chore: align slurp config and scaffolding
This commit is contained in:
anthonyrawlins
@@ -15,48 +15,48 @@ import (
|
||||
|
||||
// MonitoringSystem provides comprehensive monitoring for the distributed context system
|
||||
type MonitoringSystem struct {
|
||||
mu sync.RWMutex
|
||||
config *config.Config
|
||||
metrics *MetricsCollector
|
||||
healthChecks *HealthCheckManager
|
||||
alertManager *AlertManager
|
||||
dashboard *DashboardServer
|
||||
logManager *LogManager
|
||||
traceManager *TraceManager
|
||||
|
||||
mu sync.RWMutex
|
||||
config *config.Config
|
||||
metrics *MetricsCollector
|
||||
healthChecks *HealthCheckManager
|
||||
alertManager *AlertManager
|
||||
dashboard *DashboardServer
|
||||
logManager *LogManager
|
||||
traceManager *TraceManager
|
||||
|
||||
// State
|
||||
running bool
|
||||
monitoringPort int
|
||||
updateInterval time.Duration
|
||||
retentionPeriod time.Duration
|
||||
running bool
|
||||
monitoringPort int
|
||||
updateInterval time.Duration
|
||||
retentionPeriod time.Duration
|
||||
}
|
||||
|
||||
// MetricsCollector collects and aggregates system metrics
|
||||
type MetricsCollector struct {
|
||||
mu sync.RWMutex
|
||||
timeSeries map[string]*TimeSeries
|
||||
counters map[string]*Counter
|
||||
gauges map[string]*Gauge
|
||||
histograms map[string]*Histogram
|
||||
customMetrics map[string]*CustomMetric
|
||||
aggregatedStats *AggregatedStatistics
|
||||
exporters []MetricsExporter
|
||||
lastCollection time.Time
|
||||
mu sync.RWMutex
|
||||
timeSeries map[string]*TimeSeries
|
||||
counters map[string]*Counter
|
||||
gauges map[string]*Gauge
|
||||
histograms map[string]*Histogram
|
||||
customMetrics map[string]*CustomMetric
|
||||
aggregatedStats *AggregatedStatistics
|
||||
exporters []MetricsExporter
|
||||
lastCollection time.Time
|
||||
}
|
||||
|
||||
// TimeSeries represents a time-series metric
|
||||
type TimeSeries struct {
|
||||
Name string `json:"name"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
DataPoints []*TimeSeriesPoint `json:"data_points"`
|
||||
Name string `json:"name"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
DataPoints []*TimeSeriesPoint `json:"data_points"`
|
||||
RetentionTTL time.Duration `json:"retention_ttl"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
}
|
||||
|
||||
// TimeSeriesPoint represents a single data point in a time series
|
||||
type TimeSeriesPoint struct {
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Value float64 `json:"value"`
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Value float64 `json:"value"`
|
||||
Labels map[string]string `json:"labels,omitempty"`
|
||||
}
|
||||
|
||||
@@ -64,7 +64,7 @@ type TimeSeriesPoint struct {
|
||||
type Counter struct {
|
||||
Name string `json:"name"`
|
||||
Value int64 `json:"value"`
|
||||
Rate float64 `json:"rate"` // per second
|
||||
Rate float64 `json:"rate"` // per second
|
||||
Labels map[string]string `json:"labels"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
}
|
||||
@@ -82,13 +82,13 @@ type Gauge struct {
|
||||
|
||||
// Histogram represents distribution of values
|
||||
type Histogram struct {
|
||||
Name string `json:"name"`
|
||||
Buckets map[float64]int64 `json:"buckets"`
|
||||
Count int64 `json:"count"`
|
||||
Sum float64 `json:"sum"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
Name string `json:"name"`
|
||||
Buckets map[float64]int64 `json:"buckets"`
|
||||
Count int64 `json:"count"`
|
||||
Sum float64 `json:"sum"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
Percentiles map[float64]float64 `json:"percentiles"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
}
|
||||
|
||||
// CustomMetric represents application-specific metrics
|
||||
@@ -114,81 +114,81 @@ const (
|
||||
|
||||
// AggregatedStatistics provides high-level system statistics
|
||||
type AggregatedStatistics struct {
|
||||
SystemOverview *SystemOverview `json:"system_overview"`
|
||||
PerformanceMetrics *PerformanceOverview `json:"performance_metrics"`
|
||||
HealthMetrics *HealthOverview `json:"health_metrics"`
|
||||
ErrorMetrics *ErrorOverview `json:"error_metrics"`
|
||||
ResourceMetrics *ResourceOverview `json:"resource_metrics"`
|
||||
NetworkMetrics *NetworkOverview `json:"network_metrics"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
SystemOverview *SystemOverview `json:"system_overview"`
|
||||
PerformanceMetrics *PerformanceOverview `json:"performance_metrics"`
|
||||
HealthMetrics *HealthOverview `json:"health_metrics"`
|
