handle case when Patroni returns that lag is unknown

manifests/postgresql.crd.yaml

@@ -462,11 +462,11 @@ spec:
                 type: integer
               standby:
                 type: object
-                required:
-                  - s3_wal_path
                 properties:
                   s3_wal_path:
                     type: string
+                  gs_wal_path:
+                    type: string
               teamId:
                 type: string
               tls:

pkg/cluster/k8sres_test.go

@@ -24,7 +24,6 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
-	"k8s.io/apimachinery/pkg/util/intstr"
 	"k8s.io/client-go/kubernetes/fake"
 	v1core "k8s.io/client-go/kubernetes/typed/core/v1"
 )
@@ -47,11 +46,6 @@ type ExpectedValue struct {
 	envVarValue    string
 }
 
-func toIntStr(val int) *intstr.IntOrString {
-	b := intstr.FromInt(val)
-	return &b
-}
-
 func TestGenerateSpiloJSONConfiguration(t *testing.T) {
 	var cluster = New(
 		Config{
@@ -311,7 +305,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 					Labels:    map[string]string{"team": "myapp", "cluster-name": "myapp-database"},
 				},
 				Spec: policyv1beta1.PodDisruptionBudgetSpec{
-					MinAvailable: toIntStr(1),
+					MinAvailable: util.ToIntStr(1),
 					Selector: &metav1.LabelSelector{
 						MatchLabels: map[string]string{"spilo-role": "master", "cluster-name": "myapp-database"},
 					},
@@ -335,7 +329,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 					Labels:    map[string]string{"team": "myapp", "cluster-name": "myapp-database"},
 				},
 				Spec: policyv1beta1.PodDisruptionBudgetSpec{
-					MinAvailable: toIntStr(0),
+					MinAvailable: util.ToIntStr(0),
 					Selector: &metav1.LabelSelector{
 						MatchLabels: map[string]string{"spilo-role": "master", "cluster-name": "myapp-database"},
 					},
@@ -359,7 +353,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 					Labels:    map[string]string{"team": "myapp", "cluster-name": "myapp-database"},
 				},
 				Spec: policyv1beta1.PodDisruptionBudgetSpec{
-					MinAvailable: toIntStr(0),
+					MinAvailable: util.ToIntStr(0),
 					Selector: &metav1.LabelSelector{
 						MatchLabels: map[string]string{"spilo-role": "master", "cluster-name": "myapp-database"},
 					},
@@ -383,7 +377,7 @@ func TestGeneratePodDisruptionBudget(t *testing.T) {
 					Labels:    map[string]string{"team": "myapp", "cluster-name": "myapp-database"},
 				},
 				Spec: policyv1beta1.PodDisruptionBudgetSpec{
-					MinAvailable: toIntStr(1),
+					MinAvailable: util.ToIntStr(1),
 					Selector: &metav1.LabelSelector{
 						MatchLabels: map[string]string{"spilo-role": "master", "cluster-name": "myapp-database"},
 					},

pkg/cluster/pod.go

@@ -514,11 +514,15 @@ func (c *Cluster) getSwitchoverCandidate(master *v1.Pod) (spec.NamespacedName, e
 	// pick candidate with lowest lag
 	// if sync_standby replicas were found assume synchronous_mode is enabled and ignore other candidates list
 	if len(syncCandidates) > 0 {
-		sort.Slice(syncCandidates, func(i, j int) bool { return syncCandidates[i].LagInMb < syncCandidates[j].LagInMb })
+		sort.Slice(syncCandidates, func(i, j int) bool {
+			return util.IntFromIntStr(syncCandidates[i].Lag) < util.IntFromIntStr(syncCandidates[j].Lag)
+		})
 		return spec.NamespacedName{Namespace: master.Namespace, Name: syncCandidates[0].Name}, nil
 	}
 	if len(candidates) > 0 {
-		sort.Slice(candidates, func(i, j int) bool { return candidates[i].LagInMb < candidates[j].LagInMb })
+		sort.Slice(candidates, func(i, j int) bool {
+			return util.IntFromIntStr(candidates[i].Lag) < util.IntFromIntStr(candidates[j].Lag)
+		})
 		return spec.NamespacedName{Namespace: master.Namespace, Name: candidates[0].Name}, nil
 	}
 

