feat(ops): add job center for cross-source job operations
Provide a unified ops console for inspecting, retrying and cancelling jobs across generation, template/kol assist, knowledge parse, desktop publish/task, compliance review and monitoring collect sources. Wires RabbitMQ for retry republish and consolidates the desktop_publish_jobs columns into the base migration. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -0,0 +1,417 @@
|
||||
package app
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"fmt"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"github.com/geo-platform/tenant-api/internal/ops/domain"
|
||||
"github.com/geo-platform/tenant-api/internal/ops/repository"
|
||||
sharedconfig "github.com/geo-platform/tenant-api/internal/shared/config"
|
||||
"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
|
||||
"github.com/geo-platform/tenant-api/internal/shared/response"
|
||||
"github.com/geo-platform/tenant-api/internal/shared/stream"
|
||||
)
|
||||
|
||||
const (
|
||||
ActionJobRetry = "job.retry"
|
||||
ActionJobCancel = "job.cancel"
|
||||
)
|
||||
|
||||
type JobService struct {
|
||||
jobs *repository.JobRepository
|
||||
audits *AuditService
|
||||
mq *rabbitmq.Client
|
||||
}
|
||||
|
||||
func NewJobService(jobs *repository.JobRepository, audits *AuditService, mq *rabbitmq.Client) *JobService {
|
||||
return &JobService{jobs: jobs, audits: audits, mq: mq}
|
||||
}
|
||||
|
||||
func (s *JobService) List(ctx context.Context, filter domain.JobFilter, page, size int) (*domain.JobListResult, error) {
|
||||
if page <= 0 {
|
||||
page = 1
|
||||
}
|
||||
if size <= 0 || size > 500 {
|
||||
size = 50
|
||||
}
|
||||
filter.Limit = size
|
||||
filter.Offset = (page - 1) * size
|
||||
|
||||
items, total, counts, err := s.jobs.List(ctx, filter)
|
||||
if err != nil {
|
||||
return nil, response.ErrInternal(52101, "ops_jobs_query_failed", "failed to list jobs")
|
||||
}
|
||||
return &domain.JobListResult{
|
||||
Items: items,
|
||||
Total: total,
|
||||
Page: page,
|
||||
Size: size,
|
||||
Counts: counts,
|
||||
}, nil
|
||||
}
|
||||
|
||||
func (s *JobService) Get(ctx context.Context, source, id string) (*domain.JobDetail, error) {
|
||||
item, err := s.jobs.Get(ctx, source, id)
|
||||
if err != nil {
|
||||
return nil, mapJobError(err)
|
||||
}
|
||||
return item, nil
|
||||
}
|
||||
|
||||
func (s *JobService) Retry(ctx context.Context, actor *Actor, source, id string) (*domain.JobDetail, error) {
|
||||
before, err := s.jobs.Get(ctx, source, id)
|
||||
if err != nil {
|
||||
return nil, mapJobError(err)
|
||||
}
|
||||
if !before.CanRetry {
|
||||
return nil, response.ErrConflict(40970, "job_retry_unavailable", "当前任务状态不支持重试")
|
||||
}
|
||||
|
||||
item, err := s.jobs.Retry(ctx, source, id)
|
||||
if err != nil {
|
||||
return nil, mapJobError(err)
|
||||
}
|
||||
if err := s.publishRetry(ctx, item); err != nil {
|
||||
_ = s.markRetryPublishFailed(context.Background(), item, err)
|
||||
return nil, response.ErrServiceUnavailable(50370, "job_retry_queue_unavailable", "任务已回滚为失败或取消状态,队列暂不可用")
|
||||
}
|
||||
|
||||
if actor != nil && s.audits != nil {
|
||||
_ = s.audits.Append(ctx, actor.audit(ActionJobRetry, "job", 0, jobAuditMetadata(item)))
|
||||
}
|
||||
return item, nil
|
||||
}
|
||||
|
||||
func (s *JobService) Cancel(ctx context.Context, actor *Actor, source, id, reason string) (*domain.JobDetail, error) {
|
||||
before, err := s.jobs.Get(ctx, source, id)
|
||||
if err != nil {
|
||||
return nil, mapJobError(err)
|
||||
}
|
||||
if !before.CanCancel {
|
||||
return nil, response.ErrConflict(40971, "job_cancel_unavailable", "当前任务状态不支持取消")
