Remove pycache to save space (20:4 MiB pre:post squashfs-ing)

[mvidner]

Related to:

• https://bugzilla.suse.com/show_bug.cgi?id=1172139
• https://trello.com/c/wnyocCGR/1961-5-reduce-installer-memory-needs-disk-files

Files in the __pycache__ directories are compiled versions of the sources, used for faster loading. In the inst-sys, saving space is more important than the small speedup, so we delete them.

Size Before (Tumbleweed of 2020-07-04)

$  wget https://download.opensuse.org/tumbleweed/repo/oss/boot/x86_64/{common,root}
(...)
$ du -hc common root
61M     common
73M     root
133M    total
$ pysize() { unsquashfs -ll $1 | grep __pycache__ | awk '{s += $3} END {print s/1024/1024}'; }
$ pysize common
18.8646
$ pysize root
0.151923

Size After

$ du -hc common root
57M     common
72M     root
129M    total
$ pysize root
0
$ pysize common
0

[lslezak]

Start dir?

[lslezak]

I do not know where the cached files can be located, but do we need to scan the whole tree? How long does it take in OBS? I mean scanning e.g. just usr could be a bit faster... But maybe it's irrelevant considering how long the i-i build actually takes....

[wfeldt]

I don't think it would make much of a difference. /usr is the largest blob anyway and the file tree has just been created and is likely still cached.

[mvidner]

Luckily the measurement is easily obtained, and the answer is: it takes 1 second or less:

$ grep -A1 pycache /var/tmp/build-root-openSUSE_Factory/.build.log
[  333s] running "find . -name __pycache__ -print0| xargs -0 rm -rf"
[  334s] running "touch etc/sysconfig/kernel"
--
[  475s] running "find . -name __pycache__ -print0| xargs -0 rm -rf"
[  475s] running "find var -type f"
--
[ 1348s] running "find . -name __pycache__ -print0| xargs -0 rm -rf"
[ 1348s] running "touch etc/sysconfig/kernel"

[mvidner]

@wfeldt Anything else? Can we merge?

[wfeldt]

lgtm