[arch-general] Arch's move to systemd integration
I've followed this discussion closely and, as can be seen in one of the posts below, I initially objected to it. I had previously tried systemd with little success. Next I carefully re-read all the objections, complaints, rants, raves, etc, and decided to try systemd again. This isn't the place for personal info but I think it's relevant here. I suffer from early onset dementia so it takes me more time to understand documentation and retain information. Most of the time I have to start over everytime I work on something because no matter how many times I've done it, it doesn't stick. That's one reason I prefer the rc.conf file and initscipts, it's easy to understand and keep track off. I posted last night that it took me around 30 minutes reading and editing to get my system booted properly. I was wrong. I still have a couple of minor glitches, but nothing that breaks my system. I firmly believe part of the problem is perception. Previously rc.conf held all the information. On my system rc.conf is 716 B. The three files (hostname 8 B, vconsole.conf 47 B, and locale.conf 30 B) take up less space. Granted at the moment it's more but that will be offset when the move to systemd can be completed. If I understand everything correctly the conf.d and rc.d directorys will no longer be needed as everything will be in /etc/systemd. The amount of disk space taken up should, major caveat, be less because most of the files in /etc/systemd/system/??? are soft links to the acutal files in /usr/lib/systemd/system. Yes, apparently, this changes(?) some of Arch's KISS principle, but maybe it doesn't. Maybe it actually makes it easier in the long run. If you keep /etc backed up, you do don't you, it's a simple matter to set everything up whenever where ever. I think the biggest problems here are "perception", resistance to change - me, possibly some FUD, and no one likes being told what to do. My main motivation for trying systemd again stemmed from realizing I was using the same arguments I fought in another industry for years "We've done it that way for 30 years and it works, why change now". Don't think of it as change, think of it as "evolution". Everyone believes in evolution, don't you? If this provides better integration with upstream developers might that mean less breakage, patched software, less complaints, and less bug reports to deal with for the Arch Devs. I'm proof even old dogs can learn new tricks. I'm open to discussion on any points I made about assumptions of what will and won't be necessary when the systemd integration is completed, ie conf.d and rc.d etc. Links to previous discussions. https://mail.google.com/mail/?shva=1#label/archlinux%2Fgeneral/138ad0e074447... https://mail.google.com/mail/?shva=1#label/archlinux%2Fgeneral/138ab21f72d9b... https://mail.google.com/mail/?shva=1#label/archlinux%2Fdev-public/138a6cee47... Myra -- Life's fun when your sick and psychotic!
On Sun, Jul 22, 2012 at 2:11 PM, Myra Nelson <myra.nelson@hughes.net> wrote:
I've followed this discussion closely and, as can be seen in one of the posts below, I initially objected to it. I had previously tried systemd with little success. Next I carefully re-read all the objections, complaints, rants, raves, etc, and decided to try systemd again. This isn't the place for personal info but I think it's relevant here. I suffer from early onset dementia so it takes me more time to understand documentation and retain information. Most of the time I have to start over everytime I work on something because no matter how many times I've done it, it doesn't stick. That's one reason I prefer the rc.conf file and initscipts, it's easy to understand and keep track off.
I posted last night that it took me around 30 minutes reading and editing to get my system booted properly. I was wrong. I still have a couple of minor glitches, but nothing that breaks my system. I firmly believe part of the problem is perception. Previously rc.conf held all the information. On my system rc.conf is 716 B. The three files (hostname 8 B, vconsole.conf 47 B, and locale.conf 30 B) take up less space. Granted at the moment it's more but that will be offset when the move to systemd can be completed. If I understand everything correctly the conf.d and rc.d directorys will no longer be needed as everything will be in /etc/systemd. The amount of disk space taken up should, major caveat, be less because most of the files in /etc/systemd/system/??? are soft links to the acutal files in /usr/lib/systemd/system.
Yes, apparently, this changes(?) some of Arch's KISS principle, but maybe it doesn't. Maybe it actually makes it easier in the long run. If you keep /etc backed up, you do don't you, it's a simple matter to set everything up whenever where ever.
I think the biggest problems here are "perception", resistance to change - me, possibly some FUD, and no one likes being told what to do. My main motivation for trying systemd again stemmed from realizing I was using the same arguments I fought in another industry for years "We've done it that way for 30 years and it works, why change now". Don't think of it as change, think of it as "evolution". Everyone believes in evolution, don't you? If this provides better integration with upstream developers might that mean less breakage, patched software, less complaints, and less bug reports to deal with for the Arch Devs. I'm proof even old dogs can learn new tricks.
