slh64 has quit [Quit: gone]
black_ant has quit [Ping timeout: 260 seconds]
kakaka has quit [Ping timeout: 240 seconds]
Tapper has joined #openwrt-devel
cheakoirccloud has joined #openwrt-devel
hbug___ has joined #openwrt-devel
hbug__ has quit [Ping timeout: 240 seconds]
xback has quit [Ping timeout: 268 seconds]
xback has joined #openwrt-devel
xback has quit [Ping timeout: 260 seconds]
xback has joined #openwrt-devel
tobleminer-tSYS has quit [Quit: AS4242423214]
tobleminer-tSYS has joined #openwrt-devel
Dracos-Carazza has joined #openwrt-devel
victhor has quit [Ping timeout: 248 seconds]
Grommish has quit [Read error: Connection reset by peer]
Grommish has joined #openwrt-devel
Grommish has quit [Read error: Connection reset by peer]
Grommish has joined #openwrt-devel
ashkan has joined #openwrt-devel
snh has quit [Read error: Connection reset by peer]
snh has joined #openwrt-devel
nast has joined #openwrt-devel
cheakoirccloud has quit [Quit: Connection closed for inactivity]
<Grommish> When I try to do a Package/postinst, it tries to run it at compile time. Anyone give me a hint as to what I'm missing? https://gist.github.com/Grommish/8ec171ec0a49187a76c7879cd89e3be8
ashkan has quit [Ping timeout: 248 seconds]
valku has quit [Quit: valku]
Tapper has quit [Ping timeout: 268 seconds]
<mangix> turns out the docker openwrt package alone is 14MB
<mangix> that is shockingly large
[mbm] has joined #openwrt-devel
mattsm has quit [Quit: The Lounge - https://thelounge.chat]
<mangix> wonder how much of a bad idea it is to use a flash drive as /overlay
<aparcar[m]> mangix: talk to dango, he's building a much smaller docker replacement
mattsm has joined #openwrt-devel
<Thermi> mangix: docker is written in Go and that has no dynamic libs.
<Thermi> What else did you expect?
Grommish has quit [Read error: Connection reset by peer]
Grommish_ has joined #openwrt-devel
Grommish_ is now known as Grommish
nitroshift has joined #openwrt-devel
Acinonyx_ has joined #openwrt-devel
rmilecki has joined #openwrt-devel
Acinonyx has quit [Ping timeout: 260 seconds]
Grommish has quit [Read error: Connection reset by peer]
Grommish has joined #openwrt-devel
ashkan has joined #openwrt-devel
Grommish has quit [Read error: Connection reset by peer]
Grommish has joined #openwrt-devel
paulcarroty has joined #openwrt-devel
slh64 has joined #openwrt-devel
Grommish has quit [Read error: Connection reset by peer]
Grommish has joined #openwrt-devel
Acinonyx_ has quit [Ping timeout: 265 seconds]
Acinonyx has joined #openwrt-devel
<mangix> aparcar[m]: do tell
<mangix> Thermi: I run docker on a different host that does not run OpenWrt. Don't really pay attention to sizes there.
<mangix> lxc could be used as an alternative but that's not docker :)
<mangix> alright. make package/x/refresh is officially difficult
Acinonyx_ has joined #openwrt-devel
Acinonyx has quit [Ping timeout: 246 seconds]
dedeckeh has joined #openwrt-devel
kakaka has joined #openwrt-devel
ivanich has joined #openwrt-devel
black_ant has joined #openwrt-devel
black_ant has quit [Changing host]
black_ant has joined #openwrt-devel
dottedmag has joined #openwrt-devel
<dottedmag> I'm interested in building a firmware based on OpenWRT for a non-network-router device — a e-book with eInk screen. That includes a number of packages (these can go to a separate feed) and a new (actually, old resurrected) target. I'd like to keep this firmware as close to OpenWRT as possible and merge everything relevant back to OpenWRT.
<dottedmag> So, should I submit to OpenWRT support for a target that's out of scope, or should I keep it separate?
