Hi Emilio, On 2025-09-23 10:14, Emilio Pozuelo Monfort wrote: > Hi Aurelien, > > On Sun, 31 Aug 2025 10:37:19 +0200 Aurelien Jarno <[email protected]> wrote: > > control: tag - 1 + patch > > > > Hi, > > > > On 2025-08-26 11:18, Graham Inggs wrote: > > > Source: guestfs-tools > > > Version: 1.52.3-1 > > > Severity: serious > > > Tags: ftbfs > > > X-Debbugs-Cc: [email protected] > > > > Hi Maintainer > > > > Since the recent binNMU for libxml2, guestfs-tools FTBFS on riscv64 > > > [1], where it built previously. > > > > I think this is the only test failure: > > > > FAIL test-virt-alignment-scan-guests.sh (exit status: 1) > > > > I initially got difficulties to reproduce the issue locally, but I have > > finally been able to reproduce the issue on ricci.d.o, the riscv64 > > porterbox. It is reproducible most of the time (8/10) in a sid chroot, > > but more rarely (1/10) in a trixie chroot. It appeared to be a timing > > issue, i.e. things are running a tiny bit slower in sid than in trixie > > (presumably due to new version of packages, either QEMU or things > > running inside the VM), causing the failure to happen way more often. > > > > Looking at it more in details it appears that the testsuite launches 12 > > qemu-system processes in parallel, which make things quite slow. It > > makes sense when KVM acceleration is available (the test is probably IO > > bound), but not with TCG as used on the build daemons (the test is > > definitely CPU bound). test-virt-df-guests.sh is also affected by the > > same issue. > > > > Therefore I came up with the following patch to limit the number of VMs > > to the number of CPUs: > > > > --- guestfs-tools-1.52.3.orig/align/test-virt-alignment-scan-guests.sh > > +++ guestfs-tools-1.52.3/align/test-virt-alignment-scan-guests.sh > > @@ -21,7 +21,7 @@ skip_if_skipped > > > > libvirt_uri="test://$abs_top_builddir/test-data/phony-guests/guests-all-good.xml" > > -$VG virt-alignment-scan -c "$libvirt_uri" > > +$VG virt-alignment-scan -P$(nproc) -c "$libvirt_uri" > > r=$? > > # 0, 2 and 3 are reasonable non-error exit codes. Others are errors. > > --- guestfs-tools-1.52.3.orig/df/test-virt-df-guests.sh > > +++ guestfs-tools-1.52.3/df/test-virt-df-guests.sh > > @@ -25,4 +25,4 @@ skip_if_skipped > > libvirt_uri="test://$abs_top_builddir/test-data/phony-guests/guests.xml" > > -$VG virt-df -c "$libvirt_uri" > > +$VG virt-df -P $(nproc) -c "$libvirt_uri" > > > > In addition to fixing the issue, it also makes the tests around 70% > > faster on a system where the tests were already passing. > > I wonder if instead of nproc, we shouldn't use DEB_BUILD_OPTION_PARALLEL > threads instead. But it works here, so I'll upload it to unblock the > libxml2.9 removal.
I agree with you from the Debian point of view. But if we want this patch to be upstreamble, nproc is probably a better choice. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B [email protected] http://aurel32.net
