deploy: support for read affinity options per cluster

Implemented the capability to include read affinity options
for individual clusters within the ceph-csi-config ConfigMap.
This allows users to configure the crush location for each
cluster separately. The read affinity options specified in
the ConfigMap will supersede those provided via command line arguments.

Signed-off-by: Praveen M <m.praveen@ibm.com>

charts/ceph-csi-rbd/values.yaml

@@ -27,6 +27,11 @@ serviceAccounts:
 #       - "<MONValue2>"
 #     rbd:
 #       netNamespaceFilePath: "{{ .kubeletDir }}/plugins/{{ .driverName }}/net"
+#     readAffinity:
+#       enabled: true
+#       crushLocationLabels:
+#         - topology.kubernetes.io/region
+#         - topology.kubernetes.io/zone
 csiConfig: []
 
 # Configuration details of clusterID,PoolID and FscID mapping

cmd/cephcsi.go

@@ -88,6 +88,8 @@ func init() {
 		"",
 		"list of Kubernetes node labels, that determines the"+
 			" CRUSH location the node belongs to, separated by ','")
+	flag.StringVar(&conf.ContainerOrchestrator, "container-orchestrator", "kubernetes",
+		"container orchestrator where cephcsi is running")
 
 	// cephfs related flags
 	flag.BoolVar(

deploy/csi-config-map-sample.yaml

@@ -32,12 +32,15 @@ kind: ConfigMap
 # path for the Ceph cluster identified by the <cluster-id>, This will be used
 # by the RBD CSI plugin to execute the rbd map/unmap in the
 # network namespace specified by the "rbd.netNamespaceFilePath".
+# The "readAffinity" fields are used to enable read affinity and pass the crush
+# location map for the Ceph cluster identified by the cluster <cluster-id>,
+# enabling this will add
+# "read_from_replica=localize,crush_location=<label:value>" to the map option.
 # If a CSI plugin is using more than one Ceph cluster, repeat the section for
 # each such cluster in use.
 # NOTE: Changes to the configmap is automatically updated in the running pods,
 # thus restarting existing pods using the configmap is NOT required on edits
 # to the configmap.
-
 # Lets see the different configuration under cluster-mapping.json key.
 # This configuration is needed when volumes are mirrored using the Ceph-CSI.
 # clusterIDMapping holds the mapping between two clusterId's of storage
@@ -66,6 +69,15 @@ data:
         }
         "nfs": {
           "netNamespaceFilePath": "<kubeletRootPath>/plugins/nfs.csi.ceph.com/net",
+        },
+        "readAffinity": {
+          "enabled": "false",
+          "crushLocationLabels": [
+            "<Label1>",
+            "<Label2>"
+            ...
+            "<Label3>"
+          ]
         }
       }
     ]

docs/deploy-rbd.md

@@ -49,7 +49,7 @@ make image-cephcsi
 | `--maxsnapshotsonimage`  | `450`                         | Maximum number of snapshots allowed on rbd image without flattening                                                                                                                                                                                                                  |
 | `--setmetadata`          | `false`                       | Set metadata on volume                                                                                                                                                                                                                                                               |
 | `--enable-read-affinity` | `false`                       | enable read affinity                                                                                                                                                                                                                                                                 |
-| `--crush-location-labels`| _empty_                       | Kubernetes node labels that determine the CRUSH location the node belongs to, separated by ','                                                                                                                                                                                       |
+| `--crush-location-labels`| _empty_                       | Kubernetes node labels that determine the CRUSH location the node belongs to, separated by ','.<br>`Note: These labels will be replaced if crush location labels are defined in the ceph-csi-config ConfigMap for the specific cluster.`                                                                                                                                                                                       |
 
 **Available volume parameters:**
 
@@ -224,6 +224,12 @@ If enabled, this option will be added to all RBD volumes mapped by Ceph CSI.
 Well known labels can be found
 [here](https://kubernetes.io/docs/reference/labels-annotations-taints/).
 
