fix: 初验针对修改

This commit is contained in:
zxr
2026-09-09 16:45:01 +08:00
parent 38a86e1fd8
commit b0ffde3100
25 changed files with 1531 additions and 158 deletions

View File

@@ -0,0 +1,502 @@
package otlp
import (
"crypto/subtle"
"encoding/base64"
"encoding/hex"
"encoding/json"
"fmt"
"net/http"
"strconv"
"strings"
"time"
"unicode/utf8"
"git.apinb.com/ops/logs/internal/config"
"git.apinb.com/ops/logs/internal/impl"
"git.apinb.com/ops/logs/internal/models"
"github.com/gin-gonic/gin"
"gorm.io/gorm/clause"
)
const (
maxRecordsPerRequest = 10000
maxRequestBytes = 32 * 1024 * 1024
)
type anyValue struct {
StringValue *string `json:"stringValue"`
BoolValue *bool `json:"boolValue"`
IntValue json.RawMessage `json:"intValue"`
DoubleValue *float64 `json:"doubleValue"`
BytesValue string `json:"bytesValue"`
ArrayValue *arrayValue `json:"arrayValue"`
KVListValue *keyValueList `json:"kvlistValue"`
}
type arrayValue struct {
Values []anyValue `json:"values"`
}
type keyValueList struct {
Values []keyValue `json:"values"`
}
type keyValue struct {
Key string `json:"key"`
Value anyValue `json:"value"`
}
type resource struct {
Attributes []keyValue `json:"attributes"`
}
type traceRequest struct {
ResourceSpans []struct {
Resource resource `json:"resource"`
ScopeSpans []struct {
Spans []spanPayload `json:"spans"`
} `json:"scopeSpans"`
} `json:"resourceSpans"`
}
type spanPayload struct {
TraceID string `json:"traceId"`
SpanID string `json:"spanId"`
ParentSpanID string `json:"parentSpanId"`
Name string `json:"name"`
Kind json.RawMessage `json:"kind"`
StartTimeUnixNano json.RawMessage `json:"startTimeUnixNano"`
EndTimeUnixNano json.RawMessage `json:"endTimeUnixNano"`
Attributes []keyValue `json:"attributes"`
Status struct {
Code json.RawMessage `json:"code"`
Message string `json:"message"`
} `json:"status"`
}
type logRequest struct {
ResourceLogs []struct {
Resource resource `json:"resource"`
ScopeLogs []struct {
LogRecords []logPayload `json:"logRecords"`
} `json:"scopeLogs"`
} `json:"resourceLogs"`
}
type logPayload struct {
TimeUnixNano json.RawMessage `json:"timeUnixNano"`
ObservedTimeUnixNano json.RawMessage `json:"observedTimeUnixNano"`
SeverityText string `json:"severityText"`
SeverityNumber json.RawMessage `json:"severityNumber"`
Body anyValue `json:"body"`
Attributes []keyValue `json:"attributes"`
TraceID string `json:"traceId"`
SpanID string `json:"spanId"`
}
func ReceiveTraces(ctx *gin.Context) {
if !authorize(ctx) {
return
}
ctx.Request.Body = http.MaxBytesReader(ctx.Writer, ctx.Request.Body, maxRequestBytes)
var request traceRequest
if err := ctx.ShouldBindJSON(&request); err != nil {
ctx.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "OTLP Trace 请求无效"})
return
}
rows, rejected, message := decodeSpans(request)
if len(rows) > 0 {
if err := impl.DBService.Clauses(clause.OnConflict{Columns: []clause.Column{{Name: "trace_id"}, {Name: "span_id"}}, DoNothing: true}).
