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

perf(postgres): 分离 API 池并降低连接预热

Browse Source 
生产同构千并发请求在干净队列上复现 API Pool 31/32、idle=1 且 canceled acquire 持续增长，Worker Pool 仅使用少量连接，确认瓶颈位于 API 数据库连接预算而非 Worker 执行槽。

新增可配置的 AI_GATEWAY_DATABASE_MIN_IDLE_CONNS，生产仅预热 4 条连接；API Pool 独立固定为 48，Worker Pool 继续跟随 P24/P28/P32 的 32/36/40 档位。发布工具、Kubernetes 清单、验收门禁和文档同步更新。

验证：Go 全量测试、gofmt、bash -n、ShellCheck、kubectl kustomize 和 git diff --check 通过。
This commit is contained in:
wangbo
2026-07-31 03:28:43 +08:00
parent 2606cc8223
commit 2529679b67
11 changed files with 78 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -164,7 +164,7 @@ AI_GATEWAY_DATABASE_URL=postgresql://easyai:easyai2025@localhost:5432/easyai_ai_
如果现有 `easyai-pgvector` 没有把 `5432` 映射到宿主机，就需要补端口映射，或者把 AI Gateway 后端容器化后接入同一个 `easyai` Docker network。
异步队列 worker 不使用固定业务并发。服务分别汇总启用平台模型和活跃用户组的有效 `concurrent` 策略，采用两者中更严格的集群容量，默认每 5 秒在线调整 River 执行容量。策略解析兼容历史 `platformLimits/modelLimits.max_concurrent_requests`，运行时统一转换为 `rules`。`AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT` 默认 `2048`，限制策略推导出的集群目标；`AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT` 默认 `32`，按单 Worker 的内存安全容量限制实例分配，即使其他实例失活也不会突破。Worker 的 `AI_GATEWAY_DATABASE_MAX_CONNS` 必须高于实例执行容量，为心跳、选主、租约续期和健康检查保留连接；生产环境按每实例执行容量 `24` 配置连接池 `32`。平台模型和用户组的 PostgreSQL concurrency lease 仍是业务并发真值。可通过 `AI_GATEWAY_ASYNC_WORKER_REFRESH_INTERVAL_SECONDS` 调整刷新周期。
异步队列 worker 不使用固定业务并发。服务分别汇总启用平台模型和活跃用户组的有效 `concurrent` 策略，采用两者中更严格的集群容量，默认每 5 秒在线调整 River 执行容量。策略解析兼容历史 `platformLimits/modelLimits.max_concurrent_requests`，运行时统一转换为 `rules`。`AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT` 默认 `2048`，限制策略推导出的集群目标；`AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT` 默认 `32`，按单 Worker 的内存安全容量限制实例分配，即使其他实例失活也不会突破。Worker 的 `AI_GATEWAY_DATABASE_MAX_CONNS` 必须高于实例执行容量，为心跳、选主、租约续期和健康检查保留连接；生产环境按每实例执行容量 `24` 配置连接池上限 `32`。`AI_GATEWAY_DATABASE_MIN_IDLE_CONNS` 控制启动预热连接数，默认 `0`，生产 K3s 配置为 `4`，避免把连接池上限全部常驻占用。平台模型和用户组的 PostgreSQL concurrency lease 仍是业务并发真值。可通过 `AI_GATEWAY_ASYNC_WORKER_REFRESH_INTERVAL_SECONDS` 调整刷新周期。
## 迁移原则
apps/api/cmd/gateway/main.go
+1
View File
@@ -39,6 +39,7 @@ func main() {
db, err := store.ConnectWithPoolOptions(ctx, cfg.DatabaseURL, store.PostgresPoolOptions{
MaxConns: cfg.DatabaseMaxConns,
MinIdleConns: cfg.DatabaseMinIdleConns,
IdleInTransactionTimeout: time.Duration(cfg.DatabaseIdleInTransactionTimeoutSeconds) * time.Second,
LockTimeout: time.Duration(cfg.DatabaseLockTimeoutSeconds) * time.Second,
})
apps/api/internal/config/config.go
+6
View File
@@ -59,6 +59,7 @@ type Config struct {
BillingEngineMode string
ProcessRole string
DatabaseMaxConns int
DatabaseMinIdleConns int
DatabaseIdleInTransactionTimeoutSeconds int
DatabaseLockTimeoutSeconds int
MediaRequestConcurrency int
@@ -122,6 +123,7 @@ func Load() Config {
BillingEngineMode: strings.ToLower(env("BILLING_ENGINE_MODE", "observe")),
ProcessRole: strings.ToLower(strings.TrimSpace(env("AI_GATEWAY_PROCESS_ROLE", "all"))),
DatabaseMaxConns: envInt("AI_GATEWAY_DATABASE_MAX_CONNS", 0),
DatabaseMinIdleConns: envOptionalIntValidated("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS", 0),
DatabaseIdleInTransactionTimeoutSeconds: envOptionalIntValidated(
"AI_GATEWAY_DATABASE_IDLE_IN_TRANSACTION_TIMEOUT_SECONDS",
@@ -147,6 +149,10 @@ func (c Config) Validate() error {
if c.DatabaseMaxConns < 0 || c.DatabaseMaxConns > 1000 {
return errors.New("AI_GATEWAY_DATABASE_MAX_CONNS must be between 1 and 1000 when configured")
}
if c.DatabaseMinIdleConns < 0 || c.DatabaseMinIdleConns > 1000 ||
