chore: checkpoint 第一版

2026-09-06 17:04:11 +08:00
parent a8cdec98d1
commit 0c2151ace9
659 changed files with 185855 additions and 554 deletions
@@ -214,13 +214,16 @@ func buildPilotDailySnapshot(ctx context.Context, tx pgx.Tx, tenantPublicID, ten
 	rows, err = tx.Query(ctx, `WITH providers AS (
 		SELECT provider FROM provider_dependency_state WHERE tenant_id=$1::uuid
 		UNION SELECT pe.provider FROM provider_event pe WHERE pe.tenant_id=$1::uuid AND pe.received_at >= $2 AND pe.received_at < $3
-		UNION SELECT rt.provider FROM conversation c JOIN rtc_task rt ON rt.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.created_at >= $2 AND c.created_at < $3)
+		UNION SELECT rt.provider FROM conversation c JOIN rtc_task rt ON rt.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.created_at >= $2 AND c.created_at < $3
+		UNION SELECT hs.provider FROM conversation c JOIN hardware_agent_session hs ON hs.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.created_at >= $2 AND c.created_at < $3)
 		SELECT p.provider,COALESCE(pds.status,'UNKNOWN'),COALESCE(pds.consecutive_failures,0),
-		(SELECT count(*) FROM provider_event pe WHERE pe.tenant_id=$1::uuid AND pe.provider=p.provider AND pe.processing_status='failed' AND pe.received_at >= $2 AND pe.received_at < $3),
+		(SELECT count(*) FROM provider_event pe WHERE pe.tenant_id=$1::uuid AND pe.provider=p.provider AND pe.processing_status='failed' AND pe.received_at >= $2 AND pe.received_at < $3)+
+		(SELECT count(*) FROM hardware_agent_event he JOIN hardware_agent_session hs ON hs.conversation_id=he.conversation_id WHERE he.tenant_id=$1::uuid AND hs.provider=p.provider AND he.event_type='session_failed' AND he.occurred_at >= $2 AND he.occurred_at < $3),
 		(SELECT count(DISTINCT c.id) FROM conversation c
-		 JOIN rtc_task rt ON rt.conversation_id=c.id
+		 LEFT JOIN rtc_task rt ON rt.conversation_id=c.id
+		 LEFT JOIN hardware_agent_session hs ON hs.conversation_id=c.id
 		 JOIN conversation_state_history csh ON csh.conversation_id=c.id
-		 WHERE c.tenant_id=$1::uuid AND rt.provider=p.provider AND csh.to_status='FAILED' AND csh.occurred_at >= $2 AND csh.occurred_at < $3),
+		 WHERE c.tenant_id=$1::uuid AND COALESCE(rt.provider,hs.provider)=p.provider AND csh.to_status='FAILED' AND csh.occurred_at >= $2 AND csh.occurred_at < $3),
 		(SELECT count(*) FROM provider_event pe WHERE pe.tenant_id=$1::uuid AND pe.provider=p.provider AND pe.processing_status IN ('pending','failed'))
 		FROM providers p LEFT JOIN provider_dependency_state pds ON pds.tenant_id=$1::uuid AND pds.provider=p.provider ORDER BY p.provider`, tenantInternalID, windowStart, windowEnd)
 	if err != nil {
@@ -191,7 +191,7 @@ func (s *Service) Search(ctx context.Context, principal identity.Principal, tena
 		"device_id":         "d.public_id=$2::uuid",
 		"conversation_id":   "c.public_id=$2::uuid",
 		"room_id":           "rt.room_id=$2",
-		"provider_task_ref": "rt.provider_task_ref=$2",
+		"provider_task_ref": "COALESCE(rt.provider_task_ref,hs.provider_session_ref)=$2",
 		"request_id":        "EXISTS(SELECT 1 FROM audit_event ae WHERE ae.tenant_id=c.tenant_id AND ae.request_id=$2 AND ae.resource_id IN(c.id,c.public_id,d.id,d.public_id))",
