feat: add runner failover policies and traces

This commit is contained in:
2026-05-12 02:16:42 +08:00
parent be31923e74
commit 05632172d0
26 changed files with 2033 additions and 140 deletions
+20
View File
@@ -229,6 +229,20 @@ type RuntimePolicySet struct {
UpdatedAt time.Time `json:"updatedAt"`
}
type RunnerPolicy struct {
ID string `json:"id"`
PolicyKey string `json:"policyKey"`
Name string `json:"name"`
Description string `json:"description,omitempty"`
FailoverPolicy map[string]any `json:"failoverPolicy,omitempty"`
HardStopPolicy map[string]any `json:"hardStopPolicy,omitempty"`
PriorityDemotePolicy map[string]any `json:"priorityDemotePolicy,omitempty"`
Metadata map[string]any `json:"metadata,omitempty"`
Status string `json:"status"`
CreatedAt time.Time `json:"createdAt"`
UpdatedAt time.Time `json:"updatedAt"`
}
type PricingRule struct {
ID string `json:"id"`
RuleSetID string `json:"ruleSetId,omitempty"`
@@ -446,6 +460,7 @@ type TaskAttempt struct {
Retryable bool `json:"retryable"`
Simulated bool `json:"simulated"`
RequestID string `json:"requestId,omitempty"`
StatusCode int `json:"statusCode,omitempty"`
Usage map[string]any `json:"usage,omitempty"`
Metrics map[string]any `json:"metrics,omitempty"`
RequestSnapshot map[string]any `json:"requestSnapshot,omitempty"`
@@ -1796,6 +1811,11 @@ func IsUniqueViolation(err error) bool {
return isUniqueViolation(err)
}
func IsUndefinedDatabaseObject(err error) bool {
var pgErr *pgconn.PgError
return errors.As(err, &pgErr) && (pgErr.Code == "42P01" || pgErr.Code == "42703")
}
func isUniqueViolation(err error) bool {
var pgErr *pgconn.PgError
return errors.As(err, &pgErr) && pgErr.Code == "23505"
+168
View File
@@ -0,0 +1,168 @@
package store
import (
"context"
"encoding/json"
"strings"
"time"
"github.com/jackc/pgx/v5"
)
const runnerPolicyColumns = `
id::text, policy_key, name, COALESCE(description, ''), failover_policy,
hard_stop_policy, priority_demote_policy, metadata, status, created_at, updated_at`
type RunnerPolicyInput struct {
PolicyKey string `json:"policyKey"`
Name string `json:"name"`
Description string `json:"description"`
FailoverPolicy map[string]any `json:"failoverPolicy"`
HardStopPolicy map[string]any `json:"hardStopPolicy"`
PriorityDemotePolicy map[string]any `json:"priorityDemotePolicy"`
Metadata map[string]any `json:"metadata"`
Status string `json:"status"`
}
type runnerPolicyScanner interface {
Scan(dest ...any) error
}
func (s *Store) GetActiveRunnerPolicy(ctx context.Context) (RunnerPolicy, error) {
item, err := scanRunnerPolicy(s.pool.QueryRow(ctx, `
SELECT `+runnerPolicyColumns+`
FROM gateway_runner_policies
ORDER BY CASE WHEN policy_key = 'default-runner-v1' THEN 0 ELSE 1 END,
CASE WHEN status = 'active' THEN 0 ELSE 1 END,
updated_at DESC
LIMIT 1`))
if err != nil {
if err == pgx.ErrNoRows || IsUndefinedDatabaseObject(err) {
return defaultRunnerPolicy(), nil
}
return RunnerPolicy{}, err
}
return item, nil
}
func (s *Store) UpsertDefaultRunnerPolicy(ctx context.Context, input RunnerPolicyInput) (RunnerPolicy, error) {
input = normalizeRunnerPolicyInput(input)
failoverPolicy, _ := json.Marshal(emptyObjectIfNil(input.FailoverPolicy))
hardStopPolicy, _ := json.Marshal(emptyObjectIfNil(input.HardStopPolicy))
priorityDemotePolicy, _ := json.Marshal(emptyObjectIfNil(input.PriorityDemotePolicy))
metadata, _ := json.Marshal(emptyObjectIfNil(input.Metadata))
return scanRunnerPolicy(s.pool.QueryRow(ctx, `
INSERT INTO gateway_runner_policies (
policy_key, name, description, failover_policy, hard_stop_policy, priority_demote_policy, metadata, status
