1 Commits

Author	SHA1	Message	Date
Vishesh	ea90848429	Feature: Add support for DRS in a Cluster (#7723) ... This pull request (PR) implements a Distributed Resource Scheduler (DRS) for a CloudStack cluster. The primary objective of this feature is to enable automatic resource optimization and workload balancing within the cluster by live migrating the VMs as per configuration. Administrators can also execute DRS manually for a cluster, using the UI or the API. Adds support for two algorithms - condensed & balanced. Algorithms are pluggable allowing ACS Administrators to have customized control over scheduling. Implementation There are three top level components: Scheduler A timer task which: Generate DRS plan for clusters Process DRS plan Remove old DRS plan records DRS Execution We go through each VM in the cluster and use the specified algorithm to check if DRS is required and to calculate cost, benefit & improvement of migrating that VM to another host in the cluster. On the basis of cost, benefit & improvement, the best migration is selected for the current iteration and the VM is migrated. The maximum number of iterations (live migrations) possible on the cluster is defined by drs.iterations which is defined as a percentage (as a value between 0 and 1) of total number of workloads. Algorithm Every algorithms implements two methods: needsDrs - to check if drs is required for cluster getMetrics - to calculate cost, benefit & improvement of a migrating a VM to another host. Algorithms Condensed - Packs all the VMs on minimum number of hosts in the cluster. Balanced - Distributes the VMs evenly across hosts in the cluster. Algorithms use drs.level to decide the amount of imbalance to allow in the cluster. APIs Added listClusterDrsPlan id - ID of the DRS plan to list clusterid - to list plans for a cluster id generateClusterDrsPlan id - cluster id iterations - The maximum number of iterations in a DRS job defined as a percentage (as a value between 0 and 1) of total number of workloads. Defaults to value of cluster's drs.iterations setting. executeClusterDrsPlan id - ID of the cluster for which DRS plan is to be executed. migrateto - This parameter specifies the mapping between a vm and a host to migrate that VM. Format of this parameter: migrateto[vm-index].vm=<uuid>&migrateto[vm-index].host=<uuid>. Config Keys Added ClusterDrsPlanExpireInterval Key drs.plan.expire.interval Scope Global Default Value 30 days Description The interval in days after which old DRS records will be cleaned up. ClusterDrsEnabled Key drs.automatic.enable Scope Cluster Default Value false Description Enable/disable automatic DRS on a cluster. ClusterDrsInterval Key drs.automatic.interval Scope Cluster Default Value 60 minutes Description The interval in minutes after which a periodic background thread will schedule DRS for a cluster. ClusterDrsIterations Key drs.max.migrations Scope Cluster Default Value 50 Description Maximum number of live migrations in a DRS execution. ClusterDrsAlgorithm Key drs.algorithm Scope Cluster Default Value condensed Description DRS algorithm to execute on the cluster. This PR implements two algorithms - balanced & condensed. ClusterDrsLevel Key drs.imbalance Scope Cluster Default Value 0.5 Description Percentage (as a value between 0.0 and 1.0) of imbalance allowed in the cluster. 1.0 means no imbalance is allowed and 0.0 means imbalance is allowed. ClusterDrsMetric Key drs.imbalance.metric Scope Cluster Default Value memory Description The cluster imbalance metric to use when checking the drs.imbalance.threshold. Possible values are memory and cpu.	2023-10-26 11:48:18 +05:30