Users/Developers Threatening Fork Of Debian GNU/Linux

[birdie]

So you lession learned should be: Stop calling yourself a poweruser and stop doing stupid actions without even informing yourself, then blaming others for it! But my guess is you're going to ignore this post, just as you did with my last.

Show me a single test/benchmark where a system with SWAP enabled beats the same system without SWAP.

No, I'm not talking about embedded systems which are absolutely low on memory.

I'm talking about servers/PCs with 4GB of RAM and more.

And please don't give me benchmarks of radically rare use cases - we are talking about normals tasks, like browsing the web, IM'ing, audio/video/images authoring/editing, watching movies, etc. etc. etc.

[curaga]

just pass it through strings to get rid of any non-printable characters

I don't think anybody mentioned this yet, but journald compresses by default. Strings will be useless in recovering corruption when the file is compressed.

[pal666]

compressing text logs and writing in binary format are two different things. if you can't understand the difficulties and differencies then there's no need for a discussion (and since you listed it as a counter argument, it seems you do not understand).

there is zero difference from pov of file corruption.it seems you do not understand

and why is the indexed text db so important ? if I want that, I can import to suitable database via an import interface/conversion script/daemon ...

indexed text db is not important. important is indexing, because grep is slow and unreliable. all indexed dbs are binary. systemd already wants that and imports via journald. then it can show you log lines related to service. your lack of imagination is not going to make the world stop. btw, journal logs differ from text lols not only in indexing. they are signed and intruder will not be able to edit them without breaking signature.

export to syslog is nice, but somebody earlier mentioned that the existence of journald is justified by it logging the boot process before syslog is running, how do you export to syslog if none is running ?

by buffering ffs. why we are having this silly discussion ?

You only need logs when something goes wrong (most common case). And if something goes wrong in the stage where journald has its reason for existence (very early boot) you usually cannot boot up the system. This in turn means, anything special needed to read the log files SLOWs you down or prohibits any reasonable troubleshooting (you have to have a system where you can read journald files). grep/cat/sed/awk are standard utilities available everywhere so you can boot up the failing system from rescue media, mount the filesystems and go hunting for the issue. in case of a binary log, you are screwed unless you have the correct utility to read it.

you live in some fantasy world. journalctl is standard utility and is available everywhere where its logs are available. in case of text log you are screwed because there is no text log until you boot syslogd.

[V10lator]

Show me a single test/benchmark where a system with SWAP enabled beats the same system without SWAP.

I won't give you benchmarks cause all you're replying to from my 2 posts is something i didn't even claim in the post you quoted! I guess you're talking about the disc I/O I was talking about in another post? Anyway, I already told you where it beats, again, just for you:
- Hybernation - The system without SWAP can't do this!
- Reduced disc I/O without swap (especially when available RAM is low and/or there's much I/O). No, I still won't give you benchmarks, especially as this is a known fact: When the kernel has to drop something from the disc cache and you want to acces (read) it again you will get the speed of the harddrive/SSD, not the speed of your RAM. Now please show me a benchmark where a HDD/SDD beats a ramdisc. With SWAP the kernel is able to keep the cache (if he feels it's important, like you readed the file 500x in the last 2 minutes) by moving other data that's almost never accessed to swap. How to do this without swap and why do you need benchmarks for that?

And, as you want to kill kswapd completely the system without swap will also have:
- Horrible memory management, as a resull reduced memory I/O or less usable RAM.
- Less usable RAM (no, I'n not talking about the swap space)

like browsing the web, IM'ing, audio/video/images authoring/editing, watching movies, etc. etc. etc.

Yea, cause that are things where an impact on disc I/O is easily measurable cause it's not like these things do most of their work without disc I/O... I mean browser, IMing, really? How do you want to benchmark these things anyway? There are variables you can't reproduce: Speed of user input and Network latency/bandwith.

So... Reply to all my points instead of completely wrongly quoting one thing and then asking for impossible things or I won't feed you anymore little troll.