CreateInBatches(rows, 500).Error; err != nil {
ctx.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": "Trace 入库失败"})
return
}
}
ctx.JSON(http.StatusOK, gin.H{"partialSuccess": gin.H{"rejectedSpans": rejected, "errorMessage": message}})
}
func ReceiveLogs(ctx *gin.Context) {
if !authorize(ctx) {
return
}
ctx.Request.Body = http.MaxBytesReader(ctx.Writer, ctx.Request.Body, maxRequestBytes)
var request logRequest
if err := ctx.ShouldBindJSON(&request); err != nil {
ctx.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "OTLP Log 请求无效"})
return
}
rows, rejected, message := decodeLogs(request)
if len(rows) > 0 {
if err := impl.DBService.CreateInBatches(rows, 500).Error; err != nil {
ctx.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": "Log 入库失败"})
return
}
}
ctx.JSON(http.StatusOK, gin.H{"partialSuccess": gin.H{"rejectedLogRecords": rejected, "errorMessage": message}})
}
func authorize(ctx *gin.Context) bool {
provided := strings.TrimSpace(ctx.GetHeader("X-OTLP-Internal-Key"))
expected := config.Spec.OTLP.InternalKey
if provided == "" || len(provided) != len(expected) || subtle.ConstantTimeCompare([]byte(provided), []byte(expected)) != 1 {
ctx.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{"error": "OTLP 接入密钥无效"})
return false
}
return true
}
func decodeSpans(request traceRequest) ([]models.TraceSpan, int, string) {
rows := make([]models.TraceSpan, 0)
rejected := 0
for _, group := range request.ResourceSpans {
resourceAttrs := attributesMap(group.Resource.Attributes)
resourceJSON, _ := json.Marshal(resourceAttrs)
for _, scope := range group.ScopeSpans {
for _, input := range scope.Spans {
if len(rows)+rejected >= maxRecordsPerRequest {
rejected++
continue
}
row, err := spanModel(input, resourceAttrs, string(resourceJSON))
if err != nil {
rejected++
continue
}
rows = append(rows, row)
}
}
}
message := ""
if rejected > 0 {
message = "部分 Span 因标识或时间字段无效被拒绝"
}
return rows, rejected, message
}
func spanModel(input spanPayload, resourceAttrs map[string]interface{}, resourceJSON string) (models.TraceSpan, error) {
traceID, err := normalizeID(input.TraceID, 16)
if err != nil {
return models.TraceSpan{}, err
}
spanID, err := normalizeID(input.SpanID, 8)
if err != nil {
return models.TraceSpan{}, err
}
parentID := ""
if strings.TrimSpace(input.ParentSpanID) != "" {
parentID, err = normalizeID(input.ParentSpanID, 8)
if err != nil {
return models.TraceSpan{}, err
}
}
start, err := nanoTime(input.StartTimeUnixNano)
if err != nil {
return models.TraceSpan{}, err
}
end, err := nanoTime(input.EndTimeUnixNano)
if err != nil || end.Before(start) {
return models.TraceSpan{}, fmt.Errorf("Span 时间无效")
}
attrs := attributesMap(input.Attributes)
attrsJSON, _ := json.Marshal(attrs)
serviceName := textAttribute(resourceAttrs, "service.name")
if serviceName == "" {
serviceName = "unknown_service"
}
serviceName = limitCharacters(serviceName, 255)
operation := strings.TrimSpace(input.Name)
if operation == "" {
operation = "unnamed"
}
operation = limitCharacters(operation, 512)
statusCode := statusText(input.Status.Code)
if httpStatus := intAttribute(attrs, "http.response.status_code", "http.status_code"); statusCode == "unset" && httpStatus >= 500 {
statusCode = "error"
}
now := time.Now().UTC()
return models.TraceSpan{
TraceID: traceID, SpanID: spanID, ParentSpanID: parentID, ServiceName: serviceName,
OperationName: operation, SpanKind: spanKindText(input.Kind), StartTime: start, EndTime: end,
