jasongrossman has quit [Ping timeout: 252 seconds]
lassulus has quit [Ping timeout: 268 seconds]
tertl3 has quit [Quit: Connection closed for inactivity]
sir_guy_carleton has quit [Quit: WeeChat 2.2]
jasongrossman has joined #nixos-chat
<MichaelRaskin>
drakonis: re: FHS boot: so you want just to create a user with login shell being «go into FHS»?
<MichaelRaskin>
That sounds quite easy to script and may solve your desired use case.
jD91mZM2 has joined #nixos-chat
jackdk has quit [Ping timeout: 250 seconds]
<drakonis>
yeah?
<MichaelRaskin>
Do you use declarative or imperative user management? I think you could write a shell script that runs the enter-FHS script (dunno, installed into profile maybe?), then set this script as the login shell of a user
<MichaelRaskin>
If you are really reckless, you can give that user the same home directory and the same UID as another user
lassulus has joined #nixos-chat
<gchristensen>
wow
<gchristensen>
this is amazing
<MichaelRaskin>
Any details?
MichaelRaskin has quit [Quit: MichaelRaskin]
__Sander__ has joined #nixos-chat
Ralith has quit [Remote host closed the connection]
thefloweringash has quit [Write error: Connection reset by peer]
sphalerit has quit [Remote host closed the connection]
Ericson2314 has quit [Remote host closed the connection]
<gchristensen>
the fhs script as a login shell =)
sphalerit has joined #nixos-chat
<infinisil>
,locate g++
<{^_^}>
Found in packages: gcc, gcj, gccgo, gfortran, gcc_debug, ccacheWrapper, gcc-unwrapped, distccWrapper, bashCompletion, distccMasquerade, gcc-arm-embedded-4_7, mentorToolchains.armEabi, mentorToolchains.armLinuxGnuEabi
Ericson2314 has joined #nixos-chat
Ralith has joined #nixos-chat
thefloweringash has joined #nixos-chat
__monty__ has joined #nixos-chat
jD91mZM2 has quit [Quit: WeeChat 2.2]
<gchristensen>
one day I'm going to write something about combining timers with ConditionACPower to run user tasks without draining your battery.
aminechikhaoui has quit [Ping timeout: 264 seconds]
<sphalerite>
gchristensen: nice!
aminechikhaoui has joined #nixos-chat
jasongrossman has quit [Quit: ERC (IRC client for Emacs 26.1)]
jD91mZM2 has joined #nixos-chat
<andi->
Vacation is a terrible thing... I haven't eaten all day since I am able to work on my projects... I ignore everything else /o\
<infinisil>
andi-: Can relate.. also sometimes no sleep for the same reason
<andi->
I sleep from 2am - 8am and can do that for weeks on vacation.. When I have to go back to work that would totally lead to me being an unproductive zombie.. maybe work is just bad :D
Lisanna has quit [Remote host closed the connection]
<gchristensen>
deploying software in to ISP's edge nodes
qyliss has joined #nixos-chat
jasongrossman has joined #nixos-chat
<edef>
gchristensen: :3
<maurer>
gchristensen: fastly has had edge computing for a while too, dunno what the differences are between its offering and mutable's though
<gchristensen>
fastly's edge computing happens in their edge DC. as far as I understand it, the mutable edge DCs are the ones on the other end of the cable connecting your home to the ISP
<maurer>
Oh, so they make deals with ISPs?
<gchristensen>
yeah
<maurer>
You'd think the ISPs would just sell that direct, but I don't know the economics of that
<gchristensen>
you
<gchristensen>
you-the-customer would have to buy it from zillions of ISPs
<maurer>
Not that many? There are like, 3 real ones per country
<samueldr>
and how many countries?
<maurer>
and presumably mutable only has contracts with the couple big ones anyways
<__monty__>
We don't have 3 even.
<samueldr>
(woosh) pretty sure there are more than three countries
<gchristensen>
at any rate, you-the-customer probably don't want to buy N Us per super-local DC, whereas mutable says "here, we have presense in all of them"
<maurer>
samueldr: pretty sure if you've scaled to the point that you care about perf, not just "it works" in multiple countries, you probably have the time to contract out, but sure.
<maurer>
gchristensen: OK
<gchristensen>
some applications require extremely local access to users, but the company may be young and not able to afford it.
<edef>
maurer: there are about a dozen just in the US
<gchristensen>
also they use nixos :eyes:
<edef>
maurer: the number's been going down due to acquisitions, but if you look inside those larger companies you can see Conway's law at work
<edef>
maurer: so.. integrating with those is nontrivial, getting hardware at the edge is nontrivial
<maurer>
edef: Isn't the US pretty just Comcast/Verizon most places, and some places Cox?
