Le samedi 10 novembre 2007 à 14:44 +0100, Frans Pop a écrit : > D-I does allow setting up a system without swap and I have seen installation > reports where users actually did that. There's nothing we can do against users shooting themselves in the foot. > Of course, in most cases that will > be on systems with a fair amount of RAM, but maybe it would be a good idea > to not set up tmpfs in that case (and just let tmp be part of /). In that case the user is just screwed, but it is indeed better to not make things worse by setting up tmpfs. I'm attaching a new version of the corresponding script. > I've also seen some systems will get really heavy tmp usage. If you look at > gluck for example, that currently has ~500MB used and I have seen it use > the full 2.8GB, mostly because of stale files from various cron jobs and > temporary CVS server files. Well, in the case of gluck the default tmpfs size would be 2 GiB, so I say that would be enough. The administrator of a machine can still increase that size if needed. With the proposed changes, he can also still setup a /tmp partition by hand in the installer - you don't use partman-auto when you know the server will be special. > I also must say that I'm not sure if I'd want tmp to use up memory/swap > space on the two old Pentium desktops I abuse as servers here at home. > Somehow just having /tmp on physical disk just feels safer. I think having it on a physical disk that is purged upon reboot merely makes a psychological difference. > For desktop usage, using tmpfs probably makes sense, but for servers? Servers benefit from tmpfs just as much as desktops. If the root partition is full, you can't do anything on both of them. By enforcing the separation of /tmp, creating a tmpfs if there is no dedicated partition, we can guarantee that a reboot will let enough room for daemons to start. I've been using tmpfs for a long time on most systems I've owned or administered, including clusters of hundreds of nodes and giant file servers, and it only showed benefits. (Well, there was that crappy kernel module that crashed when writing debugging information in /tmp, but that was a bug that we actually helped fixing this way.) > Any idea what other distros do? *cough* I've seen some distributions not even purging their /tmp at reboot time despite what the FHS says, so they're completely screwed on tmpfs. I don't know for all distributions, though. Cheers, -- .''`. : :' : We are debian.org. Lower your prices, surrender your code. `. `' We will add your hardware and software distinctiveness to `- our own. Resistance is futile.
Attachment:
tmpfs_for_tmp
Description: application/shellscript
Attachment:
signature.asc
Description: Ceci est une partie de message =?ISO-8859-1?Q?num=E9riquement?= =?ISO-8859-1?Q?_sign=E9e?=