|
||||
}
|
||||
|
||||
item, err := s.jobs.Cancel(ctx, source, id, reason)
|
||||
if err != nil {
|
||||
return nil, mapJobError(err)
|
||||
}
|
||||
if actor != nil && s.audits != nil {
|
||||
metadata := jobAuditMetadata(item)
|
||||
if strings.TrimSpace(reason) != "" {
|
||||
metadata["reason"] = reason
|
||||
}
|
||||
_ = s.audits.Append(ctx, actor.audit(ActionJobCancel, "job", 0, metadata))
|
||||
}
|
||||
return item, nil
|
||||
}
|
||||
|
||||
func (s *JobService) publishRetry(ctx context.Context, item *domain.JobDetail) error {
|
||||
if item == nil {
|
||||
return nil
|
||||
}
|
||||
switch item.Source {
|
||||
case "generation":
|
||||
return s.publishGenerationRetry(ctx, item)
|
||||
case "template_assist":
|
||||
return s.publishTemplateAssistRetry(ctx, item)
|
||||
case "kol_assist":
|
||||
return s.publishKolAssistRetry(ctx, item)
|
||||
case "desktop_task":
|
||||
return s.publishDesktopTaskRetry(ctx, item)
|
||||
case "compliance_review":
|
||||
return s.publishComplianceReviewRetry(ctx, item)
|
||||
default:
|
||||
return nil
|
||||
}
|
||||
}
|
||||
|
||||
func (s *JobService) markRetryPublishFailed(ctx context.Context, item *domain.JobDetail, cause error) error {
|
||||
if item == nil || cause == nil {
|
||||
return nil
|
||||
}
|
||||
message := "retry publish failed: " + cause.Error()
|
||||
switch item.Source {
|
||||
case "generation", "template_assist", "kol_assist", "desktop_task", "compliance_review":
|
||||
_, err := s.jobs.Cancel(ctx, item.Source, item.ID, message)
|
||||
return err
|
||||
default:
|
||||
return nil
|
||||
}
|
||||
}
|
||||
|
||||
func (s *JobService) publishGenerationRetry(ctx context.Context, item *domain.JobDetail) error {
|
||||
if s.mq == nil {
|
||||
return fmt.Errorf("rabbitmq is not configured")
|
||||
}
|
||||
if item.NumericID == nil || item.TenantID == nil {
|
||||
return fmt.Errorf("generation task is missing ids")
|
||||
}
|
||||
body, err := json.Marshal(map[string]any{
|
||||
"task_id": *item.NumericID,
|
||||
"task_type": item.Kind,
|
||||
"tenant_id": *item.TenantID,
|
||||
"article_id": int64FromMap(item.Metadata, "article_id"),
|
||||
})
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
return s.mq.PublishGenerationTask(ctx, body)
|
||||
}
|
||||
|
||||
func (s *JobService) publishTemplateAssistRetry(ctx context.Context, item *domain.JobDetail) error {
|
||||
if s.mq == nil {
|
||||
return fmt.Errorf("rabbitmq is not configured")
|
||||
}
|
||||
if item.TenantID == nil {
|
||||
return fmt.Errorf("template assist task is missing tenant")
|
||||
}
|
||||
templateID := int64FromMap(item.Metadata, "template_id")
|
||||
body, err := json.Marshal(templateAssistRetryEnvelope(item, templateID))
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
return s.mq.PublishTemplateAssistTask(ctx, body)
|
||||
}
|
||||
|
||||
func (s *JobService) publishKolAssistRetry(ctx context.Context, item *domain.JobDetail) error {
|
||||
if s.mq == nil {
|
||||
return fmt.Errorf("rabbitmq is not configured")
|
||||
}
|
||||
if item.TenantID == nil {
|
||||
return fmt.Errorf("kol assist task is missing tenant")
|
||||
}
|
||||
body, err := json.Marshal(map[string]any{
|
||||
"task_id": item.ID,
|
||||
"tenant_id": *item.TenantID,
|
||||
})
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
return s.mq.PublishKolAssistTask(ctx, body)
|
||||
}
|
||||
|
||||
func (s *JobService) publishComplianceReviewRetry(ctx context.Context, item *domain.JobDetail) error {
|
||||
if s.mq == nil {
|
||||
return fmt.Errorf("rabbitmq is not configured")
|
||||
}
|
||||
if item.NumericID == nil || item.TenantID == nil {
|
||||
