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/templates/nodeplugin-daemonset.yaml

@@ -87,7 +87,7 @@ spec:
 {{- if .Values.nodeplugin.profiling.enabled }}
             - "--enableprofiling={{ .Values.nodeplugin.profiling.enabled }}"
 {{- end }}
-            - "--enable-read-affinity={{ and .Values.readAffinity .Values.readAffinity.enabled }}"
+- "--enable-read-affinity={{ and .Values.readAffinity .Values.readAffinity.enabled }}"
 {{- if and .Values.readAffinity .Values.readAffinity.enabled }}
             - "--crush-location-labels={{ .Values.readAffinity.crushLocationLabels | join "," }}"
 {{- end }}

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 +88,7 @@ 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 [kubernetes]")
 
 	// cephfs related flags
 	flag.BoolVar(

deploy/csi-config-map-sample.yaml

@@ -66,6 +66,15 @@ data:
         }
         "nfs": {
           "netNamespaceFilePath": "<kubeletRootPath>/plugins/nfs.csi.ceph.com/net",
+        },
+        "readAffinity": {
+          "enabled": "false",
+          "crushLocationLabels": [
+            "<Label1>",
+            "<Label2>"
+            ...
+            "<Label3>"
+          ]
         }
       }
     ]

docs/deploy-rbd.md

@@ -48,8 +48,7 @@ make image-cephcsi
 | `--skipforceflatten`     | `false`                       | skip image flattening on kernel < 5.2 which support mapping of rbd images which has the deep-flatten feature                                                                                                                                                                         |
 | `--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:**
 
@@ -215,7 +214,7 @@ for more details.
 
 This can be enabled by adding labels to Kubernetes nodes like
 `"topology.io/region=east"` and `"topology.io/zone=east-zone1"` and
-passing command line arguments `"--enable-read-affinity=true"` and
+passing command line argument
 `"--crush-location-labels=topology.io/zone,topology.io/region"` to Ceph CSI
 RBD daemonset pod "csi-rbdplugin" container, resulting in Ceph CSI adding
 `"--options read_from_replica=localize,crush_location=zone:east-zone1|region:east"`
@@ -224,6 +223,12 @@ 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)
@@ -125,8 +126,15 @@ func (r *Driver) Run(conf *util.Config) {
 			})
 	}
 
+	if conf.ContainerOrchestrator == "kubernetes" {
+		nodeLabels, err = k8s.GetNodeLabels(conf.NodeID)
+		if err != nil {
+			log.FatalLogMsg(err.Error())
+		}
+	}
+
 	if conf.EnableReadAffinity {
-		crushLocationMap, err = util.GetCrushLocationMap(conf.CrushLocationLabels, conf.NodeID)
+		crushLocationMap, err = util.GetCrushLocationMap(conf.CrushLocationLabels, nodeLabels)
 		if err != nil {
 			log.FatalLogMsg(err.Error())
 		}
@@ -140,7 +148,7 @@ 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

@@ -20,6 +20,8 @@ import (
 	"context"
 	"testing"
 
+	"github.com/ceph/ceph-csi/internal/util"
+
 	"github.com/container-storage-interface/spec/lib/go/csi"
 	"github.com/stretchr/testify/assert"
 )
@@ -107,7 +109,7 @@ func TestParseBoolOption(t *testing.T) {
 	}
 }
 
-func TestNodeServer_SetReadAffinityMapOptions(t *testing.T) {
+func TestNodeServer_ConstructReadAffinityMapOption(t *testing.T) {
 	t.Parallel()
 	tests := []struct {
 		name             string
@@ -147,9 +149,10 @@ func TestNodeServer_SetReadAffinityMapOptions(t *testing.T) {
 		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)
+			ns := &NodeServer{
+				CLIReadAffinityMapOptions: util.ConstructReadAffinityMapOption(currentTT.crushLocationmap),
+			}
+			assert.Contains(t, currentTT.wantAny, ns.CLIReadAffinityMapOptions)
 		})
 	}
 }
@@ -236,10 +239,7 @@ func TestNodeServer_appendReadAffinityMapOptions(t *testing.T) {
 				MapOptions: currentTT.args.mapOptions,
 				Mounter:    currentTT.args.mounter,
 			}
-			ns := &NodeServer{
-				readAffinityMapOptions: currentTT.args.readAffinityMapOptions,
-			}
-			ns.appendReadAffinityMapOptions(rv)
+			rv.appendReadAffinityMapOptions(currentTT.args.readAffinityMapOptions)
 			assert.Equal(t, currentTT.want, rv.MapOptions)
 		})
 	}

