agent loadbalancing doesn't work in MSses

[DaanHoogland]

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

[weizhouapache]

what's the value of global setting indirect.agent.lb.algorithm ?

[DaanHoogland]

what's the value of global setting indirect.agent.lb.algorithm ?

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 host but 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]

what's the value of global setting indirect.agent.lb.algorithm ?

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 host but 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.

yep.
You can put all mgmt server IPs in host, change indirect.agent.lb.algorithm to roundrobin or shuffle
then restart mgmt servers (or agents)

[DaanHoogland]

what's the value of global setting indirect.agent.lb.algorithm ?

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 host but 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.

yep. You can put all mgmt server IPs in host, change indirect.agent.lb.algorithm to roundrobin or shuffle then restart mgmt servers (or agents)

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]

what's the value of global setting indirect.agent.lb.algorithm ?

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 host but 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.

yep. You can put all mgmt server IPs in host, change indirect.agent.lb.algorithm to roundrobin or shuffle then restart mgmt servers (or agents)

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.

ok.
I only know the feature introduced by #2469 .
Is there another functionality ? sorry I do not remember it ..

[vishesh92]

There is also agent.load.threshold global 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]

@vishesh92 , do you think this is a major effort? It seems a bit heavy for a .1 release, agree?