VIC Fails To Balance Containers

Products

• VMware vSphere ESXi

Issue/Introduction

Symptoms:

 When powering on large number of containers, the containers power on a single ESXi host. This can cause CPU contention, causing ESXi management services to become unresponsive and hang.

Environment

VMware vSphere Integrated Containers 1.x

Cause

DRS uses snapshot based placement that updates every 5 minutes resulting in each container power on getting very similar ESXi placement recommendations. In vSphere 6.5 a new DRS algorithm is introduced for continuous evaluation which will place the containers across the ESXi cluster using real time data. Accounting for the container resources about to power on in serial power on operations.

Resolution

Requires vCenter 6.5 and ESXi 6.5. to enable continuous DRS

Disable the following features for the cluster using the VCH:

• Distributed Power Management - DPM
• Proactive HA
• HA admission control

Set DRS to fully automated mode

Deploy the VCH using a resource pool. vAPP based VCHs will use the snapshot based evaluations in vSphere 6.5

Set AlwaysCheckPowerOn to true in vCenter.

1. Open the Flex Web Client (Flash)
2. Navigate to "host and cluster" > vCenter object in the left tree > Manage > Settings > Advanced Settings.
3. Click edit and add the Key: config.vpxd.cluster.AlwaysCheckPowerOn with the Value: true and click okay.
4. The change will not take effect until vpxd is restarted. This can be done by cli restarting the "vmware-vpxd" service as described in the KB Stopping, starting, or restarting VMware vCenter Server Appliance 6.x services (2109887). Rebooting the vCenter will also apply the changes.

Additional Information

DRS PERFORMANCE VMware vSphere 6.5