The virtual machine running the LUV server was killed by the kernel OOM 4 hours ago. I didn't immediately notice because the VM running my Jabber server (which notifies me of system problems) was also killed). When we had the last problem I converted the virtual machines from Xen to KVM. With KVM the VMs are regular Linux processes and they share the same memory as regular processes. So if another process allocates too much RAM then it may cause KVM memory allocation to fail. Also if the entire system runs out of RAM the kernel may stupidly decide to kill the KVM instance instead of something else. I think that part of the problem was that BOINC was configured to use up to 90% of system RAM. That was an OK setting for a Xen server where the Dom0 had nothing of note running other than BOINC and the virtual machines had RAM reserved. When running KVM this wasn't a suitable setting. I've configured BOINC to only use 40% of RAM and increased swap size. This shouldn't happen again. Also I'm going to move the Jabber server to the Dom0 so that if the DomUs die then I can still get alerts. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
On Friday, 23 September 2016 4:48:47 PM AEST Russell Coker via luv-main wrote:
I've configured BOINC to only use 40% of RAM and increased swap size.
I've got to ask - what's the rationale for running BOINC on the LUV server? Thanks for the outage report. All the best, Chris -- Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC
On Friday, 23 September 2016 6:19:31 PM AEST Chris Samuel via luv-main wrote:
I've configured BOINC to only use 40% of RAM and increased swap size.
I've got to ask - what's the rationale for running BOINC on the LUV server?
The LUV server is not running BOINC. The LUV server is a VM hosted on a system that runs BOINC among other things. All the cheap servers are shared in some way. Cheap hosting companies like Linode sell VMs where you are sharing the physical hardware with random people. Linode used to have a policy of moving VMs when users complained about performance, I don't know if they still have such a policy as the performance hasn't been a problem since they moved to SSD. In this case LUV has a free server which is shared with other people. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/
On Friday, 23 September 2016 6:48:35 PM AEST Russell Coker via luv-main wrote:
The LUV server is not running BOINC.
The LUV server is a VM hosted on a system that runs BOINC among other things.
Ahh, thanks for the explanation. Odd that services on one VM can kill another. cheers, Chris -- Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC
Hi Russell, If a dedicated VM hosted in Melbourne would be useful, hit me up offlist. Regards Darryl On 23/09/2016 6:48 PM, Russell Coker via luv-main wrote:
On Friday, 23 September 2016 6:19:31 PM AEST Chris Samuel via luv-main wrote:
I've configured BOINC to only use 40% of RAM and increased swap size. I've got to ask - what's the rationale for running BOINC on the LUV server? The LUV server is not running BOINC.
The LUV server is a VM hosted on a system that runs BOINC among other things.
All the cheap servers are shared in some way. Cheap hosting companies like Linode sell VMs where you are sharing the physical hardware with random people. Linode used to have a policy of moving VMs when users complained about performance, I don't know if they still have such a policy as the performance hasn't been a problem since they moved to SSD.
In this case LUV has a free server which is shared with other people.
participants (4)
-
Andrew Pam -
Chris Samuel -
Darryl Ross -
Russell Coker