||||
ErrorMetrics *ErrorOverview `json:"error_metrics"`
|
||||
ResourceMetrics *ResourceOverview `json:"resource_metrics"`
|
||||
NetworkMetrics *NetworkOverview `json:"network_metrics"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
}
|
||||
|
||||
// SystemOverview provides system-wide overview metrics
|
||||
type SystemOverview struct {
|
||||
TotalNodes int `json:"total_nodes"`
|
||||
HealthyNodes int `json:"healthy_nodes"`
|
||||
TotalContexts int64 `json:"total_contexts"`
|
||||
DistributedContexts int64 `json:"distributed_contexts"`
|
||||
ReplicationFactor float64 `json:"average_replication_factor"`
|
||||
SystemUptime time.Duration `json:"system_uptime"`
|
||||
ClusterVersion string `json:"cluster_version"`
|
||||
LastRestart time.Time `json:"last_restart"`
|
||||
TotalNodes int `json:"total_nodes"`
|
||||
HealthyNodes int `json:"healthy_nodes"`
|
||||
TotalContexts int64 `json:"total_contexts"`
|
||||
DistributedContexts int64 `json:"distributed_contexts"`
|
||||
ReplicationFactor float64 `json:"average_replication_factor"`
|
||||
SystemUptime time.Duration `json:"system_uptime"`
|
||||
ClusterVersion string `json:"cluster_version"`
|
||||
LastRestart time.Time `json:"last_restart"`
|
||||
}
|
||||
|
||||
// PerformanceOverview provides performance metrics
|
||||
type PerformanceOverview struct {
|
||||
RequestsPerSecond float64 `json:"requests_per_second"`
|
||||
AverageResponseTime time.Duration `json:"average_response_time"`
|
||||
P95ResponseTime time.Duration `json:"p95_response_time"`
|
||||
P99ResponseTime time.Duration `json:"p99_response_time"`
|
||||
Throughput float64 `json:"throughput_mbps"`
|
||||
CacheHitRate float64 `json:"cache_hit_rate"`
|
||||
QueueDepth int `json:"queue_depth"`
|
||||
ActiveConnections int `json:"active_connections"`
|
||||
RequestsPerSecond float64 `json:"requests_per_second"`
|
||||
AverageResponseTime time.Duration `json:"average_response_time"`
|
||||
P95ResponseTime time.Duration `json:"p95_response_time"`
|
||||
P99ResponseTime time.Duration `json:"p99_response_time"`
|
||||
Throughput float64 `json:"throughput_mbps"`
|
||||
CacheHitRate float64 `json:"cache_hit_rate"`
|
||||
QueueDepth int `json:"queue_depth"`
|
||||
ActiveConnections int `json:"active_connections"`
|
||||
}
|
||||
|
||||
// HealthOverview provides health-related metrics
|
||||
type HealthOverview struct {
|
||||
OverallHealthScore float64 `json:"overall_health_score"`
|
||||
ComponentHealth map[string]float64 `json:"component_health"`
|
||||
FailedHealthChecks int `json:"failed_health_checks"`
|
||||
LastHealthCheck time.Time `json:"last_health_check"`
|
||||
HealthTrend string `json:"health_trend"` // improving, stable, degrading
|
||||
CriticalAlerts int `json:"critical_alerts"`
|
||||
WarningAlerts int `json:"warning_alerts"`
|
||||
OverallHealthScore float64 `json:"overall_health_score"`
|
||||
ComponentHealth map[string]float64 `json:"component_health"`
|
||||
FailedHealthChecks int `json:"failed_health_checks"`
|
||||
LastHealthCheck time.Time `json:"last_health_check"`
|
||||
HealthTrend string `json:"health_trend"` // improving, stable, degrading
|
||||
CriticalAlerts int `json:"critical_alerts"`
|
||||
WarningAlerts int `json:"warning_alerts"`
|
||||
}
|
||||
|
||||
// ErrorOverview provides error-related metrics
|
||||
type ErrorOverview struct {
|
||||
TotalErrors int64 `json:"total_errors"`
|
||||
ErrorRate float64 `json:"error_rate"`
|
||||
ErrorsByType map[string]int64 `json:"errors_by_type"`
|
||||
ErrorsByComponent map[string]int64 `json:"errors_by_component"`
|
||||
LastError *ErrorEvent `json:"last_error"`
|
||||
ErrorTrend string `json:"error_trend"` // increasing, stable, decreasing
|
||||
TotalErrors int64 `json:"total_errors"`
|
||||
ErrorRate float64 `json:"error_rate"`
|
||||
ErrorsByType map[string]int64 `json:"errors_by_type"`
|
||||
ErrorsByComponent map[string]int64 `json:"errors_by_component"`
|
||||
LastError *ErrorEvent `json:"last_error"`
|
||||
ErrorTrend string `json:"error_trend"` // increasing, stable, decreasing
|
||||
}
|
||||
|
||||
// ResourceOverview provides resource utilization metrics
|
||||
type ResourceOverview struct {
|
||||
CPUUtilization float64 `json:"cpu_utilization"`
|
||||
MemoryUtilization float64 `json:"memory_utilization"`
|
||||
DiskUtilization float64 `json:"disk_utilization"`
|
||||
NetworkUtilization float64 `json:"network_utilization"`
|
||||
StorageUsed int64 `json:"storage_used_bytes"`
|
||||
StorageAvailable int64 `json:"storage_available_bytes"`