 	}[field]
 	query := diagnosticSelect + " WHERE d.tenant_id=$1::uuid AND " + condition + " ORDER BY c.created_at DESC NULLS LAST LIMIT 50"
@@ -242,8 +242,8 @@ func (s *Service) StopConversation(ctx context.Context, principal identity.Princ
 	}
 	body, _ := json.Marshal(map[string]string{"conversation_id": conversationPublicID, "reason": reason})
 	result, err := s.idem.Execute(ctx, idempotency.Request{TenantID: &a.tenantID, ActorKey: "ops:" + a.userPublicID, Scope: "POST /api/v1/ops/diagnostics/conversations/stop", Key: key, RequestHash: idempotency.Hash(body), ExpiresAfter: 24 * time.Hour}, func(operationContext context.Context, tx pgx.Tx) (idempotency.Result, error) {
-		var internalID, deviceID, status, rtcStatus string
-		err := tx.QueryRow(operationContext, `SELECT c.id::text,d.public_id::text,c.status,rt.status FROM conversation c JOIN device d ON d.id=c.device_id JOIN rtc_task rt ON rt.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.public_id=$2::uuid FOR UPDATE OF c,rt`, a.tenantID, conversationPublicID).Scan(&internalID, &deviceID, &status, &rtcStatus)
+		var internalID, deviceID, status, runtimeStatus, runtimeType string
+		err := tx.QueryRow(operationContext, `SELECT c.id::text,d.public_id::text,c.status,c.runtime_type,COALESCE(rt.status,hs.status,'') FROM conversation c JOIN device d ON d.id=c.device_id LEFT JOIN rtc_task rt ON rt.conversation_id=c.id LEFT JOIN hardware_agent_session hs ON hs.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.public_id=$2::uuid FOR UPDATE OF c`, a.tenantID, conversationPublicID).Scan(&internalID, &deviceID, &status, &runtimeType, &runtimeStatus)
 		if errors.Is(err, pgx.ErrNoRows) {
 			return idempotency.Result{}, ErrNotFound
 		}
@@ -256,15 +256,19 @@ func (s *Service) StopConversation(ctx context.Context, principal identity.Princ
 		case "COMPLETED", "FAILED", "EXPIRED", "STOPPING":
 		case "PREPARING", "ROOM_ISSUED":
 			status = "EXPIRED"
-			if _, err = tx.Exec(operationContext, `UPDATE conversation SET status='EXPIRED',stop_reason='ops_stop_before_start',ended_at=now(),lease_expires_at=NULL,version=version+1,updated_at=now() WHERE id=$1::uuid`, internalID); err == nil {
+			if _, err = tx.Exec(operationContext, `UPDATE conversation SET status='EXPIRED',stop_reason='ops_stop_before_start',ended_at=now(),lease_expires_at=NULL,version=version+1,updated_at=now() WHERE id=$1::uuid`, internalID); err == nil && runtimeType == "volc_hardware_agent" {
+				_, err = tx.Exec(operationContext, `UPDATE hardware_agent_session SET status='EXPIRED',updated_at=now() WHERE conversation_id=$1::uuid`, internalID)
+			} else if err == nil {
 				_, err = tx.Exec(operationContext, `UPDATE rtc_task SET status='COMPLETED',lease_expires_at=NULL,updated_at=now() WHERE conversation_id=$1::uuid`, internalID)
 			}
 		case "STARTING_AGENT", "ACTIVE":
-			status, queued = "STOPPING", true
-			if _, err = tx.Exec(operationContext, `UPDATE conversation SET status='STOPPING',stop_reason='ops_stop',lease_expires_at=NULL,version=version+1,updated_at=now() WHERE id=$1::uuid`, internalID); err == nil {
+			status, queued = "STOPPING", runtimeType != "volc_hardware_agent"
