My disk (on kvr05) is acting horribly slow.
is there a system status update that might be relevant?
up_the_irons?
up_the_irons - any idea about my slow disk?
been crappy all morning
do I have a noisy neighbor again?
DAMN... this needs fixing.
1048576 bytes transferred in 22.494234 secs (46615 bytes/sec)
it took 22 seconds to write A MEGABYTE
up_the_irons ding ding ding
RandalSchwartz: yeah, i've been having intermittent problems with the host kvr05
can you migrate me to a different box then?
it's interfering with business and tasks.
mine's fairly slow too.
52428800 bytes (52 MB) copied, 71.7589 s, 731 kB/s
:/
RandalSchwartz: the first priority is to fix the issue with kvr05; i think it is a noisy neighbor, as you said
jdoe: are you able to reach your vps at all?  (i assume perhaps so, b/c you could get those test results)
up_the_irons: yeah.
jdoe: and it is on kvr05?
not sure offhand.
how would I tell?
jdoe: it's the same as your vnc host (listed in portal under vm details)
you can tell how often I log in ;)
sec.
:)
mm i'm on KVR05 and is sloww, my monitoring software is detecting the VM as down/up/down/up
i can't ssh into kvr05.  serial console is responsive, however, it also won't let me in past me typing in my login name.  i think the disk is completely locked up
kvr10 is doing fine. :D
ariel: although it is interesting that you can still get in, even though i can't get into the host.
up_the_irons: lol... I have at least two portal accounts, neither showing my vps.
nope, I'm on kvr06
jdoe: roger
seems better now, 6-ish M/s.
jdoe: i'm not touching kvr06 ;)
but 6M/s is still pretty slow
I know, I'm just saying whatever it was, it's improved slightly. And no argument ;)
kvr05, i'm in!
jdoe: roger :)
now to find out who is hogging the disk
prolly that schwartz guy.
randal actually _does_ use quite a bit of I/O, but not enough to take down the box.. ;)
I'm surprised any guest can make it *that* unresponsive.
maybe someone is swapping?
up_the_irons: sorry for the delay on centos.  I had to board up house and horses for the storm.  I'll be unboarding and such tonight and tomorrow.. then can get back to work on it.
jdoe: well, kvm/qemu doesn't really have too much in the way of I/O isolation. Like, each VM is just a Linux process and if it wants to really load up the disk, the scheduler isn't gonna stop it
jpalmer: oh problem at all, take your time.  I'm actually going to make the 5.8 version :)
I saw a presentation from Verio a few years back where they had talked about all the custom work they had done to give a level of control over disk I/O, memory and CPU usage to their VPS product
they claimed they were going to release that back to the FreeBSD project, but never did :/
up_the_irons: linux does kinda suck for that.
I have come up with a new way of dealing with disk provisioning for Linux guests that will afford me a *much* easier time supporting more distros; therefore, I'm pumped to make CentOS 5.8 templates, then Ubuntu 12.04 after that, then who knows... Fedora, Gentoo, Arch, ...? you name it
LFS PLZ.
twobithacker: jdoe yeah, i've seen several proprietary solutions talked about for that stuff, but nothing out in the open (but i admit, i  haven't looked into it in a while)
as the list grows longer so does the amount of effort to maintain such a list. I know it won't effect our bottom line but still... ;-)
<sarcasm> doesn't ionice work? </sarcasm>
up_the_irons: yeah I dunno, it doesn't seem like a problem anyone is anxious to solve, despite the work that's gone into cfq and deadline. I dunno why. I can load the shit out of the disk on fbsd, and the machine is still responsive.
try that on linux and it craters.
toddf: well, that's the thing, up until now the effort to maintain such a list has been prohibitive; but, with my new disk provisioning strategy, no longer :)
I'm not in on all the details, but with a new disk provisioning strategy, doesn't that still mean someone has to manually test install a distribution for it to work?
jdoe: i think it is a problem that people have accepted; kinda like DoS attacks, they suck but what can ya do? ;)
toddf: yes, but only once, then the template is copied for every VM down the line.  new distro versions don't come out _that_ often; openbsd is actually the most frequent publisher, where I find myself having to update my templates every 6 months ;)
up_the_irons: request multiple vps's across different kvr* hosts like randalschwartz, and make it a cluster that will lessen the effect of one kvr* with temporary io starvation ;_)
eh, ubuntu is on a 6 month cycle too isn't it?
toddf: wait wut?;)
jdoe: is it?  i haven't even noticed...   for ubuntu i tend to do only the LTS versions
if you want non-LTS, a ubuntu fan can just get an LTS and dist-upgrade, very simple process with like 2 commands
up_the_irons: my 'request multiple vps' thing was in rsponse to your 'what can ya do?' and my response involves more business for you *grin*
toddf: :)
up_the_irons: if you're tracking every version, then yeah it's every 6 months, most of the time.
XX.04 and XX.10
roger
time for kvr05 to admit failure and reboot
up 968 days, 13:33,  5 users,  load average: 11.82, 12.28, 20.87
of *course* i can't reach 1000, ever...
the load was getting better, but now it is back to worse
up_the_irons: re your comments earlier about protecting one guest from another on KVM, cgroups might be what you're after
didn't know those worked for io... wonder how well they work.
jdoe: Well, if we all start doing really heavy i/o usage, perhaps up_the_irons will find out for us? :-)
plett: i'll check it out, tnx
up_the_irons - there's no PST until november. :(
"at approximately 08/27/2012 14:30 PST" - not possible
but at least my system is working again. :)
Do we know the cause?
Also, my VPS seems to be working fine.
it was just kvr05 acting up
900+ uptime days
RandalSchwartz: I'm on kvr05
arenlor: note past tense
jdoe: So the answer is no that we don't know what the cause was.
RandalSchwartz: bah, i always end up screwing the date in some way
RandalSchwartz: so, no further issues now?  (i don't see any)
arenlor: cause is unknown, but i simply suspect after 960+ days of uptime, something just got kinked.  I've analyzed enough logs and i'm writing the resolution as "Power cycling the server fixed the high I/O wait issue" and leaving it at that.  I'll take 960+ days of uptime and move on... :)
up_the_irons: quitters never win and winners never quit...
i usually win more sleep if i quit early
:o
yeah - seems fine now
RandalSchwartz: cool
sako: so looks like you didn't make it to la dev ops? ;)
stop spying on me
lol
up_the_irons, turns out we have a few friends in common, sako being one
and Lars
pjs: you know sako and lars?  hah, nice
pjs: do u go to LADevOps at all?
perhaps pjs is there now...
we had 75 nicks in here earlier, a record!
up_the_irons: still do =)
Webhostbudd: still do wut? :)
oh, 75 nicks!
Hmm.
How many VPSes?
CaZe: I have two, what about you?
Infinity.
Nice trick, gave an interger overflow on the billing platform I'm sure.
holy crap, 76 nicks!
up_the_irons: Yeah, so I decided to check out ARP Networks.
Aerosonic: ah, your nick looked new
:)
although it sounds oddly familiar...
I used to be on Linode?
Maybe you know me from their old support chan
Aerosonic: hmm.. i was never on their old support chan; i used to be in the slicehost one...  oh well, doesn't matter )
:)