K8SPSMDB-1296: improve readiness probe

cmd/mongodb-healthcheck/db/db.go

@@ -29,25 +29,27 @@ var (
 )
 
 func Dial(ctx context.Context, conf *Config) (mongo.Client, error) {
-	if err := conf.configureTLS(); err != nil {
+	log := logf.FromContext(ctx).WithName("Dial")
+	ctx = logf.IntoContext(ctx, log)
+
+	if err := conf.configureTLS(ctx); err != nil {
 		return nil, errors.Wrap(err, "configure TLS")
 	}
 
-	log := logf.FromContext(ctx)
 	log.V(1).Info("Connecting to mongodb", "hosts", conf.Hosts, "ssl", conf.SSL.Enabled, "ssl_insecure", conf.SSL.Insecure)
 
 	if conf.Username != "" && conf.Password != "" {
 		log.V(1).Info("Enabling authentication for session", "user", conf.Username)
 	}
 
-	cl, err := mongo.Dial(&conf.Config)
+	cl, err := mongo.Dial(ctx, &conf.Config)
 	if err != nil {
 		cfg := conf.Config
 		cfg.Direct = true
 		cfg.ReplSetName = ""
-		cl, err = mongo.Dial(&cfg)
+		cl, err = mongo.Dial(ctx, &cfg)
 		if err != nil {
-			return nil, errors.Wrap(err, "filed to dial mongo")
+			return nil, errors.Wrap(err, "failed to dial mongo")
 		}
 	}
 

cmd/mongodb-healthcheck/db/ssl.go

@@ -15,6 +15,7 @@
 package db
 
 import (
+	"context"
 	"crypto/tls"
 	"crypto/x509"
 	"os"
@@ -40,8 +41,8 @@ func (sc *SSLConfig) loadCaCertificate() (*x509.CertPool, error) {
 	return certificates, nil
 }
 
-func (cnf *Config) configureTLS() error {
-	log := logf.Log
+func (cnf *Config) configureTLS(ctx context.Context) error {
+	log := logf.FromContext(ctx).WithName("configureTLS")
 
 	if !cnf.SSL.Enabled {
 		return nil
@@ -72,7 +73,7 @@ func (cnf *Config) configureTLS() error {
 			return errors.Wrapf(err, "check if file with name %s exists", cnf.SSL.CAFile)
 		}
 
-		log.V(1).Info("Loading SSL/TLS Certificate Authority: %s", "ca", cnf.SSL.CAFile)
+		log.V(1).Info("Loading SSL/TLS Certificate Authority", "ca", cnf.SSL.CAFile)
 		ca, err := cnf.SSL.loadCaCertificate()
 		if err != nil {
 			return errors.Wrapf(err, "load client CAs from %s", cnf.SSL.CAFile)

cmd/mongodb-healthcheck/db/ssl_test.go

@@ -16,7 +16,7 @@ func TestSSLNotEnabled(t *testing.T) {
 		},
 	}
 
-	if err := cfg.configureTLS(); err != nil {
+	if err := cfg.configureTLS(t.Context()); err != nil {
 		t.Fatalf("TLS configuration failed: %s", err)
 	}
 
@@ -32,7 +32,7 @@ func TestSSLEnabled(t *testing.T) {
 		},
 	}
 
-	if err := cfg.configureTLS(); err != nil {
+	if err := cfg.configureTLS(t.Context()); err != nil {
 		t.Fatalf("TLS configuration failed: %s", err)
 	}
 
@@ -49,7 +49,7 @@ func TestPEMKeyFileDoesNotExists(t *testing.T) {
 		},
 	}
 
-	err := cfg.configureTLS()
+	err := cfg.configureTLS(t.Context())
 	if err == nil {
 		t.Fatal("Expected TLS config to fail, but it returned no error")
 	}
@@ -71,7 +71,7 @@ func TestCAFileDoesNotExists(t *testing.T) {
 		},
 	}
 
-	err := cfg.configureTLS()
+	err := cfg.configureTLS(t.Context())
 	if err == nil {
 		t.Fatal("Expected TLS config to fail, but it returned no error")
 	}

cmd/mongodb-healthcheck/healthcheck/health.go

@@ -16,12 +16,9 @@ package healthcheck
 
 import (
 	"context"
-	"encoding/json"
 
-	v "github.com/hashicorp/go-version"
 	"github.com/pkg/errors"
 	"go.mongodb.org/mongo-driver/bson"
-	"go.mongodb.org/mongo-driver/bson/primitive"
 	logf "sigs.k8s.io/controller-runtime/pkg/log"
 