+			if _, err = tx.Exec(operationContext, `UPDATE conversation SET status='STOPPING',stop_reason='ops_stop',lease_expires_at=NULL,version=version+1,updated_at=now() WHERE id=$1::uuid`, internalID); err == nil && runtimeType == "volc_hardware_agent" {
+				_, err = tx.Exec(operationContext, `UPDATE hardware_agent_session SET status='STOPPING',updated_at=now() WHERE conversation_id=$1::uuid`, internalID)
+			} else if err == nil {
 				_, err = tx.Exec(operationContext, `UPDATE rtc_task SET status=CASE WHEN status='ACTIVE' THEN 'STOPPING' ELSE 'UNKNOWN' END,lease_expires_at=NULL,updated_at=now() WHERE conversation_id=$1::uuid`, internalID)
 			}
-			if err == nil {
+			if err == nil && runtimeType != "volc_hardware_agent" {
 				dedupe := "ops-stop:" + internalID
 				payload, _ := json.Marshal(map[string]string{"conversation_id": conversationPublicID, "reason": "ops_stop"})
 				_, err = s.outbox.Enqueue(operationContext, tx, outbox.NewEvent{TenantID: &a.tenantID, AggregateType: "conversation", AggregateID: internalID, EventType: "conversation.agent_stop_requested", DeduplicationKey: &dedupe, Payload: payload})
@@ -280,7 +284,7 @@ func (s *Service) StopConversation(ctx context.Context, principal identity.Princ
 				return idempotency.Result{}, err
 			}
 		}
-		if err = s.audit.Record(operationContext, tx, audit.Event{TenantID: &a.tenantID, ActorType: "user", ActorPublicID: &a.userPublicID, Action: "ops.diagnostics.conversation.stop", ResourceType: "conversation", ResourceID: &conversationPublicID, RequestID: requestID, Purpose: reason, Result: "success", Detail: map[string]any{"from_rtc_status": rtcStatus, "to_status": status}}); err != nil {
+		if err = s.audit.Record(operationContext, tx, audit.Event{TenantID: &a.tenantID, ActorType: "user", ActorPublicID: &a.userPublicID, Action: "ops.diagnostics.conversation.stop", ResourceType: "conversation", ResourceID: &conversationPublicID, RequestID: requestID, Purpose: reason, Result: "success", Detail: map[string]any{"runtime_type": runtimeType, "from_runtime_status": runtimeStatus, "to_status": status}}); err != nil {
 			return idempotency.Result{}, err
 		}
 		output := ActionResult{Action: "stop_conversation", DeviceID: deviceID, ConversationID: conversationPublicID, ConversationStatus: status, Queued: queued}
@@ -386,9 +390,9 @@ func (s *Service) RevokeDevice(ctx context.Context, principal identity.Principal
 			}
 		}

-		var conversationInternalID, conversationPublicID, conversationStatus, rtcStatus string
+		var conversationInternalID, conversationPublicID, conversationStatus, runtimeStatus, runtimeType string
 		queued := false
-		conversationQueryErr := tx.QueryRow(op, `SELECT c.id::text,c.public_id::text,c.status,rt.status FROM conversation c JOIN rtc_task rt ON rt.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.device_id=$2::uuid AND c.status NOT IN('COMPLETED','FAILED','EXPIRED') ORDER BY c.created_at DESC LIMIT 1 FOR UPDATE OF c,rt`, a.tenantID, deviceID).Scan(&conversationInternalID, &conversationPublicID, &conversationStatus, &rtcStatus)
