Description of problem:
No VM's core dumps after kill vm.
Red Hat Enterprise Virtualization Hypervisor 7.1 (20150420.0.el7ev)
# rpm -qf /var/log/core/
Red Hat Enterprise Virtualization Manager Version: 3.5.1-0.4.el6ev
Steps to Reproduce:
1. Install RHEVH-7.1-20150420.0
2. Register to RHEVM3.5.1.
3. Create VM.
4. Run command "pkill -ABRT -f 'qemu-kvm'" on host.
5. Run ll /var/log/core/ on host.
There is a vdsm patch that starts the coredump configuration  with systemd, however it's required one additional step, which is add into libvirtd.service LimitCORE=infinity as the libvirt will trigger qemu-kvm.
# cat ./usr/lib/systemd/system/libvirtd.service
LimitCORE=infinity <------------------- Added
ExecReload=/bin/kill -HUP $MAINPID
# Override the maximum number of opened files
supervdsm: enable coredumping
I don't think vdsm shouldn't touch libvirtd.service. I prefer to hear more views - ccing Danken to hear his opinion and Jiri to understand why the default is with coredump limitation and if its more reasonable to change that default.
Normal systems (RHEL, for example) run abrt daemon which will handle the crash and process or store the core dump depending on its configuration regardless on LimitCORE settings. On hosts without abrt daemon, setting LimitCore to unlimited by default is not a very good idea since core files may be really big, especially those of qemu-kvm processes running big guests. And I believe even vdsm should not enable them by default. You should rather start using abrt which should be even able to process the core file without storing it on a disk first (although this functionality is not available in RHEL 7.1, AFAIK).
However, if you still want to set LimitCORE=unlimited (either by default or on demand), you should not mess with /usr/lib/systemd/system/libvirtd.service. You should do this by creating a new file in /etc/systemd/system/libvirtd.service.d/. For example,
$ cat /etc/systemd/system/libvirtd.service.d/unlimited-core.conf
Yaniv, as Jiri hints, the best solution is to implement bug 917062. But until this happens, we should solve our el7 regression, and add the suggested libvirt.conf to our libvirt configurator.
Thanks for the replies! Moving to 3.6 and we'll consider to add it soon. I prefer not to introduce this change in 3.5
FYI I reported this as well and got closed by mskrivan@, he told me it has never been intended to have core files. See https://bugzilla.redhat.com/show_bug.cgi?id=1220860
# rpm -qf /etc/systemd/system/libvirtd.service.d/unlimited-core.conf
# cat /etc/systemd/system/libvirtd.service.d/unlimited-core.conf
## beginning of configuration section by vdsm
## end of configuration section by vdsm
# file /var/log/core/core.519.1435222212.dump
/var/log/core/core.519.1435222212.dump: ELF 64-bit LSB core file x86-64, version 1 (SYSV), SVR4-style, from '/usr/libexec/qemu-kvm -name test -S -machine rhel6.5.0,accel=kvm,usb=off -cpu S'
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.