Add ConcurrentBrokerRestartsAllowed to RollingUpgradeConfig

api/v1beta1/kafkacluster_types.go

@@ -137,6 +137,16 @@ type RollingUpgradeConfig struct {
 	// distinct broker replicas with either offline replicas or out of sync replicas and the number of alerts triggered by
 	// alerts with 'rollingupgrade'
 	FailureThreshold int `json:"failureThreshold"`
+
+	// ConcurrentBrokerRestartsAllowed controls how many brokers can be restarted in parallel during a rolling upgrade. If
+	// it is set to a value greater than 1, the operator will restart up to that amount of brokers in parallel, if the
+	// brokers are within the same rack (as specified by "broker.rack" in broker read-only configs). Since using Kafka broker
+	// racks spreads out the replicas, we know that restarting multiple brokers in the same rack will not cause more than
+	// 1/Nth of the replicas of a topic-partition to be unavailable at the same time, where N is the number of racks used.
+	// This is a safe way to speed up the rolling upgrade. Note that for the rack distribution explained above, Cruise Control
+	// requires `com.linkedin.kafka.cruisecontrol.analyzer.goals.RackAwareDistributionGoal` to be configured.
+	// +optional
+	ConcurrentBrokerRestartsAllowed int `json:"concurrentBrokerRestartsAllowed,omitempty"`
 }
 
 // DisruptionBudget defines the configuration for PodDisruptionBudget where the workload is managed by the kafka-operator

charts/kafka-operator/templates/crds.yaml

@@ -21689,6 +21689,21 @@ spec:
                 description: RollingUpgradeConfig defines the desired config of the
                   RollingUpgrade
                 properties:
+                  concurrentBrokerRestartsAllowed:
+                    description: ConcurrentBrokerRestartsAllowed controls how many
+                      brokers can be restarted in parallel during a rolling upgrade.
+                      If it is set to a value greater than 1, the operator will restart
+                      up to that amount of brokers in parallel, if the brokers are
+                      within the same rack (as specified by "broker.rack" in broker
+                      read-only configs). Since using Kafka broker racks spreads out
+                      the replicas, we know that restarting multiple brokers in the
+                      same rack will not cause more than 1/Nth of the replicas of
+                      a topic-partition to be unavailable at the same time, where
+                      N is the number of racks used. This is a safe way to speed up
+                      the rolling upgrade. Note that for the rack distribution explained
+                      above, Cruise Control requires `com.linkedin.kafka.cruisecontrol.analyzer.goals.RackAwareDistributionGoal`
+                      to be configured.
+                    type: integer
                   failureThreshold:
                     description: FailureThreshold controls how many failures the cluster
                       can tolerate during a rolling upgrade. Once the number of failures

config/base/crds/kafka.banzaicloud.io_kafkaclusters.yaml

@@ -21526,6 +21526,21 @@ spec:
                 description: RollingUpgradeConfig defines the desired config of the
                   RollingUpgrade
                 properties:
+                  concurrentBrokerRestartsAllowed:
+                    description: ConcurrentBrokerRestartsAllowed controls how many
+                      brokers can be restarted in parallel during a rolling upgrade.
+                      If it is set to a value greater than 1, the operator will restart
+                      up to that amount of brokers in parallel, if the brokers are
+                      within the same rack (as specified by "broker.rack" in broker
+                      read-only configs). Since using Kafka broker racks spreads out
+                      the replicas, we know that restarting multiple brokers in the
+                      same rack will not cause more than 1/Nth of the replicas of
+                      a topic-partition to be unavailable at the same time, where
+                      N is the number of racks used. This is a safe way to speed up
+                      the rolling upgrade. Note that for the rack distribution explained
+                      above, Cruise Control requires `com.linkedin.kafka.cruisecontrol.analyzer.goals.RackAwareDistributionGoal`
+                      to be configured.
+                    type: integer
                   failureThreshold:
                     description: FailureThreshold controls how many failures the cluster
                       can tolerate during a rolling upgrade. Once the number of failures