I'm open to discussion on any points I made about assumptions of what will and won't be necessary when the systemd integration is completed, ie conf.d and rc.d etc.
Links to previous discussions.
https://mail.google.com/mail/?shva=1#label/archlinux%2Fgeneral/138ad0e074447...
https://mail.google.com/mail/?shva=1#label/archlinux%2Fgeneral/138ab21f72d9b...
https://mail.google.com/mail/?shva=1#label/archlinux%2Fdev-public/138a6cee47...
Myra
-- Life's fun when your sick and psychotic!
Your links will only work for you, they are links to messages in your gmail inbox
Hi, Am 22.07.2012 21:11, schrieb Myra Nelson:
On my system rc.conf is 716 B. The three files (hostname 8 B, vconsole.conf 47 B, and locale.conf 30 B) take up less space. Granted at the moment it's more but that will be offset when the move to systemd can be completed.
You don't really try to argue about a few bytes like this, do you? With the default block size of 4K (for ext4), this doesn't matter anyway.
I think the biggest problems here are "perception", resistance to change
Yes, this is most definitely the problem. But this is something coming along with every change, so unless there are any technical reasons against the current movement, I don't care that much about these complains. Best regards, Karol Babioch
On Sun, Jul 22, 2012 at 2:22 PM, Karol Babioch <karol@babioch.de> wrote:
Hi,
Am 22.07.2012 21:11, schrieb Myra Nelson:
On my system rc.conf is 716 B. The three files (hostname 8 B, vconsole.conf 47 B, and locale.conf 30 B) take up less space. Granted at the moment it's more but that will be offset when the move to systemd can be completed.
You don't really try to argue about a few bytes like this, do you? With the default block size of 4K (for ext4), this doesn't matter anyway.
Though your(Karol) right, that a few bytes doesn't matter in the days of TB+ hard drives, it might be something worth looking at (how much more space will systemd use/save). And on some systems (mostly older) it may matter.
I think the biggest problems here are "perception", resistance to change
Yes, this is most definitely the problem. But this is something coming along with every change, so unless there are any technical reasons against the current movement, I don't care that much about these complains.
Best regards, Karol Babioch
On Sun, Jul 22, 2012 at 2:26 PM, Nicholas MIller <nick.kyky@gmail.com> wrote:
On Sun, Jul 22, 2012 at 2:22 PM, Karol Babioch <karol@babioch.de> wrote:
Hi,
Am 22.07.2012 21:11, schrieb Myra Nelson:
On my system rc.conf is 716 B. The three files (hostname 8 B, vconsole.conf 47 B, and locale.conf 30 B) take up less space. Granted at the moment it's more but that will be offset when the move to systemd can be completed.
You don't really try to argue about a few bytes like this, do you? With the default block size of 4K (for ext4), this doesn't matter anyway.
Though your(Karol) right, that a few bytes doesn't matter in the days of TB+ hard drives, it might be something worth looking at (how much more space will systemd use/save). And on some systems (mostly older) it may matter.
i don't have any figures/proof but i think systemd is gaining quite a bit of populatiry on the embedded linux market -- where space and minimalist is king -- at least ii feel like i've read that a few times recently :-) i'm pretty sure i've read posts to systemd-devel by the http://profusion.mobi/ guys, and some longer blogs too. also in my own endeavors with my Sheevaplug and Pandaboard i've run across a number of people who were using it with great results (as am i :-) probably beating a dead horse here, bu IMO it's just blatantly superior all ways imaginable ... this becomes increasingly obvious the longer you work with it (check out user-sessions! yay!). -- C Anthony
On 07/22/2012 02:38 PM, C Anthony Risinger wrote:
probably beating a dead horse here, bu IMO it's just blatantly superior all ways imaginable ... this becomes increasingly obvious the longer you work with it (check out user-sessions! yay!).
With anything like this consideration all I care about is what the cost/benefit of the change is. Who will benefit and how much work it will take on the communities part just to move to the next gee-whiz thing. Very simple, what does the following table look like: current init systemd init Benefits Drawbacks Benefits Drawbacks ------------- ---------------- ---------------- ------------------ If the systemd benefits outweigh the benefits of the current init and the time-cost to the community to undergo the change is small, then it is worth doing. However, if the change will take significant effort on a per-machine basis from the community, and systemd doesn't provided a _needed_ benefit that the current system doesn't then it's not worth doing. I haven't run into any limitations that the current system has, so I don't see any need to change. However, if it is basically a transparent change to me, then I don't really care. But if it takes 1/2 hour per box to dork with -- then I don't want to jack with it. -- David C. Rankin, J.D.,P.E.
