+++ This bug was initially created as a clone of Bug #967573 +++ related to bug 967572 for balloon support. this is for engine to report on guests not respecting ballooning requests (which once cgroups for memory are supported, will get swapped out for their unreserved memory) Virtual machines sharing and using memory resources - ballooning. Depending on implementation, this could affect: * Admin * CLI * Dev * PUP * UP
The following is from BZ#967573: Deatils on how to test; Setting up ============ Engine: Enable ballooning in cluster optimization Host: -- up-to-date vdsm and mom swapoff -a set rpc-port: 8080 in /etc/vdsm/mom.conf restart vdsm monitor the host (and its VMs) using: /var/log/vdsm/mom.log vdsCommand -s 0 getAllVmStats python -c "import xmlrpclib; from pprint import pprint; mom = xmlrpclib.ServerProxy('http://localhost:8080'); pprint(mom.getStatistics())" VMs (you need two): -- RAM: Host's RAM / 2 Resource allocation/Guaranteed memory: 512MB (or some other small number) yum install ovirt-guest-agent systemctl start ovirt-guest-agent systemctl enable ovirt-guest-agent mount -t tmpfs -o size=<almost all free RAM in Bytes> none /mnt swapoff -a Testing: ============ Once you have two VMs ready issue the next line in both: cat /dev/urandom >/mnt/data Then you can watch the host and once the the host's free memory drops bellow about 20% you should see ballooning commands like "mom.Controllers.Balloon - INFO - Ballooning guest:half1 from 2096400 to 1991580" being logged to mom.log. If the agent is down you will see "mom.Collectors.GuestMemory - WARNING - getVmMemoryStats() error: The ovirt-guest-agent is not active". If everything is ok on the VMs side you should see "GuestMonitor-<VM name> is ready". At that time you should stop the cat commands and start playing with stopping the guest-agent or rmmod-ing virtio_balloon module. Be aware of the fact you are filling the guest's memory and it has no swap. When the kernel needs memory and there is no free space in it, it first kills the ovirt-guest-agent and then panics... That is a feature though :)
Updated Topic [7913-553371] "Cluster Policy Settings Explained". This topic is included in the Admin Guide. The addition I made to the topic is brief, and reads: When a host's free memory drops below 20%, ballooning commands like mom.Controllers.Balloon - INFO Ballooning guest:half1 from 1096400 to 1991580 are logged to /etc/vdsm/mom.conf. /etc/vdsm/mom.conf is the Memory Overcommit Manager log file. This material must also be included in the Command Line Shell Guide. Moving to POST.
Or maybe this doesn't need to be in the Command Line Shell Guide. Andrew, do you think that this needs to be in the Command Line Shell Guide?
The Situation ------------- In the following section of the RHEV 3.4 Admin Guide, in this section: 4.2.2.4. Cluster Policy Settings Explained The following text is included: When a host's free memory drops below 20%, ballooning commands like mom.Controllers.Balloon - INFO Ballooning guest:half1 from 1096400 to 1991580 are logged to /etc/vdsm/mom.conf. /etc/vdsm/mom.conf is the Memory Overcommit Manager log file. NVR --- Red_Hat_Enterprise_Virtualization-Administration_Guide-3.4-en-US-3.4-31 This information is now in the RHEV 3.4 Admin Guide. I feel comfortable therefore closing this bug.