feat: 添加节点可观测性查询功能,支持获取节点的系统配置、指标快照、流量报告和健康事件

This commit is contained in:
ryan
2026-03-14 11:07:32 +08:00
parent d9d02e749a
commit 1f294d72b9
7 changed files with 307 additions and 0 deletions
+84
View File
@@ -0,0 +1,84 @@
package service
import (
"atsflare/model"
"errors"
"time"
"gorm.io/gorm"
)
const (
defaultObservabilityWindow = 24 * time.Hour
defaultObservabilityLimit = 120
maxObservabilityLimit = 500
)
type NodeObservabilityQuery struct {
Hours int `json:"hours"`
Limit int `json:"limit"`
}
type NodeObservabilityView struct {
NodeID string `json:"node_id"`
Profile *model.NodeSystemProfile `json:"profile"`
MetricSnapshots []*model.NodeMetricSnapshot `json:"metric_snapshots"`
TrafficReports []*model.NodeRequestReport `json:"traffic_reports"`
HealthEvents []*model.NodeHealthEvent `json:"health_events"`
}
func GetNodeObservability(id uint, query NodeObservabilityQuery) (*NodeObservabilityView, error) {
node, err := model.GetNodeByID(id)
if err != nil {
return nil, err
}
limit := normalizeObservabilityLimit(query.Limit)
since := time.Now().Add(-normalizeObservabilityWindow(query.Hours))
profile, err := model.GetNodeSystemProfile(node.NodeID)
if err != nil && !errors.Is(err, gorm.ErrRecordNotFound) {
return nil, err
}
if errors.Is(err, gorm.ErrRecordNotFound) {
profile = nil
}
snapshots, err := model.ListNodeMetricSnapshots(node.NodeID, since, limit)
if err != nil {
return nil, err
}
reports, err := model.ListNodeRequestReports(node.NodeID, since, limit)
if err != nil {
return nil, err
}
events, err := model.ListNodeHealthEvents(node.NodeID, false, limit)
if err != nil {
return nil, err
}
return &NodeObservabilityView{
NodeID: node.NodeID,
Profile: profile,
MetricSnapshots: snapshots,
TrafficReports: reports,
HealthEvents: events,
}, nil
}
func normalizeObservabilityLimit(limit int) int {
if limit <= 0 {
return defaultObservabilityLimit
}
if limit > maxObservabilityLimit {
return maxObservabilityLimit
}
return limit
}
func normalizeObservabilityWindow(hours int) time.Duration {
if hours <= 0 {
return defaultObservabilityWindow
}
return time.Duration(hours) * time.Hour
}
+97
View File
@@ -493,3 +493,100 @@ func TestHeartbeatNodeResolvesMissingHealthEvents(t *testing.T) {
t.Fatalf("expected resolved health event record, got %+v", allEvents)
}
}
func TestGetNodeObservability(t *testing.T) {
setupServiceTestDB(t)
node := &model.Node{
NodeID: "node-observability-query",
Name: "query-edge",
IP: "10.0.0.61",
AgentToken: "token-query",
AgentVersion: "v0.6.0",
NginxVersion: "1.27.1.2",
Status: NodeStatusOnline,
}
if err := node.Insert(); err != nil {
t.Fatalf("failed to insert node: %v", err)
}
if err := model.UpsertNodeSystemProfile(&model.NodeSystemProfile{
NodeID: node.NodeID,
Hostname: "query-edge",
OSName: "Ubuntu",
Architecture: "amd64",
ReportedAt: time.Now(),
}); err != nil {
t.Fatalf("failed to insert node system profile: %v", err)
}
if err := (&model.NodeMetricSnapshot{
NodeID: node.NodeID,
CapturedAt: time.Now(),
}).Insert(); err != nil {
t.Fatalf("failed to insert node metric snapshot: %v", err)
}
if err := (&model.NodeRequestReport{
NodeID: node.NodeID,
WindowStartedAt: time.Now().Add(-time.Minute),
WindowEndedAt: time.Now(),
RequestCount: 123,
}).Insert(); err != nil {
t.Fatalf("failed to insert node request report: %v", err)
}
if err := model.DB.Create(&model.NodeHealthEvent{
NodeID: node.NodeID,
EventType: "sync_error",
Severity: NodeHealthSeverityWarning,
Status: NodeHealthEventStatusActive,
Message: "checksum mismatch",
FirstTriggeredAt: time.Now().Add(-time.Minute),
LastTriggeredAt: time.Now(),
ReportedAt: time.Now(),
}).Error; err != nil {
t.Fatalf("failed to insert node health event: %v", err)
}
view, err := GetNodeObservability(node.ID, NodeObservabilityQuery{Hours: 24, Limit: 10})
if err != nil {
t.Fatalf("GetNodeObservability failed: %v", err)
}
if view.NodeID != node.NodeID {
t.Fatalf("unexpected node id: %s", view.NodeID)
}
if view.Profile == nil || view.Profile.OSName != "Ubuntu" {
t.Fatalf("unexpected profile: %+v", view.Profile)
}
if len(view.MetricSnapshots) != 1 {
t.Fatalf("expected 1 metric snapshot, got %d", len(view.MetricSnapshots))
}
if len(view.TrafficReports) != 1 || view.TrafficReports[0].RequestCount != 123 {
t.Fatalf("unexpected traffic reports: %+v", view.TrafficReports)
}
if len(view.HealthEvents) != 1 || view.HealthEvents[0].EventType != "sync_error" {
t.Fatalf("unexpected health events: %+v", view.HealthEvents)
}
}
func TestGetNodeObservabilityAllowsMissingProfile(t *testing.T) {
setupServiceTestDB(t)
node := &model.Node{
NodeID: "node-observability-empty",
Name: "empty-edge",
IP: "10.0.0.62",
AgentToken: "token-empty",
AgentVersion: "v0.6.0",
NginxVersion: "1.27.1.2",
Status: NodeStatusOnline,
}
if err := node.Insert(); err != nil {
t.Fatalf("failed to insert node: %v", err)
}
view, err := GetNodeObservability(node.ID, NodeObservabilityQuery{})
if err != nil {
t.Fatalf("GetNodeObservability failed: %v", err)
}
if view.Profile != nil {
t.Fatalf("expected nil profile when profile not reported, got %+v", view.Profile)
}
}