 	"github.com/percona/percona-server-mongodb-operator/cmd/mongodb-healthcheck/db"
@@ -32,6 +29,7 @@ var ErrNoReplsetConfigStr = "(NotYetInitialized) no replset config has been rece
 
 func HealthCheckMongosLiveness(ctx context.Context, cnf *db.Config) (err error) {
 	log := logf.FromContext(ctx).WithName("HealthCheckMongosLiveness")
+	ctx = logf.IntoContext(ctx, log)
 
 	client, err := db.Dial(ctx, cnf)
 	if err != nil {
@@ -58,6 +56,7 @@ func HealthCheckMongosLiveness(ctx context.Context, cnf *db.Config) (err error)
 
 func HealthCheckMongodLiveness(ctx context.Context, cnf *db.Config, startupDelaySeconds int64) (_ *mongo.MemberState, err error) {
 	log := logf.FromContext(ctx).WithName("HealthCheckMongodLiveness")
+	ctx = logf.IntoContext(ctx, log)
 
 	client, err := db.Dial(ctx, cnf)
 	if err != nil {
@@ -74,50 +73,14 @@ func HealthCheckMongodLiveness(ctx context.Context, cnf *db.Config, startupDelay
 		return nil, errors.Wrap(err, "get isMaster response")
 	}
 
-	buildInfo, err := client.RSBuildInfo(ctx)
+	rsStatus, err := client.RSStatus(ctx)
 	if err != nil {
-		return nil, errors.Wrap(err, "get buildInfo response")
-	}
-
-	replSetStatusCommand := bson.D{{Key: "replSetGetStatus", Value: 1}}
-	mongoVersion := v.Must(v.NewVersion(buildInfo.Version))
-	if mongoVersion.Compare(v.Must(v.NewVersion("4.2.1"))) < 0 {
-		// https://docs.mongodb.com/manual/reference/command/replSetGetStatus/#syntax
-		replSetStatusCommand = append(replSetStatusCommand, primitive.E{Key: "initialSync", Value: 1})
-	}
-
-	res := client.Database("admin").RunCommand(ctx, replSetStatusCommand)
-	if res.Err() != nil {
-		// if we come this far, it means db connection was successful
-		// standalone mongod nodes in an unmanaged cluster doesn't need
-		// to die before they added to a replset
-		if res.Err().Error() == ErrNoReplsetConfigStr {
+		if err.Error() == ErrNoReplsetConfigStr {
 			state := mongo.MemberStateUnknown
-			log.V(1).Info("replSetGetStatus failed", "err", res.Err().Error(), "state", state)
+			log.V(1).Info("replSetGetStatus failed", "err", err.Error(), "state", state)
 			return &state, nil
 		}
-		return nil, errors.Wrap(res.Err(), "get replsetGetStatus response")
-	}
-
-	// this is a workaround to fix decoding of empty interfaces
-	// https://jira.mongodb.org/browse/GODRIVER-988
-	rsStatus := ReplSetStatus{}
-	tempResult := bson.M{}
-	err = res.Decode(&tempResult)
-	if err != nil {
-		return nil, errors.Wrap(err, "decode replsetGetStatus response")
-	}
-
-	if err == nil {
-		result, err := json.Marshal(tempResult)
-		if err != nil {
-			return nil, errors.Wrap(err, "marshal temp result")
-		}
-
-		err = json.Unmarshal(result, &rsStatus)
-		if err != nil {
-			return nil, errors.Wrap(err, "unmarshal temp result")
-		}
+		return nil, errors.Wrap(err, "get replSetGetStatus response")
 	}
 
 	oplogRs := OplogRs{}
@@ -156,15 +119,12 @@ type OplogRs struct {
 	StorageSize      int64 `bson:"storageSize" json:"storageSize"`
 }
 
-type ReplSetStatus struct {
-	InitialSyncStatus InitialSyncStatus `bson:"initialSyncStatus" json:"initialSyncStatus"`
-	mongo.Status      `bson:",inline"`
-}
-
-type InitialSyncStatus interface{}
-
-func CheckState(rs ReplSetStatus, startupDelaySeconds int64, oplogSize int64) error {
-	uptime := rs.GetSelf().Uptime
+func CheckState(rs mongo.Status, startupDelaySeconds int64, oplogSize int64) error {
+	self := rs.GetSelf()
+	if self == nil {
+		return errors.New("self member is not found")
+	}
+	uptime := self.Uptime
 
 	switch rs.MyState {
 	case mongo.MemberStatePrimary, mongo.MemberStateSecondary, mongo.MemberStateArbiter:

cmd/mongodb-healthcheck/healthcheck/readiness.go

@@ -17,6 +17,7 @@ package healthcheck
 import (
 	"context"
 	"net"
+	"time"
 
 	"github.com/pkg/errors"
 	"go.mongodb.org/mongo-driver/bson"
@@ -27,21 +28,64 @@ import (
 )
 