<dottedmag> The target in question is Ingenic XBurst MIPS. It was dropped from OpenWRT some time ago, and I'm not sure how useful it is here.
rsalvaterra1 has quit [Quit: Leaving.]
rsalvaterra has joined #openwrt-devel
<paulcarroty> dottedmag, submit it for sure
<dottedmag> great, will do.
<dottedmag> The target does not have to be fully functional to be merged, right? I've got a ton of old kernel patches, so it might take some time to sort and forward-port them all.
<zorun> dottedmag, drop an email to openwrt-devel with details about the target (and how much work is likely needed to catch up), you will hopefully get more informed feedback
<dottedmag> thanks
<zorun> also, would that target have any in-tree device if merged back into openwrt? that's an important consideration
<zorun> we have limited build resources, and targets that don't have many devices tend to bit-rot much faster
poljar has joined #openwrt-devel
poljar1 has quit [Ping timeout: 260 seconds]
<dottedmag> sorry, what's in-tree device? the device openwrt supports? No. Previous incarnation had only another e-book as subtarget, bitrot (due to stale kernel) and was removed.
<dottedmag> OTOH I can set up a CI node for making sure that at least the configuration from master builds.
Tost has joined #openwrt-devel
<zorun> yes, the devices supported in openwrt
<paulcarroty> zorun, Github/Gitlab CI open source credits aren't an option?
<dottedmag> For a single build configuration I don't see a problem setting up a GH, GitLab, CircleCI or a privately hosted CI build - whatever turns out to be the easiest.
<dottedmag> I'll need that for full eBook image anyway.
<rmilecki> dottedmag: you should probably use "source-only" flag for such specific target
<rmilecki> dottedmag: if it doesn't go with tons of patches (making OpenWrt harder to maintain), i think it may be accepted
<paulcarroty> what's ebook we're talking about?
<rmilecki> dottedmag: FEATURES:=source-only
<dottedmag> paulcarroty: I'm resurrecting an old project of mine, OpenInkpot, and picking up the target it supported best (and HW for which I still have): https://wiki.mobileread.com/wiki/N516 - if this goes well, I'll try getting more eBooks supported.
<dottedmag> rmilecki: got it, thanks
Borromini has joined #openwrt-devel
poljar1 has joined #openwrt-devel
Namidairo has quit [Read error: Connection reset by peer]
poljar has quit [Ping timeout: 265 seconds]
Namidairo has joined #openwrt-devel
<paulcarroty> dottedmag, cool device but ancient. something like modern kobo or kindle will be much more interesting for contributors. guess they're locked down, sadly.
<dottedmag> Old Kindles weren't - I still have a custom board that connects to a serial port in Kindle Keyboard that does not even require one to open the device.
<dottedmag> New Kindles are.
<dottedmag> I'll get to it once I have N516 running. Maybe other eBook manufacturers are not so interested in locking down their things.
<dottedmag> paulcarroty: Yes, that's the "old kindles". I don't think Oasis and newer populate serial port header anymore, and their upgrade process seems to be locked down.
<dottedmag> Though a nice hack to get old ones for cheap to play with :)
<dottedmag> Thank you Amazon!
opal has quit [Ping timeout: 240 seconds]
ashkan has quit [Ping timeout: 260 seconds]
ashkan has joined #openwrt-devel
ashkan has quit [Ping timeout: 260 seconds]
nick[m] has quit [Quit: Idle for 30+ days]
csrf has quit [Quit: ZNC 1.7.2+deb3 - https://znc.in]
csrf has joined #openwrt-devel
Neighbor11111116 has quit [Ping timeout: 240 seconds]
opal has joined #openwrt-devel
Neighbor11111116 has joined #openwrt-devel
opal has quit [Remote host closed the connection]
opal has joined #openwrt-devel
victhor has joined #openwrt-devel
dedeckeh has quit [Quit: Ping timeout (120 seconds)]
pine127 has quit [Remote host closed the connection]
kontaxis has quit [Remote host closed the connection]
kontaxis has joined #openwrt-devel
pine127 has joined #openwrt-devel
AnimaSMR has joined #openwrt-devel
AnimaSMR has left #openwrt-devel [#openwrt-devel]
<rsalvaterra> Guys, does backport-5.4/020-backport_netfilter_rtcache.patch make sense? It was never even merged upstream, it wasn't forward-ported to 5.10, and I have this faint idea that route caches were dropped in favour of RCU route lookups, quite a while ago.
