|
Jim Silly-Balls posted:does this thread have an opinion on UnRaid? I want to pool up a bunch of disks with a dell raid card and not paying for a windows server license would be nice. ZFS but amateur and without the good parts, is my impression. And marketed at "hardcore gamers" according to their website, so that disqualifies that trash immediately. Ditch the card unless it can run as a dumb HBA. Sun had a burning hatred for raid cards and they were absolutely correct. There are two modern file systems, btrfs and zfs. If you want raid, you either do mdraid+btrfs (because btrfs has a terrible raid implementation) or zfs. Zfs is the only file system I trust and is by far the most pleasant to actually deal with. If you just want a NAS and not deal with the internal crap, I hear freenas is still okay.
|
# ¿ Feb 25, 2020 12:59 |
|
|
# ¿ May 13, 2024 15:02 |
|
We have ~0.5PB on ZFS and I use it at home in a simple mirror. The weird Windows people get access to hourly snapshots via the "previous versions" tab and the Linux people know how to copy from the snapdir. Not having to go to backup because some numpty unlinked a TB of research data is its own reward. Also, not having to deal with btrfs caveats is nice. ZFS's caveats are much more interesting and fun, but at least they don't eat your data.
|
# ¿ Feb 25, 2020 13:59 |
|
Using ceph… on top of zfs.
|
# ¿ Feb 25, 2020 14:16 |
|
Using a NAS is boring. Where's your sense for adventure?
|
# ¿ Feb 25, 2020 19:46 |
|
You bring dishonour to your ancestors.
|
# ¿ Feb 26, 2020 18:39 |
|
Every time I have to compile qtwebkit, I cry. That's all. //edit: Somehow, qtwebengine manages to be worse, though Antigravitas fucked around with this message at 19:16 on Feb 27, 2020 |
# ¿ Feb 27, 2020 19:14 |
|
qtwebkit takes a very long time to compile. 15 minutes on an 8 core Ryzen on a ramdisk. qtwebengine uses BOTH make (qt) and ninja (chromium). The project is so large that the insane C header crap makes it unbelievably slow. They added the "jumbo build" option that concatenates all the headers, but each build thread then takes >2GB of private memory. So with 16 threads you need at least 32GB RAM because ninja defaults to using all threads and ignores makeopts. At least an hour of compile time. And it keeps. growing. Building chromium is just insane, utterly insane, in every way. The only thing that's more insane to compile is freetype with harfbuzz support, where you have to compile freetype without harfbuzz support, then harfbuzz, then recompile freetype with harfbuzz support, because freetype is a build dependency of harfbuzz.
|
# ¿ Feb 29, 2020 21:42 |
|
What I'm trying to say is, all C++ programmers should be sectioned. Also, hug your distro maintainers. They need one.
|
# ¿ Feb 29, 2020 21:45 |
|
Ninja is being called by qtwebengine's qmake. I never really looked much deeper into the makefile because that way lies madness and I am not strong enough to do that right now. I just want a qtwebengine…
|
# ¿ Mar 1, 2020 01:11 |
|
I sometimes dip into other languages but honestly, Python is really nice to come home to. Dealing with venvs is garbage, but most of what I write is covered by the stdlib and what OS doesn't have requests or flask in its distro packages? So what if it's not the newest, I just make sure that my stuff works on Debian stable, deal with it. Another upside: no makefile / automake / meson / ninja / cargo / …
|
# ¿ Oct 21, 2020 21:54 |
|
ratbert90 posted:venv when containers exist???? That's a venv but also for C programs. venvs are bad.
|
# ¿ Oct 22, 2020 07:33 |
|
I have written beautiful code once. It's perfect, handles all exceptions properly, is easy to read, has very little indentation, does one thing and it does it well, is documented with help output and all. It covers a lot of edge cases without becoming complex and it uses no confusing language features. It serves as great contrast to everything else. I read it sometimes and weep in despair.
|
# ¿ Oct 23, 2020 12:44 |
|
Soricidus posted:which programming languages support zalgo variable names? just curious >>> i̢n͡ͅv͍̯o̤̭k̟͞e̗͞ = "yospos" >>> print(i̢n͡ͅv͍̯o̤̭k̟͞e̗͞) yospos Too much Zalgo breaks python: >>> Yͨ̅O̤ͤS͊̀P̓̇Oͬ̈S̈́̂ = "bithc" File "<stdin>", line 1 Yͨ̅O̤ͤS͊̀P̓̇Oͬ̈S̈́̂ = "bithc" ^ SyntaxError: invalid character in identifier
|
# ¿ Oct 23, 2020 20:57 |
|
That "distroless" container thing looks like an extremely convoluted way of circling back to the equivalent of a service launched via a normal systemd unit file with cgroup isolation and ephemeral uid/gid. At that point you can just throw away the entire container thing and build a deb package of the application and wow! it's so much simpler!
