GT-198 Cleanout calculation - picks members with the lowest number of shards (#1261)

Author: jwierzbo

CHANGELOG.md

@@ -19,6 +19,7 @@
 - (Improvement) Change Operator default ReplicasCount to 1
 - (Maintenance) Change MD content injection method
 - (Maintenance) Generate README Platforms
+- (Improvement) Cleanout calculation - picks members with the lowest number of shards
 
 ## [1.2.24](https://github.com/arangodb/kube-arangodb/tree/1.2.24) (2023-01-25)
 - (Bugfix) Fix deployment creation on ARM64

pkg/apis/deployment/v1/member_status.go

@@ -53,7 +53,7 @@ type MemberStatus struct {
 	CreatedAt meta.Time `json:"created-at"`
 	// Conditions specific to this member
 	Conditions ConditionList `json:"conditions,omitempty"`
-	// RecentTerminatons holds the times when this member was recently terminated.
+	// RecentTerminations holds the times when this member was recently terminated.
 	// First entry is the oldest. (do not add omitempty, since we want to be able to switch from a list to an empty list)
 	RecentTerminations []meta.Time `json:"recent-terminations"`
 	// IsInitialized is set after the very first time a pod was created for this member.

pkg/apis/deployment/v2alpha1/member_status.go

@@ -53,7 +53,7 @@ type MemberStatus struct {
 	CreatedAt meta.Time `json:"created-at"`
 	// Conditions specific to this member
 	Conditions ConditionList `json:"conditions,omitempty"`
-	// RecentTerminatons holds the times when this member was recently terminated.
+	// RecentTerminations holds the times when this member was recently terminated.
 	// First entry is the oldest. (do not add omitempty, since we want to be able to switch from a list to an empty list)
 	RecentTerminations []meta.Time `json:"recent-terminations"`
 	// IsInitialized is set after the very first time a pod was created for this member.

pkg/deployment/agency/state.go

@@ -132,6 +132,42 @@ func (s State) CountShards() int {
 	return count
 }
 
+// ShardsByDbServers returns a map of DBServers and the amount of shards they have
+func (s State) ShardsByDbServers() map[Server]int {
+	result := make(map[Server]int)
+
+	for _, collections := range s.Current.Collections {
+		for _, shards := range collections {
+			for _, shard := range shards {
+				for _, server := range shard.Servers {
+					result[server]++
+				}
+			}
+		}
+	}
+
+	return result
+}
+
+// GetDBServerWithLowestShards returns the DBServer with the lowest amount of shards
+func (s State) GetDBServerWithLowestShards() Server {
+	var resultServer Server = ""
+	var resultShards int
+
+	for server, shards := range s.ShardsByDbServers() {
+		// init first server as result
+		if resultServer == "" {
+			resultServer = server
+			resultShards = shards
+		} else if shards < resultShards {
+			resultServer = server
+			resultShards = shards
+		}
+	}
+	return resultServer
+}
+
+// PlanServers returns all servers which are part of the plan
 func (s State) PlanServers() Servers {
 	q := map[Server]bool{}
 

pkg/deployment/reconcile/plan_builder_scale.go

@@ -93,7 +93,13 @@ func (r *Reconciler) createScalePlan(status api.DeploymentStatus, members api.Me
 			Debug("Creating scale-up plan")
 	} else if len(members) > count {
 		// Note, we scale down 1 member at a time
-		if m, err := members.SelectMemberToRemove(getCleanedServer(context), getToBeCleanedServer(context), topologyMissingMemberToRemoveSelector(status.Topology), topologyAwarenessMemberToRemoveSelector(group, status.Topology)); err != nil {
+
+		if m, err := members.SelectMemberToRemove(
+			getCleanedServer(context),
+			getToBeCleanedServer(context),
+			topologyMissingMemberToRemoveSelector(status.Topology),
+			topologyAwarenessMemberToRemoveSelector(group, status.Topology),
+			getDbServerWithLowestShards(context, group)); err != nil {
 			r.planLogger.Err(err).Str("role", group.AsRole()).Warn("Failed to select member to remove")
 		} else {
 			ready, message := groupReadyForRestart(context, status, m, group)
@@ -194,6 +200,18 @@ func getToBeCleanedServer(ctx reconciler.ArangoAgencyGet) api.MemberToRemoveSele
 	}
 }
 
+func getDbServerWithLowestShards(ctx reconciler.ArangoAgencyGet, g api.ServerGroup) api.MemberToRemoveSelector {
+	return func(m api.MemberStatusList) (string, error) {
+		if g != api.ServerGroupDBServers {
+			return "", nil
+		}
+		if a, ok := ctx.GetAgencyCache(); ok {
+			return string(a.GetDBServerWithLowestShards()), nil
+		}
+		return "", nil
+	}
+}
+
 func (r *Reconciler) scaleDownCandidate(ctx context.Context, apiObject k8sutil.APIObject,
 	spec api.DeploymentSpec, status api.DeploymentStatus,
 	context PlanBuilderContext) api.Plan {