<maurer>
edef: Also, I'd assume that mutable.io, unless shown otherwise, probably only contracts with the majors themselves anyways...
<edef>
there's Comcast and Cox, and then there are a variety of smaller MSOs
<gchristensen>
I'm with Spectrum which is not comcast or verizon or cox
<gchristensen>
at any rate
<gchristensen>
dealing with any of them seems crazy, meanwhile I'm very interested in the application
<edef>
i forget what exactly i'm allowed to say in public and what's covered by NDAs, but i can assure you that we cutting deals with a variety of smaller MSOs
<edef>
running stuff at the edge is a problem for them internally as well, they're nowhere near being capable of selling it as a service to others
<maurer>
edef: Oh, are you with mutable?
<edef>
let alone aggregating it effectively so an average developer can make use of it easily
<edef>
i'm the CTO!
<maurer>
Ah, did not know that
<andi->
I would be very interested in the use-cases for hosting at the edge... The website lists a few and latency sensitive applications are a thing but I do not have a clear picture what those would be.
<maurer>
andi-: One well-known one that's not mutable related is netflix
<maurer>
they install caches in DCs with major ISPs in order to avoid bandwidth costs for shipping content across the open internet
<andi->
maurer: I know those.. I have dealt with them.. but isn't that just the metro area not really the edge?
<maurer>
If the cache is in the comcast DC, isn't that pretty much the edge?
<maurer>
You might be able to get slightly edgier, but you're not going to shave more than one or two more hops off
<andi->
there is a difference... their website makes that. Can't remember the term right now.
<andi->
they call it regional DCs
<andi->
and then there is the Edge
<edef>
for an operator like Comcast, there's half a dozen regional DCs or so
<maurer>
Are the netflix caches just in regional DCs then? I kind of assumed they were further out
<edef>
and then hundreds of headends which connect back up to those
<andi->
I've seen a few edge "DCs" in germany that could barely fit the DSL equipment... They cramped some of their own IPTV caches in there but those houses were usually repurposed toilets etc..
<edef>
yep
<edef>
they're crammed
<edef>
only way to get stuff in there is .. well, first a lot of people tell you to get bent, and then you figure out what of their rackspace you could help consolidate
<andi->
Interesting :)
<edef>
places like netflix aren't getting in there for a pretty significant pile of reasons
<andi->
I sometimes regret that I do not want to do any networking anymore. (closed source equipment vendors annoy me)
<maurer>
edef: Even if you guys get in there, wouldn't you not be able to move services to there for most of your customers?
<maurer>
Due to resource/space constraints?
<edef>
correct
<edef>
you end up with a very different style of deployment than "we start up an app in $location, and leave it alone as long as it doesn't die"
<edef>
things end up much closer to "we see your DNS request, and we start figuring out whether we should be bringing an application instance to this edge to meet the latency constraints requested"
<edef>
and you end up with an economics problem because you're dealing with resource scarcity, but that's arguably the least tricky bit of this all (spot markets for compute have sufficient prior art)
<andi->
Another one that interests me: Isn't the distribution of fiber connections making that somewhat obsolete? The latency from me on the couch through half-europe is well beyond 15ms.. I am not sure what kind of ("crazy") applications those must be :)
<edef>
15ms is a very long time in my eyes
<andi->
well the "edge" in my case is 0.4ms away, the wifi jitter is higher, the regional DC is 1ms away, google/aws/cloudflare 1.1ms,
<gchristensen>
&.&
<andi->
not saying there aren't use-cases. Just wondering if it will bve a thing in 10y :)
<andi->
I certainly like the idea.
<gchristensen>
in 10y mutable will be deploying to your TFU :eyes:
<andi->
:D
<edef>
if you're on DOCSIS 3.0 (cable), you're spending a decent amount of time just requesting and waiting for a transmit window
<edef>
like, exceeding 15ms
<edef>
those numbers are coming down, but at that point you're shifting the problem from the access network to the backhaul
<andi->
sure, just wondering what happens and at what point the benefits vs costs doesn't pay off anymore.
<edef>
continental europe has really solid well-peered infrastructure, moving stuff to the edge is somewhat less pertinent there
* gchristensen
is very poorly connected
<andi->
ok, I might be blinded my the "luxory" I am having..
