fix(repair): schedule and run repair on single node cluster

SM 3.2 changed repair param validation so that scheduling and running repair of a single node cluster is treated as an error. Although this is acceptable in terms of real users, the ability to schedule/run repair on a simple single node cluster is useful for tests and some experiments.

Fixes #3649

pkg/scyllaclient/client_scylla.go

@@ -470,15 +470,18 @@ func ReplicaHash(replicaSet []string) uint64 {
 
 // Repair invokes async repair and returns the repair command ID.
 func (c *Client) Repair(ctx context.Context, keyspace, table, master string, replicaSet []string, ranges []TokenRange) (int32, error) {
-	hosts := strings.Join(replicaSet, ",")
 	dr := dumpRanges(ranges)
 	p := operations.StorageServiceRepairAsyncByKeyspacePostParams{
 		Context:        forceHost(ctx, master),
 		Keyspace:       keyspace,
 		ColumnFamilies: &table,
-		Hosts:          &hosts,
 		Ranges:         &dr,
 	}
+	// Single node cluster repair fails with hosts param
+	if len(replicaSet) > 1 {
+		hosts := strings.Join(replicaSet, ",")
+		p.Hosts = &hosts
+	}
 
 	resp, err := c.scyllaOps.StorageServiceRepairAsyncByKeyspacePost(&p)
 	if err != nil {

pkg/service/repair/plan.go

@@ -25,18 +25,20 @@ type plan struct {
 }
 
 func newPlan(ctx context.Context, target Target, client *scyllaclient.Client) (*plan, error) {
-	filtered, err := filteredHosts(ctx, target, client)
+	status, err := client.Status(ctx)
 	if err != nil {
-		return nil, errors.Wrap(err, "filter hosts")
+		return nil, errors.Wrap(err, "get status")
 	}
+	filtered := filteredHosts(target, status)
 
 	p := new(plan)
 	for _, u := range target.Units {
 		ring, err := client.DescribeRing(ctx, u.Keyspace)
 		if err != nil {
 			return nil, errors.Wrapf(err, "keyspace %s: get ring description", u.Keyspace)
 		}
-		if ring.Replication == scyllaclient.LocalStrategy {
+		// Allow repairing single node cluster for better UX and tests
+		if ring.Replication == scyllaclient.LocalStrategy && len(status) > 1 {
 			continue
 		}
 
@@ -53,8 +55,9 @@ func newPlan(ctx context.Context, target Target, client *scyllaclient.Client) (*
 				Ranges:     rep.Ranges,
 			}
 
-			// Don't add keyspace with some ranges not replicated in filtered hosts
-			if len(rtr.ReplicaSet) <= 1 {
+			// Don't add keyspace with some ranges not replicated in filtered hosts,
+			// unless it's a single node cluster.
+			if len(rtr.ReplicaSet) <= 1 && len(status) > 1 {
 				skip = true
 				break
 			}
@@ -447,12 +450,7 @@ func (tp tablePlan) MarkRange(repIdx int, r scyllaclient.TokenRange) bool {
 }
 
 // filteredHosts returns hosts passing '--dc' and '--ignore-down-hosts' criteria.
-func filteredHosts(ctx context.Context, target Target, client *scyllaclient.Client) (*strset.Set, error) {
-	status, err := client.Status(ctx)
-	if err != nil {
-		return nil, errors.Wrap(err, "get status")
-	}
-
+func filteredHosts(target Target, status scyllaclient.NodeStatusInfoSlice) *strset.Set {
 	ignoredHosts := strset.New(target.IgnoreHosts...)
 	dcs := strset.New(target.DC...)
 	filtered := strset.New()
@@ -463,7 +461,7 @@ func filteredHosts(ctx context.Context, target Target, client *scyllaclient.Clie
 		}
 	}
 
-	return filtered, nil
+	return filtered
 }
 
 // filterReplicaSet returns hosts present in filteredHosts and passing '--host' criteria.