We have recently been made aware that the internal ABI changes due to LD_AUDIT fixes cause processes launched concurrently during the update to crash:
glibc: Upgrading to glibc-2.28-209.el8.x86_64 causes segfaults during concurrent process launch https://bugzilla.redhat.com/show_bug.cgi?id=2119304
In glibc-2.28-211.el8, we have made the ABI changes more compatible with *older* versions (before -208), but this might cause yet another bumpy update from -209 or -210 to -211 or later. (The version ranges are slightly different for s390x; see the bug. ppc64le has a different issue caused by RPM file writing order, but no crashes.)
CentOS 9 Stream should be less sensitive to these issues due to the merged libpthread/libdl.
We are putting some checks into place to catch this earlier. In general, it might not be possible to avoid such issues completely, particularly on CentOS 8 Stream. The workaround is to do updates in quasi-single-user mode, with as little running on the system as possible.
Thanks, Florian
On Fri, 2022-08-26 at 16:19 +0200, Florian Weimer wrote:
We have recently been made aware that the internal ABI changes due to LD_AUDIT fixes cause processes launched concurrently during the update to crash:
glibc: Upgrading to glibc-2.28-209.el8.x86_64 causes segfaults during concurrent process launch https://bugzilla.redhat.com/show_bug.cgi?id=2119304
In glibc-2.28-211.el8, we have made the ABI changes more compatible with *older* versions (before -208), but this might cause yet another bumpy update from -209 or -210 to -211 or later. (The version ranges are slightly different for s390x; see the bug. ppc64le has a different issue caused by RPM file writing order, but no crashes.)
Thanks for the heads up, this is much appreciated. From the linked bug: "Updates from -208 and earlier to -211 on aarch64, x86_64 are expected not to cause any disruptions" so it sounds like the safest course of action is to skip the intermediate builds and go straight to -211.
Cheers Davide
-
Davide Cavalca -
Florian Weimer