelf-life of mere months. But I'd do the same thing in your shoes. You might have to pay a premium but it is cheap insurance long-term.

Can you imagine how little non-events are recognized by senior management? A good infrastructure manager is the least noticed, while an incompetent one is like "Scotty" on Star Trek; always putting out fires and looking like a hero.

Yeah, typical misplaced appreciation. At least that's better than "No good deed goes unpunished." Ever been in a situation where 99.9% of the time your foresight and attentiveness keeps things running smoothly (beyond notice or expectation) and the 0.1% of the time things don't go perfectly results in a "talk"? Nevermind that you had been performing duties above and beyond your payscale.

[Ron]

All of our configurations were RAID based, mostly RAID-5 with the odd RAID-1 and very odd RAID-3, so any drive's data could be rebuilt after a failure.

Those were the cheap solutions.

Then there were the million-dollar EMC boxes, mirrored offsite via fiber-optics. The hardest part of the EMC boxes was trying to convince people that their particular application didn't require full blown real-time offsite mirroring! lol

The second hardest part was convincing the owners of a particular application that certain tables in their application didn't require real-time offsite mirroring... like Sybase temp tables. Boy, did THAT slow things down.