BCAI/easyai-ai-gateway
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Decouple stream cancellation and clarify failover rules

Browse Source
This commit is contained in:
wangbo
2026-05-12 21:32:23 +08:00
parent 98abd247d6
commit 682a491d27
8 changed files with 629 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
apps/api/internal/store/rate_limit_status.go
+124 -1
View File
@@ -24,6 +24,8 @@ type ModelRateLimitStatus struct {
PlatformID string `json:"platformId"`
PlatformName string `json:"platformName"`
Provider string `json:"provider"`
PlatformStatus string `json:"platformStatus"`
PlatformDisabledReason *PlatformPolicyEvent `json:"platformDisabledReason,omitempty"`
PlatformPriority int `json:"platformPriority"`
PlatformDynamicPriority *int `json:"platformDynamicPriority,omitempty"`
PlatformEffectivePriority int `json:"platformEffectivePriority"`
@@ -62,9 +64,27 @@ type PriorityDemotionRecord struct {
CreatedAt time.Time `json:"createdAt"`
}
type PlatformPolicyEvent struct {
ID string `json:"id"`
TaskID string `json:"taskId"`
PlatformID string `json:"platformId"`
PlatformModelID string `json:"platformModelId,omitempty"`
EventType string `json:"eventType"`
Reason string `json:"reason,omitempty"`
ErrorCode string `json:"errorCode,omitempty"`
ErrorMessage string `json:"errorMessage,omitempty"`
Category string `json:"category,omitempty"`
StatusCode int `json:"statusCode,omitempty"`
PolicySource string `json:"policySource,omitempty"`
Policy string `json:"policy,omitempty"`
PolicyRule string `json:"policyRule,omitempty"`
MatchedValue string `json:"matchedValue,omitempty"`
CreatedAt time.Time `json:"createdAt"`
}
func (s *Store) ListModelRateLimitStatuses(ctx context.Context) ([]ModelRateLimitStatus, error) {
rows, err := s.pool.Query(ctx, `
SELECT m.id::text, m.platform_id::text, p.name, p.provider,
SELECT m.id::text, m.platform_id::text, p.name, p.provider, p.status,
p.priority, p.dynamic_priority, COALESCE(p.dynamic_priority, p.priority),
m.model_name, COALESCE(NULLIF(m.provider_model_name, ''), m.model_name), COALESCE(m.model_alias, ''),
m.model_type, m.display_name, m.enabled,
@@ -161,6 +181,7 @@ ORDER BY p.priority ASC, m.model_name ASC`)
&item.PlatformID,
&item.PlatformName,
&item.Provider,
&item.PlatformStatus,
&item.PlatformPriority,
&platformDynamicPriority,
&item.PlatformEffectivePriority,
@@ -212,8 +233,15 @@ ORDER BY p.priority ASC, m.model_name ASC`)
if err != nil {
return nil, err
}
disabledReasons, err := s.listLatestPlatformDisabledReasons(ctx, items)
if err != nil {
return nil, err
}
for index := range items {
items[index].RecentPriorityDemotions = demotions[items[index].PlatformID]
if items[index].PlatformStatus != "enabled" {
items[index].PlatformDisabledReason = disabledReasons[items[index].PlatformID]
}
}
sort.SliceStable(items, func(i, j int) bool {
if items[i].LoadRatio == items[j].LoadRatio {
@@ -309,6 +337,101 @@ func priorityDemotionRecordFromEventPayload(id string, taskID string, message st
}
}
func (s *Store) listLatestPlatformDisabledReasons(ctx context.Context, statuses []ModelRateLimitStatus) (map[string]*PlatformPolicyEvent, error) {
out := map[string]*PlatformPolicyEvent{}
seen := map[string]bool{}
platformIDs := make([]string, 0, len(statuses))
for _, status := range statuses {
platformID := strings.TrimSpace(status.PlatformID)
if platformID == "" || status.PlatformStatus == "enabled" || seen[platformID] {
continue
}
seen[platformID] = true
platformIDs = append(platformIDs, platformID)
}
if len(platformIDs) == 0 {
return out, nil
}
rows, err := s.pool.Query(ctx, `
SELECT id::text, task_id::text, event_type, COALESCE(message, ''), payload, COALESCE(attempt_error_message, ''), created_at
FROM (
SELECT e.*,
a.error_message AS attempt_error_message,