return fmt.Errorf("compliance review job is missing ids")
|
||||
}
|
||||
body, err := json.Marshal(map[string]any{
|
||||
"job_id": *item.NumericID,
|
||||
"message_id": stringFromMap(item.Metadata, "message_id"),
|
||||
"tenant_id": *item.TenantID,
|
||||
"article_id": int64FromMap(item.Metadata, "article_id"),
|
||||
"article_version_id": int64FromMap(item.Metadata, "article_version_id"),
|
||||
"check_record_id": int64FromMap(item.Metadata, "check_record_id"),
|
||||
})
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
return s.mq.PublishComplianceReviewTask(ctx, body)
|
||||
}
|
||||
|
||||
func (s *JobService) publishDesktopTaskRetry(ctx context.Context, item *domain.JobDetail) error {
|
||||
if s.mq == nil {
|
||||
return fmt.Errorf("rabbitmq is not configured")
|
||||
}
|
||||
if item.UUID == nil || item.WorkspaceID == nil {
|
||||
return fmt.Errorf("desktop task is missing dispatch ids")
|
||||
}
|
||||
targetClientID := stringFromMap(item.Metadata, "target_client_id")
|
||||
if targetClientID == "" {
|
||||
return fmt.Errorf("desktop task is missing target client")
|
||||
}
|
||||
body, err := json.Marshal(map[string]any{
|
||||
"type": "task_available",
|
||||
"task_id": *item.UUID,
|
||||
"job_id": stringFromMap(item.Metadata, "job_id"),
|
||||
"workspace_id": *item.WorkspaceID,
|
||||
"target_account_id": stringFromMap(item.Metadata, "target_account_id"),
|
||||
"target_client_id": targetClientID,
|
||||
"platform": stringFromMap(item.Metadata, "platform_id"),
|
||||
"status": item.Status,
|
||||
"kind": item.Kind,
|
||||
"priority": intFromMap(item.Metadata, "priority"),
|
||||
"lane": stringFromMap(item.Metadata, "lane"),
|
||||
"interrupt_generation": intFromMap(item.Metadata, "interrupt_generation"),
|
||||
"updated_at": item.UpdatedAt,
|
||||
})
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if err := s.mq.PublishDesktopTaskEvent(ctx, body); err != nil {
|
||||
return err
|
||||
}
|
||||
routingKey := desktopDispatchRoutingKey(s.mq.Config(), item.Kind, stringFromMap(item.Metadata, "lane"), targetClientID)
|
||||
dispatchBody, err := json.Marshal(stream.DesktopDispatchEvent{
|
||||
Type: "task_available",
|
||||
TaskID: *item.UUID,
|
||||
JobID: stringFromMap(item.Metadata, "job_id"),
|
||||
WorkspaceID: *item.WorkspaceID,
|
||||
TargetAccountID: stringFromMap(item.Metadata, "target_account_id"),
|
||||
TargetClientID: targetClientID,
|
||||
Platform: stringFromMap(item.Metadata, "platform_id"),
|
||||
Status: item.Status,
|
||||
Kind: item.Kind,
|
||||
Priority: intFromMap(item.Metadata, "priority"),
|
||||
Lane: stringFromMap(item.Metadata, "lane"),
|
||||
InterruptGeneration: intFromMap(item.Metadata, "interrupt_generation"),
|
||||
UpdatedAt: time.Now().UTC(),
|
||||
})
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
return s.mq.PublishDesktopDispatch(ctx, routingKey, dispatchBody)
|
||||
}
|
||||
|
||||
func templateAssistRetryEnvelope(item *domain.JobDetail, templateID int64) map[string]any {
|
||||
templateName := ""
|
||||
if item.Title != nil {
|
||||
templateName = *item.Title
|
||||
}
|
||||
envelope := map[string]any{
|
||||
"TaskID": item.ID,
|
||||
"TaskType": item.Kind,
|
||||
"TenantID": int64Value(item.TenantID),
|
||||
"TemplateID": templateID,
|
||||
"TemplateKey": stringFromMap(item.Metadata, "template_key"),
|
||||
"TemplateName": firstNonEmptyString(stringFromMap(item.Metadata, "template_name"), templateName),
|
||||
}
|
||||
switch item.Kind {
|
||||
case "analyze":
|
||||
envelope["AnalyzePrompt"] = optionalStringFromMap(item.Metadata, "analyze_prompt")
|
||||
envelope["AnalyzeRequest"] = item.Payload
|
||||
case "title":
|
||||