 // MongodReadinessCheck runs a ping on a pmgo.SessionManager to check server readiness
-func MongodReadinessCheck(ctx context.Context, addr string) error {
+func MongodReadinessCheck(ctx context.Context, cnf *db.Config) error {
 	log := logf.FromContext(ctx).WithName("MongodReadinessCheck")
+	ctx = logf.IntoContext(ctx, log)
 
 	var d net.Dialer
 
+	if len(cnf.Hosts) == 0 {
+		return errors.New("no hosts found")
+	}
+	addr := cnf.Hosts[0]
 	log.V(1).Info("Connecting to " + addr)
 	conn, err := d.DialContext(ctx, "tcp", addr)
 	if err != nil {
 		return errors.Wrap(err, "dial")
 	}
-	return conn.Close()
+	if err := conn.Close(); err != nil {
+		return err
+	}
+
+	s, err := func() (status *mongo.Status, err error) {
+		cnf.Timeout = time.Second
+		client, err := db.Dial(ctx, cnf)
+		if err != nil {
+			// The operator waits for the StatefulSet to be ready before initializing the database.
+			// Until then, it is not possible to connect to it.
+			// We should ignore any errors from Dial to allow the cluster to be deployed.
+			return nil, nil
+		}
+		defer func() {
+			if derr := client.Disconnect(ctx); derr != nil && err == nil {
+				err = errors.Wrap(derr, "failed to disconnect")
+			}
+		}()
+		var rs mongo.Status
+		rs, err = client.RSStatus(ctx)
+		if err != nil {
+			if errors.Is(err, mongo.ErrInvalidReplsetConfig) {
+				log.Info("Couldn't connect to mongo due to invalid replset config. Ignoring", "error", err)
+				return nil, nil
+			}
+			return nil, err
+		}
+		return &rs, nil
+	}()
+	if err != nil || s == nil {
+		return errors.Wrap(err, "failed to get rs status")
+	}
+
+	if err := CheckState(*s, 0, 0); err != nil {
+		return errors.Wrap(err, "check state")
+	}
+
+	return nil
 }
 
 func MongosReadinessCheck(ctx context.Context, cnf *db.Config) (err error) {
 	log := logf.FromContext(ctx).WithName("MongosReadinessCheck")
+	ctx = logf.IntoContext(ctx, log)
 
 	client, err := db.Dial(ctx, cnf)
 	if err != nil {

cmd/mongodb-healthcheck/tool/tool.go

@@ -111,7 +111,7 @@ func (app *App) Run(ctx context.Context) error {
 		switch *component {
 
 		case "mongod":
-			err := healthcheck.MongodReadinessCheck(ctx, cnf.Hosts[0])
+			err := healthcheck.MongodReadinessCheck(ctx, cnf)
 			if err != nil {
 				return errors.Wrap(err, "member failed Kubernetes readiness check")
 			}

e2e-tests/arbiter/compare/statefulset_arbiter-clusterip-rs0-arbiter-oc.yml

@@ -113,6 +113,12 @@ spec:
               - readiness
               - --component
               - mongod
+              - --ssl
+              - --sslInsecure
+              - --sslCAFile
+              - /etc/mongodb-ssl/ca.crt
+              - --sslPEMKeyFile
+              - /tmp/tls.pem
           failureThreshold: 8
           initialDelaySeconds: 10
           periodSeconds: 3

e2e-tests/arbiter/compare/statefulset_arbiter-clusterip-rs0-arbiter.yml

@@ -107,6 +107,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 8
             initialDelaySeconds: 10
             periodSeconds: 3

e2e-tests/arbiter/compare/statefulset_arbiter-rs0-arbiter-oc.yml

@@ -114,6 +114,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 8
             initialDelaySeconds: 10
             periodSeconds: 3

e2e-tests/arbiter/compare/statefulset_arbiter-rs0-arbiter.yml

@@ -114,6 +114,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 8
             initialDelaySeconds: 10
             periodSeconds: 3

e2e-tests/custom-tls/compare/statefulset_some-name-cfg-oc.yml

@@ -114,6 +114,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 3
             initialDelaySeconds: 10
             periodSeconds: 3

e2e-tests/custom-tls/compare/statefulset_some-name-cfg.yml

@@ -114,6 +114,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 3
             initialDelaySeconds: 10
             periodSeconds: 3

e2e-tests/custom-tls/compare/statefulset_some-name-rs0-oc.yml

@@ -103,6 +103,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 8
             initialDelaySeconds: 10
             periodSeconds: 3

e2e-tests/custom-tls/compare/statefulset_some-name-rs0.yml

@@ -103,6 +103,12 @@ spec:
                 - readiness
                 - --component
                 - mongod
+                - --ssl
+                - --sslInsecure
+                - --sslCAFile
+                - /etc/mongodb-ssl/ca.crt
+                - --sslPEMKeyFile
+                - /tmp/tls.pem
             failureThreshold: 8
             initialDelaySeconds: 10
             periodSeconds: 3