[bearded_linux_admin]

Well, in any case, about systemd etc. I do find it very difficult sometimes to figure out why a particular systemd service gets started, and when I tried putting together a battery target which would automatically shut down various daemons that I don't need when I want to save power, it apparently somehow caused the brightness keys (fn-F5 and fn-F6) to mysteriously stop working, and as I expected, it was impossible to debug. So instead of using a systemd target, I'll just hack together a shell script that runs the necessary "service <foo> stop" instead of using a systemd target. If things start breaking horribly, I'll file debian bugs, and try to find ways to work around the brain damage. The fact that I won't be able to edit shell scripts to work around brain damage is still a little anxiety-producing, and the fact it's much more difficult to create a runlevel which is "just like runlevel 3 but without certain services running" is unfortunate.

[pal666]

Well, in any case, about systemd etc. I do find it very difficult sometimes to figure out why a particular systemd service gets started, and when I tried putting together a battery target which would automatically shut down various daemons that I don't need when I want to save power, it apparently somehow caused the brightness keys (fn-F5 and fn-F6) to mysteriously stop working, and as I expected, it was impossible to debug. So instead of using a systemd target, I'll just hack together a shell script that runs the necessary "service <foo> stop" instead of using a systemd target. If things start breaking horribly, I'll file debian bugs, and try to find ways to work around the brain damage. The fact that I won't be able to edit shell scripts to work around brain damage is still a little anxiety-producing, and the fact it's much more difficult to create a runlevel which is "just like runlevel 3 but without certain services running" is unfortunate.

https://www.google.com.ua/webhp?sour...+debug+systemd
you still can't understand that systemd allows you run and edit shell scripts. it just doesn't require it. but of course it will be more rewarding to educate yourself and fix problem properly. and if you have useful usecase you can even ask for it to be implemented upstream.

[bearded_linux_admin]

https://www.google.com.ua/webhp?sour...+debug+systemd
you still can't understand that systemd allows you run and edit shell scripts. it just doesn't require it. but of course it will be more rewarding to educate yourself and fix problem properly. and if you have useful usecase you can even ask for it to be implemented upstream.

That comment was made by Theodore Tso, go tell him what he does and doesn't understand. If you are not sure who Mr. Ts'o is, here is a link. I see a lot of GNOME heads and desktop jockeys sweeping away well reasoned and thoughtful comments, questions and insights made by the most respected members of the Linux community, often with anger, arrogance, and and air of "we know what is best for you, you are stupid, shut up and go away"

These are the people that are writing a core subsystem for my servers. When they act like juveniles on ritalin in public, I am *not* likely to trust their code.

[erendorn]

Well, in any case, about systemd etc. I do find it very difficult sometimes to figure out why a particular systemd service gets started, and when I tried putting together a battery target which would automatically shut down various daemons that I don't need when I want to save power, it apparently somehow caused the brightness keys (fn-F5 and fn-F6) to mysteriously stop working, and as I expected, it was impossible to debug. So instead of using a systemd target, I'll just hack together a shell script that runs the necessary "service <foo> stop" instead of using a systemd target. If things start breaking horribly, I'll file debian bugs, and try to find ways to work around the brain damage. The fact that I won't be able to edit shell scripts to work around brain damage is still a little anxiety-producing, and the fact it's much more difficult to create a runlevel which is "just like runlevel 3 but without certain services running" is unfortunate.

Keep in mind though, that "hacking shell scripts" is really not the preferred method of configuring a system. You may be comfortable with it due to extended experience in it, but it has an enormous entry cost, is quite prone to operational errors, is difficult to maintain when the packaged script change, etc..

[birdie]

With SWAP the kernel is able to keep the cache (if he feels it's important, like you readed the file 500x in the last 2 minutes) by moving other data that's almost never accessed to swap. How to do this without swap and why do you need benchmarks for that?

This is such BS.

The kernel does not distinguish between applications' own memory and IO operations, thus, and it constantly happens under heavy IO when you copy large amounts of data, applications get swapped out, thus when you switch back to them your system more or less freezes, because the kernel prioritizes reading memory pages from swap back to RAM. swap is almost always detrimental to the performance of the system, unless you use various swap implementations using compressed RAM.