|
||||
FileDescriptors int `json:"open_file_descriptors"`
|
||||
Goroutines int `json:"goroutines"`
|
||||
CPUUtilization float64 `json:"cpu_utilization"`
|
||||
MemoryUtilization float64 `json:"memory_utilization"`
|
||||
DiskUtilization float64 `json:"disk_utilization"`
|
||||
NetworkUtilization float64 `json:"network_utilization"`
|
||||
StorageUsed int64 `json:"storage_used_bytes"`
|
||||
StorageAvailable int64 `json:"storage_available_bytes"`
|
||||
FileDescriptors int `json:"open_file_descriptors"`
|
||||
Goroutines int `json:"goroutines"`
|
||||
}
|
||||
|
||||
// NetworkOverview provides network-related metrics
|
||||
type NetworkOverview struct {
|
||||
TotalConnections int `json:"total_connections"`
|
||||
ActiveConnections int `json:"active_connections"`
|
||||
BandwidthUtilization float64 `json:"bandwidth_utilization"`
|
||||
PacketLossRate float64 `json:"packet_loss_rate"`
|
||||
AverageLatency time.Duration `json:"average_latency"`
|
||||
NetworkPartitions int `json:"network_partitions"`
|
||||
DataTransferred int64 `json:"data_transferred_bytes"`
|
||||
TotalConnections int `json:"total_connections"`
|
||||
ActiveConnections int `json:"active_connections"`
|
||||
BandwidthUtilization float64 `json:"bandwidth_utilization"`
|
||||
PacketLossRate float64 `json:"packet_loss_rate"`
|
||||
AverageLatency time.Duration `json:"average_latency"`
|
||||
NetworkPartitions int `json:"network_partitions"`
|
||||
DataTransferred int64 `json:"data_transferred_bytes"`
|
||||
}
|
||||
|
||||
// MetricsExporter exports metrics to external systems
|
||||
@@ -200,49 +200,49 @@ type MetricsExporter interface {
|
||||
|
||||
// HealthCheckManager manages system health checks
|
||||
type HealthCheckManager struct {
|
||||
mu sync.RWMutex
|
||||
healthChecks map[string]*HealthCheck
|
||||
checkResults map[string]*HealthCheckResult
|
||||
schedules map[string]*HealthCheckSchedule
|
||||
running bool
|
||||
mu sync.RWMutex
|
||||
healthChecks map[string]*HealthCheck
|
||||
checkResults map[string]*HealthCheckResult
|
||||
schedules map[string]*HealthCheckSchedule
|
||||
running bool
|
||||
}
|
||||
|
||||
// HealthCheck represents a single health check
|
||||
type HealthCheck struct {
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description"`
|
||||
CheckType HealthCheckType `json:"check_type"`
|
||||
Target string `json:"target"`
|
||||
Timeout time.Duration `json:"timeout"`
|
||||
Interval time.Duration `json:"interval"`
|
||||
Retries int `json:"retries"`
|
||||
Metadata map[string]interface{} `json:"metadata"`
|
||||
Enabled bool `json:"enabled"`
|
||||
CheckFunction func(context.Context) (*HealthCheckResult, error) `json:"-"`
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description"`
|
||||
CheckType HealthCheckType `json:"check_type"`
|
||||
Target string `json:"target"`
|
||||
Timeout time.Duration `json:"timeout"`
|
||||
Interval time.Duration `json:"interval"`
|
||||
Retries int `json:"retries"`
|
||||
Metadata map[string]interface{} `json:"metadata"`
|
||||
Enabled bool `json:"enabled"`
|
||||
CheckFunction func(context.Context) (*HealthCheckResult, error) `json:"-"`
|
||||
}
|
||||
|
||||
// HealthCheckType represents different types of health checks
|
||||
type HealthCheckType string
|
||||
|
||||
const (
|
||||
HealthCheckTypeHTTP HealthCheckType = "http"
|
||||
HealthCheckTypeTCP HealthCheckType = "tcp"
|
||||
HealthCheckTypeCustom HealthCheckType = "custom"
|
||||
HealthCheckTypeComponent HealthCheckType = "component"
|
||||
HealthCheckTypeDatabase HealthCheckType = "database"
|
||||
HealthCheckTypeService HealthCheckType = "service"
|
||||
HealthCheckTypeHTTP HealthCheckType = "http"
|
||||
HealthCheckTypeTCP HealthCheckType = "tcp"
|
||||
HealthCheckTypeCustom HealthCheckType = "custom"
|
||||
HealthCheckTypeComponent HealthCheckType = "component"
|
||||
HealthCheckTypeDatabase HealthCheckType = "database"
|
||||
HealthCheckTypeService HealthCheckType = "service"
|
||||
)
|
||||
|
||||
// HealthCheckResult represents the result of a health check
|
||||
type HealthCheckResult struct {
|
||||
CheckName string `json:"check_name"`
|
||||
Status HealthCheckStatus `json:"status"`
|
||||
ResponseTime time.Duration `json:"response_time"`
|
||||
Message string `json:"message"`
|
||||
Details map[string]interface{} `json:"details"`
|
||||
Error string `json:"error,omitempty"`
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Attempt int `json:"attempt"`
|
||||
CheckName string `json:"check_name"`
|
||||