|
# ¿ Oct 27, 2020 13:58 |
|
I primarily take offence with the practice of bundling 350MB of random garbage into a tarball along with your 500 loc Flask application. And then writing another gigantic pile of code because someone came to the realisation that this is a stupid thing to do, like that's a revelation. It's extremely javascriptian. Btw., podman can't come to Debian 11 fast enough, jfc. spankmeister posted:Fedora is bad now Weird, have they forgotten that ed is the standard editor?
|
# ¿ Oct 27, 2020 15:28 |
|
I'm sorry, why are we using editors that aren't standard? ed is the standard editor for a reason. Whenever I open nano I get confused by its weird interface (What's an ^ anyway?). Look at how consistent and clear ed is:pre:ag@209:~# ed yospos ? ^C ? ^C ? ^C ? help ? ? ? quit ? Killed xtal posted:I must be super lucky because I've used btrfs as root for like 8 years, when the Arch wiki page for it had a huge warning banner at the top, and still haven't had any problems. I think I even converted it from ext4 in-place to begin with. btrfs is fine in single-disk scenarios. It's when you get to Real Work applications where it really shows its deficiencies. ZFS makes handling huge storage machines a breeze and easy to service, with nice and friendly messages and documentation. I wouldn't trust btrfs with our current storage needs where we regularly add entire chassis filled with spinning rust just to feed the ever growing needs of nearline storage.
|
# ¿ Oct 28, 2020 13:20 |
|
Last Chance posted:is this a joke I'm missing?
|
# ¿ Oct 28, 2020 16:43 |
|
The best vi thing I ever learned is ci. ci" deletes everything between the next set of ", places the cursor there, and switches to insert mode. It works with a lot of common characters, ( [ ' etc. Other than that, only s/// syntax and ranges are useful day-to-day. I don't even use hjkl, like a rube. Also, if you use ed, you wouldn't have to think about those things. ?
|
# ¿ Oct 28, 2020 18:45 |
|
starbucks hermit posted:https://access.redhat.com/blogs/766093/posts/3557091 That's a terrible example, $HOME is chmod 0600, I don't keep my home on a webserver, and the web application's home is really not interesting. Also, was it SeLinux or Apparmor developers that created the following scenario? Bob: You should enable our security thing and you are bad if you don't! Alice: Okay, I'm enabling your security thing, but the tooling is terrible and now my application fails with cryptic error messages. Bob: You can put our amazing security thing into complain mode and see in the logs what is failing! Alice: I did but it still keeps failing and there's nothing in the logs. Bob, the loving imbecile: Oh yeah, some actions are so sensitive we just block them and don't log even in complain mode! Alice: … … … And that's how Alice disabled the thing entirely and Bob was found strung up by his own intestines.
|
# ¿ Oct 29, 2020 08:12 |
|
starbucks hermit posted:Did you know that the apache web server runs as root? Because it does! The initial process is root, which spawns workers. Did you know that root can do a lot of superuser things??? Like read user directories? Mine doesn't. Also, maybe one shouldn't be doing unsafe things instead of trying to paper over bad design with another complex system. If you have to bind to 80/443, perhaps use capabilities like a normal person? starbucks hermit posted:Before you poo poo on these things, you should probably have a modicum of knowledge about them first. I know enough about Selinux that I encountered the non-logged denials…and to discover that the tooling is garbage. I've written policies myself, and the entire process is terrible, ill thought out, and the documentation assumes everything does trivial things and that developers know the access they need (lmao on that one). If you don't know in advance what the thing does, you'll be chasing hard to debug failures in prod. Apparmor is extremely similar in its terribleness, but at least distro defaults are usually reasonable.
|
# ¿ Oct 29, 2020 10:53 |
|
That thing threatens to collect my dirty bits. I don't feel qualified to review the code, but posting code for upstreaming without being able to demonstrate a clear advantage over the old approach is chancy. The numbers shown for a typical VM indicate a slowdown, so you'd have to provide a way to switch between approaches depending on which would be faster and that's a that has led to rejections in the past.
|
# ¿ Oct 29, 2020 12:11 |
|
I've snuck it into the ansible role common to every single linux computer in our org.
|
# ¿ Nov 11, 2020 22:11 |
|
Cowsay can be templated pretty easily. I've used it on login nodes to wall at people. But sl holds a special place in my heart.
|
# ¿ Nov 12, 2020 19:34 |
|
sb hermit posted:using Gentoo to learn linux is like jumping in the deep end from a fifty foot diving board though Ph'nglui mglw'nafh Linux from Scratch R'lyeh wgah'nagl fhtagn. Quite far, actually.