Hi, Am 22.07.2012 22:59, schrieb David C. Rankin:
If the systemd benefits outweigh the benefits of the current init and the time-cost to the community to undergo the change is small, then it is worth doing.
Once again: The proposed change is not about systemd. The maintainer(s) plan to support both the classic initscripts as well as systemd (see here [1]). So don't mix these things here. For now we are talking about the split up of "rc.conf". [1] https://mailman.archlinux.org/pipermail/arch-dev-public/2012-April/022803.ht... Best regards, Karol Babioch
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 07/22/2012 04:04 PM, Karol Babioch wrote:
If the systemd benefits outweigh the benefits of the current init and the time-cost to the community to undergo the change is small, then it is worth doing. Once again: The proposed change is not about systemd. The maintainer(s)
Am 22.07.2012 22:59, schrieb David C. Rankin: plan to support both the classic initscripts as well as systemd (see here [1]). So don't mix these things here. For now we are talking about the split up of "rc.conf".
[1] https://mailman.archlinux.org/pipermail/arch-dev-public/2012-April/022803.ht...
Well, if there is a split needed to support systemd, that's fine as long as rc.conf remains for those that don't go off chasing it. Further, if the 3 files you are taking about contain information that is used in rc.conf, why not work this 'integration' to a point where rc.conf simply sources the info it needs from the systemd files. Not needed, but if it makes sense long term. As long as we are talking about either rc.conf or hostname, vconsole.conf, and locale.conf, if the files are reasonably commented so that a user who knows rc.conf can find his way though the other three for a systemd setup, then there is no issue. The files are small, and easy to deal with. Like I said, if the move is basically transparent to the community and systemd has some 'must have' feature that everyone will use, then a migration there is a good way to then be able to compare and contrast user experience with both. I can't see a cleaner way than a single rc.conf, but I don't have objection to looking at 3 instead of one. - -- David C. Rankin, J.D.,P.E. -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.16 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAlAM2yoACgkQZMpuZ8Cyrci/lQCdFrciWLpUMiRZ4OfPPO4wy2wf fxMAnjNy6+83ysEvDaVw9ELp/kdBrGLw =/8s5 -----END PGP SIGNATURE-----
i don't have any figures/proof but i think systemd is gaining quite a bit of populatiry on the embedded linux market -- where space and minimalist is king -- at least ii feel like i've read that a few times recently :-)
You must mean a fraction of the embedded world as. Systemd is larger than init. I wonder if arm arch is moving to systemd, though that is still quite a large OS in embedded terms compared to non MMU systems like uclibc and lower. -- ________________________________________________________ Why not do something good every day and install BOINC. ________________________________________________________
Hi, Am 22.07.2012 21:26, schrieb Nicholas MIller:
Though your(Karol) right, that a few bytes doesn't matter in the days of TB+ hard drives, it might be something worth looking at (how much more space will systemd use/save). And on some systems (mostly older) it may matter.
No, it doesn't. A file can't take less space than a single block, which defaults to 4k (at least this is what "/etc/mke2fs.conf" tells me). So any separation of a small file like "rc.conf" into a few even smaller files will take up even a little bit more space. But I don't think that there is a system on this planet, which runs a recent version of Linux (and/or Arch), which can't afford to have a few more files like this. I know some of you will come up with some examples now that I've challenged it, but I think we can agree that a few kilobytes on the hard drive back and forth are not worth talking about. Furthermore that's not what KISS is about. KISS is about keeping it simple, not about keeping it small (in regard to files sizes). Although there are cases this might be the same, I don't see why this should be the case here. Personally I don't understand what this complaining is all about. It is much more simple to have dedicated files for tasks like setting the console font than having it in one "big" file specific to Arch. Furthermore it will be some sort of a standard, so you won't have to look up something as trivial as setting the hostname in case you are dealing with another distribution once in a while. I really don't get it: I'm willing to accept that not everyone is eager to get in touch with systemd and that there are some points you could criticize systemd for, but in this case we are talking about splitting up an Arch specific file in order to be in compliance with upstream and/or other distributions. I can't see why this is something that's worth this much discussion considering the fact that the "old" syntax will still be supported. Best regards, Karol Babioch
On Sun, Jul 22, 2012 at 10:45 PM, Karol Babioch <karol@babioch.de> wrote:
Hi,
Am 22.07.2012 21:26, schrieb Nicholas MIller:
Though your(Karol) right, that a few bytes doesn't matter in the days of TB+ hard drives, it might be something worth looking at (how much more space will systemd use/save). And on some systems (mostly older) it may matter.