Status HealthCheckStatus `json:"status"`
|
||||
ResponseTime time.Duration `json:"response_time"`
|
||||
Message string `json:"message"`
|
||||
Details map[string]interface{} `json:"details"`
|
||||
Error string `json:"error,omitempty"`
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Attempt int `json:"attempt"`
|
||||
}
|
||||
|
||||
// HealthCheckStatus represents the status of a health check
|
||||
@@ -258,45 +258,45 @@ const (
|
||||
|
||||
// HealthCheckSchedule defines when health checks should run
|
||||
type HealthCheckSchedule struct {
|
||||
CheckName string `json:"check_name"`
|
||||
Interval time.Duration `json:"interval"`
|
||||
NextRun time.Time `json:"next_run"`
|
||||
LastRun time.Time `json:"last_run"`
|
||||
Enabled bool `json:"enabled"`
|
||||
FailureCount int `json:"failure_count"`
|
||||
CheckName string `json:"check_name"`
|
||||
Interval time.Duration `json:"interval"`
|
||||
NextRun time.Time `json:"next_run"`
|
||||
LastRun time.Time `json:"last_run"`
|
||||
Enabled bool `json:"enabled"`
|
||||
FailureCount int `json:"failure_count"`
|
||||
}
|
||||
|
||||
// AlertManager manages system alerts and notifications
|
||||
type AlertManager struct {
|
||||
mu sync.RWMutex
|
||||
alertRules map[string]*AlertRule
|
||||
activeAlerts map[string]*Alert
|
||||
alertHistory []*Alert
|
||||
notifiers []AlertNotifier
|
||||
silences map[string]*AlertSilence
|
||||
running bool
|
||||
mu sync.RWMutex
|
||||
alertRules map[string]*AlertRule
|
||||
activeAlerts map[string]*Alert
|
||||
alertHistory []*Alert
|
||||
notifiers []AlertNotifier
|
||||
silences map[string]*AlertSilence
|
||||
running bool
|
||||
}
|
||||
|
||||
// AlertRule defines conditions for triggering alerts
|
||||
type AlertRule struct {
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description"`
|
||||
Severity AlertSeverity `json:"severity"`
|
||||
Conditions []*AlertCondition `json:"conditions"`
|
||||
Duration time.Duration `json:"duration"` // How long condition must persist
|
||||
Cooldown time.Duration `json:"cooldown"` // Minimum time between alerts
|
||||
Labels map[string]string `json:"labels"`
|
||||
Annotations map[string]string `json:"annotations"`
|
||||
Enabled bool `json:"enabled"`
|
||||
LastTriggered *time.Time `json:"last_triggered,omitempty"`
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description"`
|
||||
Severity AlertSeverity `json:"severity"`
|
||||
Conditions []*AlertCondition `json:"conditions"`
|
||||
Duration time.Duration `json:"duration"` // How long condition must persist
|
||||
Cooldown time.Duration `json:"cooldown"` // Minimum time between alerts
|
||||
Labels map[string]string `json:"labels"`
|
||||
Annotations map[string]string `json:"annotations"`
|
||||
Enabled bool `json:"enabled"`
|
||||
LastTriggered *time.Time `json:"last_triggered,omitempty"`
|
||||
}
|
||||
|
||||
// AlertCondition defines a single condition for an alert
|
||||
type AlertCondition struct {
|
||||
MetricName string `json:"metric_name"`
|
||||
Operator ConditionOperator `json:"operator"`
|
||||
Threshold float64 `json:"threshold"`
|
||||
Duration time.Duration `json:"duration"`
|
||||
MetricName string `json:"metric_name"`
|
||||
Operator ConditionOperator `json:"operator"`
|
||||
Threshold float64 `json:"threshold"`
|
||||
Duration time.Duration `json:"duration"`
|
||||
}
|
||||
|
||||
// ConditionOperator represents comparison operators for alert conditions
|
||||
@@ -313,39 +313,39 @@ const (
|
||||
|
||||
// Alert represents an active alert
|
||||
type Alert struct {
|
||||
ID string `json:"id"`
|
||||
RuleName string `json:"rule_name"`
|
||||
Severity AlertSeverity `json:"severity"`
|
||||
Status AlertStatus `json:"status"`
|
||||
Message string `json:"message"`
|
||||
Details map[string]interface{} `json:"details"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
Annotations map[string]string `json:"annotations"`
|
||||
StartsAt time.Time `json:"starts_at"`
|
||||
EndsAt *time.Time `json:"ends_at,omitempty"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
AckBy string `json:"acknowledged_by,omitempty"`
|
||||
AckAt *time.Time `json:"acknowledged_at,omitempty"`
|
||||
ID string `json:"id"`
|
||||
RuleName string `json:"rule_name"`
|
||||
Severity AlertSeverity `json:"severity"`
|
||||
Status AlertStatus `json:"status"`
|
||||
Message string `json:"message"`
|
||||
Details map[string]interface{} `json:"details"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
Annotations map[string]string `json:"annotations"`