)
VALUES ($1, $2, NULLIF($3, ''), $4, $5, $6, $7, $8)
ON CONFLICT (policy_key) DO UPDATE
SET name = EXCLUDED.name,
description = EXCLUDED.description,
failover_policy = EXCLUDED.failover_policy,
hard_stop_policy = EXCLUDED.hard_stop_policy,
priority_demote_policy = EXCLUDED.priority_demote_policy,
metadata = EXCLUDED.metadata,
status = EXCLUDED.status,
updated_at = now()
RETURNING `+runnerPolicyColumns,
input.PolicyKey, input.Name, input.Description, failoverPolicy, hardStopPolicy, priorityDemotePolicy, metadata, input.Status,
))
}
func scanRunnerPolicy(scanner runnerPolicyScanner) (RunnerPolicy, error) {
var item RunnerPolicy
var failoverPolicy []byte
var hardStopPolicy []byte
var priorityDemotePolicy []byte
var metadata []byte
if err := scanner.Scan(
&item.ID,
&item.PolicyKey,
&item.Name,
&item.Description,
&failoverPolicy,
&hardStopPolicy,
&priorityDemotePolicy,
&metadata,
&item.Status,
&item.CreatedAt,
&item.UpdatedAt,
); err != nil {
return RunnerPolicy{}, err
}
item.FailoverPolicy = decodeObject(failoverPolicy)
item.HardStopPolicy = decodeObject(hardStopPolicy)
item.PriorityDemotePolicy = decodeObject(priorityDemotePolicy)
item.Metadata = decodeObject(metadata)
return item, nil
}
func normalizeRunnerPolicyInput(input RunnerPolicyInput) RunnerPolicyInput {
input.PolicyKey = strings.TrimSpace(input.PolicyKey)
if input.PolicyKey == "" {
input.PolicyKey = "default-runner-v1"
}
input.Name = strings.TrimSpace(input.Name)
if input.Name == "" {
input.Name = "默认全局调度策略"
}
input.Description = strings.TrimSpace(input.Description)
input.Status = strings.TrimSpace(input.Status)
if input.Status == "" {
input.Status = "active"
}
return input
}
func defaultRunnerPolicy() RunnerPolicy {
now := time.Now()
return RunnerPolicy{
PolicyKey: "default-runner-v1",
Name: "默认全局调度策略",
Description: "控制多个候选平台之间的故障切换;模型运行策略只可覆盖 failoverPolicy,不能覆盖 hardStopPolicy。",
FailoverPolicy: defaultRunnerFailoverPolicy(),
HardStopPolicy: defaultRunnerHardStopPolicy(),
PriorityDemotePolicy: defaultRunnerPriorityDemotePolicy(),
Metadata: map[string]any{"source": "code-default"},
Status: "active",
CreatedAt: now,
UpdatedAt: now,
}
}
func defaultRunnerPriorityDemotePolicy() map[string]any {
return map[string]any{
"enabled": true,
"demoteStep": 100,
"categories": []any{"network", "timeout", "stream_error", "rate_limit", "provider_5xx", "provider_overloaded"},
"codes": []any{"network", "timeout", "stream_read_error", "rate_limit", "server_error", "overloaded"},
"statusCodes": []any{408, 429, 500, 502, 503, 504},
"keywords": []any{"timeout", "network", "rate_limit", "overloaded", "temporarily_unavailable", "server_error", "429", "5xx"},
}
}
func defaultRunnerFailoverPolicy() map[string]any {
return map[string]any{
"enabled": true,
"maxPlatforms": 99,
"maxDurationSeconds": 600,
"allowCategories": []any{"network", "timeout", "stream_error", "rate_limit", "provider_5xx", "provider_overloaded", "auth_error"},
"denyCategories": []any{"request_error", "unsupported_model", "user_permission", "insufficient_balance"},
"allowCodes": []any{"auth_failed", "invalid_api_key", "missing_credentials"},
"allowKeywords": []any{"timeout", "network", "rate_limit", "overloaded", "temporarily_unavailable", "server_error", "auth_failed", "invalid_api_key", "missing_credentials", "unauthorized", "forbidden", "429", "5xx"},
"denyKeywords": []any{"invalid_parameter", "missing required", "bad request"},
"allowStatusCodes": []any{401, 403, 408, 429, 500, 502, 503, 504},
"denyStatusCodes": []any{},
}
}