pkg/util/patroni/patroni.go

@@ -16,6 +16,7 @@ import (
 	"github.com/sirupsen/logrus"
 	acidv1 "github.com/zalando/postgres-operator/pkg/apis/acid.zalan.do/v1"
 	v1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/util/intstr"
 )
 
 const (
@@ -184,11 +185,11 @@ type ClusterMembers struct {
 
 // ClusterMember cluster member data from Patroni API
 type ClusterMember struct {
-	Name     string `json:"name"`
+	Name     string             `json:"name"`
-	Role     string `json:"role"`
+	Role     string             `json:"role"`
-	State    string `json:"state"`
+	State    string             `json:"state"`
-	Timeline int    `json:"timeline"`
+	Timeline int                `json:"timeline"`
-	LagInMb  int    `json:"lag"`
+	Lag      intstr.IntOrString `json:"lag,omitempty"`
 }
 
 // MemberDataPatroni child element

pkg/util/patroni/patroni_test.go

@@ -15,6 +15,7 @@ import (
 
 	acidv1 "github.com/zalando/postgres-operator/pkg/apis/acid.zalan.do/v1"
 	v1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/util/intstr"
 )
 
 var logger = logrus.New().WithField("test", "patroni")
@@ -95,22 +96,22 @@ func TestGetClusterMembers(t *testing.T) {
 			Role:     "leader",
 			State:    "running",
 			Timeline: 1,
-			LagInMb:  0,
+			Lag:      intstr.IntOrString{IntVal: 0},
 		}, {
 			Name:     "acid-test-cluster-1",
 			Role:     "sync_standby",
 			State:    "running",
 			Timeline: 1,
-			LagInMb:  0,
+			Lag:      intstr.IntOrString{IntVal: 0},
 		}, {
 			Name:     "acid-test-cluster-2",
 			Role:     "replica",
 			State:    "running",
 			Timeline: 1,
-			LagInMb:  0,
+			Lag:      intstr.IntOrString{Type: 1, StrVal: "unknown"},
 		}}
 
-	json := `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 1}, {"name": "acid-test-cluster-1", "role": "sync_standby", "state": "running", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 1, "lag": 0}, {"name": "acid-test-cluster-2", "role": "replica", "state": "running", "api_url": "http://192.168.100.3:8008/patroni", "host": "192.168.100.3", "port": 5432, "timeline": 1, "lag": 0}]}`
+	json := `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 1}, {"name": "acid-test-cluster-1", "role": "sync_standby", "state": "running", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 1, "lag": 0}, {"name": "acid-test-cluster-2", "role": "replica", "state": "running", "api_url": "http://192.168.100.3:8008/patroni", "host": "192.168.100.3", "port": 5432, "timeline": 1, "lag": "unknown"}]}`
 	r := ioutil.NopCloser(bytes.NewReader([]byte(json)))
 
 	response := http.Response{

pkg/util/util.go

@@ -19,6 +19,7 @@ import (
 	"github.com/motomux/pretty"
 	resource "k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/intstr"
 
 	"github.com/zalando/postgres-operator/pkg/spec"
 	"golang.org/x/crypto/pbkdf2"
@@ -322,6 +323,20 @@ func testNil(values ...*int32) bool {
 	return false
 }
 
+// Convert int to IntOrString type
+func ToIntStr(val int) *intstr.IntOrString {
+	b := intstr.FromInt(val)
+	return &b
+}
+
+// Get int from IntOrString and return max int if string
+func IntFromIntStr(intOrStr intstr.IntOrString) int {
+	if intOrStr.Type == 1 {
+		return 32 << (^int(0) >> 32 & 1) // return max int
+	}
+	return intOrStr.IntValue()
+}
+
 // MaxInt32 : Return maximum of two integers provided via pointers. If one value
 // is not defined, return the other one. If both are not defined, result is also
 // undefined, caller needs to check for that.