No, it doesn't. A file can't take less space than a single block, which defaults to 4k (at least this is what "/etc/mke2fs.conf" tells me). So
In modern filesystems (Btrfs, NTFS, Reiserfs), sufficiently small files can be stored in the inode itself, taking up zero blocks. -- Mantas Mikulėnas
Hi, Am 22.07.2012 21:54, schrieb Mantas Mikulėnas:
In modern filesystems (Btrfs, NTFS, Reiserfs), sufficiently small files can be stored in the inode itself, taking up zero blocks.
But even then: With a bunch of more files, you'll need a bunch of more inodes. So unless your original file itself was small enough to fit into the inode, you would end up needing more space as a result. But once again: That's not what KISS is about. Best regards, Karol Babioch
Hi again, Am 22.07.2012 21:54, schrieb Mantas Mikulėnas:
In modern filesystems (Btrfs, NTFS, Reiserfs), sufficiently small files can be stored in the inode itself, taking up zero blocks.
Offtopic: Calling NTFS "modern" is quite ironic ;). Best regards, Karol Babioch
On Sun, Jul 22, 2012 at 2:45 PM, Karol Babioch <karol@babioch.de> wrote:
Hi,
Am 22.07.2012 21:26, schrieb Nicholas MIller:
Though your(Karol) right, that a few bytes doesn't matter in the days of TB+ hard drives, it might be something worth looking at (how much more space will systemd use/save). And on some systems (mostly older) it may matter.
No, it doesn't. A file can't take less space than a single block, which defaults to 4k (at least this is what "/etc/mke2fs.conf" tells me). So any separation of a small file like "rc.conf" into a few even smaller files will take up even a little bit more space.
But I don't think that there is a system on this planet, which runs a recent version of Linux (and/or Arch), which can't afford to have a few more files like this. I know some of you will come up with some examples now that I've challenged it, but I think we can agree that a few kilobytes on the hard drive back and forth are not worth talking about.
Furthermore that's not what KISS is about. KISS is about keeping it simple, not about keeping it small (in regard to files sizes). Although there are cases this might be the same, I don't see why this should be the case here.
Personally I don't understand what this complaining is all about. It is much more simple to have dedicated files for tasks like setting the console font than having it in one "big" file specific to Arch. Furthermore it will be some sort of a standard, so you won't have to look up something as trivial as setting the hostname in case you are dealing with another distribution once in a while.
I really don't get it: I'm willing to accept that not everyone is eager to get in touch with systemd and that there are some points you could criticize systemd for, but in this case we are talking about splitting up an Arch specific file in order to be in compliance with upstream and/or other distributions. I can't see why this is something that's worth this much discussion considering the fact that the "old" syntax will still be supported.
Best regards, Karol Babioch
Karol: My comparison to file size was meant to be extended to the complete removal of rc.d and conf.d or the removal of several files in those directories. Maybe that concept is not that important. I didn't mean to imply the KISS principle was about size, just trying to imply this change doesn't necessarily violate that principle. My main argument was "If I can get this done anyone can. It's not that much different, it's just different". It appears to me to be just as portable as the current setup and it might just save the Devs some time, and provide better integration with upstream devs. Another salient point is just because it's been done that way since Moses was a baby doesn't make it right. Sorry if I offend anyone. Sorry about the links. Mya -- Life's fun when your sick and psychotic!
On Sunday 22 Jul 2012 15:14:13 Myra Nelson wrote:
My comparison to file size was meant to be extended to the complete removal of rc.d and conf.d or the removal of several files in those directories. Maybe that concept is not that important. I didn't mean to imply the KISS principle was about size, just trying to imply this change doesn't necessarily violate that principle.
When I read about file sizes, my immediate thought was that it's being used as a metric for simplicity of configuration. Did noone else think this? If the configuration file sizes are small, this might indicate that it won't be a nightmare to configure.