|
||||
StartsAt time.Time `json:"starts_at"`
|
||||
EndsAt *time.Time `json:"ends_at,omitempty"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
AckBy string `json:"acknowledged_by,omitempty"`
|
||||
AckAt *time.Time `json:"acknowledged_at,omitempty"`
|
||||
}
|
||||
|
||||
// AlertSeverity represents the severity level of an alert
|
||||
type AlertSeverity string
|
||||
|
||||
const (
|
||||
SeverityInfo AlertSeverity = "info"
|
||||
SeverityWarning AlertSeverity = "warning"
|
||||
SeverityError AlertSeverity = "error"
|
||||
SeverityCritical AlertSeverity = "critical"
|
||||
AlertAlertSeverityInfo AlertSeverity = "info"
|
||||
AlertAlertSeverityWarning AlertSeverity = "warning"
|
||||
AlertAlertSeverityError AlertSeverity = "error"
|
||||
AlertAlertSeverityCritical AlertSeverity = "critical"
|
||||
)
|
||||
|
||||
// AlertStatus represents the current status of an alert
|
||||
type AlertStatus string
|
||||
|
||||
const (
|
||||
AlertStatusFiring AlertStatus = "firing"
|
||||
AlertStatusResolved AlertStatus = "resolved"
|
||||
AlertStatusFiring AlertStatus = "firing"
|
||||
AlertStatusResolved AlertStatus = "resolved"
|
||||
AlertStatusAcknowledged AlertStatus = "acknowledged"
|
||||
AlertStatusSilenced AlertStatus = "silenced"
|
||||
AlertStatusSilenced AlertStatus = "silenced"
|
||||
)
|
||||
|
||||
// AlertNotifier sends alert notifications
|
||||
@@ -357,64 +357,64 @@ type AlertNotifier interface {
|
||||
|
||||
// AlertSilence represents a silenced alert
|
||||
type AlertSilence struct {
|
||||
ID string `json:"id"`
|
||||
Matchers map[string]string `json:"matchers"`
|
||||
StartTime time.Time `json:"start_time"`
|
||||
EndTime time.Time `json:"end_time"`
|
||||
CreatedBy string `json:"created_by"`
|
||||
Comment string `json:"comment"`
|
||||
Active bool `json:"active"`
|
||||
ID string `json:"id"`
|
||||
Matchers map[string]string `json:"matchers"`
|
||||
StartTime time.Time `json:"start_time"`
|
||||
EndTime time.Time `json:"end_time"`
|
||||
CreatedBy string `json:"created_by"`
|
||||
Comment string `json:"comment"`
|
||||
Active bool `json:"active"`
|
||||
}
|
||||
|
||||
// DashboardServer provides web-based monitoring dashboard
|
||||
type DashboardServer struct {
|
||||
mu sync.RWMutex
|
||||
server *http.Server
|
||||
dashboards map[string]*Dashboard
|
||||
widgets map[string]*Widget
|
||||
customPages map[string]*CustomPage
|
||||
running bool
|
||||
port int
|
||||
mu sync.RWMutex
|
||||
server *http.Server
|
||||
dashboards map[string]*Dashboard
|
||||
widgets map[string]*Widget
|
||||
customPages map[string]*CustomPage
|
||||
running bool
|
||||
port int
|
||||
}
|
||||
|
||||
// Dashboard represents a monitoring dashboard
|
||||
type Dashboard struct {
|
||||
ID string `json:"id"`
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description"`
|
||||
Widgets []*Widget `json:"widgets"`
|
||||
Layout *DashboardLayout `json:"layout"`
|
||||
ID string `json:"id"`
|
||||
Name string `json:"name"`
|
||||
Description string `json:"description"`
|
||||
Widgets []*Widget `json:"widgets"`
|
||||
Layout *DashboardLayout `json:"layout"`
|
||||
Settings *DashboardSettings `json:"settings"`
|
||||
CreatedBy string `json:"created_by"`
|
||||
CreatedAt time.Time `json:"created_at"`
|
||||
UpdatedAt time.Time `json:"updated_at"`
|
||||
CreatedBy string `json:"created_by"`
|
||||
CreatedAt time.Time `json:"created_at"`
|
||||
UpdatedAt time.Time `json:"updated_at"`
|
||||
}
|
||||
|
||||
// Widget represents a dashboard widget
|
||||
type Widget struct {
|
||||
ID string `json:"id"`
|
||||
Type WidgetType `json:"type"`
|
||||
Title string `json:"title"`
|
||||
DataSource string `json:"data_source"`
|
||||
Query string `json:"query"`
|
||||
Settings map[string]interface{} `json:"settings"`
|
||||
Position *WidgetPosition `json:"position"`
|
||||
RefreshRate time.Duration `json:"refresh_rate"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
ID string `json:"id"`
|
||||
Type WidgetType `json:"type"`
|
||||
Title string `json:"title"`
|
||||
DataSource string `json:"data_source"`
|
||||
Query string `json:"query"`
|
||||
Settings map[string]interface{} `json:"settings"`
|
||||
Position *WidgetPosition `json:"position"`
|
||||
RefreshRate time.Duration `json:"refresh_rate"`
|
||||
LastUpdated time.Time `json:"last_updated"`
|
||||
}
|
||||
|
||||
// WidgetType represents different types of dashboard widgets
|
||||
type WidgetType string
|
||||
|
||||
const (
|
||||
WidgetTypeMetric WidgetType = "metric"
|
||||
