shishengliang/lingniu-vehicle-ingest
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(platform): expose structured capacity metrics

Browse Source
This commit is contained in:
lingniu
2026-07-04 19:10:04 +08:00
parent 68b260c704
commit cbe7882d51
10 changed files with 297 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
vehicle-data-platform/apps/api/internal/platform/capacity_check.go
View File

@@ -9,17 +9,24 @@ import (
)
type CapacityCheckResult struct {
Status string `json:"status"`
ActiveConnections int64 `json:"active_connections"`
KafkaLag float64 `json:"kafka_lag"`
Findings []string `json:"findings"`
Status string `json:"status"`
CapacityMetrics CapacityMetrics `json:"capacity_metrics"`
Findings []string `json:"findings"`
}
type capacityCheckReport struct {
Status string `json:"status"`
Totals struct {
ActiveConnections int64 `json:"active_connections"`
KafkaLag float64 `json:"kafka_lag"`
ActiveConnections int64 `json:"active_connections"`
KafkaLag float64 `json:"kafka_lag"`
BridgeConsumerPending float64 `json:"bridge_consumer_pending"`
BridgeAckPending float64 `json:"bridge_ack_pending"`
BridgeBatchPendingMessages float64 `json:"bridge_batch_pending_messages"`
FastWriterConsumerPending float64 `json:"fast_writer_consumer_pending"`
FastWriterAckPending float64 `json:"fast_writer_ack_pending"`
FastWriterBatchPending float64 `json:"fast_writer_batch_pending_messages"`
HistoryBatchPending float64 `json:"history_batch_pending_messages"`
HistoryRowsPending float64 `json:"history_rows_pending"`
} `json:"totals"`
Findings []string `json:"findings"`
}
@@ -45,9 +52,19 @@ func (c *CapacityCheckCommand) CheckCapacity(ctx context.Context) (CapacityCheck
return CapacityCheckResult{}, jsonErr
}
return CapacityCheckResult{
Status: report.Status,
ActiveConnections: report.Totals.ActiveConnections,
KafkaLag: report.Totals.KafkaLag,
Findings: report.Findings,
Status: report.Status,
CapacityMetrics: CapacityMetrics{
ActiveConnections: int(report.Totals.ActiveConnections),
KafkaLag: int(report.Totals.KafkaLag),
BridgeConsumerPending: int(report.Totals.BridgeConsumerPending),
BridgeAckPending: int(report.Totals.BridgeAckPending),
BridgeBatchPendingMessages: int(report.Totals.BridgeBatchPendingMessages),
FastWriterConsumerPending: int(report.Totals.FastWriterConsumerPending),
FastWriterAckPending: int(report.Totals.FastWriterAckPending),
FastWriterBatchPending: int(report.Totals.FastWriterBatchPending),
HistoryBatchPending: int(report.Totals.HistoryBatchPending),
HistoryRowsPending: int(report.Totals.HistoryRowsPending),
},
Findings: report.Findings,
}, nil
}

25
vehicle-data-platform/apps/api/internal/platform/mock_store.go
View File

@@ -672,12 +672,27 @@ func bucketStats(values map[string]int) []QualityBucketStat {
func (m *MockStore) OpsHealth(context.Context) (OpsHealth, error) {
summary, _ := m.DashboardSummary(context.Background())
redisOnlineKeys := 92
activeConnections := 120000
capacityMetrics := CapacityMetrics{
ActiveConnections: activeConnections,
KafkaLag: 0,
BridgeConsumerPending: 0,
BridgeAckPending: 0,
BridgeBatchPendingMessages: 0,
FastWriterConsumerPending: 0,
FastWriterAckPending: 0,
FastWriterBatchPending: 0,
HistoryBatchPending: 0,
HistoryRowsPending: 0,
}
return OpsHealth{
LinkHealth: summary.LinkHealth,
KafkaLag: summary.KafkaLag,
RedisOnlineKeys: &redisOnlineKeys,
TDengineWritable: true,
MySQLWritable: true,
LinkHealth: summary.LinkHealth,
KafkaLag: summary.KafkaLag,
ActiveConnections: &activeConnections,
CapacityMetrics: capacityMetrics,
RedisOnlineKeys: &redisOnlineKeys,
TDengineWritable: true,
MySQLWritable: true,
}, nil
}

30
vehicle-data-platform/apps/api/internal/platform/model.go
View File

@@ -359,14 +359,28 @@ type QualityPriorityIssue struct {
}
type OpsHealth struct {
LinkHealth []LinkHealth `json:"linkHealth"`
KafkaLag *int `json:"kafkaLag"`
ActiveConnections *int `json:"activeConnections"`
CapacityFindings []string `json:"capacityFindings"`
RedisOnlineKeys *int `json:"redisOnlineKeys"`
TDengineWritable bool `json:"tdengineWritable"`
MySQLWritable bool `json:"mysqlWritable"`
Runtime RuntimeInfo `json:"runtime"`
LinkHealth []LinkHealth `json:"linkHealth"`
KafkaLag *int `json:"kafkaLag"`
ActiveConnections *int `json:"activeConnections"`
CapacityMetrics CapacityMetrics `json:"capacityMetrics"`
CapacityFindings []string `json:"capacityFindings"`
RedisOnlineKeys *int `json:"redisOnlineKeys"`
TDengineWritable bool `json:"tdengineWritable"`
MySQLWritable bool `json:"mysqlWritable"`
Runtime RuntimeInfo `json:"runtime"`
}
type CapacityMetrics struct {
ActiveConnections int `json:"activeConnections"`
KafkaLag int `json:"kafkaLag"`
BridgeConsumerPending int `json:"bridgeConsumerPending"`
BridgeAckPending int `json:"bridgeAckPending"`
BridgeBatchPendingMessages int `json:"bridgeBatchPendingMessages"`
FastWriterConsumerPending int `json:"fastWriterConsumerPending"`
FastWriterAckPending int `json:"fastWriterAckPending"`
FastWriterBatchPending int `json:"fastWriterBatchPending"`
HistoryBatchPending int `json:"historyBatchPending"`
HistoryRowsPending int `json:"historyRowsPending"`
}
type RuntimeInfo struct {

