Hide Forgot
Created attachment 1231194 [details] screenshot Description of problem: Some scheduling policies have redundant parameters none - HighUtiliztion, CpuOverCommitmentDuration vm_evenly_distributed - HighUtiliztion, CpuOverCommitmentDuration Version-Release number of selected component (if applicable): ovirt-engine-4.1.0-0.2.master.20161212172238.gitea103bd.el7.centos.noarch How reproducible: Always Steps to Reproduce: 1. Open update cluster dialog 2. Change cluster scheduling policy to none 3. Actual results: You can see redundant parameters Expected results: Policy does not have redundant parameters Additional info: See screenshot
Created attachment 1231642 [details] None policy configuration The none policy disables load balancing, but the other filters (especially CPU and CPUOverloaded) still use those values during manual VM starts and migrations.
We checked the docs about this and the CPUOverloaded unit is properly present according to docs: Additional virtual machines attached to a host will not start if that host has reached the defined high utilization value. There is a small documentation bug here though, the above sentence is present for power saving and equally balanced policies: Table 5.7 in https://access.redhat.com/documentation/en/red-hat-virtualization/4.0/single/administration-guide/ But there is nothing like that mentioned for the other policies. But the "none" policy is basically identical to equally balanced with no automatic load balancing (manual VM starts and migrations still obey the rules though). The same applies to the VM count based policy. All standard filters apply.
Can you please describe exactly what you want changed and why? Comment #2 is not clear.
I think we just need to update the documentation to properly state that no VMs will be started on a host that is in CPU overloaded condition (the default is load of more than 80% for 5 minutes). The rule is already expressed in the docs, but in a very confusing and out of date way by mentioning Maximum Service Level, which has not been called like that since 3.1 or 3.2. The today's equivalent is expressed using HighUtilization and CpuOverCommitDurationMinutes parameters.
Assigning to Megan. Megan, this bug may have been already covered by your recent updates to the scheduling policy content.
2.2.1 I would maybe point out that the 80%/5 minutes rule is the default and can be changed (per cluster even) 5.7. Load Balancing Policy: InClusterUpgrade - Please point out very visibly that this is related to the major version upgrade only. It was meant to be used for RHEL 6 to RHEL 7 migration. It does not have any effect in 4.1 and should not be used (we are actually discussing hiding or removing it on the engineering side of the house)
Hi Megan, both changes look great and are exactly what was needed. I do not remember seeing anything else so I think we are good. Thanks.
Reviewed and verified.
The content for this bug is now available: 4.1 https://access.redhat.com/documentation/en-us/red_hat_virtualization/4.1/html-single/administration_guide/#sect-Scheduling_Policies https://access.redhat.com/documentation/en-us/red_hat_virtualization/4.1/html-single/technical_reference/#chap-Load_Balancing_Scheduling_and_Migration 4.0 https://access.redhat.com/documentation/en-us/red_hat_virtualization/4.0/html-single/administration_guide/#sect-Scheduling_Policies https://access.redhat.com/documentation/en-us/red_hat_virtualization/4.0/html-single/technical_reference/#chap-Load_Balancing_Scheduling_and_Migration