God, people are so much full of BS. Like those segfaults in SysVinit which no one can show me (meanwhile I can easily find several hundred real bug reports about segfaults, crashes and freezes in systemd). When you're telling them that SysVinit basically has no way of segfaulting they say that _real_ _multiple_ bug reports about crashes in systemd are a figment of my imagination. Wow.

And I owe you nothing. If it's so difficult for you to repeat your previous messages, then so be it.

Huh, you've just told me that swap is required for hibernation. Well, f*ck me, but if some kernel developer had a hangover when he was designing this feature, then it's not my problem - it's a bad ugly implementation. People often increase their RAM which means in Linux you must resize your swap partition (of course some crazy guys create a swap partition which is several times their physical RAM but that's not exactly an ordinary practice). I'm sorry but that's utterly f*cked up; that's not a thing to be proud of, or an argument for using swap.

[jrch2k8]

interesting
i never cried configuring anything on my system, for some... 9-10 years now
from bash through vsftpd, apache, irssi to the X server, all more or less sane
(openbox-es XML config was one of the worst, because its XML and i don't want/know how to use an XML editor)
alsa is one of the more confusing ones
it's more of a programming language then a config
but even alsa makes sense

then i read the .service spec
it's horrible if you ask me
ExecReload stuck to mind, just the name makes no sense considering what it does
and why does everything have to have Caps ?
why are there a "[Unit]" and "[Install]" sections ?
why are there sections at all if all declaration names are unique ?
(if they are unique, i didn't check)
if a .service is "A unit configuration file" why is there also a .unit file that the documentation says it is "A unit configuration file" ?

and what if i want to do something advanced, like a beowolf cluster without hard drives on nodes... i have no f-in idea how to do it
if that is even possible

i fkin made a self modifying program and it was more straightforward to do then this


then again i'm no professional admin and i didn't put much time beyond writing a basic .service file
i did however ask a professional admin that tried it and he said something like "that is not going on my servers"
reading around the net there seems to be more admins that don't quite like that way of doing things
although to be fair most sysadmins probably wont care as long as it runs a program as it's supposed to


i still say a sysadmin should know how to write a shell script...

1.) well like i said is for specific scenarios, i named some but i guess i wasn't specific enough in the first paragraph
2.) well about syntax is always a personal matter, i kinda like it except maybe slices nomenclature but at the end of the day is just syntax and can always be improved as long as the functionality is there
3.) well i have ages without using beowulf so i can't say for certain because i prefer software specific clustering this days but well is a matter of taste and specific needs
4.) about the ddos protection is quite impressive thanks to cgroup+selinux+namespace, see in the blog and the links i posted before about process slice, i was able to even crash the service inside that slice but never used 1 more bit of resources than those i set(cpu, iops, ram) and selinux deals with any try to access unauthorised path in the drive for write(not outside the cgroups instance but files inside of it), of course this need manual configuration because as far as i know no distro do this kind of hardening by default(would be nice tho)
5.) about advanced security i mean audit(as in-kernel audit infrastructure)+selinux+seccomp+cgroups to make sure that even if the process itself get hacked externally it can never reach the actual OS from it and get all the audit information of it on journald(syslog don't support audit/seccomp data as far as i remember) but again this is not set by default, additionally the root user inside cgroups is a fake and only valid inside that slice, the PID virtual tablespace(namespaces) and the network stack process table(network namespaces) are isolated, meaning no child rogue(or infected) process can reparent with any other process only with its original parent because the other PIDs simply don't exist(sort of a super chroot)
6.) about dbus yeah it has some issues but the marshalling is quite decent and the current DBUS code can't pass big data or get DMA and in systemd is inside an slice and with low privileges even tho in theory can be used as an attack vector, it is a lot harder with systemd(you need to escape the super chroot and elevate privis somehow before you can do something interesting with it) than without it, KDBUS will improve this by a lot since the kernel will handle the marshalling through all its security infrastructure with RAM isolated file descriptors or controlled copies using kernel infrastructure to user space