16
vehicle-data-platform/apps/api/internal/platform/production_store.go
View File

@@ -929,7 +929,7 @@ func (s *ProductionStore) OpsHealth(ctx context.Context) (OpsHealth, error) {
locationHealth := s.mysqlTableReadHealth(ctx, "vehicle_realtime_location", "读取实时位置表正常")
tdengineHealth := s.tdengineRawFrameHealth(ctx)
redisHealth, redisOnlineKeys := s.redisOnlineKeyHealth(ctx)
capacityHealth, kafkaLag, activeConnections, capacityFindings := s.capacityCheckHealth(ctx)
capacityHealth, kafkaLag, activeConnections, capacityMetrics, capacityFindings := s.capacityCheckHealth(ctx)
mysqlWritable := mysqlStatus == "ok" && snapshotHealth.Status == "ok" && locationHealth.Status == "ok"
return OpsHealth{
LinkHealth: []LinkHealth{
@@ -942,6 +942,7 @@ func (s *ProductionStore) OpsHealth(ctx context.Context) (OpsHealth, error) {
},
KafkaLag: kafkaLag,
ActiveConnections: activeConnections,
CapacityMetrics: capacityMetrics,
CapacityFindings: capacityFindings,
RedisOnlineKeys: redisOnlineKeys,
TDengineWritable: tdengineHealth.Status == "ok",
@@ -949,19 +950,20 @@ func (s *ProductionStore) OpsHealth(ctx context.Context) (OpsHealth, error) {
}, nil
}
func (s *ProductionStore) capacityCheckHealth(ctx context.Context) (LinkHealth, *int, *int, []string) {
func (s *ProductionStore) capacityCheckHealth(ctx context.Context) (LinkHealth, *int, *int, CapacityMetrics, []string) {
health := LinkHealth{Name: "Capacity check", Status: "warning", Detail: "平台暂未接入 capacity-check"}
if s.capacityCheck == nil {
return health, nil, nil, []string{}
return health, nil, nil, CapacityMetrics{}, []string{}
}
result, err := s.capacityCheck.CheckCapacity(ctx)
if err != nil {
health.Status = "error"
health.Detail = err.Error()
return health, nil, nil, []string{}
return health, nil, nil, CapacityMetrics{}, []string{}
}
kafkaLag := int(result.KafkaLag)
activeConnections := int(result.ActiveConnections)
capacityMetrics := result.CapacityMetrics
kafkaLag := capacityMetrics.KafkaLag
activeConnections := capacityMetrics.ActiveConnections
detail := "active_connections=" + strconv.Itoa(activeConnections) + ", kafka_lag=" + strconv.Itoa(kafkaLag)
if len(result.Findings) > 0 {
detail += ", findings=" + strings.Join(result.Findings, "; ")
@@ -972,7 +974,7 @@ func (s *ProductionStore) capacityCheckHealth(ctx context.Context) (LinkHealth,
health.Status = "warning"
}
health.Detail = detail
return health, &kafkaLag, &activeConnections, append([]string{}, result.Findings...)
return health, &kafkaLag, &activeConnections, capacityMetrics, append([]string{}, result.Findings...)
}
func (s *ProductionStore) redisOnlineKeyHealth(ctx context.Context) (LinkHealth, *int) {

23
vehicle-data-platform/apps/api/internal/platform/production_store_test.go
View File

@@ -213,10 +213,14 @@ func TestOpsHealthUsesCapacityCheckProbeForKafkaLag(t *testing.T) {
store := &ProductionStore{
db: db,
capacityCheck: fakeCapacityCheckProbe{result: CapacityCheckResult{
Status: "degraded",
ActiveConnections: 120000,
KafkaLag: 42,
Findings: []string{"kafka lag 42"},
Status: "degraded",
CapacityMetrics: CapacityMetrics{
ActiveConnections: 120000,
KafkaLag: 42,
BridgeConsumerPending: 1024,
BridgeAckPending: 2,
},
Findings: []string{"kafka lag 42"},
}},
}
@@ -230,6 +234,9 @@ func TestOpsHealthUsesCapacityCheckProbeForKafkaLag(t *testing.T) {
if health.ActiveConnections == nil || *health.ActiveConnections != 120000 {
t.Fatalf("OpsHealth should expose capacity-check active connections, got %+v", health.ActiveConnections)
}
if health.CapacityMetrics.BridgeConsumerPending != 1024 || health.CapacityMetrics.BridgeAckPending != 2 {
t.Fatalf("OpsHealth should expose structured capacity metrics, got %+v", health.CapacityMetrics)
}
if len(health.CapacityFindings) != 1 || health.CapacityFindings[0] != "kafka lag 42" {
t.Fatalf("OpsHealth should expose structured capacity findings, got %+v", health.CapacityFindings)
}
@@ -254,9 +261,11 @@ func TestOpsHealthReturnsEmptyCapacityFindingsArray(t *testing.T) {
store := &ProductionStore{
db: db,
capacityCheck: fakeCapacityCheckProbe{result: CapacityCheckResult{
Status: "ok",
ActiveConnections: 254,
KafkaLag: 0,
Status: "ok",
CapacityMetrics: CapacityMetrics{
ActiveConnections: 254,
KafkaLag: 0,
},
}},
}