pr'd :3 github.com/TecharoHQ/an...

also: i'm sorry for spamming upstream issues and prs by just pushing commits (that links to issues and co., which makes those "referenced by this commit" stuff in issues).

i'd love to disable that but i don't think github lets me do that atm

xh (httpie-like command) to localhost:8923 (port anubis is listening on) which proxies a local echo server listening on a unix socket. it is now correctly proxying requests to the actual path, instead of having the target unix socket path prepended.

it is now Wrok.(-ish? i still need to do deployment tests)

commit pushed to github.com/soopyc/anubi..., audits welcome because im sure this is crap (less crap than before since i ran it though oomf)

why are these images so fucking crusty i paid for my own hosting /hj

screenshot of a terminal showing curl requests to a local request echo server. "curling" localhost:8923 (proxied by anubis) results in a GET request to the actual path of the unix socket, while curling the socket directly results in the intended GET request to root.

ooauggh

turns out that gave me a bit of insight into whats actually going on and was quite useful lol, thanks for the suggestion

banning v6 dns queries via resolv.conf's `options no-aaaa` """""fixed""""" the problem, so at least i know where to investigate further.

as always, it was dns.

i feel like this is a nodejs issue. both npm and pnpm times out, other tools like curl and xh falls back to v4.

setting gai.conf didn't help for some reason.

this is weird because pnpm can connect to registry.npmjs.org just fine, and that has a bazillion ip6 a⁴ records.

I've also tried mitmproxy'ing that and it works fine, so I'm really not sure whats going on. could this be a pnpm thing?

turns out it's reproducible outside of nix, it just gives me a socket timeout.

I think it's trying to connect to an ip6 address, of which I don't have an internet facing one so it just hangs forever.

feels a bit too heavy duty for this xd, but I guess it's also a valid method :p

nix build output of derivation "nuscht-search-pnpm-deps", in which pnpm logged 4 warnings about not being able to fetch 2 packages from codeberg, then an error about that but it's a failure instead

i have a feeling nix and codeberg don't really go together, this happens 9 out of 10 times and there's a really slim chance that it actually succeeds instead of hanging for 4 minutes and failing

downloading the exact url with xh completes immediately though

grafana node dashboard for ryo.d.soopy.moe, showing 98.9% cpu usage (of which 76% is labeled as Other?? 10% for both system and user.), 309% system load and 70% swap usage (i think it's zramswap).

the image does not match the alt text, i referenced the wrong image, sorry. this was the actual image.

grafana node dashboard for ryo.d.soopy.moe, showing 98.9% cpu usage (of which 76% is labeled as Other?? 10% for both system and user.), 309% system load and 70% swap usage (i think it's zramswap).

nyaboom emoji

i didn't eval nixpkgs on it this time, just copying stuff made it explode. maybe its time to not use zram as swap

grafana dashboard same as the last post. this time it shows more detail, the ram usage is at 91.3%, swap at 35.9% and the graphs just stopped abruptly at 17:04:00(+08:00), leaving about 4 minutes of empty graph space. memory usage was shown spiking up from ~256MiB to ~768MiB shortly before the cut-off.

ok it just died. do NOT evaluate nixpkgs or nixosSystem in an oracle VM.Standard.E2.1.Micro unless you want to explod

grafana graph, showing cpu usage: idle: 46%; iowait: 37.3%; other: 15.8%; system: 7.45%; user: 3.08%; irq: 0.167%

i thought oracle gave me an ssd but im back in iowait hell

output of nix-output-manager, root package is immich 1.120.1, the dependency tree is extremely deep. 4 log lines show that llvm is currently being built.

it's now building llvm now, as if i didn't do it enough when doing lfs

terminal with tmux open, 3 windows visible. on the left, nix with nix-output-monitor, currently building gettext, readline, xvidcore, immich 1.121.0. a horrendously long dependency tree is shown. on the right, 2 panes arranged vertically. the top pane shows the bottom tool (htop-like tool) showing 100% CPU usage. the bottom pane is short, displaying the output of `git bisect log`.

nixos bisecting: automatic LFS to debug a broken package

one of duckduckgo search results: react-resizable-panels. description is all garbled text, likely due to an encoding issue

encoding errors in 2024???????

test reply 2

test

🥴

did:key:zQ3shph8F8QSBdQ8BbjEwrdZBx65FpJYkMFbQFdd1Z9XuwxSL
p77oyax2fxw3swococf3hhz5e6mb5hxw6ucl56wbbllcohxlk7ta

did:plc:erx5wg4jdj3rdzv336pto44w

during a test account migration i went to sleep but didn't save the refreshToken... the accessToken is expired when i woke up.

always remember to save both or just do everything in one go :upside_down:

half a dozen emails from github as a result of pushing new code to a branch with a PR open. the titles are the same: "re: soopyc/mystia nixos/bsky-pds: fix secret loading". the content slightly different, but apart from the first email which reads "merged #109 into mistress", other reads "@soopyc pushed 1 commit. [git commit hash] nixos/bsky-pds"

perils of testing on prod

i love testing on prod

command ran: nix fmt command output: nixfmt: ./.direnv/flake-inputs/[hash]-source/deprecated: openTempFileWithDefaultPermissions: permission denied (Read-only file system) context: the entire .direnv is in .gitignore

nixfmt doesn't ignore .gitignore files??

i guess i dug myself into this hole /shrug

getting started with a new project with nix (esp. on nixos) is so unbearably annoying because you have to write a wrapping nix shell/flake for *anything* to work, but once it works it's amazing