--- Begin Message ---
- To: submit@bugs.debian.org
- Subject: overwriting {,src}pkgcache.bin on near-every APT call strains flash
- From: Ivan Shmakov <ivan@siamics.net>
- Date: Sun, 30 Jan 2022 19:10:34 +0000
- Message-id: <20220130191033.4tduts6hl65zlnq2@violet.siamics.net>
Package: apt
Version: 2.3.14
Severity: minor
Control: found -1 1.8.2.3
[Please do not Cc: me, for I’m “on the list,” so to say, and
I try to reserve my inbox for private communication only.
I’d have set up Mail-Followup-To:, but there doesn’t seem to
be a way to make it point to the bug being filed.]
As of this writing, pkgcache.bin and srcpkgcache.bin files for
an APT instance configured for testing / Sid are 61 MiB combined.
The way those caches are updated is by writing a new version
into a temporary file and then calling rename(2) to atomically
replace the old version with the new. As such, every time the
caches are updated, some 61 MiB gets written to the filesystem
While it could be argued that such an amount written makes
no measurable impact on HDDs, and rather marginally affects
the lifetime of contemporary SSDs, it’s not uncommon to start
Debian from a micro-SD card on single-board computers, where
write cycles may be considerably more limited.
Similarly, when filesystems-level snapshots are supported
(Btrfs, Nilfs, etc.), the issue described prevents space (and
bandwidth, when snapshots are transferred with btrfs-send(8) /
btrfs-receive(8)) savings by getting in the way of the
filesystem’s copy-on-write behavior.
(FWIW, apt is by no means the only package with such an issue:
dpkg’s status file is updated the same way, and so is the
templates.dat file debconf uses by default; yet dpkg/status is
smaller, and debconf can be configured to use Driver: PackageDir,
also lowering the impact. Outside of the Debian infrastructure,
tor caches are also ill-suited for write-limited media.)
Consider, for example, the following chroot environment with a
freshly installed Debian Bookworm (--arch=amd64 --variant=minbase
and a couple of packages on top of that):
chroot# apt-get update
Hit:1 http://cdn-fastly.deb.debian.org/debian bookworm InRelease
Hit:2 http://security.debian.org/debian-security bookworm-security/updates InRelease
Reading package lists...
chroot#
Now, let’s try to install some lightweight packages, checking
the ‘lifetime writes’ meter before and after the apt-get call:
base# mount -o remount,ro -- /dev/vgfoo/lvchild-z61f590 \
&& dumpe2fs -- /dev/vgfoo/lvchild-z61f590 | grep -E -- ^Lifetime
Lifetime writes: 5418 MB
base#
chroot# apt-get install -- jpeginfo
Reading package lists...
Building dependency tree...
Reading state information...
The following additional packages will be installed:
libjpeg62-turbo
The following NEW packages will be installed:
jpeginfo libjpeg62-turbo
0 upgraded, 2 newly installed, 0 to remove and 5 not upgraded.
Need to get 0 B/177 kB of archives.
After this operation, 725 kB of additional disk space will be used.
Do you want to continue? [Y/n]
debconf: delaying package configuration, since apt-utils is not installed
E: Can not write log (Is /dev/pts mounted?) - posix_openpt (19: No such device)
Selecting previously unselected package libjpeg62-turbo:amd64.
(Reading database ... 6737 files and directories currently installed.)
Preparing to unpack .../libjpeg62-turbo_1%3a2.1.2-1_amd64.deb ...
Unpacking libjpeg62-turbo:amd64 (1:2.1.2-1) ...
Selecting previously unselected package jpeginfo.
Preparing to unpack .../jpeginfo_1.6.1-1_amd64.deb ...
Unpacking jpeginfo (1.6.1-1) ...
Setting up libjpeg62-turbo:amd64 (1:2.1.2-1) ...
Setting up jpeginfo (1.6.1-1) ...
Processing triggers for libc-bin (2.33-3) ...
chroot#
base# mount -o remount,ro -- /dev/vgfoo/lvchild-z61f590 \
&& dumpe2fs -- /dev/vgfoo/lvchild-z61f590 | grep -E -- ^Lifetime
Lifetime writes: 5740 MB
base#
By the looks of it, installing 725 KiB worth of Debian packages
resulted in some 322 MiB getting written to the filesystem,
for which I’m inclined to think that APT cache update behavior
is in no small part responsible. (I suppose running apt-get
under strace would allow for a more accurate estimate.)