(c.DatabaseMaxConns > 0 && c.DatabaseMinIdleConns > c.DatabaseMaxConns) {
return errors.New("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS must be between 0 and AI_GATEWAY_DATABASE_MAX_CONNS")
}
if c.DatabaseIdleInTransactionTimeoutSeconds < 0 || c.DatabaseIdleInTransactionTimeoutSeconds > 3600 {
return errors.New("AI_GATEWAY_DATABASE_IDLE_IN_TRANSACTION_TIMEOUT_SECONDS must be between 0 and 3600")
}
apps/api/internal/config/config_test.go
+15 -2
View File
@@ -144,9 +144,12 @@ func TestProcessRolePrecedenceAndCompatibility(t *testing.T) {
func TestValidateProcessRoleAndDatabasePool(t *testing.T) {
cfg := Load()
if cfg.DatabaseIdleInTransactionTimeoutSeconds != 60 || cfg.DatabaseLockTimeoutSeconds != 30 {
if cfg.DatabaseMinIdleConns != 0 ||
cfg.DatabaseIdleInTransactionTimeoutSeconds != 60 ||
cfg.DatabaseLockTimeoutSeconds != 30 {
t.Fatalf(
"database transaction timeouts = %d/%d, want 60/30",
"database pool minimum/transaction timeouts = %d/%d/%d, want 0/60/30",
cfg.DatabaseMinIdleConns,
cfg.DatabaseIdleInTransactionTimeoutSeconds,
cfg.DatabaseLockTimeoutSeconds,
)
@@ -161,6 +164,16 @@ func TestValidateProcessRoleAndDatabasePool(t *testing.T) {
t.Fatalf("Validate() error = %v, want invalid database max conns", err)
}
cfg.DatabaseMaxConns = 16
cfg.DatabaseMinIdleConns = 17
if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "DATABASE_MIN_IDLE_CONNS") {
t.Fatalf("Validate() error = %v, want invalid database min idle conns", err)
}
cfg.DatabaseMinIdleConns = 4
t.Setenv("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS", "not-an-integer")
if err := Load().Validate(); err == nil || !strings.Contains(err.Error(), "DATABASE_MIN_IDLE_CONNS") {
t.Fatalf("Validate() error = %v, want invalid non-integer database min idle conns", err)
}
t.Setenv("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS", "4")
cfg.DatabaseIdleInTransactionTimeoutSeconds = 3601
if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "IDLE_IN_TRANSACTION") {
t.Fatalf("Validate() error = %v, want invalid idle transaction timeout", err)
apps/api/internal/store/postgres.go
+2 -2
View File
@@ -27,11 +27,11 @@ type Store struct {
const (
postgresApplicationName = "easyai-ai-gateway"
postgresConnectTimeout = 5 * time.Second
postgresPoolWarmLimit = 64
)
type PostgresPoolOptions struct {
MaxConns int
MinIdleConns int
IdleInTransactionTimeout time.Duration
LockTimeout time.Duration
}
@@ -130,8 +130,8 @@ func postgresPoolConfigWithOptions(databaseURL string, options PostgresPoolOptio
}
if options.MaxConns > 0 {
config.MaxConns = int32(options.MaxConns)
config.MinIdleConns = int32(min(options.MaxConns, postgresPoolWarmLimit))
}
config.MinIdleConns = int32(min(max(options.MinIdleConns, 0), int(config.MaxConns)))
return config, nil
}
apps/api/internal/store/postgres_config_test.go
+11 -10
View File
@@ -27,7 +27,7 @@ func TestPostgresPoolConfigRejectsMalformedURL(t *testing.T) {
}
}
func TestPostgresPoolConfigWarmsBoundedIdleConnections(t *testing.T) {
func TestPostgresPoolConfigUsesExplicitMinimumIdleConnections(t *testing.T) {
config, err := postgresPoolConfig(
"postgresql://gateway:password@localhost:5432/gateway?sslmode=disable",
24,
@@ -35,23 +35,23 @@ func TestPostgresPoolConfigWarmsBoundedIdleConnections(t *testing.T) {
if err != nil {
t.Fatalf("parse PostgreSQL pool config: %v", err)
}
if config.MaxConns != 24 || config.MinIdleConns != 24 {
if config.MaxConns != 24 || config.MinIdleConns != 0 {
t.Fatalf(
"pool bounds max=%d minIdle=%d, want 24/24",
"pool bounds max=%d minIdle=%d, want 24/0",