<stintel> rsalvaterra: when in doubt, measure the difference?
<karlp> well, ask who committed it to that directory?
<karlp> backports are pretty explicitly meant to be backports...
plntyk has quit [Remote host closed the connection]
<rmilecki> karlp: someone took it for a backport as it was sent for upstream inclusion
<rmilecki> but patch was rejected after all apparently
Borromini has quit [Quit: Lost terminal]
<stintel> so first we should figure out if it's still applicable / relevant for 5.10. if it is, test performance impact. if it's substantial it should be resubmitted and moved to pending-5.10 ?
<rmilecki> sounds sane
<rmilecki> testing should involve few targets
<rmilecki> and comparing with offloading probabl
<rsalvaterra> I don't know (yet, at least) how to test that, though. I stumbled upon that because I found some unused kconfig symbols.
<rsalvaterra> Namely, CONFIG_IP6_NF_QUEUE (dropped in 3.5), CONFIG_IP6_NF_TARGET_LOG (dropped in 3.4), CONFIG_IP_NF_MATCH_DSCP (dropped in 2.6.19), CONFIG_NF_CONNTRACK_IPV4 (dropped in 4.19), CONFIG_NF_CONNTRACK_IPV6 (dropped in 4.19) and CONFIG_NF_CONNTRACK_RTCACHE (which was never merged).
plntyk has joined #openwrt-devel
<rsalvaterra> stintel, rmilecki, do we really want to pay the price of that patch? It's 32 bytes *per conntrack entry* (on x86-64). Source: https://marc.info/?l=linux-netdev&m=141805298717637&w=4
ashkan has joined #openwrt-devel
astylian_ has joined #openwrt-devel
<rmilecki> rsalvaterra: how can we tell without testing speed improvement first? :)
<stintel> rsalvaterra: 112 to 147 Mbps tells me yes
<rmilecki> we need more tests I believe
<rmilecki> and maybe verify if it still works at all
<rsalvaterra> rmilecki: I have my doubts it's still relevant… but without measuring, it's just handwaving.
astylian has quit [Read error: Connection reset by peer]
paulcarroty has quit [Remote host closed the connection]
<rmilecki> rsalvaterra: get two fast machines, connect 1 to WAN, connect one to LAN, setup IPs, make router do NAT between LAN & WAN
<rmilecki> rsalvaterra: test speed without and with patch
<stintel> no need
<stintel> rmmod the module
<stintel> and you might be able to leverage netns and test on a single machine but I've yet to get some real experience with that
<rsalvaterra> rmilecki: Not really feasible on my side… :( If someone else tests it, I won't mind do the other part of the job (backporting or dropping, depending on the result). How about it? :)
<stintel> my fast machines are using my L3/10GbE switch as GW so not super easy for me either
<stintel> hmmz
Tapper has joined #openwrt-devel
<rsalvaterra> One day, I'll have a home lab. Today isn't the day, unfortunately. :)
<stintel> oh wait, I just revived my previous box
<stintel> I can move it to a different VLAN
<rmilecki> rsalvaterra: i'll keep this patch in mind, but can't promise when I find time to test it
<rsalvaterra> stintel: Oh, I decided to go with cat.6a cabling on the new flat. Best bang for the €, at the moment… :P
<rsalvaterra> rmilecki: No worries, thanks! :)
<stintel> yeah I have some ugly 30m cat6a running next the walls :P
<stintel> but I'm way past being bothered by that
<rsalvaterra> Eheheh!
<stintel> running 7.1.2 dolby atmos in a rental apartment ...
<stintel> cables, cables everywhere!
<rsalvaterra> You're officially crazy.