DurationMs: float64(end.Sub(start).Nanoseconds()) / 1e6, StatusCode: statusCode,
StatusMessage: strings.TrimSpace(input.Status.Message), BusinessSystemID: uintAttribute(resourceAttrs, attrs, "business.system.id", "business_system_id"),
ResourceUID: limitCharacters(firstTextAttribute(resourceAttrs, attrs, "resource.uid", "resource_uid"), 255),
HTTPMethod: limitCharacters(firstTextAttribute(attrs, nil, "http.request.method", "http.method"), 16),
HTTPRoute: limitCharacters(firstTextAttribute(attrs, nil, "http.route", "url.path", "http.target"), 1024),
HTTPStatusCode: intAttribute(attrs, "http.response.status_code", "http.status_code"),
Attributes: string(attrsJSON), ResourceAttrs: resourceJSON, IngestedAt: &now,
}, nil
}
func decodeLogs(request logRequest) ([]models.LogEvent, int, string) {
rows := make([]models.LogEvent, 0)
rejected := 0
for _, group := range request.ResourceLogs {
resourceAttrs := attributesMap(group.Resource.Attributes)
for _, scope := range group.ScopeLogs {
for _, input := range scope.LogRecords {
if len(rows)+rejected >= maxRecordsPerRequest {
rejected++
continue
}
row, err := logModel(input, resourceAttrs)
if err != nil {
rejected++
continue
}
rows = append(rows, row)
}
}
}
message := ""
if rejected > 0 {
message = "部分 LogRecord 因标识或内容无效被拒绝"
}
return rows, rejected, message
}
func logModel(input logPayload, resourceAttrs map[string]interface{}) (models.LogEvent, error) {
attrs := attributesMap(input.Attributes)
detailJSON, _ := json.Marshal(attrs)
body := scalarText(anyValueToInterface(input.Body))
if body == "" {
return models.LogEvent{}, fmt.Errorf("日志正文为空")
}
if len(body) > 64*1024 {
body = limitUTF8Bytes(body, 64*1024)
}
traceID := ""
spanID := ""
var err error
if strings.TrimSpace(input.TraceID) != "" {
traceID, err = normalizeID(input.TraceID, 16)
if err != nil {
return models.LogEvent{}, err
}
}
if strings.TrimSpace(input.SpanID) != "" {
spanID, err = normalizeID(input.SpanID, 8)
if err != nil {
return models.LogEvent{}, err
}
}
serviceName := limitCharacters(textAttribute(resourceAttrs, "service.name"), 512)
resourceUID := limitCharacters(firstTextAttribute(resourceAttrs, attrs, "resource.uid", "resource_uid"), 255)
return models.LogEvent{
CreatedAt: timeFromLog(input), SourceKind: "otlp", RawPayload: body, NormalizedSummary: body,
NormalizedDetail: string(detailJSON), DeviceName: serviceName, ResourceType: "service", ResourceUID: resourceUID,
ResourceID: resourceUID, ResourceName: serviceName, BusinessSystemID: uintAttribute(resourceAttrs, attrs, "business.system.id", "business_system_id"),
TraceID: traceID, SpanID: spanID, MatchMethod: "otlp", DispatchStatus: "not_applicable",
SeverityCode: limitCharacters(severityText(input.SeverityText, input.SeverityNumber), 32),
}, nil
}
func limitCharacters(value string, maximum int) string {
characters := []rune(value)
if len(characters) <= maximum {
return value
}
return string(characters[:maximum])
}
func limitUTF8Bytes(value string, maximum int) string {
if len(value) <= maximum {
return value
}
end := maximum
for end > 0 && !utf8.ValidString(value[:end]) {
end--
}
return value[:end]
}
func attributesMap(values []keyValue) map[string]interface{} {
result := make(map[string]interface{}, len(values))
for _, item := range values {
if key := strings.TrimSpace(item.Key); key != "" {
result[key] = anyValueToInterface(item.Value)
}
}
return result
}
func anyValueToInterface(value anyValue) interface{} {