My main argument was "If I can get this done anyone can. It's not that much different, it's just different". It appears to me to be just as portable as the current setup and it might just save the Devs some time, and provide better integration with upstream devs. Another salient point is just because it's been done that way since Moses was a baby doesn't make it right. Sorry if I offend anyone.
Thank you for sharing this Myra; I think this is a helpful argument. The lesson seems to be: systemd is not difficult to use; it's just new, shiny, and different. Don't be scared of it. As for me, I haven't made the switch yet mainly because I haven't had the time to look into it properly. Paul
systemd is not difficult to use;
But it isn't as easy untill the gui comes along is it. I'd guess an ncurses gui is planned but apparently Fedora is only meant as a desktop? -- ________________________________________________________ Why not do something good every day and install BOINC. ________________________________________________________
On Mon, Jul 23, 2012 at 3:38 AM, Paul Gideon Dann <pdgiddie@gmail.com> wrote:
On Sunday 22 Jul 2012 15:14:13 Myra Nelson wrote:
My comparison to file size was meant to be extended to the complete removal of rc.d and conf.d or the removal of several files in those directories. Maybe that concept is not that important. I didn't mean to imply the KISS principle was about size, just trying to imply this change doesn't necessarily violate that principle.
When I read about file sizes, my immediate thought was that it's being used as a metric for simplicity of configuration. Did noone else think this? If the configuration file sizes are small, this might indicate that it won't be a nightmare to configure.
My main argument was "If I can get this done anyone can. It's not that much different, it's just different". It appears to me to be just as portable as the current setup and it might just save the Devs some time, and provide better integration with upstream devs. Another salient point is just because it's been done that way since Moses was a baby doesn't make it right. Sorry if I offend anyone.
Thank you for sharing this Myra; I think this is a helpful argument. The lesson seems to be: systemd is not difficult to use; it's just new, shiny, and different. Don't be scared of it.
As for me, I haven't made the switch yet mainly because I haven't had the time to look into it properly.
Paul
Paul: A voice of reason in the darkness. The size comparisons were a silly comparison to such BS in another thread about systemd. You seem to be the only one who picked up on the central point of my argument. If a half brain dead old woman can do this with minimal problems the younger generation shouldn't have any problem with it. Thank you very much. Myra -- Life's fun when your sick and psychotic!
Hi all, I've used Arch for the last 4 years and I'm increasingly comfortable using it On the switch to systemd, people interested in the topic should give a look at the Fosdem presentation about the project a year ago ;) systemd, beyond init http://www.youtube.com/watch?v=TyMLi8QF6sw&feature=plcp Best, Antoine 2012/7/23 Myra Nelson <myra.nelson@hughes.net>:
On Mon, Jul 23, 2012 at 3:38 AM, Paul Gideon Dann <pdgiddie@gmail.com> wrote:
On Sunday 22 Jul 2012 15:14:13 Myra Nelson wrote:
My comparison to file size was meant to be extended to the complete removal of rc.d and conf.d or the removal of several files in those directories. Maybe that concept is not that important. I didn't mean to imply the KISS principle was about size, just trying to imply this change doesn't necessarily violate that principle.
When I read about file sizes, my immediate thought was that it's being used as a metric for simplicity of configuration. Did noone else think this? If the configuration file sizes are small, this might indicate that it won't be a nightmare to configure.
My main argument was "If I can get this done anyone can. It's not that much different, it's just different". It appears to me to be just as portable as the current setup and it might just save the Devs some time, and provide better integration with upstream devs. Another salient point is just because it's been done that way since Moses was a baby doesn't make it right. Sorry if I offend anyone.
Thank you for sharing this Myra; I think this is a helpful argument. The lesson seems to be: systemd is not difficult to use; it's just new, shiny, and different. Don't be scared of it.
As for me, I haven't made the switch yet mainly because I haven't had the time to look into it properly.
Paul
Paul:
A voice of reason in the darkness. The size comparisons were a silly comparison to such BS in another thread about systemd. You seem to be the only one who picked up on the central point of my argument. If a half brain dead old woman can do this with minimal problems the younger generation shouldn't have any problem with it.
Thank you very much.
Myra
-- Life's fun when your sick and psychotic!
participants (9)
-
Antoine Jardin
-
C Anthony Risinger
-
David C. Rankin
-
Karol Babioch
-
Kevin Chadwick
-
Mantas Mikulėnas
-
Myra Nelson
-
Nicholas MIller
-
Paul Gideon Dann