<stintel> not to mention long USB cables to power ESP32/ESP8266/RPi0W's with various sensors etc :P
<rsalvaterra> I'm not going past 5.1 in a flat. Ever.
<stintel> well I went from 7.1 to 5.1.2 and that was awesome... for the few atmos sources + the amp didn't support the front height upfiring channel with DTS:X
<stintel> so I got tempted and bought a new amp that can do 7.1.2 (the previous one was choice between 7.1 or 5.1.2)
<karlp> stereo's where it's at :)
<rsalvaterra> I'm still at stereo too. :P
<stintel> actually it can do 7.2.4 but then I need 1 stereo PA extra
<stintel> which I already built (DIY kit) few weeks ago
<stintel> so now I am tempted to test front wide with the speakers from the bedroom 😂
<rsalvaterra> stintel: I'd hate to be your neighbour. :P
<stintel> well I have a duplex penthouse, the fun stuff is in the top floor, and no neighbor directly below me so there's 2 floors between me and the next
<rsalvaterra> Ah, ok, that's fine.
<stintel> and I put the sub on acoustic isolation stands, which greatly reduces the bass spreading via the walls
<stintel> ah, that reminds me I'm still waiting for my new speakers :(
<rsalvaterra> stintel: Do you use Kodi as media centre?
<stintel> rsalvaterra: what else is there ;)
<stintel> I'm using XBMC since Xbox1
<rsalvaterra> :D
<stintel> I even have the XBMC hoodie!
<stintel> the one they were selling at the time of the rebrand
<rsalvaterra> I started with OpenELEC, now I'm on LibreELEC. :P
<stintel> I'm with CoreELEC because it worked best on VIM3 at the time I got it
<stintel> but I'm not a fan of any of them
<stintel> I should continue my meson target and update my kodi for openwrt packages
<rsalvaterra> Kodi on OpenWrt? Heresy. o_O
<rsalvaterra> *facepalm*
<stintel> my main problem with *ELEC is that it's read-only and can't really use it for anything else than kodi
<stintel> I'm allergic to those kind of things :P
<stintel> there's a reason I'm running OpenWrt
<rsalvaterra> Oh, I love statelessness. ;)
<rsalvaterra> But that's the whole point… It's supposed to be an appliance.
<stintel> well hence my reason for "porting" it to OpenWrt
<stintel> but there was very little interest, and some demotivating response even
<rsalvaterra> Sure, it you want to run OpenWrt on everything… :)
Acinonyx_ has quit [Ping timeout: 240 seconds]
<stintel> OpenWrt on embedded, Gentoo on powerful stuff
<rsalvaterra> I don't know… for everything involving GPUs, I prefer to run bleeding edge kernels. 5.10 is a bit long in the tooth already.
<stintel> well all my rpi's run headless for example
<rsalvaterra> Heh… I had a 1 B+ as a print/scan server, running OpenWrt. Unfortunately I had issues with SANE (the scanner would hang if you cancelled the scan, had to unplug the USB to fix it), so I had to decommission it.
Acinonyx has joined #openwrt-devel
<stintel> ah I have a network printer/scanner
<stintel> that does double-sided printing in double-pass and double-sided scanning in single pass. love that thing
dedeckeh has joined #openwrt-devel
valku has joined #openwrt-devel
nitroshift has quit [Quit: Gone that way --->]
muhaha has joined #openwrt-devel
Tapper has quit [Ping timeout: 240 seconds]
plntyk has quit [Remote host closed the connection]
plntyk has joined #openwrt-devel
Slimey__ is now known as Slimey
<rsalvaterra> Yay! My zstd-as-default patch for ubifs was accepted upstream. :) https://lore.kernel.org/linux-mtd/CAFLxGvwhtRY-6kT-sN=AgjvyssHb5qdTS6WQHkhKu3YrPuTkdw@mail.gmail.com/
<stintel> *clapclap*
* rsalvaterra is on a crusade to kill both lzo and zlib …:P
* karlp appreciates the portability of some boring traditional uncool things...