+		conversationQueryErr := tx.QueryRow(op, `SELECT c.id::text,c.public_id::text,c.status,c.runtime_type,COALESCE(rt.status,hs.status,'') FROM conversation c LEFT JOIN rtc_task rt ON rt.conversation_id=c.id LEFT JOIN hardware_agent_session hs ON hs.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.device_id=$2::uuid AND c.status NOT IN('COMPLETED','FAILED','EXPIRED') ORDER BY c.created_at DESC LIMIT 1 FOR UPDATE OF c`, a.tenantID, deviceID).Scan(&conversationInternalID, &conversationPublicID, &conversationStatus, &runtimeType, &runtimeStatus)
 		if conversationQueryErr != nil && !errors.Is(conversationQueryErr, pgx.ErrNoRows) {
 			return idempotency.Result{}, conversationQueryErr
 		}
@@ -400,21 +404,31 @@ func (s *Service) RevokeDevice(ctx context.Context, principal identity.Principal
 				if _, updateErr := tx.Exec(op, `UPDATE conversation SET status='EXPIRED',stop_reason='device_revoked_before_start',ended_at=now(),lease_expires_at=NULL,version=version+1,updated_at=now() WHERE id=$1::uuid`, conversationInternalID); updateErr != nil {
 					return idempotency.Result{}, updateErr
 				}
-				if _, updateErr := tx.Exec(op, `UPDATE rtc_task SET status='COMPLETED',lease_expires_at=NULL,updated_at=now() WHERE conversation_id=$1::uuid`, conversationInternalID); updateErr != nil {
+				query := `UPDATE rtc_task SET status='COMPLETED',lease_expires_at=NULL,updated_at=now() WHERE conversation_id=$1::uuid`
+				if runtimeType == "volc_hardware_agent" {
+					query = `UPDATE hardware_agent_session SET status='EXPIRED',updated_at=now() WHERE conversation_id=$1::uuid`
+				}
+				if _, updateErr := tx.Exec(op, query, conversationInternalID); updateErr != nil {
 					return idempotency.Result{}, updateErr
 				}
 			case "STARTING_AGENT", "ACTIVE":
-				conversationStatus, queued = "STOPPING", true
+				conversationStatus, queued = "STOPPING", runtimeType != "volc_hardware_agent"
 				if _, updateErr := tx.Exec(op, `UPDATE conversation SET status='STOPPING',stop_reason='device_revoked',lease_expires_at=NULL,version=version+1,updated_at=now() WHERE id=$1::uuid`, conversationInternalID); updateErr != nil {
 					return idempotency.Result{}, updateErr
 				}
-				if _, updateErr := tx.Exec(op, `UPDATE rtc_task SET status=CASE WHEN status='ACTIVE' THEN 'STOPPING' ELSE 'UNKNOWN' END,lease_expires_at=NULL,updated_at=now() WHERE conversation_id=$1::uuid`, conversationInternalID); updateErr != nil {
+				query := `UPDATE rtc_task SET status=CASE WHEN status='ACTIVE' THEN 'STOPPING' ELSE 'UNKNOWN' END,lease_expires_at=NULL,updated_at=now() WHERE conversation_id=$1::uuid`
+				if runtimeType == "volc_hardware_agent" {
+					query = `UPDATE hardware_agent_session SET status='STOPPING',updated_at=now() WHERE conversation_id=$1::uuid`
+				}
+				if _, updateErr := tx.Exec(op, query, conversationInternalID); updateErr != nil {
 					return idempotency.Result{}, updateErr
 				}
-				dedupe := "ops-stop:" + conversationInternalID
-				payload, _ := json.Marshal(map[string]string{"conversation_id": conversationPublicID, "reason": "device_revoked"})
-				if _, enqueueErr := s.outbox.Enqueue(op, tx, outbox.NewEvent{TenantID: &a.tenantID, AggregateType: "conversation", AggregateID: conversationInternalID, EventType: "conversation.agent_stop_requested", DeduplicationKey: &dedupe, Payload: payload}); enqueueErr != nil {