envelope["TitlePrompt"] = optionalStringFromMap(item.Metadata, "title_prompt")
|
||||
envelope["TitleRequest"] = item.Payload
|
||||
case "outline":
|
||||
envelope["OutlinePrompt"] = optionalStringFromMap(item.Metadata, "outline_prompt")
|
||||
envelope["OutlineRequest"] = item.Payload
|
||||
default:
|
||||
envelope["AnalyzeRequest"] = item.Payload
|
||||
}
|
||||
return envelope
|
||||
}
|
||||
|
||||
func mapJobError(err error) error {
|
||||
if errors.Is(err, repository.ErrJobNotFound) {
|
||||
return response.ErrNotFound(40470, "job_not_found", "任务不存在")
|
||||
}
|
||||
return err
|
||||
}
|
||||
|
||||
func jobAuditMetadata(item *domain.JobDetail) map[string]any {
|
||||
metadata := map[string]any{
|
||||
"job_uid": item.UID,
|
||||
"source": item.Source,
|
||||
"id": item.ID,
|
||||
"kind": item.Kind,
|
||||
"status": item.Status,
|
||||
"phase": item.Phase,
|
||||
}
|
||||
if item.TenantID != nil {
|
||||
metadata["tenant_id"] = *item.TenantID
|
||||
}
|
||||
if item.WorkspaceID != nil {
|
||||
metadata["workspace_id"] = *item.WorkspaceID
|
||||
}
|
||||
return metadata
|
||||
}
|
||||
|
||||
func int64Value(value *int64) int64 {
|
||||
if value == nil {
|
||||
return 0
|
||||
}
|
||||
return *value
|
||||
}
|
||||
|
||||
func int64FromMap(values map[string]any, key string) int64 {
|
||||
if values == nil {
|
||||
return 0
|
||||
}
|
||||
switch value := values[key].(type) {
|
||||
case int64:
|
||||
return value
|
||||
case int:
|
||||
return int64(value)
|
||||
case float64:
|
||||
return int64(value)
|
||||
case json.Number:
|
||||
n, _ := value.Int64()
|
||||
return n
|
||||
case string:
|
||||
n, _ := strconvParseInt64(value)
|
||||
return n
|
||||
default:
|
||||
return 0
|
||||
}
|
||||
}
|
||||
|
||||
func intFromMap(values map[string]any, key string) int {
|
||||
return int(int64FromMap(values, key))
|
||||
}
|
||||
|
||||
func stringFromMap(values map[string]any, key string) string {
|
||||
if values == nil {
|
||||
return ""
|
||||
}
|
||||
switch value := values[key].(type) {
|
||||
case string:
|
||||
return strings.TrimSpace(value)
|
||||
default:
|
||||
return fmt.Sprint(value)
|
||||
}
|
||||
}
|
||||
|
||||
func optionalStringFromMap(values map[string]any, key string) *string {
|
||||
value := stringFromMap(values, key)
|
||||
if value == "" {
|
||||
return nil
|
||||
}
|
||||
return &value
|
||||
}
|
||||
|
||||
func firstNonEmptyString(values ...string) string {
|
||||
for _, value := range values {
|
||||
value = strings.TrimSpace(value)
|
||||
if value != "" {
|
||||
return value
|
||||
}
|
||||
}
|
||||
return ""
|
||||
}
|
||||
|
||||
func strconvParseInt64(value string) (int64, error) {
|
||||
return strconv.ParseInt(strings.TrimSpace(value), 10, 64)
|
||||
}
|
||||
|
||||
func desktopDispatchRoutingKey(cfg sharedconfig.RabbitMQConfig, kind, lane, targetClientID string) string {
|
||||
prefix := strings.TrimSpace(cfg.DesktopDispatchPublishPrefix)
|
||||
if strings.TrimSpace(kind) == "monitor" {
|
||||
prefix = strings.TrimSpace(cfg.DesktopDispatchMonitorPrefix)
|
||||
}
|
||||
if prefix == "" {
|
||||
if strings.TrimSpace(kind) == "monitor" {
|
||||
prefix = "monitor"
|
||||
} else {
|
||||
prefix = "publish"
|
||||
}
|
||||
}
|
||||
if strings.TrimSpace(kind) == "monitor" {
|
||||
laneSegment := "normal"
|
||||
switch strings.TrimSpace(lane) {
|
||||
case "high":
|
||||
laneSegment = "high"
|
||||
case "retry":
|
||||
laneSegment = "retry"
|
||||
}
|
||||
return fmt.Sprintf("%s.%s.%s", prefix, laneSegment, targetClientID)
|
||||
}
|
||||
return fmt.Sprintf("%s.%s", prefix, targetClientID)
|
||||
}
|
||||
Reference in New Issue
Block a user