<karlp> I mean, "kill zlib" seems excessive...
<rsalvaterra> It's not killing… it's putting it out of its misery. :P
* karlp infers from the lack of alterantive c impls, and from the omitted benchmarks that program size and memory usage are not goals of zstd, and may be a reason why some people still use "useless, inferior, legacy" trash :)
Zoxc has joined #openwrt-devel
<rsalvaterra> karlp: Those are configurable… If you want high compression ratios, you pay with RAM and CPU time. ;)
* ldir wants the moon on a stick - take loads of ram and inordinate amounts of cpu
rsalvaterra1 has joined #openwrt-devel
rsalvaterra has quit [Ping timeout: 248 seconds]
Neighbor11111116 has quit [Ping timeout: 260 seconds]
ashkan has quit [Quit: Konversation terminated!]
Neighbor11111116 has joined #openwrt-devel
rsalvaterra1 has quit [Quit: Leaving.]
rsalvaterra has joined #openwrt-devel
Dracos-Carazza has quit [Ping timeout: 248 seconds]
<karlp> rsalvaterra: it's goign to replace things like https://github.com/atomicobject/heatshrink anytime soon though :)
<karlp> it's _not_ going to replace....
<karlp> (admittedly, not zlib either, but there's tiny zlib compatible implementations too)
paulcarroty has joined #openwrt-devel
Tapper has joined #openwrt-devel
Zero_Chaos has joined #openwrt-devel
<Zero_Chaos> Package iptables-nft is missing dependencies for the following libraries:
<Zero_Chaos> libiptext6.so
<Zero_Chaos> trying to build on the 21.02 branch
Tost has quit [Ping timeout: 265 seconds]
danitool has quit [Quit: Cubum autem in duos cubos, aut quadratoquadratum in duos quadratoquadratos]
pgwipeout[m] has quit [Quit: Idle for 30+ days]
finsternis has quit [Remote host closed the connection]
mm__ has joined #openwrt-devel
<rsalvaterra> karlp: Uhh… that's a library for tiny MCUs, for sure. I'm positive zstd isn't supposed to be used on 80C51 and friends… :P
<karlp> wel, zlib does get used in those places, or at least, zlib compatible :)
muhaha has quit [Quit: Connection closed]
<karlp> so you can use the same tooling in both ends easily
<karlp> anywya, zstd is rad, no complaints, just maybe we don't need to kill zlib :)
<rsalvaterra> Oh, I don't mean killing in the general case! :)
Dracos-Carazza has joined #openwrt-devel
Slimey has quit [Read error: Connection reset by peer]
Slimey has joined #openwrt-devel
Neighbor11111117 has joined #openwrt-devel
Neighbor11111116 has quit [Ping timeout: 240 seconds]
Dracos-Carazza_ has joined #openwrt-devel
Dracos-Carazza has quit [Ping timeout: 265 seconds]
Neighbor11111117 has quit [Ping timeout: 248 seconds]
Neighbor11111117 has joined #openwrt-devel
Neighbor11111117 has quit [Read error: Connection reset by peer]
<rsalvaterra> stintel, rmilecki, on a deeper analysis, I'm lead to believe 020-backport_netfilter_rtcache.patch is just broken.
<rsalvaterra> I see several instances of #if IS_ENABLED(CONFIG_NF_CONNTRACK_IPV6) in that patch…
<stintel> rsalvaterra: I have that module loaded on all my routers
<rsalvaterra> … CONFIG_NF_CONNTRACK_IPV6 has been dropped in Linux 4.19.
<rsalvaterra> stintel: Running which OpenWrt versions?
<stintel> rsalvaterra: master
<rsalvaterra> Well, it surely isn't doing anything for IPv6, then.
<stintel> with the strongswan breakage I noticed I haven't upgraded in a while (like 45 days) but recent enough
<stintel> rsalvaterra: does anyone use IPv6 then?