For instance, largest recently changed files on the FS right
after the # apt-get install call were (note that the .deb files
to be installed were obtained earlier, and also that
/var/cache/apt/archives was on a separate FS anyway):
chroot# find / -xdev -cmin -4 -printf %k\\t%p\\n | sort -srn
31568 /var/cache/apt/pkgcache.bin
31548 /var/cache/apt/srcpkgcache.bin
580 /usr/lib/x86_64-linux-gnu/libjpeg.so.62.3.0
88 /var/lib/dpkg/status-old
88 /var/lib/dpkg/status
52 /var/log/dpkg.log
40 /usr/share/doc/libjpeg62-turbo/copyright
A proper solution would be to move to some file format that
allows in-place updates, such as Berkeley DB or SQLite.
As a work-around, it’s possible to move the cache to tmpfs;
consider, e. g.:
$ cat < /etc/apt/apt.conf.d/99-run-pkgcache
Dir::Cache::pkgcache "/run/apt/pkgcache.bin";
Dir::Cache::srcpkgcache "/run/apt/srcpkgcache.bin";
$ grep -F -- /run/apt < /etc/fstab
run.apt /run/apt tmpfs rw,nodev,noexec,nosuid,mode=0755,size=256M,X-mount.mkdir 0 0
$
When RAM is also limited (which is to say, the filesystem size=
limit above isn’t negligible), it may make sense to remove the
files when not in use, such as by enabling the atime filesystem
option and using the following (untested) Cron job:
$ grep -F -- /run/apt < /etc/fstab
run.apt /run/apt tmpfs rw,nodev,noexec,nosuid,mode=0755,size=256M,X-mount.mkdir,atime 0 0
$ cat < /etc/cron.hourly/cleanup-run-apt
#!/bin/sh
set -e
set -C -u
test -d /run/apt \
|| exit
## .
exec find /run/apt/ -maxdepth 1 \
-type f -amin +113 -cmin +113 \
-size +63k -name \*pkgcache.bin\* \
-execdir rm -- {} +
$
--
FSF associate member #7257 http://am-1.org/~ivan/
--- End Message ---
--- Begin Message ---
- To: 1004583-close@bugs.debian.org
- Subject: Re: Bug#1004583: overwriting {,src}pkgcache.bin on near-every APT call strains flash
- From: Julian Andres Klode <jak@debian.org>
- Date: Tue, 1 Feb 2022 12:14:54 +0100
- Message-id: <20220201121150.GA1245217@debian.org>
- In-reply-to: <20220130191033.4tduts6hl65zlnq2@violet.siamics.net>
- References: <20220130191033.4tduts6hl65zlnq2@violet.siamics.net>
On Sun, Jan 30, 2022 at 07:10:34PM +0000, Ivan Shmakov wrote:
> Package: apt
> Version: 2.3.14
> Severity: minor
> Control: found -1 1.8.2.3
>
> [Please do not Cc: me, for I’m “on the list,” so to say, and
> I try to reserve my inbox for private communication only.
> I’d have set up Mail-Followup-To:, but there doesn’t seem to
> be a way to make it point to the bug being filed.]
>
> As of this writing, pkgcache.bin and srcpkgcache.bin files for
> an APT instance configured for testing / Sid are 61 MiB combined.
>
> The way those caches are updated is by writing a new version
> into a temporary file and then calling rename(2) to atomically
> replace the old version with the new. As such, every time the
> caches are updated, some 61 MiB gets written to the filesystem
This is the correct way to do it, and it's not going to change.
The srcpkgcache.bin is rewritten on update time only, the
pkgcache.bin also after an install.
This means, with daily updates and upgrades enabled and 60 MB caches,
you end up with
2 * 60 MB # update
+ 1 * 60 MB # upgrade
-----------------------------
180 MB
written per day, or 5.4 GB per month.
There is nothing we can do about that.
--
debian developer - deb.li/jak | jak-linux.org - free software dev
ubuntu core developer i speak de, en
--- End Message ---