PMM-13477 Support MongoDB 8.0 (#943)

BupycHuk · ademidoff · web-flow · commit 1e917bf2a650 · 2024-11-05T12:35:08.000+03:00
* PMM-13477 Support MongoDB 8.0 * PMM-13477 Support MongoDB 8.0 * PMM-13477 Support MongoDB 8.0 * PMM-13477 Fix tests. * PMM-13477 Fix imports and test matrix. * PMM-13477 A little bit of refactoring. * PMM-13477 Fix tests. * PMM-13477 Fix fcv tests. * PMM-13477 see logs on failure. * PMM-13477 fix tests. * PMM-13477 fix tests. * PMM-13477 fix imports. * PMM-13477 a bit of refactoring. * PMM-13477 fix test. * Update exporter/diagnostic_data_collector.go Co-authored-by: Alex Demidoff <alexander.demidoff@percona.com> --------- Co-authored-by: Alex Demidoff <alexander.demidoff@percona.com>
diff --git a/.github/workflows/go.yml b/.github/workflows/go.yml
@@ -22,8 +22,15 @@ jobs:
         image:
           - mongo:4.4
           - mongo:5.0
+          - mongo:6.0
+          - mongo:7.0
+          - mongo:8.0
+          - mongo:latest
           - percona/percona-server-mongodb:4.4
           - percona/percona-server-mongodb:5.0
+          - percona/percona-server-mongodb:6.0
+          - percona/percona-server-mongodb:7.0
+          - percona/percona-server-mongodb:latest
 
     runs-on: ubuntu-latest
 
@@ -42,3 +49,17 @@ jobs:
           sleep 10
           make test-race
           make test-cluster-clean
+
+      - name: Run debug commands on failure
+        if: ${{ failure() }}
+        run: |
+          echo "--- Environment variables ---"
+          env | sort
+          echo "--- GO Environment ---"
+          go env | sort
+          echo "--- Git status ---"
+          git status
+          echo "--- Docker logs ---"
+          docker compose logs
+          echo "--- Docker ps ---"
+          docker compose ps -a
diff --git a/exporter/diagnostic_data_collector.go b/exporter/diagnostic_data_collector.go
@@ -35,12 +35,14 @@ type diagnosticDataCollector struct {
 	ctx  context.Context
 	base *baseCollector
 
+	buildInfo buildInfo
+
 	compatibleMode bool
 	topologyInfo   labelsGetter
 }
 
 // newDiagnosticDataCollector creates a collector for diagnostic information.
-func newDiagnosticDataCollector(ctx context.Context, client *mongo.Client, logger *logrus.Logger, compatible bool, topology labelsGetter) *diagnosticDataCollector {
+func newDiagnosticDataCollector(ctx context.Context, client *mongo.Client, logger *logrus.Logger, compatible bool, topology labelsGetter, buildInfo buildInfo) *diagnosticDataCollector {
 	nodeType, err := getNodeType(ctx, client)
 	if err != nil {
 		logger.WithFields(logrus.Fields{
@@ -57,6 +59,8 @@ func newDiagnosticDataCollector(ctx context.Context, client *mongo.Client, logge
 		ctx:  ctx,
 		base: newBaseCollector(client, logger.WithFields(logrus.Fields{"collector": "diagnostic_data"})),
 
+		buildInfo: buildInfo,
+
 		compatibleMode: compatible,
 		topologyInfo:   topology,
 	}
@@ -112,6 +116,21 @@ func (d *diagnosticDataCollector) collect(ch chan<- prometheus.Metric) {
 		logger.Debug("getDiagnosticData result")
 		debugResult(logger, m)
 
+		// MongoDB 8.0 splits the diagnostic data into multiple blocks, so we need to merge them
+		if d.buildInfo.VersionArray[0] >= 8 { //nolint:gomnd
+			b := bson.M{}
+			for _, mv := range m {
+				block, ok := mv.(bson.M)
+				if !ok {
+					continue
+				}
+				for k, v := range block {
+					b[k] = v
+				}
+			}
+			m = b
+		}
+
 		metrics = makeMetrics("", m, d.topologyInfo.baseLabels(), d.compatibleMode)
 		metrics = append(metrics, locksMetrics(logger, m)...)
 
