| 1 |
On Tue, Jul 28, 2020 at 03:19:26PM -0400, Mike Gilbert wrote: |
| 2 |
> On Tue, Jul 28, 2020 at 3:13 PM Mike Gilbert <floppym@g.o> wrote: |
| 3 |
> > |
| 4 |
> > On Tue, Jul 28, 2020 at 2:50 PM Zoltan Puskas <zoltan@×××××××××.info> wrote: |
| 5 |
> > > |
| 6 |
> > > Hi, |
| 7 |
> > > |
| 8 |
> > > I've upgraded to and running systemd-246_rc2 on one of my systems and |
| 9 |
> > > noticed that tmpfs mounted directories are significantly smaller. |
| 10 |
> > > |
| 11 |
> > > This is because with commit |
| 12 |
> > > https://github.com/systemd/systemd/commit/7d85383edbab73274dc81cc888d884bb01070bc2 |
| 13 |
> > > they have changed them to be 10% of the physical memory instead of the |
| 14 |
> > > default of 50%. |
| 15 |
> > > |
| 16 |
> > > This is a potentially breaking, or at least an unexpected behaviour |
| 17 |
> > > change, especially for people using tmpfs on /tmp for compiling. |
| 18 |
> > > |
| 19 |
> > > Maybe we should make a news item to let people know that they either |
| 20 |
> > > need to add an fstab entry with size option set, or better, create a |
| 21 |
> > > systemd local override with relevant content. |
| 22 |
> > |
| 23 |
> > Don't use /tmp for PORTAGE_TMPDIR. /tmp is meant for small temporary |
| 24 |
> > storage. If you want to compile in a tmpfs, set up a separate mount |
| 25 |
> > point for it. |
| 26 |
> > |
| 27 |
> > I don't intend to create a news item for this, but I would not object |
| 28 |
> > to someone else doing it. |
| 29 |
> |
| 30 |
> Also, the limit for /tmp is likely to change again before the 246 final release. |
| 31 |
> |
| 32 |
> https://github.com/systemd/systemd/pull/16576 |
| 33 |
> |
| 34 |
|
| 35 |
I volunteer to write the news item. |
| 36 |
|
| 37 |
It seems other distros also have met the overly restricted /tmp size |
| 38 |
issue, due to yet another legitimate use case (see the RedHat bug |
| 39 |
referenced in the PR: https://bugzilla.redhat.com/show_bug.cgi?id=1856514). |
| 40 |
|
| 41 |
It's worth noting that above PR only resets only /tmp size, but will |
| 42 |
keep /dev/shm, /run, etc. at the lower limit. While we have to wait and |
| 43 |
see what the final form will be for systemd-246, I think it'd be useful |
| 44 |
to notify users. Systemd changing the age old convention of non |
| 45 |
configured tmpfs mount sizes (from a user's perspective non configured) |
| 46 |
it means openrc and systemd boxes will end up with different behaviours |
| 47 |
(e.g. /dev/shm will now be sized differently). |
| 48 |
|
| 49 |
Cheers, |
| 50 |
Zoltan |
Archives