+Read affinity can be configured for individual clusters within the
+`ceph-csi-config` ConfigMap. This allows configuring the crush location labels
+for each ceph cluster separately. The crush location labels specified in the
+ConfigMap will supersede  those provided via command line argument
+`--crush-location-labels`.
+
 >Note: Label values will have all its dots `"."` normalized with dashes `"-"`
 in order for it to work with ceph CRUSH map.
 

internal/rbd/driver/driver.go

@@ -26,6 +26,7 @@ import (
 	csicommon "github.com/ceph/ceph-csi/internal/csi-common"
 	"github.com/ceph/ceph-csi/internal/rbd"
 	"github.com/ceph/ceph-csi/internal/util"
+	"github.com/ceph/ceph-csi/internal/util/k8s"
 	"github.com/ceph/ceph-csi/internal/util/log"
 
 	"github.com/container-storage-interface/spec/lib/go/csi"
@@ -68,14 +69,14 @@ func NewControllerServer(d *csicommon.CSIDriver) *rbd.ControllerServer {
 func NewNodeServer(
 	d *csicommon.CSIDriver,
 	t string,
-	topology map[string]string,
-	crushLocationMap map[string]string,
+	nodeLabels, topology, crushLocationMap map[string]string,
 ) (*rbd.NodeServer, error) {
 	ns := rbd.NodeServer{
-		DefaultNodeServer: csicommon.NewDefaultNodeServer(d, t, topology),
-		VolumeLocks:       util.NewVolumeLocks(),
+		DefaultNodeServer:         csicommon.NewDefaultNodeServer(d, t, topology),
+		VolumeLocks:               util.NewVolumeLocks(),
+		NodeLabels:                nodeLabels,
+		CLIReadAffinityMapOptions: util.ConstructReadAffinityMapOption(crushLocationMap),
 	}
-	ns.SetReadAffinityMapOptions(crushLocationMap)
 
 	return &ns, nil
 }
@@ -87,8 +88,8 @@ func NewNodeServer(
 // setupCSIAddonsServer().
 func (r *Driver) Run(conf *util.Config) {
 	var (
-		err                        error
-		topology, crushLocationMap map[string]string
+		err                                    error
+		nodeLabels, topology, crushLocationMap map[string]string
 	)
 	// update clone soft and hard limit
 	rbd.SetGlobalInt("rbdHardMaxCloneDepth", conf.RbdHardMaxCloneDepth)
@@ -125,13 +126,17 @@ func (r *Driver) Run(conf *util.Config) {
 			})
 	}
 
-	if conf.EnableReadAffinity {
-		crushLocationMap, err = util.GetCrushLocationMap(conf.CrushLocationLabels, conf.NodeID)
+	if conf.ContainerOrchestrator == "kubernetes" {
+		nodeLabels, err = k8s.GetNodeLabels(conf.NodeID)
 		if err != nil {
 			log.FatalLogMsg(err.Error())
 		}
 	}
 
+	if conf.EnableReadAffinity {
+		crushLocationMap = util.GetCrushLocationMap(conf.CrushLocationLabels, nodeLabels)
+	}
+
 	// Create GRPC servers
 	r.ids = NewIdentityServer(r.cd)
 
@@ -140,7 +145,7 @@ func (r *Driver) Run(conf *util.Config) {
 		if err != nil {
 			log.FatalLogMsg(err.Error())
 		}
-		r.ns, err = NewNodeServer(r.cd, conf.Vtype, topology, crushLocationMap)
+		r.ns, err = NewNodeServer(r.cd, conf.Vtype, nodeLabels, topology, crushLocationMap)
 		if err != nil {
 			log.FatalLogMsg("failed to start node server, err %v\n", err)
 		}

internal/rbd/nodeserver.go

@@ -45,8 +45,10 @@ type NodeServer struct {
 	// A map storing all volumes with ongoing operations so that additional operations
 	// for that same volume (as defined by VolumeID) return an Aborted error
 	VolumeLocks *util.VolumeLocks
-	// readAffinityMapOptions contains map options to enable read affinity.
-	readAffinityMapOptions string
+	// NodeLabels stores the node labels
+	NodeLabels map[string]string
+	// CLIReadAffinityMapOptions contains map options passed through command line to enable read affinity.
+	CLIReadAffinityMapOptions string
 }
 
 // stageTransaction struct represents the state a transaction was when it either completed
@@ -258,11 +260,10 @@ func (ns *NodeServer) populateRbdVol(
 		rv.Mounter = rbdNbdMounter
 	}
 
