Files
yovision/Sense/server/app/sense/edge_node/service.go
T

168 lines
6.4 KiB
Go

package edge_node
import (
"context"
"errors"
"fmt"
"strings"
"time"
coreService "github.com/go-admin-team/go-admin-core/sdk/service"
"gorm.io/gorm"
"gorm.io/gorm/clause"
)
var (
ErrInvalidQuery = errors.New("边缘节点查询条件不符合要求")
ErrNotFound = errors.New("边缘节点不存在")
ErrInvalidSample = errors.New("边缘节点心跳样本不符合要求")
)
type Service struct {
coreService.Service
Now func() time.Time
}
func NewService(db *gorm.DB) *Service {
return &Service{Service: coreService.Service{Orm: db}, Now: time.Now}
}
func (s *Service) List(request PageRequest) (PageResponse, error) {
if request.GetPageSize() > 100 {
return PageResponse{}, ErrInvalidQuery
}
var count int64
if err := s.Orm.Model(&Node{}).Count(&count).Error; err != nil {
return PageResponse{}, fmt.Errorf("count edge nodes: %w", err)
}
var models []Node
if err := s.Orm.Order("name ASC, id ASC").Limit(request.GetPageSize()).Offset((request.GetPageIndex() - 1) * request.GetPageSize()).Find(&models).Error; err != nil {
return PageResponse{}, fmt.Errorf("list edge nodes: %w", err)
}
result := PageResponse{List: make([]Response, 0, len(models)), Count: count}
result.Summary.HeartbeatTimeout = int64(HeartbeatTimeout / time.Second)
var all []Node
if err := s.Orm.Find(&all).Error; err != nil {
return PageResponse{}, fmt.Errorf("summarize edge nodes: %w", err)
}
result.Summary.Total = len(all)
for _, item := range all {
status := s.status(item)
switch status {
case StatusOffline:
result.Summary.Offline++
case StatusRecovering:
result.Summary.Recovering++
default:
result.Summary.Online++
}
result.Summary.LoadUsed += item.LoadUsed
result.Summary.LoadCapacity += item.LoadCapacity
result.Summary.BackfillQueueDepth += item.BackfillQueueDepth
}
for _, item := range models {
result.List = append(result.List, s.response(item, nil))
}
return result, nil
}
func (s *Service) Get(id string) (Response, error) {
id = strings.TrimSpace(id)
if id == "" || len(id) > 64 {
return Response{}, ErrInvalidQuery
}
var model Node
if err := s.Orm.First(&model, "id = ?", id).Error; err != nil {
if errors.Is(err, gorm.ErrRecordNotFound) {
return Response{}, ErrNotFound
}
return Response{}, fmt.Errorf("get edge node: %w", err)
}
var events []Event
if err := s.Orm.Where("node_id = ?", id).Order("occurred_at DESC, id DESC").Limit(20).Find(&events).Error; err != nil {
return Response{}, fmt.Errorf("list edge node events: %w", err)
}
return s.response(model, events), nil
}
// ApplyHeartbeat updates the Sense-owned projection atomically and records
// only state transitions. Callers are internal adapters, not public APIs.
func (s *Service) ApplyHeartbeat(ctx context.Context, sample HeartbeatSample) error {
now := sample.CollectedAt.UTC()
if now.IsZero() {
now = s.now()
}
sample.ID = strings.TrimSpace(sample.ID)
if sample.ID == "" || len(sample.ID) > 64 || strings.TrimSpace(sample.Name) == "" || sample.LastHeartbeatAt.IsZero() || sample.StartedAt.IsZero() || sample.LoadUsed < 0 || sample.LoadCapacity < 0 || sample.LoadUsed > sample.LoadCapacity || sample.BackfillQueueDepth < 0 {
return ErrInvalidSample
}
sample.LastHeartbeatAt = sample.LastHeartbeatAt.UTC()
sample.LastCollectedAt = now
return s.Orm.WithContext(ctx).Transaction(func(tx *gorm.DB) error {
var previous Node
err := tx.Clauses(clause.Locking{Strength: "UPDATE"}).First(&previous, "id = ?", sample.ID).Error
if err != nil && !errors.Is(err, gorm.ErrRecordNotFound) {
return fmt.Errorf("lock edge node projection: %w", err)
}
from := ""
if err == nil {
from = statusAt(previous, now)
sample.CreatedAt = previous.CreatedAt
sample.ProjectionVersion = previous.ProjectionVersion + 1
} else {
sample.ProjectionVersion = 1
}
if err = tx.Save(&sample.Node).Error; err != nil {
return fmt.Errorf("save edge node projection: %w", err)
}
to := statusAt(sample.Node, now)
if from == to && from != "" {
return nil
}
if from == StatusOffline {
timeoutAt := previous.LastHeartbeatAt.UTC().Add(HeartbeatTimeout)
if err = tx.Create(&Event{NodeID: sample.ID, EventType: "heartbeat_timeout", FromStatus: StatusOnline, ToStatus: StatusOffline, Detail: "节点心跳超过 90 秒未更新", OccurredAt: timeoutAt}).Error; err != nil {
return err
}
}
eventType, detail := "heartbeat_received", "收到节点首次心跳"
if from == StatusOffline {
eventType, detail = "node_recovered", "节点心跳恢复,通道状态继续独立收敛"
} else if to == StatusRecovering {
eventType, detail = "recovery_started", "节点进入恢复阶段"
}
return tx.Create(&Event{NodeID: sample.ID, EventType: eventType, FromStatus: from, ToStatus: to, Detail: detail, OccurredAt: now}).Error
})
}
func (s *Service) response(model Node, events []Event) Response {
now := s.now()
freshness := int64(now.Sub(model.LastHeartbeatAt.UTC()).Seconds())
if freshness < 0 {
freshness = 0
}
uptime := int64(model.LastHeartbeatAt.UTC().Sub(model.StartedAt.UTC()).Seconds())
if uptime < 0 {
uptime = 0
}
return Response{ID: model.ID, Name: model.Name, Location: model.Location, Status: statusAt(model, now), RuntimeVersion: model.RuntimeVersion, ModelVersion: model.ModelVersion, StartedAt: model.StartedAt, UptimeSeconds: uptime, LastHeartbeatAt: model.LastHeartbeatAt, LastCollectedAt: model.LastCollectedAt, FreshnessSeconds: freshness, Stale: freshness > int64(HeartbeatTimeout/time.Second), LoadUsed: model.LoadUsed, LoadCapacity: model.LoadCapacity, ControlTunnelStatus: model.ControlTunnelStatus, ControlTunnelDetail: model.ControlTunnelDetail, ControlLatencyMs: model.ControlLatencyMs, VideoPlaneStatus: model.VideoPlaneStatus, VideoPlaneDetail: model.VideoPlaneDetail, ActiveStreamCount: model.ActiveStreamCount, BackfillStatus: model.BackfillStatus, BackfillQueueDepth: model.BackfillQueueDepth, RecoveryPhase: model.RecoveryPhase, ProjectionVersion: model.ProjectionVersion, Events: events}
}
func (s *Service) status(model Node) string { return statusAt(model, s.now()) }
func (s *Service) now() time.Time {
if s.Now != nil {
return s.Now().UTC()
}
return time.Now().UTC()
}
func statusAt(model Node, now time.Time) string {
if now.UTC().Sub(model.LastHeartbeatAt.UTC()) > HeartbeatTimeout {
return StatusOffline
}
if strings.TrimSpace(model.RecoveryPhase) != "" && model.RecoveryPhase != "converged" {
return StatusRecovering
}
return StatusOnline
}