<edef>
like, in continental Europe i can play an FPS that's running on EC2 in Frankfurt and be competitive
<edef>
in the US .. not so much, outside the rare spot where you've got fiber in a major urban centre
<edef>
i'm sitting in London, and hitting my ISP's network edge is 5.5ms best-case (the wifi is sub-millisecond), 6.5ms best-case to a London DC, but the jitter is pretty absurd (nearing 30ms)
<edef>
within the ISP's network, there's submillisecond jitter and i'm spending ~9ms to their network border
<edef>
(9ms worst-case that is)
<edef>
the numbers get way more exciting if you go to the US, and places like ad networks and CDNs are pretty interested in shaving their milliseconds down
<edef>
that's excluding use cases that aren't really possible if you haven't got compute nearby with solid latency guarantees
<andi->
in anyway keep in doing awesome things like that edef! :-)
<andi->
I like that someone is touching that and is going away from the traditional clouds
<edef>
i started out with "fuck, deploying distributed systems really sucks huh" after doing PaaS ops for a while
<edef>
the smug part of me believes orchestration won't get any better as long as it's left to developers rather than ops folk, and most ops folk are siloed into firefighting rather than innovating
<andi->
100% agreed.
<edef>
i've spent .. half a decade on this stuff now, and i finally feel like i'm getting somewhere on real-world deployment of the stuff that seemed obvious years ago
<edef>
and having a lot more of the necessary tooling exist helps a lot too .. not that long ago the only viable way of doing a multitenant container platform at scale was illumos
<edef>
(and thus i remain an illumos nut to this very day - i'm IRCing from inside an illumos zone!)
ekleog has quit [Quit: back soon]
<edef>
(also aside from latency, matching backhaul bandwidth to the ever-increasing amount of bandwidth offered to end users thanks to innovation in the access network is .. not gonna keep up)
<__monty__>
That's the oss continuation of solaris, right?
<edef>
oracle bought sun and killed opensolaris, and illumos forked off from the last release, and is worked on by a lot of the people who quit oracle
ekleog has joined #nixos-chat
<edef>
NixOS is the first thing that has managed to take me away from my first love (and even so, i've run Nix on illumos for shits 'n giggles)
<gchristensen>
in a linux flavored zone?
<qyliss>
NixOS on illumos when
<gchristensen>
as soon as edef does it I guess
<edef>
gchristensen: also that, which has been going more poorly over time as things get hardened
<edef>
vizanto and i had Nix running natively on illumos a few years back
<gchristensen>
edef: as nix becomes hardened, linux hardens, or illumos hardens?
<edef>
nixos hardening
<edef>
my sshd broke and stuff
<edef>
no seccomp in illumos LX zones
<gchristensen>
ahh
<gchristensen>
I'll take that as good news/bad news
<edef>
it's very nice to be able to run DTrace against Linux applications
<edef>
and like .. everything mostly works, systemd works, cgroups (being the knockoff of Solaris Contracts) are implemented
<edef>
if i had the time i'd probably spend some time on making NixOS run well in there, and i'd love to have a Nix-based illumos distro
<edef>
Dyson is pretty much Debian with an illumos kernel and userland
<edef>
but as much as i miss having decent, understandable virtual networking as baseline, and solid containers, and proper storage.. Linux has a pretty solid pile of primitives by now
<edef>
and if the primitive i seek doesn't exist.. there's a lot of infrastructure to use when you're doing kernel programming on Linux
<edef>
illumos very much carries that old-world charm of being a genuine UNIX descendant while still innovating a lot, but it shows its age at times
aminechikhaoui has quit [Ping timeout: 252 seconds]
aminechikhaoui has joined #nixos-chat
drakonis1 has joined #nixos-chat
__Sander__ has quit [Quit: Konversation terminated!]
Lisanna has joined #nixos-chat
Taneb is now known as GHOSTLY_SPOOK
GHOSTLY_SPOOK is now known as Taneb
drakonis1 has quit [Quit: WeeChat 2.2]
sir_guy_carleton has joined #nixos-chat
drakonis_ has joined #nixos-chat
drakonis has quit [Ping timeout: 268 seconds]
MichaelRaskin has joined #nixos-chat
<drakonis_>
MichaelRaskin: i went to sleep
<drakonis_>
i use declarative user management
drakonis has joined #nixos-chat
drakonis_ has quit [Ping timeout: 264 seconds]
<MichaelRaskin>
drakonis_: I dunno, maybe UID collision could be forbidden in such a case
<MichaelRaskin>
But setting login shell of a user to a FHS-chrooting script should still work.
jD91mZM2 has quit [Quit: WeeChat 2.2]
drakonis has quit [Read error: Connection reset by peer]