<stintel> (SCNR)
<stintel> let
<rsalvaterra> :P
<stintel> let's see if I can do that setup to test
<stintel> hmmmz should have ordered some extra Cat6a with those DACs :/
<stintel> and I need to find the sheet of paper with the VLAN/subnet convention I came up with when configuring my 10GbE switch
<stintel> welp
* rsalvaterra was very happy to know Cat.6A doesn't require special tools for termination…
<rsalvaterra> You're going to test conntrack at 10 Gb/s? Now that's torture. :P
<stintel> :D
<stintel> no no, on my spare ERL
<stintel> so that there's no influence from real traffic
<stintel> oh, I actually made an ods!
<rsalvaterra> Uh?
<stintel> open document spreadsheet (with the VLAN/subnet info)
<rsalvaterra> Ah! I was totally out of context. :P
<stintel> uhhh I wanna play with the CN6640 :(
astylian_ has quit [Ping timeout: 240 seconds]
<Grommish> rsalvaterra: I've got a 350m spool of cat6a sitting next to me :D But i only use 2-part rj45 connectors, single piece connectors aren't worth the hassle
<Grommish> Or, can always cheat and just get a keystone jack and punch it in
Acinonyx_ has joined #openwrt-devel
Acinonyx has quit [Ping timeout: 265 seconds]
shibboleth has joined #openwrt-devel
mm__ has quit [Quit: Connection closed for inactivity]
lukedashjr has joined #openwrt-devel
luke-jr has quit [Ping timeout: 265 seconds]
lukedashjr is now known as luke-jr
Borromini has joined #openwrt-devel
Neighbor11111117 has joined #openwrt-devel
cheakoirccloud has joined #openwrt-devel
goliath has quit [Quit: SIGSEGV]
Acinonyx has joined #openwrt-devel
j007 has joined #openwrt-devel
Acinonyx_ has quit [Ping timeout: 240 seconds]
paulcarroty has quit [Ping timeout: 248 seconds]
j007 has quit [Quit: Leaving]
madwoota has quit [Ping timeout: 265 seconds]
madwoota has joined #openwrt-devel
Tost has joined #openwrt-devel
zkrx has quit [Ping timeout: 265 seconds]
astylian_ has joined #openwrt-devel
<stintel> argh, been on the phone for over 2h, let's see if I can get to testing
zkrx has joined #openwrt-devel
ivanich_ has joined #openwrt-devel
ivanich has quit [Remote host closed the connection]
Hayate has quit [Ping timeout: 268 seconds]
Tycale has quit [Ping timeout: 268 seconds]
Nyakajima has joined #openwrt-devel
Tycale has joined #openwrt-devel
astylian_ has quit [Ping timeout: 240 seconds]
<stintel> pffft can't even get the firewall rules straight
<rsalvaterra> Ouch… what happened? :P
<stintel> I've been using this device for trying to reproduce the horrible WG breakage, with doing that renamed some interfaces etc
<stintel> I allowed forwarding from test to wan but in fact I needed to allow from test to lan
<stintel> I should probably start documenting my network a bit
<stintel> had been eyeing netbox but overly complex imo
<stintel> I mean to maintain
<stintel> different components etc
<stintel> [ 4] 0.00-10.00 sec 55.3 MBytes 46.4 Mbits/sec 30 sender
<stintel> lulz
<rsalvaterra> That's… not good… :P
<stintel> sqm :P
<stintel> now even with sqm disabled it's not good: [ 4] 0.00-10.00 sec 141 MBytes 119 Mbits/sec 29 sender
<stintel> and that's with flow_offloading + rtcache
<stintel> I don't remember the ERL being *that* slow
<stintel> it's without NAT even
<stintel> ah no it's without rtcache
<stintel> modprobed it, no noticeable difference
astylian_ has joined #openwrt-devel
<stintel> it seems to make a difference without flow offloading
dedeckeh has quit [Ping timeout: 240 seconds]
<stintel> let me make some ascii art gist with some stats
Borromini has quit [Quit: Lost terminal]
<stintel> also added with flow_offload now
<stintel> so it seems to me that nf_conntrack_rtcache is useless in combination with flow offload
<stintel> and flow offload improves performance much better
<stintel> so given that + fact that nf_conntrack_rtcache was never accepted upstream and flow_offload was ... we should probably just drop nf_conntrack_rtcache and focus on flow_offload
<stintel> ahhh and rmmod nf_conntrack_rtcache on x86/64 -> crash
<stintel> well maybe not a crash
<stintel> but I almost can't believe those speeds without rtcache and without offloading
<stintel> 35Mbps?!
