Hi!

A user reported a significant slowdown of an operation which used to
work quite a bit faster.  This time, however, the regression is not
on the qemu side but on the guest OS side (debian bookworm to debian
trixie), running in the same qemu-user.  It is more, - it doesn't
really matter which version of qemu-user to use (from a few recent
versions).

For example, building a particular debian source package (in this case,
bird3) in qemu-user-backed aarch64 debian chroot for current debian
release (trixie) takes at least as twice time than the same operation
on bookworm (previous debian release).

I can confirm this slowdown, however, I see about 2x time difference,
while the OP reports 10x slowdown.  https://bugs.debian.org/1113951
has a few more details about this.

I observe this difference (omitting the build logs):

 $ sbuild -d bookworm --arch arm64 bird3_3.1.0-1.dsc
 Build needed 00:14:45, 265340k disk space

 $ sbuild -d trixie --arch arm64 bird3_3.1.0-1.dsc
 Build needed 00:27:00, 267508k disk space

One of the most notable contributors in this time difference seems
to be tex setup procedure:

 Processing triggers for tex-common (6.19) ...
 Running updmap-sys. This may take some time... done.
 Running mktexlsr /var/lib/texmf ... done.
 Building format(s) --all.

at this point it runs pdftex, luatex, luahbtex and a few other
tex-related programs, which are visible in top(1) output for minutes
when running in trixie chroot.

Other programs are also running slower, but not that significant.

I'll try to provide a minimal reproducer here.

But overall, how to debug such situation, when it is not qemu which
is at issue?

Thanks,

/mjt

Reply via email to