Latest build has been failing in tests but this seems to be a real problem in the code. Failure is: galerautils/tests/gu_rset_test.cpp:164:E:RecordSet v1:ver1:0: (after this point) Received signal 7 (Bus error) The faulty code is here, there is a comment (copied from reference implementation) saying alignment needs to be handled there, and it is not: static GU_FORCE_INLINE void _mmh3_128_blocks (const uint64_t* const blocks, size_t const nblocks, uint64_t* h1, uint64_t* h2) { //---------- // body size_t i; for(i = 0; i < nblocks; i++) { //----------------------------------------------------------------------------- // Block read - if your platform needs to do endian-swapping or can only // handle aligned reads, do the conversion here uint64_t k1 = gu_le64(blocks[i]); i++; uint64_t k2 = gu_le64(blocks[i]); _mmh3_128_block (k1, k2, h1, h2); } } I failed to find a place for upstream bug reports on https://galeracluster.com/ so recording it here.
Someone wrote a portable library for murmur, I wonder if it could be used instead https://github.com/kloetzl/libmurmurhash That one does it right https://github.com/kloetzl/libmurmurhash/blob/master/PMurHash.c#L214
Thank you for reporting this; and the detective work to find the cause. > I failed to find a place for upstream bug reports on > https://galeracluster.com/ so recording it here. I also could not see a place on their site for reporting a bug. For starters, please ask info@codership.com (their contact e-mail) about how to do so; then when you have some sort of bug reference, please post that on *this* bug in the URL field in the header. And notify when you hear it has been fixed. Assigning to DavidG who has this package; I imagine he will await the resolution upstream.
Keywords: (none) => UPSTREAMAssignee: bugsquad => geiger.david68210
That email unfortunately goes to some Sales Manager and after explaining that no I am not working on a solution for a customer I didn't get another reply, so I have no idea if it was passed to technical people.
Looking around, Debian is still on 25.3.* (they updated to 25.3.29 few weeks ago) so I can't steal a fix from there. I'll try to work on it at some point but this is lower priority than many things...
Maybe we can report the arm build issue here: https://github.com/codership/galera
So Debian has a different compile as ours, they use: # Use bundled asio, since system asio is too new (> 1.10.8) SCONS_ARGS += system_asio=0 and: # Bus errors on arm64 kernels prevent running testsuite when building armhf ifeq (armhf,$(DEB_HOST_ARCH)) SCONS_ARGS += tests=0 deterministic_tests=0 and: # Use strict compilation flags (ie -Werror) if requested ifneq (,$(filter strict_build_flags,$(DEB_BUILD_OPTIONS))) SCONS_ARGS += strict_build_flags=1 WDYT?
Hmm so their solution is to disable the tests on arm to not fail despite the problem being in the code and not the tests :(
I opened https://github.com/codership/galera/issues/570
latest 26.4.4 release fixes armv7hl build, so closing of this bug!
Status: NEW => RESOLVEDResolution: (none) => FIXED