WidgetTypeChart WidgetType = "chart"
|
||||
WidgetTypeTable WidgetType = "table"
|
||||
WidgetTypeAlert WidgetType = "alert"
|
||||
WidgetTypeHealth WidgetType = "health"
|
||||
WidgetTypeTopology WidgetType = "topology"
|
||||
WidgetTypeLog WidgetType = "log"
|
||||
WidgetTypeCustom WidgetType = "custom"
|
||||
WidgetTypeMetric WidgetType = "metric"
|
||||
WidgetTypeChart WidgetType = "chart"
|
||||
WidgetTypeTable WidgetType = "table"
|
||||
WidgetTypeAlert WidgetType = "alert"
|
||||
WidgetTypeHealth WidgetType = "health"
|
||||
WidgetTypeTopology WidgetType = "topology"
|
||||
WidgetTypeLog WidgetType = "log"
|
||||
WidgetTypeCustom WidgetType = "custom"
|
||||
)
|
||||
|
||||
// WidgetPosition defines widget position and size
|
||||
@@ -427,11 +427,11 @@ type WidgetPosition struct {
|
||||
|
||||
// DashboardLayout defines dashboard layout settings
|
||||
type DashboardLayout struct {
|
||||
Columns int `json:"columns"`
|
||||
RowHeight int `json:"row_height"`
|
||||
Margins [2]int `json:"margins"` // [x, y]
|
||||
Spacing [2]int `json:"spacing"` // [x, y]
|
||||
Breakpoints map[string]int `json:"breakpoints"`
|
||||
Columns int `json:"columns"`
|
||||
RowHeight int `json:"row_height"`
|
||||
Margins [2]int `json:"margins"` // [x, y]
|
||||
Spacing [2]int `json:"spacing"` // [x, y]
|
||||
Breakpoints map[string]int `json:"breakpoints"`
|
||||
}
|
||||
|
||||
// DashboardSettings contains dashboard configuration
|
||||
@@ -446,43 +446,43 @@ type DashboardSettings struct {
|
||||
|
||||
// CustomPage represents a custom monitoring page
|
||||
type CustomPage struct {
|
||||
Path string `json:"path"`
|
||||
Title string `json:"title"`
|
||||
Content string `json:"content"`
|
||||
ContentType string `json:"content_type"`
|
||||
Handler http.HandlerFunc `json:"-"`
|
||||
Path string `json:"path"`
|
||||
Title string `json:"title"`
|
||||
Content string `json:"content"`
|
||||
ContentType string `json:"content_type"`
|
||||
Handler http.HandlerFunc `json:"-"`
|
||||
}
|
||||
|
||||
// LogManager manages system logs and log analysis
|
||||
type LogManager struct {
|
||||
mu sync.RWMutex
|
||||
logSources map[string]*LogSource
|
||||
logEntries []*LogEntry
|
||||
logAnalyzers []LogAnalyzer
|
||||
mu sync.RWMutex
|
||||
logSources map[string]*LogSource
|
||||
logEntries []*LogEntry
|
||||
logAnalyzers []LogAnalyzer
|
||||
retentionPolicy *LogRetentionPolicy
|
||||
running bool
|
||||
running bool
|
||||
}
|
||||
|
||||
// LogSource represents a source of log data
|
||||
type LogSource struct {
|
||||
Name string `json:"name"`
|
||||
Type LogSourceType `json:"type"`
|
||||
Location string `json:"location"`
|
||||
Format LogFormat `json:"format"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
Enabled bool `json:"enabled"`
|
||||
LastRead time.Time `json:"last_read"`
|
||||
Name string `json:"name"`
|
||||
Type LogSourceType `json:"type"`
|
||||
Location string `json:"location"`
|
||||
Format LogFormat `json:"format"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
Enabled bool `json:"enabled"`
|
||||
LastRead time.Time `json:"last_read"`
|
||||
}
|
||||
|
||||
// LogSourceType represents different types of log sources
|
||||
type LogSourceType string
|
||||
|
||||
const (
|
||||
LogSourceTypeFile LogSourceType = "file"
|
||||
LogSourceTypeHTTP LogSourceType = "http"
|
||||
LogSourceTypeStream LogSourceType = "stream"
|
||||
LogSourceTypeDatabase LogSourceType = "database"
|
||||
LogSourceTypeCustom LogSourceType = "custom"
|
||||
LogSourceTypeFile LogSourceType = "file"
|
||||
LogSourceTypeHTTP LogSourceType = "http"
|
||||
LogSourceTypeStream LogSourceType = "stream"
|
||||
LogSourceTypeDatabase LogSourceType = "database"
|
||||
LogSourceTypeCustom LogSourceType = "custom"
|
||||
)
|
||||
|
||||
// LogFormat represents log entry format
|
||||
@@ -497,14 +497,14 @@ const (
|
||||
|
||||
// LogEntry represents a single log entry
|
||||
type LogEntry struct {
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Level LogLevel `json:"level"`
|
||||
Source string `json:"source"`
|
||||
Message string `json:"message"`
|
||||
Fields map[string]interface{} `json:"fields"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
TraceID string `json:"trace_id,omitempty"`
|
||||
SpanID string `json:"span_id,omitempty"`
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Level LogLevel `json:"level"`
|
||||
Source string `json:"source"`
|
||||
Message string `json:"message"`
|
||||
Fields map[string]interface{} `json:"fields"`
|
||||