row_number() OVER (
PARTITION BY e.payload->>'platformId'
ORDER BY e.created_at DESC, e.seq DESC
) AS disabled_rank
FROM gateway_task_events e
LEFT JOIN LATERAL (
SELECT error_message
FROM gateway_task_attempts attempt
WHERE attempt.task_id = e.task_id
AND attempt.platform_id::text = e.payload->>'platformId'
ORDER BY attempt.attempt_no DESC, attempt.started_at DESC
LIMIT 1
) a ON TRUE
WHERE e.event_type IN ('task.policy.failover_disabled', 'task.policy.auto_disabled')
AND e.payload->>'platformId' = ANY($1::text[])
) ranked
WHERE disabled_rank = 1`, platformIDs)
if err != nil {
return nil, err
}
defer rows.Close()
for rows.Next() {
var id string
var taskID string
var eventType string
var message string
var payloadBytes []byte
var attemptErrorMessage string
var createdAt time.Time
if err := rows.Scan(&id, &taskID, &eventType, &message, &payloadBytes, &attemptErrorMessage, &createdAt); err != nil {
return nil, err
}
record := platformPolicyEventFromPayload(id, taskID, eventType, message, attemptErrorMessage, decodeObject(payloadBytes), createdAt)
if record.PlatformID == "" {
continue
}
out[record.PlatformID] = &record
}
return out, rows.Err()
}
func platformPolicyEventFromPayload(id string, taskID string, eventType string, message string, attemptErrorMessage string, payload map[string]any, createdAt time.Time) PlatformPolicyEvent {
errorMessage := stringValue(payload["errorMessage"])
if errorMessage == "" {
errorMessage = stringValue(payload["message"])
}
if errorMessage == "" {
errorMessage = strings.TrimSpace(attemptErrorMessage)
}
if errorMessage == "" {
errorMessage = strings.TrimSpace(message)
}
errorCode := stringValue(payload["errorCode"])
if errorCode == "" {
errorCode = stringValue(payload["code"])
}
return PlatformPolicyEvent{
ID: id,
TaskID: taskID,
PlatformID: stringValue(payload["platformId"]),
PlatformModelID: stringValue(payload["platformModelId"]),
EventType: eventType,
Reason: stringValue(payload["reason"]),
ErrorCode: errorCode,
ErrorMessage: errorMessage,
Category: stringValue(payload["category"]),
StatusCode: intValue(payload["statusCode"]),
PolicySource: stringValue(payload["policySource"]),
Policy: stringValue(payload["policy"]),
PolicyRule: stringValue(payload["policyRule"]),
MatchedValue: stringValue(payload["matchedValue"]),
CreatedAt: createdAt,
}
}
func effectiveModelRateLimitPolicy(platformPolicy map[string]any, runtimePolicy map[string]any, runtimeOverride map[string]any, modelPolicy map[string]any) map[string]any {
policy := platformPolicy
if hasRateLimitRules(runtimePolicy) {
apps/api/internal/store/rate_limit_status_test.go
+23
View File
@@ -61,3 +61,26 @@ func TestPriorityDemotionRecordFromEventPayloadKeepsReason(t *testing.T) {
t.Fatalf("expected createdAt %s, got %s", createdAt, record.CreatedAt)
}
}
func TestPlatformPolicyEventFromPayloadUsesAttemptErrorMessage(t *testing.T) {
createdAt := time.Date(2026, 5, 12, 10, 30, 0, 0, time.UTC)
record := platformPolicyEventFromPayload("event-1", "task-1", "task.policy.failover_disabled", "fallback event message", "upstream invalid api key", map[string]any{
"platformId": "platform-1",
"platformModelId": "platform-model-1",
"reason": "failover_allow_policy",
"errorCode": "auth_failed",
"category": "auth_error",
"statusCode": float64(401),
"policySource": "gateway_runner_policies.failover_policy",
"policy": "failoverPolicy",
"policyRule": "allowCategories",
"matchedValue": "auth_error",
}, createdAt)
if record.EventType != "task.policy.failover_disabled" || record.Reason != "failover_allow_policy" {
t.Fatalf("expected disabled event identity to survive, got %+v", record)
}
if record.ErrorMessage != "upstream invalid api key" || record.StatusCode != 401 {
t.Fatalf("expected disabled reason details from attempt, got %+v", record)
}
}