<stintel> even with rtcache 65Mbps seems crazy
shibboleth has quit [Quit: shibboleth]
<stintel> 26|07:14:04< stintel> I'm doing 100/60 on EdgeRouter Lite
<stintel> 26|07:14:08< stintel> with sqm
<stintel> that was december 2016. smells like a regression. then again I have always suspected a general network regression at some point
<stintel> also with my apu2
<stintel> I can probably try some official images, wiped config, and connect the spare ERL directly to 2 physical machines to do some comparisons between older openwrt/lede releases and now
<rsalvaterra> stintel: Back, sorry for the delay.
<rsalvaterra> Heh… rmmod is mostly best-effort, I think. :P
<rsalvaterra> So, there's the reason why the rtcache patch wasn't merged. :)
<stintel> well it predates flow offload afaik, and it does improve performance
ivanich_ has quit [Quit: Konversation terminated!]
<stintel> pretty sure that's the reason we kept it
<rsalvaterra> In any case, it's probably incompatible with SQM, just like flow offloading…
<stintel> that's an interesting thought
* stintel tries speedtest without sqm and without flow offload on his main router
<rsalvaterra> Wait, you had SQM enabled?
<stintel> no no the gist is without sqm
<rsalvaterra> Phew… :P
<stintel> and on a spare ERL, not doing an real traffic
<stintel> but if disabling flow offloading shows a similar slowdown as enabling sqm on my main router ...
<stintel> which kind of looks like it
<stintel> so maybe enabling sqm just breaks flow offloading
rmilecki has quit [Read error: Connection reset by peer]
rmilecki has joined #openwrt-devel
black_ant has quit [Ping timeout: 240 seconds]
<rsalvaterra> It does. I had some really weird latency spikes and wild throughput variability with SQM and flow offloading both enabled. Took me a while to figure out the cause.
rmilecki has quit [Ping timeout: 265 seconds]
indy has quit [Read error: No route to host]
indy has joined #openwrt-devel
<stintel> hmmm.. would it be awkward if I ordered another pizza just 10 minutes after the first one was delivered :P
<rsalvaterra> I don't think they're in the business of judging… :P
<stintel> ERROR: package/feeds/routing/bird2 failed to build.
<stintel> derp
Tapper has quit [Ping timeout: 260 seconds]
<jow> stintel: rtcache is indeed obsolete since there's upstream endorsed flow offloading
<jow> it was kind of a stop gap solution to speed up slow ar71xx/ath79
<stintel> rsalvaterra: feel like sending a patch? jow can ACK it and I can merge it ;)
<rsalvaterra> jow: Yay, thanks for the authoritative explanation. :)
<rsalvaterra> stintel: Sure, let me cook it up. Should be faster than that pizza… :P
<stintel> should we even nuke it from 21.02?
<stintel> rsalvaterra: hahaa
<jow> at some point netfilter performance nosedived with newer kernel versions
<stintel> rsalvaterra: yeah, no dr oetker in storrage :P
<rsalvaterra> Oh…
<jow> but before flow offloading existed
* rsalvaterra remembers already having bumped 5.10 to 5.10.28.
<jow> so nbd merged the rtcache stuff which never ended up upstream
<stintel> I might do some effort tomorrow to rip out the spare ERL from my rack, throw it between two beefy physical machines, and do some proper testing with default config
<stintel> I'm no longer working full time, so I should have some more time for OSS, but will try to spread it between that and life
<jow> as for flow offloading and sqm/qos... isn't that conflicting goals by definition?