-	err = getMapOptions(req, rv)
+	err = ns.getMapOptions(req, rv)
 	if err != nil {
 		return nil, err
 	}
-	ns.appendReadAffinityMapOptions(rv)
 
 	rv.VolID = volID
 
@@ -280,14 +281,14 @@ func (ns *NodeServer) populateRbdVol(
 
 // appendReadAffinityMapOptions appends readAffinityMapOptions to mapOptions
 // if mounter is rbdDefaultMounter and readAffinityMapOptions is not empty.
-func (ns NodeServer) appendReadAffinityMapOptions(rv *rbdVolume) {
+func (rv *rbdVolume) appendReadAffinityMapOptions(readAffinityMapOptions string) {
 	switch {
-	case ns.readAffinityMapOptions == "" || rv.Mounter != rbdDefaultMounter:
+	case readAffinityMapOptions == "" || rv.Mounter != rbdDefaultMounter:
 		return
 	case rv.MapOptions != "":
-		rv.MapOptions += "," + ns.readAffinityMapOptions
+		rv.MapOptions += "," + readAffinityMapOptions
 	default:
-		rv.MapOptions = ns.readAffinityMapOptions
+		rv.MapOptions = readAffinityMapOptions
 	}
 }
 
@@ -1395,22 +1396,3 @@ func getDeviceSize(ctx context.Context, devicePath string) (uint64, error) {
 
 	return size, nil
 }
-
-func (ns *NodeServer) SetReadAffinityMapOptions(crushLocationMap map[string]string) {
-	if len(crushLocationMap) == 0 {
-		return
-	}
-
-	var b strings.Builder
-	b.WriteString("read_from_replica=localize,crush_location=")
-	first := true
-	for key, val := range crushLocationMap {
-		if first {
-			b.WriteString(fmt.Sprintf("%s:%s", key, val))
-			first = false
-		} else {
-			b.WriteString(fmt.Sprintf("|%s:%s", key, val))
-		}
-	}
-	ns.readAffinityMapOptions = b.String()
-}

internal/rbd/nodeserver_test.go

@@ -18,8 +18,12 @@ package rbd
 
 import (
 	"context"
+	"encoding/json"
+	"os"
 	"testing"
 
+	"github.com/ceph/ceph-csi/internal/util"
+
 	"github.com/container-storage-interface/spec/lib/go/csi"
 	"github.com/stretchr/testify/assert"
 )
@@ -107,53 +111,6 @@ func TestParseBoolOption(t *testing.T) {
 	}
 }
 