Labels map[string]string `json:"labels"`
|
||||
TraceID string `json:"trace_id,omitempty"`
|
||||
SpanID string `json:"span_id,omitempty"`
|
||||
}
|
||||
|
||||
// LogLevel represents log entry severity
|
||||
@@ -527,22 +527,22 @@ type LogAnalyzer interface {
|
||||
|
||||
// LogAnalysisResult represents the result of log analysis
|
||||
type LogAnalysisResult struct {
|
||||
AnalyzerName string `json:"analyzer_name"`
|
||||
Anomalies []*LogAnomaly `json:"anomalies"`
|
||||
Patterns []*LogPattern `json:"patterns"`
|
||||
Statistics *LogStatistics `json:"statistics"`
|
||||
Recommendations []string `json:"recommendations"`
|
||||
AnalyzedAt time.Time `json:"analyzed_at"`
|
||||
AnalyzerName string `json:"analyzer_name"`
|
||||
Anomalies []*LogAnomaly `json:"anomalies"`
|
||||
Patterns []*LogPattern `json:"patterns"`
|
||||
Statistics *LogStatistics `json:"statistics"`
|
||||
Recommendations []string `json:"recommendations"`
|
||||
AnalyzedAt time.Time `json:"analyzed_at"`
|
||||
}
|
||||
|
||||
// LogAnomaly represents detected log anomaly
|
||||
type LogAnomaly struct {
|
||||
Type AnomalyType `json:"type"`
|
||||
Severity AlertSeverity `json:"severity"`
|
||||
Description string `json:"description"`
|
||||
Entries []*LogEntry `json:"entries"`
|
||||
Confidence float64 `json:"confidence"`
|
||||
DetectedAt time.Time `json:"detected_at"`
|
||||
Type AnomalyType `json:"type"`
|
||||
Severity AlertSeverity `json:"severity"`
|
||||
Description string `json:"description"`
|
||||
Entries []*LogEntry `json:"entries"`
|
||||
Confidence float64 `json:"confidence"`
|
||||
DetectedAt time.Time `json:"detected_at"`
|
||||
}
|
||||
|
||||
// AnomalyType represents different types of log anomalies
|
||||
@@ -558,38 +558,38 @@ const (
|
||||
|
||||
// LogPattern represents detected log pattern
|
||||
type LogPattern struct {
|
||||
Pattern string `json:"pattern"`
|
||||
Frequency int `json:"frequency"`
|
||||
LastSeen time.Time `json:"last_seen"`
|
||||
Sources []string `json:"sources"`
|
||||
Confidence float64 `json:"confidence"`
|
||||
Pattern string `json:"pattern"`
|
||||
Frequency int `json:"frequency"`
|
||||
LastSeen time.Time `json:"last_seen"`
|
||||
Sources []string `json:"sources"`
|
||||
Confidence float64 `json:"confidence"`
|
||||
}
|
||||
|
||||
// LogStatistics provides log statistics
|
||||
type LogStatistics struct {
|
||||
TotalEntries int64 `json:"total_entries"`
|
||||
EntriesByLevel map[LogLevel]int64 `json:"entries_by_level"`
|
||||
EntriesBySource map[string]int64 `json:"entries_by_source"`
|
||||
ErrorRate float64 `json:"error_rate"`
|
||||
AverageRate float64 `json:"average_rate"`
|
||||
TimeRange [2]time.Time `json:"time_range"`
|
||||
TotalEntries int64 `json:"total_entries"`
|
||||
EntriesByLevel map[LogLevel]int64 `json:"entries_by_level"`
|
||||
EntriesBySource map[string]int64 `json:"entries_by_source"`
|
||||
ErrorRate float64 `json:"error_rate"`
|
||||
AverageRate float64 `json:"average_rate"`
|
||||
TimeRange [2]time.Time `json:"time_range"`
|
||||
}
|
||||
|
||||
// LogRetentionPolicy defines log retention rules
|
||||
type LogRetentionPolicy struct {
|
||||
RetentionPeriod time.Duration `json:"retention_period"`
|
||||
MaxEntries int64 `json:"max_entries"`
|
||||
CompressionAge time.Duration `json:"compression_age"`
|
||||
ArchiveAge time.Duration `json:"archive_age"`
|
||||
Rules []*RetentionRule `json:"rules"`
|
||||
RetentionPeriod time.Duration `json:"retention_period"`
|
||||
MaxEntries int64 `json:"max_entries"`
|
||||
CompressionAge time.Duration `json:"compression_age"`
|
||||
ArchiveAge time.Duration `json:"archive_age"`
|
||||
Rules []*RetentionRule `json:"rules"`
|
||||
}
|
||||
|
||||
// RetentionRule defines specific retention rules
|
||||
type RetentionRule struct {
|
||||
Name string `json:"name"`
|
||||
Condition string `json:"condition"` // Query expression
|
||||
Retention time.Duration `json:"retention"`
|
||||
Action RetentionAction `json:"action"`
|
||||
Name string `json:"name"`
|
||||
Condition string `json:"condition"` // Query expression
|
||||
Retention time.Duration `json:"retention"`
|
||||
Action RetentionAction `json:"action"`
|
||||
}
|
||||
|
||||
// RetentionAction represents retention actions
|
||||
@@ -603,47 +603,47 @@ const (
|
||||
|
||||
// TraceManager manages distributed tracing
|
||||
type TraceManager struct {
|
||||
mu sync.RWMutex
|
||||
traces map[string]*Trace
|
||||
spans map[string]*Span
|
||||
samplers []TraceSampler
|
||||
exporters []TraceExporter
|
||||