|
# ¿ Feb 24, 2021 09:46 |
|
Using Mint is dumb though. Everything I've seen from that distro is amateur-hour bullshit. It's also downstream of Ubuntu, which is downstream of Debian, so you get some serious distro centipede bullshit on top. OpenSuse, Fedora, mayyybe Ubuntu if you have a desktop and don't want to deal with dumb stuff all the time. Debian if you want dependable two years of peace and quiet between upheaval, Gentoo if you are a mad person. code:
|
# ¿ Feb 24, 2021 10:04 |
|
I, for one, appreciate Debian's contribution in the fight against spectrum pollution.
|
# ¿ Feb 25, 2021 08:35 |
|
Antigravitas posted:OpenSuse, Fedora, mayyybe Ubuntu
|
# ¿ Feb 25, 2021 09:24 |
|
I admit my opinion of arch the distro is forever tainted by the people who use arch.
|
# ¿ Feb 25, 2021 09:29 |
|
Gentoo legit has some pretty cool tooling. I maintain an overlay with software that isn't in the main repository and writing ebuilds is much easier than writing deb or rpm packages. If you're involved in the low-level guts of the Linux ecosystem it's good. If you are not: Don't even think about using Gentoo.
|
# ¿ Feb 25, 2021 21:43 |
|
Don't forget budget. Redundancy costs money. Sometimes you just don't have the money. Sometimes redundancy is just not economical. Sometimes you could architect redundancy but it complicates operations. "Yes, we could use glusterfs everywhere, but are we confident we can recover glusterfs if it dies and the gluster guru is on vacation/got hit by a bus?" is a real concern. Often you can take the hit of going down unscheduled once a decade but interrupting service when you need to patch right now isn't a good option.
|
# ¿ Feb 26, 2021 09:50 |
|
Never work at a university then. Pro: Some seriously cool poo poo, some of our stuff is currently in a container on a boat in the arctic circle. A real container, not some cgrouped chroot. Neg: People tell you to re-architect your "app" in node.js on aws. We also have ssh bastion hosts that let you connect to multi-million euro equipment via rsh…
|
# ¿ Feb 26, 2021 10:37 |
|
Soricidus posted:zero uptime is where it’s at There's medication for that.
|
# ¿ Feb 26, 2021 10:59 |
|
How can you do right when all the tools are wrong?
|
# ¿ Feb 26, 2021 16:28 |
|
Now delete something with a newline in its filename. My favourite teaching moment is asking someone to delete all dotfiles in a directory. (But making sure they don't have anything irreplaceable in harm's way)
|
# ¿ Feb 26, 2021 21:15 |
|
Careful though, a Linux source tree can get pretty large after building. 5GB for 5.11. Probably a good idea to use gentoo-kernel-bin anyway.
|
# ¿ Feb 27, 2021 13:34 |
|
Writing my "stdio.h considered harmful" blog post now.
|
# ¿ Mar 5, 2021 15:48 |
|
It's honestly sad at this point. I'd love to have "ZFS, but in-tree", but btrfs is just treading water. It has all these bizarre caveats and options and footguns. ZFS isn't exactly trivial, but at least its footguns are well understood and far less severe. And also, and I know this is somewhat petty, ZFS has friendlier messages when something goes wrong.
|
# ¿ Mar 7, 2021 20:39 |
|
I don't mean more useful, though that is also the case. I mean that they are written in large, friendly letters. A zfs status with a pool that has a missing drive produces this: quote:state: DEGRADED It doesn't really tell me more than the more austere btrfs messages, but it feels reassuring. "Don't worry", ZFS says, "I got this". (ZFS is also better in every area I care about but that's beside the point )
|
# ¿ Mar 7, 2021 21:25 |
|
Be glad that drive poo poo the bed before you had any effort and data invested. However, a drive making GBS threads that many errors could also be a bad connection. If it's socketed, it may be worth taking it out and putting it in again. Just be gentle doing it, especially if it's your first time handling it. That is, if the lunatics at Dell consider this user serviceable. If not, don't touch it. e: I've never looked at mine. quote:Smart Log for NVME device:nvme0 namespace-id:ffffffff drat, that thing may last me a long time with my use… Antigravitas fucked around with this message at 17:09 on Mar 8, 2021 |
# ¿ Mar 8, 2021 17:06 |
|
|
# ¿ May 13, 2024 15:02 |
|
Actually, it just occurred to me that a smart error wouldn't be a connection issue, or rather the connection would be the solder. So disregard what I wrote, my brain has turned to mush from work…
|
# ¿ Mar 8, 2021 17:12 |