@@ -132,12 +151,7 @@ func (d *diagnosticDataCollector) collect(ch chan<- prometheus.Metric) {
 	}
 
 	if d.compatibleMode {
-		buildInfo, err := retrieveMongoDBBuildInfo(d.ctx, client, logger)
-		if err != nil {
-			logger.Errorf("cannot retrieve MongoDB buildInfo: %s", err)
-		}
-
-		metrics = append(metrics, serverVersion(buildInfo))
+		metrics = append(metrics, serverVersion(d.buildInfo))
 
 		if nodeType == typeArbiter {
 			if hm := arbiterMetrics(d.ctx, client, logger); hm != nil {
diff --git a/exporter/diagnostic_data_collector_test.go b/exporter/diagnostic_data_collector_test.go
@@ -46,31 +46,36 @@ func TestDiagnosticDataCollector(t *testing.T) {
 	logger := logrus.New()
 	ti := labelsGetterMock{}
 
-	c := newDiagnosticDataCollector(ctx, client, logger, false, ti)
+	dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+	require.NoError(t, err)
+
+	c := newDiagnosticDataCollector(ctx, client, logger, false, ti, dbBuildInfo)
+
+	prefix := "local.oplog.rs.stats.storageStats.wiredTiger"
+	if dbBuildInfo.VersionArray[0] < 7 {
+		prefix = "local.oplog.rs.stats.wiredTiger"
+	}
 
 	// The last \n at the end of this string is important
-	expected := strings.NewReader(`
-	# HELP mongodb_oplog_stats_ok local.oplog.rs.stats.
-	# TYPE mongodb_oplog_stats_ok untyped
-	mongodb_oplog_stats_ok 1
-	# HELP mongodb_oplog_stats_wt_btree_fixed_record_size local.oplog.rs.stats.wiredTiger.btree.
+	expectedString := fmt.Sprintf(`
+	# HELP mongodb_oplog_stats_wt_btree_fixed_record_size %s.btree.
 	# TYPE mongodb_oplog_stats_wt_btree_fixed_record_size untyped
 	mongodb_oplog_stats_wt_btree_fixed_record_size 0
-	# HELP mongodb_oplog_stats_wt_transaction_update_conflicts local.oplog.rs.stats.wiredTiger.transaction.
+	# HELP mongodb_oplog_stats_wt_transaction_update_conflicts %s.transaction.
 	# TYPE mongodb_oplog_stats_wt_transaction_update_conflicts untyped
-	mongodb_oplog_stats_wt_transaction_update_conflicts 0` + "\n")
+	mongodb_oplog_stats_wt_transaction_update_conflicts 0`, prefix, prefix)
+	expected := strings.NewReader(expectedString + "\n")
 
 	// Filter metrics for 2 reasons:
 	// 1. The result is huge
 	// 2. We need to check against know values. Don't use metrics that return counters like uptime
 	//    or counters like the number of transactions because they won't return a known value to compare
 	filter := []string{
-		"mongodb_oplog_stats_ok",
 		"mongodb_oplog_stats_wt_btree_fixed_record_size",
 		"mongodb_oplog_stats_wt_transaction_update_conflicts",
 	}
 
-	err := testutil.CollectAndCompare(c, expected, filter...)
+	err = testutil.CollectAndCompare(c, expected, filter...)
 	assert.NoError(t, err)
 }
 
@@ -188,7 +193,10 @@ func TestCollectorWithCompatibleMode(t *testing.T) {
 			logger := logrus.New()
 			ti := labelsGetterMock{}
 
-			c := newDiagnosticDataCollector(ctx, client, logger, true, ti)
+			dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+			require.NoError(t, err)
+
+			c := newDiagnosticDataCollector(ctx, client, logger, true, ti, dbBuildInfo)
 
 			err = testutil.CollectAndCompare(c, tt.expectedMetrics(), tt.metricsFilter...)
 			assert.NoError(t, err)
@@ -202,12 +210,16 @@ func TestAllDiagnosticDataCollectorMetrics(t *testing.T) {
 
 	client := tu.DefaultTestClient(ctx, t)
 
-	ti := newTopologyInfo(ctx, client, logrus.New())
+	logger := logrus.New()
+	ti := newTopologyInfo(ctx, client, logger)
 
-	c := newDiagnosticDataCollector(ctx, client, logrus.New(), true, ti)
+	dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+	require.NoError(t, err)
+
+	c := newDiagnosticDataCollector(ctx, client, logger, true, ti, dbBuildInfo)
 
 	reg := prometheus.NewRegistry()
-	err := reg.Register(c)
+	err = reg.Register(c)
 	require.NoError(t, err)
 	metrics := helpers.CollectMetrics(c)
 	actualMetrics := helpers.ReadMetrics(metrics)
@@ -281,7 +293,11 @@ func TestDiagnosticDataErrors(t *testing.T) {
 
 			logger, hook := logrustest.NewNullLogger()
 			ti := newTopologyInfo(ctx, client, logger)
-			c := newDiagnosticDataCollector(ctx, client, logger, true, ti)
+
+			dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+			require.NoError(t, err)
+
+			c := newDiagnosticDataCollector(ctx, client, logger, true, ti, dbBuildInfo)
 
 			reg := prometheus.NewRegistry()
 			err = reg.Register(c)
@@ -318,19 +334,23 @@ func TestContextTimeout(t *testing.T) {
 
 	client := tu.DefaultTestClient(ctx, t)
 
-	ti := newTopologyInfo(ctx, client, logrus.New())
+	logger := logrus.New()
+	ti := newTopologyInfo(ctx, client, logger)
+
+	dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+	require.NoError(t, err)
 
 	dbCount := 100
 
-	err := addTestData(ctx, client, dbCount)
+	err = addTestData(ctx, client, dbCount)
 	assert.NoError(t, err)
 
 	defer cleanTestData(ctx, client, dbCount) //nolint:errcheck
 
 	cctx, ccancel := context.WithCancel(context.Background())
 	ccancel()
 
-	c := newDiagnosticDataCollector(cctx, client, logrus.New(), true, ti)
+	c := newDiagnosticDataCollector(cctx, client, logger, true, ti, dbBuildInfo)
 	// it should not panic
 	helpers.CollectMetrics(c)
 }
@@ -403,7 +423,7 @@ func cleanTestData(ctx context.Context, client *mongo.Client, count int) error {
 }
 
 func TestDisconnectedDiagnosticDataCollector(t *testing.T) {
-	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
 
 	client := tu.DefaultTestClient(ctx, t)
@@ -415,7 +435,10 @@ func TestDisconnectedDiagnosticDataCollector(t *testing.T) {
 
 	ti := labelsGetterMock{}
 
-	c := newDiagnosticDataCollector(ctx, client, logger, true, ti)
+	dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+	require.Error(t, err)
+
+	c := newDiagnosticDataCollector(ctx, client, logger, true, ti, dbBuildInfo)
 
 	// The last \n at the end of this string is important
 	expected := strings.NewReader(`
diff --git a/exporter/encryption_info_test.go b/exporter/encryption_info_test.go
@@ -25,6 +25,7 @@ import (
 	"github.com/prometheus/client_golang/prometheus/testutil"
 	"github.com/sirupsen/logrus"
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 
 	"github.com/percona/mongodb_exporter/internal/tu"
 )
@@ -45,7 +46,10 @@ func TestGetEncryptionInfo(t *testing.T) {
 
 	ti := labelsGetterMock{}
 
-	c := newDiagnosticDataCollector(ctx, client, logger, true, ti)
+	dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, logger.WithField("component", "test"))
+	require.NoError(t, err)
+
+	c := newDiagnosticDataCollector(ctx, client, logger, true, ti, dbBuildInfo)
 
 	// The last \n at the end of this string is important
 	expected := strings.NewReader(`
@@ -61,6 +65,6 @@ func TestGetEncryptionInfo(t *testing.T) {
 		"mongodb_version_info",
 	}
 
-	err := testutil.CollectAndCompare(c, expected, filter...)
+	err = testutil.CollectAndCompare(c, expected, filter...)
 	assert.NoError(t, err)
 }
diff --git a/exporter/exporter.go b/exporter/exporter.go
@@ -134,16 +134,17 @@ func (e *Exporter) makeRegistry(ctx context.Context, client *mongo.Client, topol
 
 	nodeType, err := getNodeType(ctx, client)
 	if err != nil {
-		e.logger.Errorf("Registry - Cannot get node type to check if this is a mongos : %s", err)
+		e.logger.Errorf("Registry - Cannot get node type : %s", err)
+	}
+
+	dbBuildInfo, err := retrieveMongoDBBuildInfo(ctx, client, e.logger.WithField("component", "buildInfo"))
+	if err != nil {
+		e.logger.Warnf("Registry - Cannot get MongoDB buildInfo: %s", err)
 	}
 
 	gc := newGeneralCollector(ctx, client, nodeType, e.opts.Logger)
 	registry.MustRegister(gc)
 
-	if client == nil {
-		return registry
-	}
-
 	// Enable collectors like collstats and indexstats depending on the number of collections
 	// present in the database.
 	limitsOk := false
@@ -203,7 +204,7 @@ func (e *Exporter) makeRegistry(ctx context.Context, client *mongo.Client, topol
 
 	if e.opts.EnableDiagnosticData && requestOpts.EnableDiagnosticData {
 		ddc := newDiagnosticDataCollector(ctx, client, e.opts.Logger,
-			e.opts.CompatibleMode, topologyInfo)
+			e.opts.CompatibleMode, topologyInfo, dbBuildInfo)
 		registry.MustRegister(ddc)
 	}
 
@@ -335,13 +336,18 @@ func (e *Exporter) Handler() http.Handler {
 			gatherers = append(gatherers, prometheus.DefaultGatherer)
 		}
 
+		var registry *prometheus.Registry
 		var ti *topologyInfo
 		if client != nil {
 			// Topology can change between requests, so we need to get it every time.
 			ti = newTopologyInfo(ctx, client, e.logger)
+			registry = e.makeRegistry(ctx, client, ti, requestOpts)
+		} else {
+			registry = prometheus.NewRegistry()
+			gc := newGeneralCollector(ctx, client, "", e.opts.Logger)
+			registry.MustRegister(gc)
 		}
 
-		registry := e.makeRegistry(ctx, client, ti, requestOpts)
 		gatherers = append(gatherers, registry)
 
 		// Delegate http serving to Prometheus client library, which will call collector.Collect.
diff --git a/exporter/exporter_test.go b/exporter/exporter_test.go
@@ -186,9 +186,7 @@ func TestMongoS(t *testing.T) {
 		assert.NoError(t, err)
 
 		e := New(exporterOpts)
-
-		rsgsc := newReplicationSetStatusCollector(ctx, client, e.opts.Logger,
-			e.opts.CompatibleMode, new(labelsGetterMock))
+		rsgsc := newReplicationSetStatusCollector(ctx, client, e.opts.Logger, e.opts.CompatibleMode, new(labelsGetterMock))
 
 		r := e.makeRegistry(ctx, client, new(labelsGetterMock), *e.opts)
 
diff --git a/exporter/feature_compatibility_version_collector_test.go b/exporter/feature_compatibility_version_collector_test.go
@@ -55,6 +55,14 @@ func TestFCVCollector(t *testing.T) {
 		mversion = "4.4"
 	case mmv == "4.4":
 		mversion = "4.2"
+	case mmv == "6.0":
+		mversion = "5.0"
+	case mmv == "7.0":
+		mversion = "6.0"
+	case mmv == "8.0":
+		mversion = "7.0"
+	default:
+		mversion = mmv
 	}
 
 	// The last \n at the end of this string is important
diff --git a/exporter/general_collector.go b/exporter/general_collector.go
@@ -63,14 +63,11 @@ func mongodbUpMetric(ctx context.Context, client *mongo.Client, nodeType mongoDB
 		} else {
 			log.Errorf("error while checking mongodb connection: %s. mongo_up is set to 0", err.Error())
 		}
-
 		switch nodeType { //nolint:exhaustive
-		case typeMongos:
-			clusterRole = typeMongos
-		case typeArbiter:
-			clusterRole = typeArbiter
-		default:
+		case typeShardServer:
 			clusterRole = typeMongod
+		default:
+			clusterRole = nodeType
 		}
 	}
 
diff --git a/exporter/metrics.go b/exporter/metrics.go
@@ -52,11 +52,14 @@ var (
 	prefixes = [][]string{
 		{"serverStatus.wiredTiger.transaction", "ss_wt_txn"},
 		{"serverStatus.wiredTiger", "ss_wt"},
+		{"serverStatus.queues.execution", "ss_wt_concurrentTransactions"},
 		{"serverStatus", "ss"},
 		{"replSetGetStatus", "rs"},
 		{"systemMetrics", "sys"},
 		{"local.oplog.rs.stats.wiredTiger", "oplog_stats_wt"},
+		{"local.oplog.rs.stats.storageStats.wiredTiger", "oplog_stats_wt"},
 		{"local.oplog.rs.stats", "oplog_stats"},
+		{"local.oplog.rs.stats.storageStats", "oplog_stats"},
 		{"collstats_storage.wiredTiger", "collstats_storage_wt"},
 		{"collstats_storage.indexDetails", "collstats_storage_idx"},
 		{"collStats.storageStats", "collstats_storage"},
@@ -105,6 +108,7 @@ var (
 		"serverStatus.opcountersRepl.":                    "legacy_op_type",
 		"serverStatus.transactions.commitTypes.":          "commit_type",
 		"serverStatus.wiredTiger.concurrentTransactions.": "txn_rw_type",
+		"serverStatus.queues.execution.":                  "txn_rw_type",
 		"serverStatus.wiredTiger.perf.":                   "perf_bucket",
 		"systemMetrics.disks.":                            "device_name",
 	}
diff --git a/exporter/secondary_lag_test.go b/exporter/secondary_lag_test.go
@@ -23,6 +23,7 @@ import (
 
 	"github.com/prometheus/client_golang/prometheus"
 	dto "github.com/prometheus/client_model/go"
+	"github.com/sirupsen/logrus"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"go.mongodb.org/mongo-driver/bson"
@@ -126,7 +127,7 @@ func TestSecondaryLag(t *testing.T) {
 	assert.NoError(t, err)
 
 	m, _ = m["data"].(bson.M)
-	metrics := replSetMetrics(m)
+	metrics := replSetMetrics(m, logrus.WithField("component", "test"))
 	var lag prometheus.Metric
 	for _, m := range metrics {
 		if strings.HasPrefix(m.Desc().String(), `Desc{fqName: "mongodb_mongod_replset_member_replication_lag"`) {
diff --git a/exporter/server.go b/exporter/server.go
diff --git a/exporter/topology_info.go b/exporter/topology_info.go
diff --git a/exporter/topology_info_test.go b/exporter/topology_info_test.go
diff --git a/exporter/v1_compatibility.go b/exporter/v1_compatibility.go

Original file line number	Diff line number	Diff line change
`@@ -63,14 +63,11 @@ func mongodbUpMetric(ctx context.Context, client *mongo.Client, nodeType mongoDB`
`63`	`63`	`} else {`
`64`	`64`	`log.Errorf("error while checking mongodb connection: %s. mongo_up is set to 0", err.Error())`
`65`	`65`	`}`
`66`		`-`
`67`	`66`	`switch nodeType { //nolint:exhaustive`
`68`		`- case typeMongos:`
`69`		`- clusterRole = typeMongos`
`70`		`- case typeArbiter:`
`71`		`- clusterRole = typeArbiter`
`72`		`- default:`
	`67`	`+ case typeShardServer:`
`73`	`68`	`clusterRole = typeMongod`
	`69`	`+ default:`
	`70`	`+ clusterRole = nodeType`
`74`	`71`	`}`
`75`	`72`	`}`
`76`	`73`