<gchristensen>
they actually can't be deprovisioned :)
<gchristensen>
ryantm: yes, everything has to be in ram
<__monty__>
Is that a pun on slow loris?
<gchristensen>
which is no doubt a significant limitation
<gchristensen>
__monty__: it literally utilizes the slow loris attack to make these servers boot quickly
<__monty__>
Ah, think I've seen this before. You keep the PXE boot busy until you actually want to boot the machine?
<gchristensen>
right
justan0theruser has joined #nixos-dev
justanotheruser has quit [Ping timeout: 246 seconds]
psyanticy has joined #nixos-dev
__monty__ has quit [Ping timeout: 258 seconds]
__monty__ has joined #nixos-dev
<yorick>
does packet like their servers idling at max power in the pxe thing instead of off?
* gchristensen
shrugs
<yorick>
hm, you could port coreboot to these things, the chipset is supported
<gchristensen>
yeah, I don't know. they already make decisions about which machines are "hot" with ubuntu, and which are in standby and which are off
jtojnar has joined #nixos-dev
layus[m] has joined #nixos-dev
avn has joined #nixos-dev
orivej has joined #nixos-dev
orivej has quit [Ping timeout: 272 seconds]
cransom has joined #nixos-dev
__monty__ has quit [Ping timeout: 245 seconds]
__monty__ has joined #nixos-dev
ajs124 has quit [Quit: Gateway shutdown]
ajs124 has joined #nixos-dev
FRidh2 has quit [Quit: Konversation terminated!]
orivej has joined #nixos-dev
<thoughtpolice>
lol @ slow loris'ing your PXE server
<gchristensen>
anything to save those 5m
<ryantm>
gchristensen: How much of the RAM does the base OS take up?
<gchristensen>
ryantm: maybe we should PM about your project and its requirements?
v0|d has joined #nixos-dev
Jackneill has quit [Remote host closed the connection]
orivej has quit [Ping timeout: 245 seconds]
<aristid>
gchristensen: woah, how does slow loris make servers boot faster?
<gchristensen>
it skips the slow part about booting: booting. it keeps the server right at the moment of deciding how to boot
<simpson>
gchristensen: I am reminded of the classic HTTP technique, "long polling".
<gchristensen>
yeah very similar
<gchristensen>
but I think that typically requires the client to consent
<simpson>
Or at least to not time out.
<aristid>
gchristensen: and you can still use the server in that state?!
<gchristensen>
aristid: what do you mean?
<gchristensen>
it is not usable until it receives the "do the boot" instruction
<aristid>
but that takes less than 5 minutes, unlike the normal boot?
<gchristensen>
it takes only as long as it takes to download and execute your iPXE image. which in my tests is about 45s
<aristid>
so what does the server do normally in the 5 minutes?
<aristid>
let timeouts expire?
<averell>
no, the trick is to wait the 5 minutes at the beginning, then poll until an actual provisioning which is fast
<gchristensen>
normally, the first five minutes are spent initializing the CPU, RAM, disks, NICs, etc.
<gchristensen>
the 5 minutes still happen, but if you keep some hardware hot and at this spot, you experience a 45s boot
<aristid>
gchristensen: is averell's explanation right? because i understand that, but i'm not sure if it's what you mean :)
<aristid>
gchristensen: ok and the reason you don't just boot the system earlier is that you might not have the right ipxe image yet?
<gchristensen>
right, you may not know what you want to boot yet
<aristid>
or does packet only charge after the bootup is complete?
<gchristensen>
as a customer on Packet, you probably don't want to implement this yourself, as you'll be billed for all the time it is sitting at this slow loris stage
<gchristensen>
but Packet might want to implement this
<aristid>
gchristensen: oh so you suggest that they just never powerdown their machines, and instead boot them into slowloris
<gchristensen>
right (but maybe not "never", but keeping some % of their hardware hot in this spot would make a difference)
<averell>
do you contract for them? i think the engineers might hang a wanted poster.
<gchristensen>
I don't
<gchristensen>
and yes, they are hiring :)
<averell>
haha, not that kind.
<gchristensen>
oh! haha
<samueldr>
WANTED for crimes of high availability. offline or online. reward: 99.999% SLA
<arianvp>
who here was collecting totally useless --show-trace's ?
<gchristensen>
the Nix bug tracker please, and CC me :)
<worldofpeace>
i can't really remember if that's really important since that is different from how it was before
<jtojnar>
worldofpeace: the sed-insertion looks fine
<jtojnar>
I am not sure about “GDM dconf profile changed, please update gdm.nix”, how is one supposed to update it?
<worldofpeace>
I don't really see if it would ever change that "user-db:user" wouldn't be the first line, but I guess if it does it would prevent someone from building their system
<worldofpeace>
which reminds me that our gnome-gdm test doesn't block the channel jtojnar
<jtojnar>
worldofpeace: I misread it, now I understand it checks for upstream changes
<zimbatm>
simpson: the achievement is more that they are able to serve a snapshot of 11 million files. in DAT 1.0 a site could only contain a small number of files. IPFS has similar problems I think.
<simpson>
zimbatm: Ah, sure. It's definitely improvement, for sure.
<worldofpeace>
anyone know why the lightdm test is commented out of release-combined.nix?
<worldofpeace>
it's the default DM and it's untested