Created attachment 796691 [details] allow automatic migration engine log Description of problem: There are two hosts in cluster and cluster use only RAM filter policy. host1 has overloaded RAM, host2 is not overloaded. There is vm which is pinned to host1, if this vm is started, it tries to run on host1, which has overloded RAM. Version-Release number of selected component (if applicable): is12 How reproducible: always Steps to Reproduce: 1. Have two hosts in one cluster. This cluster use policy which use only RAM filter. 2. Create vm which is pinned to host1 3. Overload host1 RAM resources ( I had 99% ) (host2 had ~6%) 4. Start vm. Actual results: Vm tries to run on host1. Expected results: Vm run on host2 Additional info:
Created attachment 796692 [details] dont allow migration engine log
Ondra, what is the difference between the 2 logs? The 2nd one being failed by the engine: 2013-09-12 10:47:35,204 WARN [org.ovirt.engine.core.bll.RunVmCommand] (pool-5-thread-49) CanDoAction of action RunVm failed. Reasons:VAR__ACTION__RUN,VAR__TYPE__VM,VAR__ACTION__RUN,VAR__TYPE__VM,VAR__ACTION__RUN,VAR__TYPE__VM Also, we need more data on the actual memory calculations; what optimizations did you use (rqm load / balloon / ksm / etc).
I tried it on is15, and I was able to reproduce only when vm has checked option 'do not allow migration'. When there is vm with this checked, then vm try to run on host1 even when the scheduler don't use pin-to-host filter, but only RAM filter. When vm has checked 'allow auto/man migration', then vm starts on host2. I didn't use any memory optimizations, disabled memory page sharing, no ksm, no balooning. The first log is with vm which has checked 'allow auto/man migration'. The second log is with vm which has checked 'do not allow migration'.
tested in is20.2
Closing - RHEV 3.3 Released