Date: Thu, 10 Jan 2008 14:43:57 -0800 From: "Derrick Ryalls" <ryallsd@gmail.com> To: "Derek Ragona" <derek@computinginnovations.com> Cc: FreeBSD Questions <freebsd-questions@freebsd.org> Subject: Re: Nut and RAID on FreeBSD 7.0 Message-ID: <d5eb95fc0801101443l6b4cd69do6d69d5b90cd91702@mail.gmail.com> In-Reply-To: <6.0.0.22.2.20080110155430.023d29d8@mail.computinginnovations.com> References: <d5eb95fc0801091651h1788afd1m256a824e4722906d@mail.gmail.com> <6.0.0.22.2.20080110025841.023a9bb8@mail.computinginnovations.com> <d5eb95fc0801101140w717095dfhed569274c261e4a6@mail.gmail.com> <6.0.0.22.2.20080110155430.023d29d8@mail.computinginnovations.com>
next in thread | previous in thread | raw e-mail | index | archive | help
> > > > Greetings, > > > > I have a RAID fileserver plugged into a UPS and nut is able to > > communicate with it successfully. With the winds making the lights > > flicker, I started looking into having the computer shut down when > > power goes out for more than say 5 minutes or so. Looking at the > > documentation, I found that the 'true' solution is more like the > > system goes into a safe state when the battery gets low, then the ups > > eventually dies. When power is restored, the UPS and computer are > > supposed to both come back to life. This would be a great system to > > have in place, but it does sound a bit risky and so may not be worth > > doing just to save my home fileserver. > > > > The instructions and the conf file have the shutdown command of > > 'shutdown -h +0' which will halt the system. The man page for halt > > says the the disk cache will be flushed, but doesn't mention anything > > about going to read-only or anything. I suppose my first question is > > whether or not flushing the cache is sufficient to save the RAID (5) > > array, or if I need to find a way to get the file systems into read > > only mode? > > > > The second question has to do with a rc.d script that nut recommends > > creating. The script does a 'upsdrvctl shutdown' and then a sleep > > 120, basically waiting for the machine to die while in the script. > > Won't this block the other rc.d scripts? Also, is this the magic part > > that enables the machine to auto power up when power is restored? > > > > Changing the shutdown command in nut to 'shutdown -p +0' looks like > > the sure fire way to get the system down clean before the power is > > lost, but if my concerns are not valid, then I could be missing out on > > some nice functionality for no reason. > > > > Does anyone have experience with this? > > I have my servers all using nut to safely shutdown. My configuration is > > the servers are set up with one as master for nut, that master connected > to > > the UPS. The other servers are slaves and get their nut information from > > the master. > > > > My setup has the servers wait until the UPS is on low battery, then they > > all shutdown. > > > > As a separate part of the setup, the servers are set in their BIOS to > power > > on, after a power failure. This is in the BIOS power setup. > > > > So if there is a minor power problem, the servers run from battery. In > a > > larger power outage, they are shutdown cleanly once the battery level is > > low, and power up automatically once power is restored. > > > > In my upsmon.conf file I have this: > > SHUTDOWNCMD "/sbin/shutdown -h +0" > > > > If you want more specifics, I can look through the configuration files > and > > email you relevant settings. > > > > After doing more reading, I am confident that a shutdown -h would be > sufficient, but am a bit concern on the order of operations. The nut > documentation has a recommendation to add a kill script as such: > > > #!/bin/sh > > if [ "$1" == "stop" ] > then > > if [ -f /etc/killpower ] > then > echo "Killing the power, bye!" > /usr/local/libexec/nut/upsdrvctl shutdown > > sleep 120 > fi > fi > > </copy> > > Even if I name this zz_killpower.sh to make it run last, depending on > how long it takes FreeBSD to flush the cash after all rc.d scripts are > run, I could end up doing a dirty power down, right? Without this, if > the power does come back while before the battery finally dies, the > system won't restart since the power was never fully interrupted at > the computer side? > You are reading the old documentation. The current nut, 2.2, has complete > rc scripts that are installed in /usr/local/etc/rc.d > > You need only define the flag file you want to use in upsmon.conf > > Also define what actions you want in that file as well. You need to use > the sample files installed in /usr/local/etc/nut and be sure to read the > comments. > I have 2.2 installed and am using the existing scripts. In the comments in uspmon.conf, there is this part: # -------------------------------------------------------------------------- # POWERDOWNFLAG - Flag file for forcing UPS shutdown on the master system # # upsmon will create a file with this name in master mode when it's time # to shut down the load. You should check for this file's existence in # your shutdown scripts and run 'upsdrvctl shutdown' if it exists. # # See the shutdown.txt file in the docs subdirectory for more information. POWERDOWNFLAG /etc/killpower Which in the related documentation means I need the custom shutdown script mentioned above which checks for the existence of the /etc/killpower file before doing the upsdrvctl shutdown command to kill the UPS before the battery is completely dead. I suppose in your situation you won't need this extra script as you run until the UPS is critical whereas I am trying to kill the system a bit early, before it is critical. Perhaps I need to re-evaluate my line of thinking. Light sometime flicker, but power almost never goes out. When it does it is either back on in less than 1 minute, or out for hours. If the UPS detects critical correctly and gives me at least a minute before death, then that should be plenty of time for the system to auto-shutdown. Guess I will have to do some experimentation tonight.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?d5eb95fc0801101443l6b4cd69do6d69d5b90cd91702>