Repository navigation
agent attache rebalancing doesn't work in MSses #9640
Description
ISSUE TYPE
- Bug Report
COMPONENT NAME
MS-cluster
CLOUDSTACK VERSION
CONFIGURATION
OS / ENVIRONMENT
SUMMARY
STEPS TO REPRODUCE
2 MS
2 hosts+SSVM+CPVM == 4 agents
EXPECTED RESULTS
on 4 agents at most 3 would be on one of the two MSses
ACTUAL RESULTS
all 4 agents on one MS
Activity
weizhouapache commentedon Sep 5, 2024
what's the value of global setting
indirect.agent.lb.algorithm?DaanHoogland commentedon Sep 5, 2024
that is only for indirect agents is it? not for agents in general. It would explain, as it is static and there is only one ip in
hostbut that still would constitude a bug as the other MS could easily take responsibility for the SVMs, if all hosts are with the first MS.weizhouapache commentedon Sep 5, 2024
yep.
You can put all mgmt server IPs in
host, changeindirect.agent.lb.algorithmtoroundrobinorshufflethen restart mgmt servers (or agents)
DaanHoogland commentedon Sep 5, 2024
This will configure the agents but not let the MSses renegotiate their load. These are two different functionalities. I will do some testing with direct attached host agents (vmware or xen) to see what and when this broke.
weizhouapache commentedon Sep 5, 2024
ok.
I only know the feature introduced by #2469 .
Is there another functionality ? sorry I do not remember it ..
vishesh92 commentedon Sep 25, 2024
There is also
agent.load.thresholdglobal setting which load balances the agents (only hosts [type == Routing]). Check for balancing is done at regular intervals. And once balancing has been triggered, it is not executed again.This balancing is done per cluster i.e. if migrating all the hosts of the cluster will balance the cluster, then all hosts are migrated to another MS. The balancing happens per cluster probably to support balancing across different hypervisors.
I am not sure about the effort, but we should allow this migration to happen at host level as well depending on the hypervisor.
DaanHoogland commentedon Nov 8, 2024
@vishesh92 , do you think this is a major effort? It seems a bit heavy for a .1 release, agree?
vishesh92 commentedon Nov 11, 2024
I think it is a major effort at least in terms of QA and handling the edge cases if any.
sureshanaparti commentedon Mar 7, 2025
@DaanHoogland Tested this with CS 4.19.1.3 - 3 MS + 10 KVM hosts + 2 SystemVMs (12 Agents), it works as per the values configured in
indirect.agent.lb.algorithmandindirect.agent.lb.check.intervalsettings.hostsetting should be updated with all the MSses. (by default,indirect.agent.lb.algorithmis static, all agents would connect to the first MS in the provided list inhost)the settings
agent.lb.enabled,agent.load.thresholdwould be applicable for the direct agents (VMware, XenServer/XCP-ng) as per the wiki.15 remaining items
weizhouapache commentedon Nov 4, 2025
the issue still exists in 4.22.0.0/vmware
github-actions commentedon Aug 28, 2026
This issue is stale because it has been open for 120 days with no activity. It may be removed by administrators of this project at any time. Remove the stale label or comment to request for removal of it to prevent this.