-					return idempotency.Result{}, enqueueErr
+				if runtimeType != "volc_hardware_agent" {
+					dedupe := "ops-stop:" + conversationInternalID
+					payload, _ := json.Marshal(map[string]string{"conversation_id": conversationPublicID, "reason": "device_revoked"})
+					if _, enqueueErr := s.outbox.Enqueue(op, tx, outbox.NewEvent{TenantID: &a.tenantID, AggregateType: "conversation", AggregateID: conversationInternalID, EventType: "conversation.agent_stop_requested", DeduplicationKey: &dedupe, Payload: payload}); enqueueErr != nil {
+						return idempotency.Result{}, enqueueErr
+					}
 				}
 			case "STOPPING":
 			default:
@@ -431,7 +445,8 @@ func (s *Service) RevokeDevice(ctx context.Context, principal identity.Principal
 		if conversationPublicID != "" {
 			detail["conversation_id"] = conversationPublicID
 			detail["conversation_status"] = conversationStatus
-			detail["from_rtc_status"] = rtcStatus
+			detail["runtime_type"] = runtimeType
+			detail["from_runtime_status"] = runtimeStatus
 		}
 		if auditErr := s.audit.Record(op, tx, audit.Event{TenantID: &a.tenantID, ActorType: "user", ActorPublicID: &a.userPublicID, Action: "ops.diagnostics.device.revoke", ResourceType: "device", ResourceID: &devicePublicID, RequestID: requestID, Purpose: reason, Result: "success", Detail: detail}); auditErr != nil {
 			return idempotency.Result{}, auditErr
@@ -493,8 +508,8 @@ func (s *Service) ReconcileConversation(ctx context.Context, principal identity.
 	}
 	body, _ := json.Marshal(map[string]string{"conversation_id": conversationPublicID, "reason": reason})
 	result, err := s.idem.Execute(ctx, idempotency.Request{TenantID: &a.tenantID, ActorKey: "ops:" + a.userPublicID, Scope: "POST /api/v1/ops/diagnostics/conversations/reconcile", Key: key, RequestHash: idempotency.Hash(body), ExpiresAfter: 24 * time.Hour}, func(operationContext context.Context, tx pgx.Tx) (idempotency.Result, error) {
-		var internalID, deviceID, status, taskRef string
-		if queryErr := tx.QueryRow(operationContext, `SELECT c.id::text,d.public_id::text,c.status,rt.provider_task_ref FROM conversation c JOIN device d ON d.id=c.device_id JOIN rtc_task rt ON rt.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.public_id=$2::uuid FOR UPDATE OF c`, a.tenantID, conversationPublicID).Scan(&internalID, &deviceID, &status, &taskRef); errors.Is(queryErr, pgx.ErrNoRows) {
+		var internalID, deviceID, status, taskRef, runtimeType string
+		if queryErr := tx.QueryRow(operationContext, `SELECT c.id::text,d.public_id::text,c.status,COALESCE(rt.provider_task_ref,hs.provider_session_ref),c.runtime_type FROM conversation c JOIN device d ON d.id=c.device_id LEFT JOIN rtc_task rt ON rt.conversation_id=c.id LEFT JOIN hardware_agent_session hs ON hs.conversation_id=c.id WHERE c.tenant_id=$1::uuid AND c.public_id=$2::uuid FOR UPDATE OF c`, a.tenantID, conversationPublicID).Scan(&internalID, &deviceID, &status, &taskRef, &runtimeType); errors.Is(queryErr, pgx.ErrNoRows) {
 			return idempotency.Result{}, ErrNotFound
 		} else if queryErr != nil {
 			return idempotency.Result{}, queryErr
@@ -505,7 +520,7 @@ func (s *Service) ReconcileConversation(ctx context.Context, principal identity.