internal/rbd/rbd_attach.go

@@ -295,7 +295,7 @@ func parseMapOptions(mapOptions string) (string, string, error) {
 
 // getMapOptions is a wrapper func, calls parse map/unmap funcs and feeds the
 // rbdVolume object.
-func getMapOptions(req *csi.NodeStageVolumeRequest, rv *rbdVolume) error {
+func (ns *NodeServer) getMapOptions(req *csi.NodeStageVolumeRequest, rv *rbdVolume) error {
 	krbdMapOptions, nbdMapOptions, err := parseMapOptions(req.GetVolumeContext()["mapOptions"])
 	if err != nil {
 		return err
@@ -312,6 +312,12 @@ func getMapOptions(req *csi.NodeStageVolumeRequest, rv *rbdVolume) error {
 		rv.UnmapOptions = nbdUnmapOptions
 	}
 
+	readAffinityMapOptions, err := util.GetReadAffinityMapOptions(rv.ClusterID, ns.CLIReadAffinityMapOptions, ns.NodeLabels)
+	if err != nil {
+		return err
+	}
+	rv.appendReadAffinityMapOptions(readAffinityMapOptions)
+
 	return nil
 }
 

internal/util/crushlocation.go

@@ -23,19 +23,14 @@ import (
 )
 
 // GetCrushLocationMap returns the crush location map, determined from
-// the crush location labels and their values from the CO system.
+// the crush location labels and their values from the node labels passed in arg.
 // Expects crushLocationLabels in arg to be in the format "[prefix/]<name>,[prefix/]<name>,...",.
 // Returns map of crush location types with its array of associated values.
-func GetCrushLocationMap(crushLocationLabels, nodeName string) (map[string]string, error) {
+func GetCrushLocationMap(crushLocationLabels string, nodeLabels map[string]string) (map[string]string, error) {
 	if crushLocationLabels == "" {
 		return nil, nil
 	}
 
-	nodeLabels, err := k8sGetNodeLabels(nodeName)
-	if err != nil {
-		return nil, err
-	}
-
 	return getCrushLocationMap(crushLocationLabels, nodeLabels), nil
 }
 

internal/util/csiconfig.go

@@ -64,6 +64,11 @@ type ClusterInfo struct {
 		// symlink filepath for the network namespace where we need to execute commands.
 		NetNamespaceFilePath string `json:"netNamespaceFilePath"`
 	} `json:"nfs"`
+	// Read affinity map options
+	ReadAffinity struct {
+		Enabled             bool     `json:"enabled"`
+		CrushLocationLabels []string `json:"crushLocationLabels"`
+	} `json:"readAffinity"`
 }
 
 // Expected JSON structure in the passed in config file is,
@@ -209,3 +214,20 @@ func GetNFSNetNamespaceFilePath(pathToConfig, clusterID string) (string, error)
 
 	return cluster.NFS.NetNamespaceFilePath, nil
 }
+
+// GetCrushLocationLabels returns the crushLocationLabels from csi config for the given clusterID
+// If `readAffinity.enabled` is set to true, else returns an empty string.
+func GetCrushLocationLabels(pathToConfig, clusterID string) (bool, string, error) {
+	cluster, err := readClusterInfo(pathToConfig, clusterID)
+	if err != nil {
+		return false, "", err
+	}
+
+	if !cluster.ReadAffinity.Enabled {
+		return false, "", nil
+	}
+
+	crushLocationLabels := strings.Join(cluster.ReadAffinity.CrushLocationLabels, ",")
+
+	return true, crushLocationLabels, nil
+}