func defaultRunnerHardStopPolicy() map[string]any {
return map[string]any{
"enabled": true,
"categories": []any{"request_error", "unsupported_model", "user_permission", "insufficient_balance"},
"codes": []any{"bad_request", "invalid_request", "invalid_parameter", "missing_required", "unsupported_kind", "unsupported_model", "insufficient_balance", "permission_denied"},
"statusCodes": []any{},
"keywords": []any{"invalid_parameter", "missing required", "bad request", "insufficient balance"},
}
}
@@ -134,6 +134,28 @@ WHERE id = $1::uuid`, platformID, cooldownSeconds)
return err
}
func (s *Store) DemoteCandidatePlatformPriority(ctx context.Context, platformID string, demoteStep int) error {
if strings.TrimSpace(platformID) == "" {
return nil
}
if demoteStep <= 0 {
demoteStep = 100
}
_, err := s.pool.Exec(ctx, `
UPDATE integration_platforms target
SET dynamic_priority = GREATEST(
COALESCE(target.dynamic_priority, target.priority),
COALESCE((
SELECT MAX(COALESCE(peer.dynamic_priority, peer.priority))
FROM integration_platforms peer
WHERE peer.deleted_at IS NULL
), target.priority) + $2::int
),
updated_at = now()
WHERE target.id = $1::uuid`, platformID, demoteStep)
return err
}
func scanRuntimePolicySet(scanner runtimePolicyScanner) (RuntimePolicySet, error) {
var item RuntimePolicySet
var rateLimitPolicy []byte
+42
View File
@@ -3,6 +3,7 @@ package store
import (
"context"
"encoding/json"
"strconv"
"strings"
"time"
@@ -234,6 +235,27 @@ func (s *Store) ListTaskAttempts(ctx context.Context, taskID string) ([]TaskAtte
return attemptsByTaskID[taskID], nil
}
func (s *Store) AppendTaskAttemptTrace(ctx context.Context, taskID string, attemptNo int, entry map[string]any) error {
entryJSON, _ := json.Marshal(emptyObjectIfNil(entry))
_, err := s.pool.Exec(ctx, `
UPDATE gateway_task_attempts
SET metrics = jsonb_set(
COALESCE(metrics, '{}'::jsonb),
'{trace}',
(
CASE
WHEN jsonb_typeof(COALESCE(metrics->'trace', '[]'::jsonb)) = 'array'
THEN COALESCE(metrics->'trace', '[]'::jsonb)
ELSE '[]'::jsonb
END
) || jsonb_build_array($3::jsonb),
true
)
WHERE task_id = $1::uuid
AND attempt_no = $2`, taskID, attemptNo, string(entryJSON))
return err
}
func (s *Store) listTaskAttemptsByTaskIDs(ctx context.Context, taskIDs []string) (map[string][]TaskAttempt, error) {
itemsByTaskID := map[string][]TaskAttempt{}
if len(taskIDs) == 0 {
@@ -331,6 +353,7 @@ func enrichTaskAttemptFromMetrics(item *TaskAttempt) {
item.ModelAlias = firstNonEmpty(item.ModelAlias, taskAttemptMetricString(item.Metrics, "modelAlias"))
item.ModelType = firstNonEmpty(item.ModelType, taskAttemptMetricString(item.Metrics, "modelType"))
item.ClientID = firstNonEmpty(item.ClientID, taskAttemptMetricString(item.Metrics, "clientId"))
item.StatusCode = taskAttemptMetricInt(item.Metrics, "statusCode")
}
func taskAttemptMetricString(metrics map[string]any, key string) string {
@@ -338,6 +361,25 @@ func taskAttemptMetricString(metrics map[string]any, key string) string {
return strings.TrimSpace(value)
}
func taskAttemptMetricInt(metrics map[string]any, key string) int {
switch value := metrics[key].(type) {
case int:
return value
case int64:
return int(value)
case float64:
return int(value)
case json.Number:
next, _ := value.Int64()
return int(next)
case string:
next, _ := strconv.Atoi(strings.TrimSpace(value))
return next
default:
return 0
}
}
func (s *Store) FinishTaskAttempt(ctx context.Context, input FinishTaskAttemptInput) error {
responseJSON, _ := json.Marshal(emptyObjectIfNil(input.ResponseSnapshot))
usageJSON, _ := json.Marshal(emptyObjectIfNil(input.Usage))