config.MaxConns,
config.MinIdleConns,
)
}
small, err := postgresPoolConfig(
warmed, err := postgresPoolConfigWithOptions(
"postgresql://gateway:password@localhost:5432/gateway?sslmode=disable",
4,
PostgresPoolOptions{MaxConns: 24, MinIdleConns: 4},
)
if err != nil {
t.Fatalf("parse small PostgreSQL pool config: %v", err)
t.Fatalf("parse warmed PostgreSQL pool config: %v", err)
}
if small.MinIdleConns != 4 {
t.Fatalf("small pool minIdle=%d, want 4", small.MinIdleConns)
if warmed.MinIdleConns != 4 {
t.Fatalf("warmed pool minIdle=%d, want 4", warmed.MinIdleConns)
}
}
@@ -60,6 +60,7 @@ func TestPostgresPoolConfigSetsBoundedTransactionTimeouts(t *testing.T) {
"postgresql://gateway:password@localhost:5432/gateway?sslmode=disable",
PostgresPoolOptions{
MaxConns: 32,
MinIdleConns: 4,
IdleInTransactionTimeout: 60 * time.Second,
LockTimeout: 30 * time.Second,
},
@@ -73,8 +74,8 @@ func TestPostgresPoolConfigSetsBoundedTransactionTimeouts(t *testing.T) {
if got := config.ConnConfig.RuntimeParams["lock_timeout"]; got != "30000ms" {
t.Fatalf("lock timeout = %q, want 30000ms", got)
}
if config.MaxConns != 32 || config.MinIdleConns != 32 {
t.Fatalf("pool bounds max=%d minIdle=%d, want 32/32", config.MaxConns, config.MinIdleConns)
if config.MaxConns != 32 || config.MinIdleConns != 4 {
t.Fatalf("pool bounds max=%d minIdle=%d, want 32/4", config.MaxConns, config.MinIdleConns)
}
}
deploy/kubernetes/easyai-ai-gateway-cluster-release
+16 -3
View File
@@ -19,6 +19,8 @@ source "$config_file"
: "${AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT:=24}"
: "${AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT:=48}"
: "${AI_GATEWAY_DATABASE_MAX_CONNS:=32}"
: "${AI_GATEWAY_API_DATABASE_MAX_CONNS:=48}"
: "${AI_GATEWAY_DATABASE_MIN_IDLE_CONNS:=4}"
: "${AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY:=24}"
: "${AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY:=24}"
[[ $RELEASES_DIR == /* && $NAMESPACE =~ ^[a-z0-9-]+$ ]] || {
@@ -31,6 +33,7 @@ for capacity_value in \
"$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT" \
"$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"$AI_GATEWAY_DATABASE_MAX_CONNS" \
"$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
"$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY" \
"$AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY"; do
[[ $capacity_value =~ ^[1-9][0-9]*$ ]] || {
@@ -38,11 +41,18 @@ for capacity_value in \
exit 1
}
done
[[ $AI_GATEWAY_DATABASE_MIN_IDLE_CONNS =~ ^[0-9]+$ ]] || {
echo 'database minimum idle connection configuration must be a non-negative integer' >&2
exit 1
}
(( AI_GATEWAY_WORKER_REPLICAS_NINGBO <= 16 &&
AI_GATEWAY_WORKER_REPLICAS_HONGKONG <= 16 &&
AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT <= 256 &&
AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT <= 512 &&
AI_GATEWAY_DATABASE_MAX_CONNS <= 256 &&
AI_GATEWAY_API_DATABASE_MAX_CONNS <= 256 &&
AI_GATEWAY_DATABASE_MIN_IDLE_CONNS <= AI_GATEWAY_DATABASE_MAX_CONNS &&
AI_GATEWAY_DATABASE_MIN_IDLE_CONNS <= AI_GATEWAY_API_DATABASE_MAX_CONNS &&
AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY <= 256 &&
AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY <= 1024 )) || {
echo 'worker capacity configuration exceeds the release safety bounds' >&2
@@ -127,6 +137,7 @@ rollout_worker_site() {
"AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_DATABASE_MAX_CONNS" \
"AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS" \
"AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY" \
"AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=$AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY"
"${kubectl[@]}" scale "deployment/easyai-worker-$site" -n "$NAMESPACE" \
@@ -146,7 +157,8 @@ rollout_api_capacity_site() {
local site=$1
"${kubectl[@]}" set env "deployment/easyai-api-$site" -n "$NAMESPACE" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_DATABASE_MAX_CONNS"
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
"AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS"
"${kubectl[@]}" rollout status "deployment/easyai-api-$site" -n "$NAMESPACE" --timeout=300s
}
@@ -158,7 +170,7 @@ apply_capacity_config() {
verify_site hongkong
verify_site ningbo
wait_for_url 'public readiness' "$PUBLIC_BASE_URL/api/v1/readyz" '"ok":true'
echo "production_capacity=PASS ningbo_replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO hongkong_replicas=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG instance_limit=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT global_limit=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT database_pool=$AI_GATEWAY_DATABASE_MAX_CONNS media_concurrency=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY"
echo "production_capacity=PASS ningbo_replicas=$AI_GATEWAY_WORKER_REPLICAS_NINGBO hongkong_replicas=$AI_GATEWAY_WORKER_REPLICAS_HONGKONG instance_limit=$AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT global_limit=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT worker_database_pool=$AI_GATEWAY_DATABASE_MAX_CONNS api_database_pool=$AI_GATEWAY_API_DATABASE_MAX_CONNS database_min_idle=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS media_concurrency=$AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY"
}
rollout_site() {
@@ -170,7 +182,8 @@ rollout_site() {
if [[ $api_changed == true ]]; then
"${kubectl[@]}" set env "deployment/easyai-api-$site" -n "$NAMESPACE" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_DATABASE_MAX_CONNS"
"AI_GATEWAY_DATABASE_MAX_CONNS=$AI_GATEWAY_API_DATABASE_MAX_CONNS" \
"AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$AI_GATEWAY_DATABASE_MIN_IDLE_CONNS"
"${kubectl[@]}" set image "deployment/easyai-api-$site" -n "$NAMESPACE" \
"api=$api_image"
"${kubectl[@]}" rollout status "deployment/easyai-api-$site" -n "$NAMESPACE" --timeout=300s
deploy/kubernetes/easyai-ai-gateway-cluster-release.conf.example
+2
View File
@@ -10,5 +10,7 @@ AI_GATEWAY_WORKER_REPLICAS_HONGKONG=1
AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=24
AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=48
AI_GATEWAY_DATABASE_MAX_CONNS=32
AI_GATEWAY_API_DATABASE_MAX_CONNS=48
AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=4
AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=24
AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=24
deploy/kubernetes/production/application.yaml
+10 -2
View File
@@ -53,7 +53,9 @@ spec:
- name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
value: "false"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
value: "48"
- name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
value: "4"
- name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
value: "16"
- name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
@@ -182,7 +184,9 @@ spec:
- name: AI_GATEWAY_ASYNC_QUEUE_WORKER_ENABLED
value: "false"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
value: "48"
- name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
value: "4"
- name: AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
value: "16"
- name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