if value.StringValue != nil {
return *value.StringValue
}
if value.BoolValue != nil {
return *value.BoolValue
}
if value.DoubleValue != nil {
return *value.DoubleValue
}
if len(value.IntValue) > 0 {
var number json.Number
if json.Unmarshal(value.IntValue, &number) == nil {
return number
}
var text string
if json.Unmarshal(value.IntValue, &text) == nil {
return text
}
}
if value.BytesValue != "" {
return value.BytesValue
}
if value.ArrayValue != nil {
items := make([]interface{}, 0, len(value.ArrayValue.Values))
for _, item := range value.ArrayValue.Values {
items = append(items, anyValueToInterface(item))
}
return items
}
if value.KVListValue != nil {
return attributesMap(value.KVListValue.Values)
}
return nil
}
func normalizeID(value string, size int) (string, error) {
value = strings.TrimSpace(value)
if raw, err := hex.DecodeString(value); err == nil && len(raw) == size {
return strings.ToLower(value), nil
}
raw, err := base64.StdEncoding.DecodeString(value)
if err != nil || len(raw) != size {
return "", fmt.Errorf("OTLP 标识无效")
}
return hex.EncodeToString(raw), nil
}
func nanoTime(raw json.RawMessage) (time.Time, error) {
value := strings.Trim(strings.TrimSpace(string(raw)), `"`)
nanos, err := strconv.ParseInt(value, 10, 64)
if err != nil || nanos <= 0 {
return time.Time{}, fmt.Errorf("OTLP 时间无效")
}
return time.Unix(0, nanos).UTC(), nil
}
func enumValue(raw json.RawMessage) string {
return strings.Trim(strings.TrimSpace(string(raw)), `"`)
}
func spanKindText(raw json.RawMessage) string {
switch strings.ToUpper(enumValue(raw)) {
case "2", "SPAN_KIND_SERVER":
return "server"
case "3", "SPAN_KIND_CLIENT":
return "client"
case "4", "SPAN_KIND_PRODUCER":
return "producer"
case "5", "SPAN_KIND_CONSUMER":
return "consumer"
default:
return "internal"
}
}
func statusText(raw json.RawMessage) string {
switch strings.ToUpper(enumValue(raw)) {
case "1", "STATUS_CODE_OK":
return "ok"
case "2", "STATUS_CODE_ERROR":
return "error"
default:
return "unset"
}
}
func severityText(text string, raw json.RawMessage) string {
text = strings.ToLower(strings.TrimSpace(text))
if text != "" {
if strings.Contains(text, "fatal") {
return "critical"
}
if strings.Contains(text, "error") {
return "major"
}
if strings.Contains(text, "warn") {
return "warning"
}
return "info"
}
value, _ := strconv.Atoi(enumValue(raw))
if value >= 21 {
return "critical"
}
if value >= 17 {
return "major"
}
if value >= 13 {
return "warning"
}
return "info"
}
func timeFromLog(input logPayload) time.Time {
if value, err := nanoTime(input.TimeUnixNano); err == nil {
return value
}
if value, err := nanoTime(input.ObservedTimeUnixNano); err == nil {
return value
}
return time.Now().UTC()
}
func textAttribute(values map[string]interface{}, key string) string {
return scalarText(values[key])
}
func firstTextAttribute(primary, secondary map[string]interface{}, keys ...string) string {
for _, key := range keys {
if primary != nil {
if value := scalarText(primary[key]); value != "" {
return value
}
}
if secondary != nil {
if value := scalarText(secondary[key]); value != "" {
return value
}
}
}
return ""
}
func intAttribute(values map[string]interface{}, keys ...string) int {
for _, key := range keys {
value, err := strconv.Atoi(scalarText(values[key]))
if err == nil && value != 0 {
return value
}
}
return 0
}
func uintAttribute(primary, secondary map[string]interface{}, keys ...string) *uint {
value := firstTextAttribute(primary, secondary, keys...)
