[mythtv-users] Slightly OT: Powering up remotely
linux at thehobsons.co.uk
Wed Sep 23 10:16:25 UTC 2020
This probably isn't going to be all that helpful, but as the others have already said - "it's complicated", and it can easily end up costing more than it's worth where "cost" can be in monetary terms or in other problems. But a few thoughts in no particular order ...
RAID 1 for your OS (/, /boot, /var, ...) etc - that's a no-brainer. I can't begin to count the number of times that's saved my backside over the years.
A key factor is your local conditions - what may work for me (all underground power supplies, very reliable mains power) may not be suitable for someone else (e.g. long overhead power supplied, "out in the sticks", unreliable mains). My last two jobs have encompassed both of those !
At my last job (IT services company) we were like I am at home - all underground power supplies and very reliable mains. For a long time we had no working UPS in the server room and I had servers with uptimes of over a year. In some ways, we actually had more problems caused by the UPSs over the years than we had due to mains failure - and that was echoed at some of our customers where UPS failure was sometimes more frequent than mains failures.
Where servers had dual power supplies (and were capable of running off one), we'd generally recommend connecting one via the UPS, and the other direct to the mains. That way, if the mains failed the UPS could keep the server going, but of the UPS failed the mains could keep the server going. But that has it's own problems as the UPS doesn't see the true load until the power fails, so you get erroneous estimates of run time.
At the other end of the spectrum, at my previous job (small manufacturing company) we were on the end of overhead power lines that seemed (I once got to have a peek at the network maps) to come via every small village in the area. Power cuts were frequent, and working UPS(s) were essential - though still a tough sell to manglement who had to sign off the expenditure.
As an aside, at one point when things were particularly bad, I was able to classify power cuts into 3 main groups :
The very short one's where power would come back after a few seconds (auto-reclose on breakers)
Ones that were within a few minutes of 90 minutes long. The DNO engineers (a family member was one of them) were given a target of 90 minutes to restore supplies where a physical visit to a substation was needed. I suspect that they had an unwritten rule not to respond too quickly lest their manglement cut the allowed time - "you did it in 50 minutes the last few times, your target is now cut to 60 minutes"
Ones that were "long" - where presumably there was more to restoring supplies than just switching it back on. I think the longest we had was something like 4 hours.
At that earlier job, I had looked into some automation, but in the end it just ended up too much hassle for too little reward. I had got a UPS that could run everything for 30 to 60 mins, but if I shut down "non-essentials" could keep the phone system and main server (plus remote access to it) going for a few more hours. Look into many businesses and you'll find stuff that isn't needed if the lights are out - there won't be desktop users with power so they can't use it, and back then we had few laptops. But we did have remote sites that used the main system - so keeping the core network and that going meant they could do work.
At the time, a "sales person" from APC absolutely assured me that their ShareUPS unit could do what I wanted - shut down selected servers AND REMOVE POWER*, leaving others going until the batteries were lower, and then start up anything that had been shut down. Unfortunately, what he omitted to mention that their way of starting things back up again was to shut everything down and power cycle the UPS output - and no it couldn't power off individual items. In effect, all it dod was multiply the number of servers that could connect to the UPS for basic signalling.
They did have a product in the US which I got hold of which combined the signalling with power switching - but as I mentioned, you get into a situation where the risks of failure of that get to be higher than the risks you are mitigating - and with the variety of systems (some of them Macs with no serial ports) made utilising the basic signalling difficult.
* Our main server didn't power itself off after an OS shutdown - so little power saving.
So you end up with trade offs like :
You give yourself remote access to server management ports. Great, you've solved one problem, but added a big security issue to be managed. Given the number of systems with integrated management these days, I've read suggestions that this can be a very high security risk as they can be easily overlooked for security updated etc - that's IF the manufacturer actually provides them.
Similarly with remote power switching. You solve one problem, but add another point of failure and a security risk.
So you may be better off just arranging with a friend to be able to go in and read the screen/press buttons for you. For us, we need someone to come in and feed the pets if we're away anyway.
More information about the mythtv-users