Files
easyai-ai-gateway/apps/api/internal/store/async_worker_capacity.go
T
wangbo c28bf74230 feat(worker): 实现集群限流与自适应负载
保留平台模型 RPM、TPM 和并发策略语义,增加 PostgreSQL 集群级租约、饱和候选重选和多平台自动负载,避免突发任务固定等待首个平台。\n\n新增 Worker 实时负载采样、自适应 active/heavy 容量、心跳与管理端指标,并扩展本地 acceptance runner,覆盖三 Worker、同模型三平台 2/4/6 并发和 48 个带图视频突发任务。\n\n验证:go test ./...、go vet ./...、PostgreSQL 跨 Store 集成测试、gofmt、bash -n、ShellCheck 及本地集群 provider-burst 验收通过;48/48 成功,无越限、重复提交、重复计费、重复回调或终态资源泄漏。
2026-08-03 00:13:46 +08:00

166 lines
5.0 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
LoadMode string
LocalSafeCapacity int
LocalHeavyCapacity int
LocalActiveTasks int
LocalPreparingTasks int
LocalWaitingTasks int
LocalFinalizingTasks int
LocalPressureState string
LocalPressureReason 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
}