Files
easyai-ai-gateway/apps/api/internal/store/async_worker_capacity.go
T
wangbo 9e4fc7362d feat(queue): 增加非文本模型分布式准入队列
使用 PostgreSQL 统一同步与异步非文本任务的并发准入、持久化等待和 Worker 容量分配,并将生产 API 与独立 Worker 角色拆分。

补充策略管理、共享契约、OpenAPI、Kubernetes 双节点 Worker 清单及跨节点验收脚本;未默认启用任何生产 queue_size 策略。

已在原基线完成 Go、前端、迁移、Shell、Kustomize 与长任务容量验收;合入最新主干后将重新执行发布门禁。
2026-07-29 16:15:43 +08:00

157 lines
4.7 KiB
Go

package store
import (
"context"
"fmt"
)
type AsyncWorkerCapacitySnapshot struct {
Capacity int
GlobalCapacity int
Desired int
HardLimit int
Capped bool
EnabledModels int
UnlimitedModels int
EnabledGroups int
UnlimitedGroups int
ModelDesired int
GroupDesired int
ActiveInstances int
InstanceID string
}
func (s *Store) AsyncWorkerCapacity(ctx context.Context, hardLimit int) (AsyncWorkerCapacitySnapshot, error) {
if hardLimit < 1 {
return AsyncWorkerCapacitySnapshot{}, fmt.Errorf("async worker hard limit must be positive")
}
rows, err := s.pool.Query(ctx, `
SELECT COALESCE(b.default_rate_limit_policy, '{}'::jsonb),
p.rate_limit_policy,
COALESCE(rp.rate_limit_policy, '{}'::jsonb),
(m.runtime_policy_set_id IS NOT NULL),
COALESCE(m.runtime_policy_override, '{}'::jsonb),
m.rate_limit_policy,
m.rate_limit_policy_mode
FROM platform_models m
JOIN integration_platforms p ON p.id = m.platform_id
LEFT JOIN base_model_catalog b ON b.id = m.base_model_id
LEFT JOIN model_runtime_policy_sets rp ON rp.id = COALESCE(m.runtime_policy_set_id, b.runtime_policy_set_id)
WHERE p.status = 'enabled'
AND p.deleted_at IS NULL
AND m.enabled = true`)
if err != nil {
return AsyncWorkerCapacitySnapshot{}, err
}
defer rows.Close()
modelPolicies := make([]map[string]any, 0)
for rows.Next() {
var basePolicyBytes, platformPolicyBytes, runtimePolicyBytes []byte
var runtimeOverrideBytes, modelPolicyBytes []byte
var runtimeExplicit bool
var modelPolicyMode string
if err := rows.Scan(
&basePolicyBytes,
&platformPolicyBytes,
&runtimePolicyBytes,
&runtimeExplicit,
&runtimeOverrideBytes,
&modelPolicyBytes,
&modelPolicyMode,
); err != nil {
return AsyncWorkerCapacitySnapshot{}, err
}
modelPolicies = append(modelPolicies, EffectiveRateLimitPolicy(EffectiveRateLimitPolicyInput{
BasePolicy: decodeObject(basePolicyBytes),
PlatformPolicy: decodeObject(platformPolicyBytes),
RuntimePolicy: decodeObject(runtimePolicyBytes),
RuntimePolicyExplicit: runtimeExplicit,
RuntimePolicyOverride: decodeObject(runtimeOverrideBytes),
ModelPolicy: decodeObject(modelPolicyBytes),
ModelPolicyMode: modelPolicyMode,
}))
}
if err := rows.Err(); err != nil {
return AsyncWorkerCapacitySnapshot{}, err
}
groupRows, err := s.pool.Query(ctx, `
SELECT rate_limit_policy
FROM gateway_user_groups
WHERE status = 'active'`)
if err != nil {
return AsyncWorkerCapacitySnapshot{}, err
}
defer groupRows.Close()
groupPolicies := make([]map[string]any, 0)
for groupRows.Next() {
var policyBytes []byte
if err := groupRows.Scan(&policyBytes); err != nil {
return AsyncWorkerCapacitySnapshot{}, err
}
groupPolicies = append(groupPolicies, NormalizeRateLimitPolicy(decodeObject(policyBytes)))
}
if err := groupRows.Err(); err != nil {
return AsyncWorkerCapacitySnapshot{}, err
}
return asyncWorkerCapacityFromPolicySets(modelPolicies, groupPolicies, hardLimit), nil
}
func asyncWorkerCapacityFromPolicies(policies []map[string]any, hardLimit int) AsyncWorkerCapacitySnapshot {
return asyncWorkerCapacityFromPolicySets(policies, nil, hardLimit)
}
func asyncWorkerCapacityFromPolicySets(modelPolicies []map[string]any, groupPolicies []map[string]any, hardLimit int) AsyncWorkerCapacitySnapshot {
snapshot := AsyncWorkerCapacitySnapshot{
HardLimit: hardLimit,
EnabledModels: len(modelPolicies),
EnabledGroups: len(groupPolicies),
}
modelDesired, modelFinite, unlimitedModels := concurrentPolicySetCapacity(modelPolicies)
groupDesired, groupFinite, unlimitedGroups := concurrentPolicySetCapacity(groupPolicies)
snapshot.ModelDesired = modelDesired
snapshot.GroupDesired = groupDesired
snapshot.UnlimitedModels = unlimitedModels
snapshot.UnlimitedGroups = unlimitedGroups
desired := hardLimit
switch {
case snapshot.EnabledModels == 0:
desired = 1
case modelFinite && groupFinite:
desired = min(modelDesired, groupDesired)
case modelFinite:
desired = modelDesired
case groupFinite:
desired = groupDesired
}
if desired < 1 {
desired = 1
}
snapshot.Desired = desired
snapshot.Capacity = desired
if snapshot.Capacity > hardLimit {
snapshot.Capacity = hardLimit
snapshot.Capped = true
}
snapshot.GlobalCapacity = snapshot.Capacity
return snapshot
}
func concurrentPolicySetCapacity(policies []map[string]any) (total int, finite bool, unlimited int) {
if len(policies) == 0 {
return 0, false, 0
}
finite = true
for _, policy := range policies {
capacity, policyFinite := ConcurrentPolicyCapacity(policy)
if !policyFinite {
unlimited++
finite = false
continue
}
total += capacity
}
return total, finite, unlimited
}