Ok, so a little bit further. This seems to be that stress-ng is
triggering OOMkiller which I'm not sure it should be, unless it's just
really writing a LOT of data to memory.
My setup has been the same on all architectures and I've seen the same
behaviour on Artful deployments on amd64 (Xeon Phi), Power8 and s390x:
created swap space that is 2x the amount of RAM per the output of free
ran the stress-ng based memory stress test from the certification suite
stress-ng tells us that the tests all pass
dmesg reveals invocations of OOMKiller
This is all handled by the memory_stress_ng script in the cert suite.
The tests we run are defined by this block:
crt_stressors=("bsearch" "context" "hsearch" "lsearch" "matrix" \
"memcpy" "null" "pipe" "qsort" "stack" "str" "stream" \
"tsearch" "vm-rw" "wcs" "zero" "mlock" "mmapfork" "mmapmany" \
"mremap" "shm-sysv" "vm-splice")
if [ "$numa_nodes" -gt 1 ]; then
crt_stressors+=("numa")
fi
and we run them individually, one test case at a time.
I've attached a tarball with the stderr and stdout from stress-ng, dmesg
contents and the test script that runs all this.
** Attachment added: "s390x-memory-oomkiller.tgz"
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1729878/+attachment/5006068/+files/s390x-memory-oomkiller.tgz
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1729878
Title:
stress-ng triggering call-traces on s390x z/VM instances
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1729878/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
