<div dir="ltr"><div class="gmail_extra">On Thu, Jan 31, 2013 at 9:37 AM, Munir Nassar <span dir="ltr"><<a href="mailto:tclug@beitsahour.net" target="_blank">tclug@beitsahour.net</a>></span> wrote:<br><div class="gmail_quote">
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div id=":dl">If linux crashed hard, it may not have had the ability to write to<br>
disk. sometimes it is preceeded by oopses or non-fatal panics, but in<br>
my experience not often enough.</div></blockquote></div><br>
</div><div class="gmail_extra" style>This reminded me of something interesting I discovered accidentally a couple years ago. My employer at that time had a fibre channel fabric, connecting all of our headless servers to their storage.</div>
<div class="gmail_extra" style><br></div><div class="gmail_extra" style>I was performing some maintenance on the switch fabric, the nature of which I can't remember at the moment, but regardless, it required me to disconnect a couple fiber patch cables. I pulled out one of the cables and then got distracted for a couple minutes by something else. During that time I started getting pages about a system being inaccessible. I started looking into the situation and quickly discovered that instead of disconnecting the cable I had intended to, I accidentally pulled the cable connecting this server to our FC network, effectively pulling its storage subsystem out from under its feet.</div>
<div class="gmail_extra" style><br></div><div class="gmail_extra" style>I re-patched that cable, then pulled up this server's console, fully expecting to see a kernel panic. Instead of seeing that, I saw what appeared to be a completely functional system. Not believing what I was seeing, I checked the system's uptime and sure enough, it indeed had not rebooted, but rather just picked up right where it left off before.</div>
<div class="gmail_extra" style><br></div><div class="gmail_extra" style>Armed with this information, I started playing around with a test system to see how long the kernel would stay alive with *no* storage connected, and was able to stretch things out well past the 15 minute mark with no ill effects.</div>
<div class="gmail_extra" style><br></div><div class="gmail_extra" style>This all was a revelation to me, as I had always assumed that, absent a working root partition, the system would immediately crash. I now know that's not the case.</div>
<div class="gmail_extra" style><br></div><div class="gmail_extra" style>Now, in this situation, the disks were connected via fibre, so there were no electrical/grounding ramifications to disconnecting while in use. I certainly would not advocate trying this on non-hotplug, electrically-connected hardware, as I'm sure you'd release the magic smoke from *some* component. :)</div>
<div class="gmail_extra" style><br></div><div class="gmail_extra" style>-Erik</div><div class="gmail_extra" style><br></div><div class="gmail_extra" style><br></div></div>