Hello list! I'm new here. At first, thank you for KVM!
I run KVM under 2.6.32-279.2.1.el6.x86_64 on CentOS 6.3. My box has 8 GB
RAM, a Quad Core CPU and there are 6 KVM domains that use 5 GB of RAM
and have OS type hvm. Each domain has access to all 4 CPU cores.
Today I was suddenly getting strange effects on one of these domains.
MySQL database queries would take forever. And ocassionally a process
(Varnish web accelerator) would simply exit. The load was about 3.5 but
all 4 CPU cores showed as 90% idle. I checked vmstat and iostat and
there was virtually no load. I could not figure out where that 3.5 load
came from as the domain was pretty much idle.
Then I realized that one of the other 5 domains had become unresponsive.
I destroyed it and immediately afterwards the ill-effects on the first
domain disappeared. I then went through the logs and saw that this
particular domain had a load of 120-200 for a couple of hours, probably
some processes had gone berserk.
All of the other domains behaved fine during that time.
The KVM host itself had a load of 2-3 and about 70% idle on each CPU core.
I'm wondering what went wrong here. I would have expected the KVM host
show a high load for that "berserk" domain (the qemu-kvm process), and
really puzzled why it didn't have any impact on the other domains. Did
maybe only these two domains share a single CPU core or something like that?
Am I supposing right that if I only assign 2 CPU cores to that "berserk"
domain I would be a bit more safe of this happening again in the future?
If this is a FAQ I would appreciate any pointers.
Thank you!
Markus
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
