Merge pull request #1102 from percona/PBM-1512-fix-node-validation-mixed-deploy

PBM-1512: Unable to take physical backup in mixed deployments

Author: boris-ilijic

pbm/topo/cluster.go

@@ -2,6 +2,7 @@ package topo
 
 import (
 	"context"
+	"fmt"
 	"strings"
 
 	"go.mongodb.org/mongo-driver/bson"
@@ -120,6 +121,7 @@ func ClusterMembers(ctx context.Context, m *mongo.Client) ([]Shard, error) {
 		return nil, errors.Wrap(err, "define cluster state")
 	}
 
+	// sharded cluster topo
 	var shards []Shard
 	if inf.IsMongos() || inf.IsSharded() {
 		members, err := getShardMapImpl(ctx, m)
@@ -135,10 +137,15 @@ func ClusterMembers(ctx context.Context, m *mongo.Client) ([]Shard, error) {
 		return shards, nil
 	}
 
+	// RS topo
+	hosts, err := GetReplsetHosts(ctx, m)
+	if err != nil {
+		return nil, errors.Wrap(err, "get all hosts for RS")
+	}
 	shards = []Shard{{
 		ID:   inf.SetName,
 		RS:   inf.SetName,
-		Host: inf.SetName + "/" + strings.Join(inf.Hosts, ","),
+		Host: fmt.Sprintf("%s/%s", inf.SetName, strings.Join(hosts, ",")),
 	}}
 	return shards, nil
 }
@@ -153,18 +160,44 @@ func getShardMapImpl(ctx context.Context, m *mongo.Client) (map[ReplsetName]Shar
 	// if shard name is not set, mongodb will provide unique name for it
 	// (e.g. the replset name of the shard)
 	// for configsvr, key name is "config"
-	var shardMap struct{ Map map[string]string }
+	// hosts field is used to discover hidden members, which are not present in map field
+	var shardMap struct {
+		Map   map[string]string
+		Hosts map[string]string
+	}
 	if err := res.Decode(&shardMap); err != nil {
 		return nil, errors.Wrap(err, "decode")
 	}
 
+	// Example of the hosts field from command output:
+	// hosts: {
+	//   'rs103:27017': 'rs1',
+	//   'rs101:27017': 'rs1',
+	//   'cfg02:27017': 'config',
+	//   ...
+	// }
+	// hostsByShard will contain even hidden RS members
+	hostsByShard := map[string][]string{}
+	for host, shardID := range shardMap.Hosts {
+		hostsByShard[shardID] = append(hostsByShard[shardID], host)
+	}
+
+	// for PSMDB 6 and below, config srv is not reported within hosts field
+	if len(hostsByShard["config"]) == 0 {
+		cfgHostgs, err := GetReplsetHosts(ctx, m)
+		if err != nil {
+			return nil, errors.Wrap(err, "get all hosts for config RS")
+		}
+		hostsByShard["config"] = cfgHostgs
+	}
+
 	shards := make(map[string]Shard, len(shardMap.Map))
 	for id, host := range shardMap.Map {
 		rs, _, _ := strings.Cut(host, "/")
 		shards[rs] = Shard{
 			ID:   id,
 			RS:   rs,
-			Host: host,
+			Host: fmt.Sprintf("%s/%s", rs, strings.Join(hostsByShard[id], ",")),
 		}
 	}
 

pbm/topo/topo.go

@@ -110,7 +110,7 @@ func collectTopoCheckErrors(
 		anyAvail := false
 		for _, host := range hosts {
 			a, ok := agents[host]
-			if !ok || a.Arbiter || a.Passive {
+			if !ok || a.Arbiter {
 				continue
 			}
 
@@ -195,6 +195,21 @@ func GetReplsetStatus(ctx context.Context, m *mongo.Client) (*ReplsetStatus, err
 	return status, nil
 }
 
+// GetReplsetHosts returns host names for all RS members.
+// It includes also hidden and passive RS members.
+func GetReplsetHosts(ctx context.Context, m *mongo.Client) ([]string, error) {
+	s, err := GetReplsetStatus(ctx, m)
+	if err != nil {
+		return nil, errors.Wrap(err, "get replset status")
+	}
+
+	hosts := []string{}
+	for _, m := range s.Members {
+		hosts = append(hosts, m.Name)
+	}
+	return hosts, nil
+}
+
 func GetNodeStatus(ctx context.Context, m *mongo.Client, name string) (*NodeStatus, error) {
 	s, err := GetReplsetStatus(ctx, m)
 	if err != nil {