The findMaxSingleRun runStrategy now runs all of the appInstances in sync. In each interval a failure of any instance is treated as the failure of all. As a result, the findMaxSingleRun strategy now finds the maximum user load at which all instances can pass.
The old behavior is still available in the findMaxSingleRunAsync runStrategy.
The synchronized findMaxSingleRun takes advantage of the synchronization to have more compact console output.
Added an anti-affinity rule for the auctionDataManager to nodes labeled wvrole=driver
Added CPU and memory requests to the auctionDataManager pod
This change and the previous will help the scheduler make better placement decisions for the auctionDataManager.
Increased the number of retries for waiting for up and running on services. This should help with cases with large numbers of instances.
Combined the postgresql data and log volumes into a single volume. This will reduce the number of PVCs and PVs that get created for each instance.
You should delete all existing PVCs before running with this change. Otherwise the existing PVC/PVs for postgresql will unnecessarily take up disk space.
The parameters postgresqlDataStorageClass and postgresqlLogsStorageClass have been removed and replaced with a single parameter postgresqlStorageClass. You will need to update your config files with this change.
Added a parameter instanceNodeLabels which, if present, sets a required affinity rule so that pods will only be scheduled on nodes labeled for their appInstance (e.g. wvauctionw1i1=).
This pull request has the following changes: