Merge pull request #7 from rokernel/feat/fdb-summary-collector

feat: add protected FDB summary collector

Author: rokernel

README.md

@@ -10,6 +10,7 @@ Currently supported collectors:
 - [LLDP collector](internal/collector/lldp_collector.go): collects LLDP neighbor information from SONiC Redis.
 - [VLAN collector](internal/collector/vlan_collector.go): collects VLAN and VLAN member state from SONiC Redis.
 - [LAG collector](internal/collector/lag_collector.go): collects PortChannel and member state from SONiC Redis.
+- [FDB collector](internal/collector/fdb_collector.go): collects FDB summary metrics from SONiC ASIC DB.
 
 # Usage
 
@@ -45,6 +46,12 @@ Environment variables:
 - `LAG_TIMEOUT` - timeout for one LAG refresh cycle. Default: `2s`.
 - `LAG_MAX_LAGS` - maximum number of LAGs exported per refresh. Default: `512`.
 - `LAG_MAX_MEMBERS` - maximum number of LAG members exported per refresh. Default: `4096`.
+- `FDB_ENABLED` - enable FDB collector. Default: `false`.
+- `FDB_REFRESH_INTERVAL` - FDB cache refresh interval. Default: `60s`.
+- `FDB_TIMEOUT` - timeout for one FDB refresh cycle. Default: `2s`.
+- `FDB_MAX_ENTRIES` - maximum number of ASIC FDB entries processed per refresh. Default: `50000`.
+- `FDB_MAX_PORTS` - maximum number of per-port FDB series exported. Default: `1024`.
+- `FDB_MAX_VLANS` - maximum number of per-VLAN FDB series exported. Default: `4096`.
 
 ## Validated Platforms
 

cmd/sonic-exporter/main.go

@@ -60,6 +60,7 @@ func main() {
 	lldpCollector := collector.NewLldpCollector(logger)
 	vlanCollector := collector.NewVlanCollector(logger)
 	lagCollector := collector.NewLagCollector(logger)
+	fdbCollector := collector.NewFdbCollector(logger)
 	prometheus.MustRegister(interfaceCollector)
 	prometheus.MustRegister(hwCollector)
 	prometheus.MustRegister(crmCollector)
@@ -73,6 +74,9 @@ func main() {
 	if lagCollector.IsEnabled() {
 		prometheus.MustRegister(lagCollector)
 	}
+	if fdbCollector.IsEnabled() {
+		prometheus.MustRegister(fdbCollector)
+	}
 
 	// Node exporter collectors
 	nodeCollector, err := nodecollector.NewNodeCollector(logger,

fixtures/test/asic_db_data.json

@@ -0,0 +1,37 @@
+{
+  "id": "ASIC_DB",
+  "data": {
+    "ASIC_STATE:SAI_OBJECT_TYPE_VLAN:oid:0x260000000000111": {
+      "SAI_VLAN_ATTR_VLAN_ID": "1000"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_VLAN:oid:0x260000000000222": {
+      "SAI_VLAN_ATTR_VLAN_ID": "2000"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_BRIDGE_PORT:oid:0x3a000000000001": {
+      "SAI_BRIDGE_PORT_ATTR_PORT_ID": "oid:0x1000000000002"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_BRIDGE_PORT:oid:0x3a000000000002": {
+      "SAI_BRIDGE_PORT_ATTR_PORT_ID": "oid:0x1000000000003"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_FDB_ENTRY:{\"bvid\":\"oid:0x260000000000111\",\"mac\":\"AA:BB:CC:00:00:01\",\"switch_id\":\"oid:0x21000000000000\"}": {
+      "SAI_FDB_ENTRY_ATTR_BRIDGE_PORT_ID": "oid:0x3a000000000001",
+      "SAI_FDB_ENTRY_ATTR_TYPE": "SAI_FDB_ENTRY_TYPE_DYNAMIC"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_FDB_ENTRY:{\"bvid\":\"oid:0x260000000000111\",\"mac\":\"AA:BB:CC:00:00:02\",\"switch_id\":\"oid:0x21000000000000\"}": {
+      "SAI_FDB_ENTRY_ATTR_BRIDGE_PORT_ID": "oid:0x3a000000000001",
+      "SAI_FDB_ENTRY_ATTR_TYPE": "SAI_FDB_ENTRY_TYPE_DYNAMIC"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_FDB_ENTRY:{\"bvid\":\"oid:0x260000000000222\",\"mac\":\"AA:BB:CC:00:00:03\",\"switch_id\":\"oid:0x21000000000000\"}": {
+      "SAI_FDB_ENTRY_ATTR_BRIDGE_PORT_ID": "oid:0x3a000000000002",
+      "SAI_FDB_ENTRY_ATTR_TYPE": "SAI_FDB_ENTRY_TYPE_STATIC"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_FDB_ENTRY:{\"bvid\":\"oid:0x260000000000999\",\"mac\":\"AA:BB:CC:00:00:04\",\"switch_id\":\"oid:0x21000000000000\"}": {
+      "SAI_FDB_ENTRY_ATTR_BRIDGE_PORT_ID": "oid:0x3a000000000002",
+      "SAI_FDB_ENTRY_ATTR_TYPE": "SAI_FDB_ENTRY_TYPE_DYNAMIC"
+    },
+    "ASIC_STATE:SAI_OBJECT_TYPE_FDB_ENTRY:not-json": {
+      "SAI_FDB_ENTRY_ATTR_BRIDGE_PORT_ID": "oid:0x3a000000000001",
+      "SAI_FDB_ENTRY_ATTR_TYPE": "SAI_FDB_ENTRY_TYPE_DYNAMIC"
+    }
+  }
+}

internal/collector/collector_test.go

@@ -27,6 +27,7 @@ func populateRedisData() error {
 		"../../fixtures/test/counters_db_data.json",
 		"../../fixtures/test/config_db_data.json",
 		"../../fixtures/test/appl_db_data.json",
+		"../../fixtures/test/asic_db_data.json",
 		"../../fixtures/test/state_db_data.json",
 	}
 
@@ -79,6 +80,7 @@ func TestMain(m *testing.M) {
 	os.Setenv("LLDP_INCLUDE_MGMT", "true")
 	os.Setenv("VLAN_ENABLED", "true")
 	os.Setenv("LAG_ENABLED", "true")
+	os.Setenv("FDB_ENABLED", "true")
 	err = populateRedisData()
 	if err != nil {
 		slog.Error("failed to populate redis data", "error", err)
@@ -93,6 +95,7 @@ func TestMain(m *testing.M) {
 	os.Unsetenv("LLDP_INCLUDE_MGMT")
 	os.Unsetenv("VLAN_ENABLED")
 	os.Unsetenv("LAG_ENABLED")
+	os.Unsetenv("FDB_ENABLED")
 	os.Exit(exitCode)
 }
 
@@ -375,3 +378,97 @@ func TestLagCollector(t *testing.T) {
 		t.Errorf("unexpected collecting result:\n%s", err)
 	}
 }
+
+func TestFdbCollector(t *testing.T) {
+	promslogConfig := &promslog.Config{}
+	logger := promslog.New(promslogConfig)
+
+	fdbCollector := NewFdbCollector(logger)
+
+	problems, err := testutil.CollectAndLint(fdbCollector)
+	if err != nil {
+		t.Error("metric lint completed with errors")
+	}
+
+	for _, problem := range problems {
+		t.Errorf("metric %v has a problem: %v", problem.Metric, problem.Text)
+	}
+
+	metadata := `
+		# HELP sonic_fdb_collector_success Whether FDB collector succeeded
+		# TYPE sonic_fdb_collector_success gauge
+		# HELP sonic_fdb_entries Number of FDB entries
+		# TYPE sonic_fdb_entries gauge
+		# HELP sonic_fdb_entries_unknown_vlan Number of FDB entries with unknown VLAN mapping
+		# TYPE sonic_fdb_entries_unknown_vlan gauge
+	`
+
+	expected := `
+		sonic_fdb_collector_success 1
+		sonic_fdb_entries 4
+		sonic_fdb_entries_unknown_vlan 1
+	`
+
+	if err := testutil.CollectAndCompare(fdbCollector, strings.NewReader(metadata+expected), "sonic_fdb_collector_success", "sonic_fdb_entries", "sonic_fdb_entries_unknown_vlan"); err != nil {
+		t.Errorf("unexpected collecting result:\n%s", err)
+	}
+
+	portMetadata := `
+		# HELP sonic_fdb_entries_by_port Number of FDB entries by port
+		# TYPE sonic_fdb_entries_by_port gauge
+	`
+
+	portExpected := `
+		sonic_fdb_entries_by_port{port="Ethernet0"} 2
+		sonic_fdb_entries_by_port{port="Ethernet39"} 2
+	`
+
+	if err := testutil.CollectAndCompare(fdbCollector, strings.NewReader(portMetadata+portExpected), "sonic_fdb_entries_by_port"); err != nil {
+		t.Errorf("unexpected collecting result:\n%s", err)
+	}
+
+	vlanMetadata := `
+		# HELP sonic_fdb_entries_by_vlan Number of FDB entries by VLAN
+		# TYPE sonic_fdb_entries_by_vlan gauge
+	`
+
+	vlanExpected := `
+		sonic_fdb_entries_by_vlan{vlan="1000"} 2
+		sonic_fdb_entries_by_vlan{vlan="2000"} 1
+		sonic_fdb_entries_by_vlan{vlan="unknown"} 1
+	`
+
+	if err := testutil.CollectAndCompare(fdbCollector, strings.NewReader(vlanMetadata+vlanExpected), "sonic_fdb_entries_by_vlan"); err != nil {
+		t.Errorf("unexpected collecting result:\n%s", err)
+	}
+
+	typeMetadata := `
+		# HELP sonic_fdb_entries_by_type Number of FDB entries by entry type
+		# TYPE sonic_fdb_entries_by_type gauge
+	`
+
+	typeExpected := `
+		sonic_fdb_entries_by_type{entry_type="dynamic"} 3
+		sonic_fdb_entries_by_type{entry_type="static"} 1
+	`
+
+	if err := testutil.CollectAndCompare(fdbCollector, strings.NewReader(typeMetadata+typeExpected), "sonic_fdb_entries_by_type"); err != nil {
+		t.Errorf("unexpected collecting result:\n%s", err)
+	}
+
+	statusMetadata := `
+		# HELP sonic_fdb_entries_skipped Number of FDB entries skipped during latest refresh
+		# TYPE sonic_fdb_entries_skipped gauge
+		# HELP sonic_fdb_entries_truncated Whether FDB collection hit max entries limit (1=yes, 0=no)
+		# TYPE sonic_fdb_entries_truncated gauge
+	`
+
+	statusExpected := `
+		sonic_fdb_entries_skipped 1
+		sonic_fdb_entries_truncated 0
+	`
+
+	if err := testutil.CollectAndCompare(fdbCollector, strings.NewReader(statusMetadata+statusExpected), "sonic_fdb_entries_skipped", "sonic_fdb_entries_truncated"); err != nil {
+		t.Errorf("unexpected collecting result:\n%s", err)
+	}
+}