Bug#1051649: libportal: FTBFS on riscv64 due to test timeout
Hi, On 2023-09-15 13:31, Simon McVittie wrote: > Control: tags 1051649 + pending > > On Mon, 11 Sep 2023 at 00:17:31 +0200, Aurelien Jarno wrote: > > libportal fails to build from source on riscv64 (and a few other slow > > architectures) with a timeout in a test > ... > > After investigation, it appeared the test actually passes, but needs > > 85 seconds instead of the 60 seconds it got allocated. The > > following patch uses the --timeout-multiplier feature of meson to > > increase the timeout. > > Thanks, I'll upload a similar fix after the current version has migrated > to testing (it's only 1 day off, so it would seem a shame to reset the > clock for this). I increased the timeout by a factor of 3 rather than 2, > to give some margin of error. Thanks! > However, I'm concerned that this implies riscv64 might be our new slowest > release architecture, even slower than mips64el, which is going to put > it at risk of delaying migrations, security fixes and other release > stuff (as a result of builds taking a long time, arbitrary timeouts in > build-time tests becoming insufficient, or race conditions in build-time > tests being hit when they wouldn't have been seen on faster buildds). It depends how you count. While the build time for a single package takes longer than some mips64el buildds, we plan to use more buildds. We currently have 7 buildds, and 2 more are waiting to be installed. This should ensure that build queues do not fill up, at least once the whole set of packages has been rebuilt. > Do you expect faster riscv64 buildds to become available by the time > trixie is the stable release, or is what we have now what we are going > to continue to have? We do have a set of faster buildds available for a few months already, using the VisionFive 2 boards. They are around 80% faster than the current HiFive Unmatched based buildds for building packages, and libportal's testsuite passes on them without changing the timeout factor. Unfortunately they lack kernel support in mainline, so they can't be used and thus are just stored in a box. With the kernel 6.6-rc1, we are down to 19 missing patches to support these boards, we expect full support will arrive in linux 6.7 or 6.8. We also hope that the Milk-V Pioneer board (already released) and HiFive Pro P550 board (planned) will enable us to get way faster riscv64 buildds, but it is still uncertain until they get mainline support, they get tested in a configuration similar to the buildds and that we are sure that we can host them in a datacenter. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://aurel32.net
Bug#1051649: libportal: FTBFS on riscv64 due to test timeout
Control: tags 1051649 + pending On Mon, 11 Sep 2023 at 00:17:31 +0200, Aurelien Jarno wrote: > libportal fails to build from source on riscv64 (and a few other slow > architectures) with a timeout in a test ... > After investigation, it appeared the test actually passes, but needs > 85 seconds instead of the 60 seconds it got allocated. The > following patch uses the --timeout-multiplier feature of meson to > increase the timeout. Thanks, I'll upload a similar fix after the current version has migrated to testing (it's only 1 day off, so it would seem a shame to reset the clock for this). I increased the timeout by a factor of 3 rather than 2, to give some margin of error. However, I'm concerned that this implies riscv64 might be our new slowest release architecture, even slower than mips64el, which is going to put it at risk of delaying migrations, security fixes and other release stuff (as a result of builds taking a long time, arbitrary timeouts in build-time tests becoming insufficient, or race conditions in build-time tests being hit when they wouldn't have been seen on faster buildds). Do you expect faster riscv64 buildds to become available by the time trixie is the stable release, or is what we have now what we are going to continue to have? This might also resolve #1051702, but obviously I haven't tested on hppa hardware to find out how long this test actually takes there, so I'm not marking that one as pending. smcv
Bug#1051649: libportal: FTBFS on riscv64 due to test timeout
Source: libportal Version: 0.7.1-1 Severity: important Tags: patch ftbfs Justification: fails to build from source (but built successfully in the past) User: debian-ri...@lists.debian.org Usertags: riscv64 X-Debbugs-Cc: debian-ri...@lists.debian.org Dear maintainer, libportal fails to build from source on riscv64 (and a few other slow architectures) with a timeout in a test: | 2/2 = | test: pytest | start time: 16:03:45 | duration: 60.05s | result: killed by signal 15 SIGTERM | command: MALLOC_PERTURB_=44 LD_LIBRARY_PATH=/<>/obj-riscv64-linux-gnu/libportal GI_TYPELIB_PATH=/<>/obj-riscv64-linux-gnu/libportal /usr/bin/pytest-3 --verbose --log-level=DEBUG | --- stdout --- | = test session starts == | platform linux -- Python 3.11.5, pytest-7.4.1, pluggy-1.3.0 -- /usr/bin/python3 | cachedir: .pytest_cache | rootdir: /<>/tests | collecting ... collected 30 items | | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_close_session PASSED [ 3%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_connect_to_eis PASSED [ 6%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_connect_to_eis_fail_connect_before_start PASSED [ 10%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_connect_to_eis_fail_reconnect PASSED [ 13%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_connect_to_eis_v1 PASSED [ 16%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_create_session PASSED [ 20%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_create_session_no_outputs PASSED [ 23%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_axis PASSED [ 26%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_axis_discrete PASSED [ 30%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_button PASSED [ 33%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_key PASSED [ 36%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_motion PASSED [ 40%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_motion_absolute PASSED [ 43%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_notify_touch PASSED [ 46%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_session_start PASSED [ 50%] | pyportaltest/test_remotedesktop.py::TestRemoteDesktop::test_version PASSED [ 53%] | pyportaltest/test_screencast.py::TestScreenCast::test_close_session PASSED [ 56%] | pyportaltest/test_screencast.py::TestScreenCast::test_create_session PASSED [ 60%] | pyportaltest/test_screencast.py::TestScreenCast::test_create_session_v3 | == | | | Summary of Failures: | | 2/2 pytest TIMEOUT60.05s killed by signal 15 SIGTERM | | Ok: 1 | Expected Fail: 0 | Fail: 0 | Unexpected Pass:0 | Skipped:0 | Timeout:1 | dh_auto_test: error: cd obj-riscv64-linux-gnu && DEB_PYTHON_INSTALL_LAYOUT=deb LC_ALL=C.UTF-8 MESON_TESTTHREADS=4 meson test returned exit code 1 | make[1]: *** [debian/rules:60: override_dh_auto_test] Error 25 | make[1]: Leaving directory '/<>' | make: *** [debian/rules:51: binary-arch] Error 2 | dpkg-buildpackage: error: debian/rules binary-arch subprocess returned exit status 2 The full build log is available there: https://buildd.debian.org/status/fetch.php?pkg=libportal&arch=riscv64&ver=0.7.1-1&stamp=1694361910&raw=0 After investigation, it appeared the test actually passes, but needs 85 seconds instead of the 60 seconds it got allocated. The following patch uses the --timeout-multiplier feature of meson to increase the timeout. I didn't try to use a different multiplier depending on the architecture as it has no impact on working tests anyway. --- libportal-0.7.1/debian/rules +++ libportal-0.7.1/debian/rules @@ -57,4 +57,4 @@ $(NULL) override_dh_auto_test: - xvfb-run -a dh_auto_test + xvfb-run -a dh_auto_test -- --timeout-multiplier 2 Regards Aurelien