parsed, err := strconv.ParseUint(value, 10, 32)
if err != nil || parsed == 0 {
return nil
}
result := uint(parsed)
return &result
}
func scalarText(value interface{}) string {
switch typed := value.(type) {
case string:
return typed
case json.Number:
return typed.String()
case float64, bool:
return fmt.Sprint(typed)
default:
if typed == nil {
return ""
}
encoded, _ := json.Marshal(typed)
return string(encoded)
}
}

View File

@@ -0,0 +1,192 @@
package otlp
import (
"encoding/json"
"errors"
"sort"
"strconv"
"strings"
"time"
"git.apinb.com/bsm-sdk/core/infra"
"git.apinb.com/ops/logs/internal/impl"
"git.apinb.com/ops/logs/internal/models"
"github.com/gin-gonic/gin"
"gorm.io/gorm"
)
type traceSummary struct {
TraceID string `json:"trace_id"`
StartTime time.Time `json:"start_time"`
EndTime time.Time `json:"end_time"`
DurationMs float64 `json:"duration_ms"`
SpanCount int64 `json:"span_count"`
ErrorCount int64 `json:"error_count"`
Services string `json:"services"`
BusinessSystemID *uint `json:"business_system_id,omitempty"`
ResourceUID string `json:"resource_uid,omitempty"`
}
type dependencyEdge struct {
SourceService string `json:"source_service"`
TargetService string `json:"target_service"`
CallCount int64 `json:"call_count"`
ErrorCount int64 `json:"error_count"`
AverageMs float64 `json:"average_ms"`
}
type dependencyAggregate struct {
Calls int64
Errors int64
DurationMs float64
}
func ListTraces(ctx *gin.Context) {
page, size := pageParams(ctx)
query := applyTraceFilters(impl.DBService.Model(&models.TraceSpan{}), ctx)
var total int64
if err := query.Distinct("trace_id").Count(&total).Error; err != nil {
infra.Response.Error(ctx, err)
return
}
var rows []traceSummary
err := query.Select(`trace_id, MIN(start_time) AS start_time, MAX(end_time) AS end_time,
EXTRACT(EPOCH FROM (MAX(end_time) - MIN(start_time))) * 1000 AS duration_ms, COUNT(*) AS span_count,
COUNT(*) FILTER (WHERE status_code = 'error') AS error_count,
STRING_AGG(DISTINCT service_name, ', ' ORDER BY service_name) AS services,
MAX(business_system_id) AS business_system_id, MAX(resource_uid) AS resource_uid`).
Group("trace_id").Order("start_time DESC").Offset((page - 1) * size).Limit(size).Scan(&rows).Error
if err != nil {
infra.Response.Error(ctx, err)
return
}
infra.Response.Success(ctx, gin.H{"total": total, "page": page, "page_size": size, "items": rows})
}
func GetTrace(ctx *gin.Context) {
traceID := strings.ToLower(strings.TrimSpace(ctx.Param("trace_id")))
if _, err := normalizeID(traceID, 16); err != nil {
infra.Response.Error(ctx, errors.New("trace_id 无效"))
return
}
var rows []models.TraceSpan
if err := impl.DBService.Where("trace_id = ?", traceID).Order("start_time ASC, id ASC").Find(&rows).Error; err != nil {
infra.Response.Error(ctx, err)
return
}
if len(rows) == 0 {
infra.Response.Error(ctx, gorm.ErrRecordNotFound)
return
}
infra.Response.Success(ctx, gin.H{"trace_id": traceID, "spans": rows, "count": len(rows)})
}
func ListDependencies(ctx *gin.Context) {
query := applyTraceFilters(impl.DBService.Model(&models.TraceSpan{}), ctx)
var rows []models.TraceSpan
if err := query.Order("start_time DESC").Limit(100000).Find(&rows).Error; err != nil {
infra.Response.Error(ctx, err)
return
}
spanIndex := make(map[string]models.TraceSpan, len(rows))
for _, row := range rows {
spanIndex[row.TraceID+":"+row.SpanID] = row
}
aggregates := make(map[string]*dependencyAggregate)
for _, row := range rows {
target := ""
if row.ParentSpanID != "" {
if parent, exists := spanIndex[row.TraceID+":"+row.ParentSpanID]; exists && parent.ServiceName != row.ServiceName {