-func TestNodeServer_SetReadAffinityMapOptions(t *testing.T) {
-	t.Parallel()
-	tests := []struct {
-		name             string
-		crushLocationmap map[string]string
-		wantAny          []string
-	}{
-		{
-			name:             "nil crushLocationmap",
-			crushLocationmap: nil,
-			wantAny:          []string{""},
-		},
-		{
-			name:             "empty crushLocationmap",
-			crushLocationmap: map[string]string{},
-			wantAny:          []string{""},
-		},
-		{
-			name: "single entry in crushLocationmap",
-			crushLocationmap: map[string]string{
-				"region": "east",
-			},
-			wantAny: []string{"read_from_replica=localize,crush_location=region:east"},
-		},
-		{
-			name: "multiple entries in crushLocationmap",
-			crushLocationmap: map[string]string{
-				"region": "east",
-				"zone":   "east-1",
-			},
-			wantAny: []string{
-				"read_from_replica=localize,crush_location=region:east|zone:east-1",
-				"read_from_replica=localize,crush_location=zone:east-1|region:east",
-			},
-		},
-	}
-	for _, tt := range tests {
-		currentTT := tt
-		t.Run(tt.name, func(t *testing.T) {
-			t.Parallel()
-			ns := &NodeServer{}
-			ns.SetReadAffinityMapOptions(currentTT.crushLocationmap)
-			assert.Contains(t, currentTT.wantAny, ns.readAffinityMapOptions)
-		})
-	}
-}
-
 func TestNodeServer_appendReadAffinityMapOptions(t *testing.T) {
 	t.Parallel()
 	type input struct {
@@ -236,11 +193,115 @@ func TestNodeServer_appendReadAffinityMapOptions(t *testing.T) {
 				MapOptions: currentTT.args.mapOptions,
 				Mounter:    currentTT.args.mounter,
 			}
+			rv.appendReadAffinityMapOptions(currentTT.args.readAffinityMapOptions)
+			assert.Equal(t, currentTT.want, rv.MapOptions)
+		})
+	}
+}
+
+func TestReadAffinity_GetReadAffinityMapOptions(t *testing.T) {
+	t.Parallel()
+
+	nodeLabels := map[string]string{
+		"topology.kubernetes.io/zone":   "east-1",
+		"topology.kubernetes.io/region": "east",
+	}
+
+	csiConfig := []util.ClusterInfo{
+		{
+			ClusterID: "cluster-1",
+			ReadAffinity: struct {
+				Enabled             bool     `json:"enabled"`
+				CrushLocationLabels []string `json:"crushLocationLabels"`
+			}{
+				Enabled: true,
+				CrushLocationLabels: []string{
+					"topology.kubernetes.io/region",
+				},
+			},
+		},
+		{
+			ClusterID: "cluster-2",
+			ReadAffinity: struct {
+				Enabled             bool     `json:"enabled"`
+				CrushLocationLabels []string `json:"crushLocationLabels"`
+			}{
+				Enabled: false,
+				CrushLocationLabels: []string{
+					"topology.kubernetes.io/region",
+				},
+			},
+		},
+	}
+
+	csiConfigFileContent, err := json.Marshal(csiConfig)
+	if err != nil {
+		t.Errorf("failed to marshal csi config info %v", err)
+	}
+	tmpConfPath := util.CsiConfigFile
+	err = os.Mkdir("/etc/ceph-csi-config", 0o600)
+	if err != nil {
+		t.Errorf("failed to create directory %s: %v", "/etc/ceph-csi-config", err)
+	}
+	err = os.WriteFile(tmpConfPath, csiConfigFileContent, 0o600)
+	if err != nil {
+		t.Errorf("failed to write %s file content: %v", util.CsiConfigFile, err)
+	}
+
+	tests := []struct {
+		name                   string
+		clusterID              string
+		CLICrushLocationLabels string
+		want                   string
+	}{
+		{
+			name:                   "Enabled in cluster-1 and Enabled in CLI",
+			clusterID:              "cluster-1",
+			CLICrushLocationLabels: "topology.kubernetes.io/region",
+			want:                   "read_from_replica=localize,crush_location=region:east",
+		},
+		{
+			name:                   "Disabled in cluster-2 and Enabled in CLI",
+			clusterID:              "cluster-2",
+			CLICrushLocationLabels: "topology.kubernetes.io/zone",
+			want:                   "read_from_replica=localize,crush_location=zone:east-1",
+		},
+		{
+			name:                   "Disabled in cluster-2 and Disabled in CLI",
+			clusterID:              "cluster-2",
+			CLICrushLocationLabels: "",
+			want:                   "",
+		},
+		{
+			name:                   "Absent is cluster-3 and Enabled in CLI",
+			clusterID:              "cluster-3",
+			CLICrushLocationLabels: "topology.kubernetes.io/zone",
+			want:                   "read_from_replica=localize,crush_location=zone:east-1",
+		},
+		{
+			name:                   "Absent in cluster-3 and Disabled in CLI",
+			clusterID:              "cluster-3",
+			CLICrushLocationLabels: "",
+			want:                   "",
+		},
+	}
+
+	for _, tt := range tests {
+		tc := tt
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			crushLocationMap := util.GetCrushLocationMap(tc.CLICrushLocationLabels, nodeLabels)
 			ns := &NodeServer{
-				readAffinityMapOptions: currentTT.args.readAffinityMapOptions,
+				CLIReadAffinityMapOptions: util.ConstructReadAffinityMapOption(crushLocationMap),
 			}
-			ns.appendReadAffinityMapOptions(rv)
-			assert.Equal(t, currentTT.want, rv.MapOptions)
+			readAffinityMapOptions, err := util.GetReadAffinityMapOptions(
+				tc.clusterID, ns.CLIReadAffinityMapOptions, nodeLabels,
+			)
+			if err != nil {
+				assert.Fail(t, err.Error())
+			}
+
+			assert.Equal(t, tc.want, readAffinityMapOptions)
 		})
 	}
 }