<jow> flow offloading is essentially kernel subsystem bypassing for most packets
<jow> so no wonder stuff can't mangle these properly anymore
<rsalvaterra> jow: Yes, especially when the SQM flow dissector does NAT lookups… :P
<rsalvaterra> The connection still "works", but the performance is all over the place.
<philipp64> how do you fail out of a init.d script that uses procd if you find a broken configuration?
<jow> philipp64: one way is using procd_set_param error "whatever"
<jow> this will still submit the service state to procd and register it, but procd will nto launch the process and indicate the errors in syslog instead
<philipp64> how does that affect the flow of execution? does that return?
<karlp> I've always just returned before procd_open_service.
<jow> yes, it will return like any other call
<jow> the other way is what karlp said, just don't submit the service
<jow> means avoid calling procd_open_service or procd_open_instance (which implies procd_open_service)
<jow> so essentially "return" before that point is reached
<rsalvaterra> stintel: Refreshing the kernel patches (after deleting the rtcache patch)…
<jow> time to hit the bed. bbl
<guidosarducci> stintel: hey there, still seeing that weird perf behaviour w/wo sqm?
<stintel> jow: nn!
<stintel> guidosarducci: well yeah
<stintel> guidosarducci: I don't have a super reliable way to test (speedtest.net), but it's pretty consistently much slower with SQM enabled
<stintel> guidosarducci: but haven't done more debugging/reading/experimenting since you threw me that forum link
<stintel> ~745/605 w/o SQM and ~477/205 w/ SQM
<stintel> but as I said, it's using speedtest.net (to a server of my own ISP though)
* stintel tries to find that forum post in his 200 browser tabs :/\
<rsalvaterra> stintel: You've got mail. :P
<stintel> I noticed :p
<stintel> too bed jow hit the sack already
<rsalvaterra> No rush… 5.4 is going the way of the dodo, anyway… XD
<stintel> but I can ack it :)
<stintel> rsalvaterra: thanks, acked
<stintel> I should probably hit the sack too. not working tomorrow but weather is improving so summer wheels
<rsalvaterra> stintel: Oh, I remembered…
<rsalvaterra> We probably should backport this to 21.x, right?
<stintel> I think it makes sense
<rsalvaterra> The patch I sent is against master.
<rsalvaterra> (But it should apply to 21.x.)
<mangix> rsalvaterra: in what way is 5.4 going the way of the dodo?
<rsalvaterra> mangix: In the master's way… :P
<rsalvaterra> … hopefully…
<mangix> right. but not in kernel.org terms
<mangix> 4.4 will apparently be supported until 2037...
<stintel> o_O
<rsalvaterra> 2037?! Jesus…!
<mangix> yeah. japan wants super long LTS
<mangix> for their street lights and whatnot
<mangix> speaking of japan, I should probably email the NILFS2 maintainer...
<rsalvaterra> Wait… they are *updating* their kernels in the street lighting system? Shouldn't that be a completely hermetic system? o_O
<mangix> beats me
Tapper has joined #openwrt-devel
<rsalvaterra> mangix: Gregkh's presentations are always brilliant. :)
<mangix> funny, he said he'll probably have a stop light on his desk eventually
<mangix> it always amazes me just how many changes get backported
Tost has quit [Ping timeout: 252 seconds]
<rsalvaterra> mangix: Wouldn't shock me. Have you seen his box of USB devices? :P
<mangix> i have not
<mangix> I wonder if ksmbd should be moved to base
<Mister_X> I have an old style traffic light with incandescent light bulbs, and thought about using it in my garage to let me know when it's good enough to close it
<mangix> doing updates in the packages feed is a nightmare
<Mister_X> there is a company that does boards to automatically switch between the different colors, but that's the boring thing to do
<stintel> hurricos: I've got two of those cards. I didn't find 1GiB NAND, but didn't look very hard either. found the 64Mb NOR, and a 256Kb and a 512Kb I2C serial flash (never even heard of those)
<stintel> hurricos: right now I'm waiting for some DAC cables so I can mess around with tftp from u-boot. I can probably help with sysupgrade