target = row.ServiceName
addDependency(aggregates, parent.ServiceName, target, row)
continue
}
}
if row.SpanKind == "client" {
var attrs map[string]interface{}
_ = json.Unmarshal([]byte(row.Attributes), &attrs)
target = firstTextAttribute(attrs, nil, "peer.service", "server.address", "network.peer.address")
if target != "" && target != row.ServiceName {
addDependency(aggregates, row.ServiceName, target, row)
}
}
}
edges := make([]dependencyEdge, 0, len(aggregates))
for key, value := range aggregates {
parts := strings.SplitN(key, "\x00", 2)
edges = append(edges, dependencyEdge{
SourceService: parts[0], TargetService: parts[1], CallCount: value.Calls,
ErrorCount: value.Errors, AverageMs: value.DurationMs / float64(value.Calls),
})
}
sort.Slice(edges, func(left, right int) bool {
if edges[left].SourceService != edges[right].SourceService {
return edges[left].SourceService < edges[right].SourceService
}
return edges[left].TargetService < edges[right].TargetService
})
infra.Response.Success(ctx, gin.H{"items": edges, "count": len(edges)})
}
func addDependency(aggregates map[string]*dependencyAggregate, source, target string, span models.TraceSpan) {
key := source + "\x00" + target
item := aggregates[key]
if item == nil {
item = &dependencyAggregate{}
aggregates[key] = item
}
item.Calls++
item.DurationMs += span.DurationMs
if span.StatusCode == "error" {
item.Errors++
}
}
func applyTraceFilters(query *gorm.DB, ctx *gin.Context) *gorm.DB {
if value := strings.TrimSpace(ctx.Query("trace_id")); value != "" {
query = query.Where("trace_id = ?", strings.ToLower(value))
}
if value := strings.TrimSpace(ctx.Query("service_name")); value != "" {
scope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("service_name ILIKE ?", "%"+value+"%")
query = query.Where("trace_id IN (?)", scope)
}
if value := strings.TrimSpace(ctx.Query("resource_uid")); value != "" {
scope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("resource_uid = ?", value)
query = query.Where("trace_id IN (?)", scope)
}
if value, err := strconv.ParseUint(ctx.Query("business_system_id"), 10, 32); err == nil && value > 0 {
scope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("business_system_id = ?", uint(value))
query = query.Where("trace_id IN (?)", scope)
}
if value := strings.ToLower(strings.TrimSpace(ctx.Query("status"))); value == "error" {
errorScope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("status_code = ?", "error")
query = query.Where("trace_id IN (?)", errorScope)
} else if value == "ok" {
errorScope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("status_code = ?", "error")
query = query.Where("trace_id NOT IN (?)", errorScope)
} else if value == "unset" {
setScope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("status_code IN ?", []string{"ok", "error"})
query = query.Where("trace_id NOT IN (?)", setScope)
}
start := time.Now().UTC().Add(-24 * time.Hour)
if value, err := time.Parse(time.RFC3339, ctx.Query("start_time")); err == nil {
start = value.UTC()
}
timeScope := impl.DBService.Model(&models.TraceSpan{}).Select("trace_id").Where("start_time >= ?", start)
if value, err := time.Parse(time.RFC3339, ctx.Query("end_time")); err == nil {
timeScope = timeScope.Where("start_time <= ?", value.UTC())
}
return query.Where("trace_id IN (?)", timeScope)
}
func pageParams(ctx *gin.Context) (int, int) {
page, _ := strconv.Atoi(ctx.DefaultQuery("page", "1"))
size, _ := strconv.Atoi(ctx.DefaultQuery("page_size", "50"))
if page < 1 {
page = 1
}
if size < 1 || size > 500 {
size = 50
}
return page, size
}