Re: PR/54886 CVS commit: src/usr.sbin/sysinst

[Martin Husemann <martin%duskware.de@localhost>]

The following reply was made to PR misc/54886; it has been noted by GNATS.

From: Martin Husemann <martin%duskware.de@localhost>
To: Andreas Gustafsson <gson%gson.org@localhost>
Cc: gnats-bugs%netbsd.org@localhost, ad%NetBSD.org@localhost
Subject: Re: PR/54886 CVS commit: src/usr.sbin/sysinst
Date: Wed, 11 Mar 2020 10:49:13 +0100

 On Wed, Mar 11, 2020 at 11:35:50AM +0200, Andreas Gustafsson wrote:
 > Martin,
 > 
 > On Jan 24, you wrote:
 > >  Log Message:
 > >  Factor out all RAM size thresholds as defines to avoid magic numbers.
 > >  To work around PR misc/54886 bump the threshold for a tmpfs /tmp mount
 > >  up to 256 MB.
 > >  
 > >  To generate a diff of this commit:
 > >  cvs rdiff -u -r1.36 -r1.37 src/usr.sbin/sysinst/bsddisklabel.c
 > >  cvs rdiff -u -r1.52 -r1.53 src/usr.sbin/sysinst/defs.h
 > >  cvs rdiff -u -r1.60 -r1.61 src/usr.sbin/sysinst/disks.c
 > 
 > Could you bump this a bit further, for example to 512 MB?  The amd64
 > port no longer reliably installs in 128 MB (it gets "out of swap"
 > errors during set extraction), and if you increase memory to 256 MB to
 > work around this, the tmpfs /tmp activates, and although the system
 > does not hang as it did in PR 54886, many of the ATF file system tests
 > fail with ENOSPC.
 
 I think there are two issues here that we should address separately.
 
 First with 128 MB RAM obviously the swap partition is created too small,
 sysinst scales it probably to 256 MB and we maybe should have a MD lower
 limit for it (if space on target disk permits).
 
 The second issue is ATF tests requiring at least a certain /tmp size, and
 25% of RAM (the default) not being enough for that. Do you have any idea
 what minimal /tmp size is needed? I can increase the default tmpfs size on
 lower ram machines to more than 25%, or as you suggested just raise the
 threshold. With enough swap available, bigger tmpfs is no problem (I have
 localy used very high percent-of-ram tmpfs setups on big memory machines
 to build completely in a tmpfs).
 
 Martin