@@ -312,6 +316,8 @@ spec:
value: "true"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
- name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
value: "4"
- name: AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT
value: "24"
- name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
@@ -443,6 +449,8 @@ spec:
value: "true"
- name: AI_GATEWAY_DATABASE_MAX_CONNS
value: "32"
- name: AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
value: "4"
- name: AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT
value: "24"
- name: AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
docs/operations/production-acceptance.md
+6
View File
@@ -64,6 +64,7 @@ WebP 和 4K 图片。128 组输入组合避免只命中相同缓存；每四个
```text
AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT
AI_GATEWAY_DATABASE_MAX_CONNS
AI_GATEWAY_DATABASE_MIN_IDLE_CONNS
AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY
AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY
AI_GATEWAY_WORKER_REPLICAS_NINGBO
@@ -79,6 +80,11 @@ AI_GATEWAY_WORKER_REPLICAS_HONGKONG
| P28 | 28 | 36 | 28 | 56 |
| P32 | 32 | 40 | 32 | 64 |
API Pod 使用独立发布配置 `AI_GATEWAY_API_DATABASE_MAX_CONNS`，生产同构验收固定为 `48`。
所有 API/Worker Pool 的 `AI_GATEWAY_DATABASE_MIN_IDLE_CONNS` 固定为 `4`，仅预热必要连接，
其余连接按负载弹性创建，避免四个 Pod 在空闲时就占满 PostgreSQL 连接预算。各档表中的
数据库池仍表示单个 Worker 的最大连接数。
每档依次运行全部五个模拟 Profile 三次。失败时恢复上一个已完整通过的档位，流量保持
`validation`，不会自动放开正式请求。
scripts/cluster/run-production-acceptance.sh
+8 -3
View File
@@ -667,6 +667,8 @@ mark_run_failed() {
apply_capacity_profile() {
local profile=$1
local slots pool media global
local api_pool=48
local min_idle=4
case $profile in
P24) slots=24; pool=32; media=24; global=48 ;;
P28) slots=28; pool=36; media=28; global=56 ;;
@@ -679,13 +681,15 @@ apply_capacity_profile() {
"AI_GATEWAY_ASYNC_WORKER_INSTANCE_HARD_LIMIT=$slots" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$global" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$pool" \
"AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$min_idle" \
"AI_GATEWAY_MEDIA_MATERIALIZATION_CONCURRENCY=$media" \
"AI_GATEWAY_MEDIA_REQUEST_CONCURRENCY=$media" >/dev/null
remote_kubectl rollout status "deployment/easyai-worker-$site" \
-n "$namespace" --timeout=300s
remote_kubectl set env "deployment/easyai-api-$site" -n "$namespace" \
"AI_GATEWAY_ASYNC_WORKER_HARD_LIMIT=$global" \
"AI_GATEWAY_DATABASE_MAX_CONNS=$pool" >/dev/null
"AI_GATEWAY_DATABASE_MAX_CONNS=$api_pool" \
"AI_GATEWAY_DATABASE_MIN_IDLE_CONNS=$min_idle" >/dev/null
remote_kubectl rollout status "deployment/easyai-api-$site" \
-n "$namespace" --timeout=300s
done
@@ -895,6 +899,7 @@ wait_for_memory_recovery() {
verify_runtime_gates() {
local expected_slots expected_global expected_pool
local expected_api_pool=48
case $active_profile in
P24) expected_slots=24; expected_pool=32; expected_global=48 ;;
P28) expected_slots=28; expected_pool=36; expected_global=56 ;;
@@ -1055,8 +1060,8 @@ WHERE status='active'
pool_max=$(awk '$1=="easyai_gateway_postgres_pool_max_connections" {print $2}' <<<"$metrics")
acquired=$(awk '$1=="easyai_gateway_postgres_pool_acquired_connections" {print $2}' <<<"$metrics")
idle=$(awk '$1=="easyai_gateway_postgres_pool_idle_connections" {print $2}' <<<"$metrics")
[[ ${pool_max%.*} == "$expected_pool" ]]
if [[ ${acquired%.*} == "$expected_pool" && ${idle%.*} == 0 ]]; then
[[ ${pool_max%.*} == "$expected_api_pool" ]]
if [[ ${acquired%.*} == "$expected_api_pool" && ${idle%.*} == 0 ]]; then
echo "API connection pool saturated for site $site" >&2
return 1
fi