running bool
|
||||
mu sync.RWMutex
|
||||
traces map[string]*Trace
|
||||
spans map[string]*Span
|
||||
samplers []TraceSampler
|
||||
exporters []TraceExporter
|
||||
running bool
|
||||
}
|
||||
|
||||
// Trace represents a distributed trace
|
||||
type Trace struct {
|
||||
TraceID string `json:"trace_id"`
|
||||
Spans []*Span `json:"spans"`
|
||||
Duration time.Duration `json:"duration"`
|
||||
StartTime time.Time `json:"start_time"`
|
||||
EndTime time.Time `json:"end_time"`
|
||||
Status TraceStatus `json:"status"`
|
||||
Tags map[string]string `json:"tags"`
|
||||
Operations []string `json:"operations"`
|
||||
TraceID string `json:"trace_id"`
|
||||
Spans []*Span `json:"spans"`
|
||||
Duration time.Duration `json:"duration"`
|
||||
StartTime time.Time `json:"start_time"`
|
||||
EndTime time.Time `json:"end_time"`
|
||||
Status TraceStatus `json:"status"`
|
||||
Tags map[string]string `json:"tags"`
|
||||
Operations []string `json:"operations"`
|
||||
}
|
||||
|
||||
// Span represents a single span in a trace
|
||||
type Span struct {
|
||||
SpanID string `json:"span_id"`
|
||||
TraceID string `json:"trace_id"`
|
||||
ParentID string `json:"parent_id,omitempty"`
|
||||
Operation string `json:"operation"`
|
||||
Service string `json:"service"`
|
||||
StartTime time.Time `json:"start_time"`
|
||||
EndTime time.Time `json:"end_time"`
|
||||
Duration time.Duration `json:"duration"`
|
||||
Status SpanStatus `json:"status"`
|
||||
Tags map[string]string `json:"tags"`
|
||||
Logs []*SpanLog `json:"logs"`
|
||||
SpanID string `json:"span_id"`
|
||||
TraceID string `json:"trace_id"`
|
||||
ParentID string `json:"parent_id,omitempty"`
|
||||
Operation string `json:"operation"`
|
||||
Service string `json:"service"`
|
||||
StartTime time.Time `json:"start_time"`
|
||||
EndTime time.Time `json:"end_time"`
|
||||
Duration time.Duration `json:"duration"`
|
||||
Status SpanStatus `json:"status"`
|
||||
Tags map[string]string `json:"tags"`
|
||||
Logs []*SpanLog `json:"logs"`
|
||||
}
|
||||
|
||||
// TraceStatus represents the status of a trace
|
||||
type TraceStatus string
|
||||
|
||||
const (
|
||||
TraceStatusOK TraceStatus = "ok"
|
||||
TraceStatusError TraceStatus = "error"
|
||||
TraceStatusOK TraceStatus = "ok"
|
||||
TraceStatusError TraceStatus = "error"
|
||||
TraceStatusTimeout TraceStatus = "timeout"
|
||||
)
|
||||
|
||||
@@ -675,18 +675,18 @@ type TraceExporter interface {
|
||||
|
||||
// ErrorEvent represents a system error event
|
||||
type ErrorEvent struct {
|
||||
ID string `json:"id"`
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Level LogLevel `json:"level"`
|
||||
Component string `json:"component"`
|
||||
Message string `json:"message"`
|
||||
Error string `json:"error"`
|
||||
Context map[string]interface{} `json:"context"`
|
||||
TraceID string `json:"trace_id,omitempty"`
|
||||
SpanID string `json:"span_id,omitempty"`
|
||||
Count int `json:"count"`
|
||||
FirstSeen time.Time `json:"first_seen"`
|
||||
LastSeen time.Time `json:"last_seen"`
|
||||
ID string `json:"id"`
|
||||
Timestamp time.Time `json:"timestamp"`
|
||||
Level LogLevel `json:"level"`
|
||||
Component string `json:"component"`
|
||||
Message string `json:"message"`
|
||||
Error string `json:"error"`
|
||||
Context map[string]interface{} `json:"context"`
|
||||
TraceID string `json:"trace_id,omitempty"`
|
||||
SpanID string `json:"span_id,omitempty"`
|
||||
Count int `json:"count"`
|
||||
FirstSeen time.Time `json:"first_seen"`
|
||||
LastSeen time.Time `json:"last_seen"`
|
||||
}
|
||||
|
||||
// NewMonitoringSystem creates a comprehensive monitoring system
|
||||
@@ -722,7 +722,7 @@ func (ms *MonitoringSystem) initializeComponents() error {
|
||||
aggregatedStats: &AggregatedStatistics{
|
||||
LastUpdated: time.Now(),
|
||||
},
|
||||
exporters: []MetricsExporter{},
|
||||
exporters: []MetricsExporter{},
|
||||
lastCollection: time.Now(),
|
||||
}
|
||||
|
||||
@@ -1134,15 +1134,15 @@ func (ms *MonitoringSystem) createDefaultDashboards() {
|
||||
|
||||
func (ms *MonitoringSystem) severityWeight(severity AlertSeverity) int {
|
||||
switch severity {
|
||||
case SeverityCritical:
|
||||
case AlertSeverityCritical:
|
||||
return 4
|
||||
case SeverityError:
|
||||
case AlertSeverityError:
|
||||
return 3
|
||||
case SeverityWarning:
|
||||
case AlertSeverityWarning:
|
||||
return 2
|
||||
case SeverityInfo:
|
||||
case AlertSeverityInfo:
|
||||
return 1
|
||||
default:
|
||||
return 0
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user