 		} else if status == "STOPPING" {
 			eventType = "conversation.agent_stop_reconcile_requested"
 		}
-		queued := eventType != ""
+		queued := eventType != "" && runtimeType != "volc_hardware_agent"
 		if queued {
 			dedupe := "ops-reconcile:" + taskRef + ":" + key
 			payload, _ := json.Marshal(map[string]string{"conversation_id": conversationPublicID, "provider_task_ref": taskRef, "reason": "ops_reconcile"})
@@ -513,7 +528,7 @@ func (s *Service) ReconcileConversation(ctx context.Context, principal identity.
 				return idempotency.Result{}, enqueueErr
 			}
 		}
-		if auditErr := s.audit.Record(operationContext, tx, audit.Event{TenantID: &a.tenantID, ActorType: "user", ActorPublicID: &a.userPublicID, Action: "ops.diagnostics.conversation.reconcile", ResourceType: "conversation", ResourceID: &conversationPublicID, RequestID: requestID, Purpose: reason, Result: "success", Detail: map[string]any{"status": status, "queued": queued}}); auditErr != nil {
+		if auditErr := s.audit.Record(operationContext, tx, audit.Event{TenantID: &a.tenantID, ActorType: "user", ActorPublicID: &a.userPublicID, Action: "ops.diagnostics.conversation.reconcile", ResourceType: "conversation", ResourceID: &conversationPublicID, RequestID: requestID, Purpose: reason, Result: "success", Detail: map[string]any{"status": status, "runtime_type": runtimeType, "queued": queued}}); auditErr != nil {
 			return idempotency.Result{}, auditErr
 		}
 		output := ActionResult{Action: "reconcile_conversation", DeviceID: deviceID, ConversationID: conversationPublicID, ConversationStatus: status, Queued: queued}
@@ -578,7 +593,7 @@ const diagnosticSelect = `SELECT
 	COALESCE(dt.desired_revision,1),dt.reported_revision,COALESCE(dt.last_error_code,''),dt.last_error_at,
 	COALESCE(c.public_id::text,''),COALESCE(c.status,''),COALESCE(c.failure_code,''),c.allocation_expires_at,
 	rt.credential_issued_at,rt.credential_expires_at,
-	COALESCE(rt.room_id,''),COALESCE(rt.provider,''),COALESCE(rt.provider_task_ref,''),COALESCE(rt.status,''),
+	COALESCE(rt.room_id,''),COALESCE(rt.provider,hs.provider,''),COALESCE(rt.provider_task_ref,hs.provider_session_ref,''),COALESCE(rt.status,hs.status,''),
 	COALESCE(rt.start_attempts,0),COALESCE(rt.stop_attempts,0),COALESCE(pe.event_count,0),
 	COALESCE(pe.event_type,''),COALESCE(pe.event_status,''),pe.received_at,COALESCE(ae.request_id,''),
 	c.created_at,ch.room_issued_at,ch.device_activated_at,ch.agent_active_at,ch.stop_requested_at,c.ended_at
@@ -587,9 +602,14 @@ LEFT JOIN device_twin dt ON dt.device_id=d.id AND dt.tenant_id=d.tenant_id
 LEFT JOIN device_presence_state dp ON dp.device_id=d.id AND dp.tenant_id=d.tenant_id
 LEFT JOIN conversation c ON c.device_id=d.id AND c.tenant_id=d.tenant_id
 LEFT JOIN rtc_task rt ON rt.conversation_id=c.id
+LEFT JOIN hardware_agent_session hs ON hs.conversation_id=c.id
 LEFT JOIN LATERAL (
-	SELECT count(*) OVER() event_count,event_type,COALESCE(sanitized_payload->>'status','') event_status,received_at
-	FROM provider_event WHERE conversation_id=c.id ORDER BY received_at DESC LIMIT 1
+	SELECT count(*) OVER() event_count,event_type,event_status,received_at FROM (
+		SELECT event_type,COALESCE(sanitized_payload->>'status','') event_status,received_at
+		FROM provider_event WHERE conversation_id=c.id
+		UNION ALL
+		SELECT event_type,event_type,occurred_at FROM hardware_agent_event WHERE conversation_id=c.id
+	) runtime_event ORDER BY received_at DESC LIMIT 1
 ) pe ON true
 LEFT JOIN LATERAL (
 	SELECT