将身份协调和安全事件心跳改为 PostgreSQL 单 Leader 执行,并从独立 Worker 进程中移除身份运行时,避免多副本重复写同一状态。 为安全事件事务增加锁等待、空闲事务超时及独立回滚上下文;Worker 需连续丢失六次心跳后才判定失效,降低跨节点抖动导致的容量反复扩缩。 验证:Go 全量测试、go vet、Race 聚焦测试、PostgreSQL 18 Leader/安全事件/Worker 分配集成测试及迁移测试通过。
168 lines
4.5 KiB
Go
168 lines
4.5 KiB
Go
package store
|
|
|
|
import (
|
|
"context"
|
|
"errors"
|
|
"strings"
|
|
"time"
|
|
|
|
"github.com/jackc/pgx/v5"
|
|
)
|
|
|
|
const workerHeartbeatStaleAfter = 30 * time.Second
|
|
|
|
type WorkerRegistrationInput struct {
|
|
InstanceID string
|
|
PodUID string
|
|
PodName string
|
|
Site string
|
|
Revision string
|
|
DesiredCapacity int
|
|
HeartbeatStaleAfter time.Duration
|
|
}
|
|
|
|
type WorkerAllocation struct {
|
|
InstanceID string
|
|
DesiredCapacity int
|
|
Allocated int
|
|
ActiveInstances int
|
|
HeartbeatAt time.Time
|
|
}
|
|
|
|
func (s *Store) RegisterWorkerInstance(ctx context.Context, input WorkerRegistrationInput) (WorkerAllocation, error) {
|
|
input.InstanceID = strings.TrimSpace(input.InstanceID)
|
|
if input.InstanceID == "" {
|
|
return WorkerAllocation{}, errors.New("worker instance ID is required")
|
|
}
|
|
if input.DesiredCapacity < 0 {
|
|
return WorkerAllocation{}, errors.New("worker desired capacity cannot be negative")
|
|
}
|
|
staleAfter := input.HeartbeatStaleAfter
|
|
if staleAfter < workerHeartbeatStaleAfter {
|
|
staleAfter = workerHeartbeatStaleAfter
|
|
}
|
|
tx, err := s.pool.Begin(ctx)
|
|
if err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
defer tx.Rollback(ctx)
|
|
if _, err := tx.Exec(ctx, `SELECT pg_advisory_xact_lock(hashtextextended('gateway-worker-capacity', 0))`); err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
if _, err := tx.Exec(ctx, `
|
|
INSERT INTO gateway_worker_instances (
|
|
instance_id, pod_uid, pod_name, site, revision, status,
|
|
desired_capacity, allocated_capacity, started_at, heartbeat_at, updated_at
|
|
)
|
|
VALUES ($1, $2, $3, $4, $5, 'active', $6, 0, now(), now(), now())
|
|
ON CONFLICT (instance_id) DO UPDATE
|
|
SET pod_uid = EXCLUDED.pod_uid,
|
|
pod_name = EXCLUDED.pod_name,
|
|
site = EXCLUDED.site,
|
|
revision = EXCLUDED.revision,
|
|
status = 'active',
|
|
desired_capacity = EXCLUDED.desired_capacity,
|
|
heartbeat_at = now(),
|
|
updated_at = now()`,
|
|
input.InstanceID,
|
|
strings.TrimSpace(input.PodUID),
|
|
strings.TrimSpace(input.PodName),
|
|
strings.TrimSpace(input.Site),
|
|
strings.TrimSpace(input.Revision),
|
|
input.DesiredCapacity,
|
|
); err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
if _, err := tx.Exec(ctx, `
|
|
UPDATE gateway_worker_instances
|
|
SET allocated_capacity = 0,
|
|
updated_at = now()
|
|
WHERE status <> 'active'
|
|
OR heartbeat_at <= now() - $1::interval`, staleAfter.String()); err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
|
|
rows, err := tx.Query(ctx, `
|
|
SELECT instance_id
|
|
FROM gateway_worker_instances
|
|
WHERE status = 'active'
|
|
AND heartbeat_at > now() - $1::interval
|
|
ORDER BY instance_id ASC
|
|
FOR UPDATE`, staleAfter.String())
|
|
if err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
activeIDs := make([]string, 0)
|
|
for rows.Next() {
|
|
var instanceID string
|
|
if err := rows.Scan(&instanceID); err != nil {
|
|
rows.Close()
|
|
return WorkerAllocation{}, err
|
|
}
|
|
activeIDs = append(activeIDs, instanceID)
|
|
}
|
|
if err := rows.Err(); err != nil {
|
|
rows.Close()
|
|
return WorkerAllocation{}, err
|
|
}
|
|
rows.Close()
|
|
if len(activeIDs) == 0 {
|
|
return WorkerAllocation{}, errors.New("worker registration was not active after heartbeat")
|
|
}
|
|
|
|
base := input.DesiredCapacity / len(activeIDs)
|
|
remainder := input.DesiredCapacity % len(activeIDs)
|
|
allocated := 0
|
|
for index, instanceID := range activeIDs {
|
|
capacity := base
|
|
if index < remainder {
|
|
capacity++
|
|
}
|
|
if _, err := tx.Exec(ctx, `
|
|
UPDATE gateway_worker_instances
|
|
SET desired_capacity = $2,
|
|
allocated_capacity = $3,
|
|
updated_at = now()
|
|
WHERE instance_id = $1`, instanceID, input.DesiredCapacity, capacity); err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
if instanceID == input.InstanceID {
|
|
allocated = capacity
|
|
}
|
|
}
|
|
var heartbeatAt time.Time
|
|
if err := tx.QueryRow(ctx, `
|
|
SELECT heartbeat_at
|
|
FROM gateway_worker_instances
|
|
WHERE instance_id = $1`, input.InstanceID).Scan(&heartbeatAt); err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
if err := tx.Commit(ctx); err != nil {
|
|
return WorkerAllocation{}, err
|
|
}
|
|
return WorkerAllocation{
|
|
InstanceID: input.InstanceID,
|
|
DesiredCapacity: input.DesiredCapacity,
|
|
Allocated: allocated,
|
|
ActiveInstances: len(activeIDs),
|
|
HeartbeatAt: heartbeatAt,
|
|
}, nil
|
|
}
|
|
|
|
func (s *Store) MarkWorkerDraining(ctx context.Context, instanceID string) error {
|
|
result, err := s.pool.Exec(ctx, `
|
|
UPDATE gateway_worker_instances
|
|
SET status = 'draining',
|
|
allocated_capacity = 0,
|
|
heartbeat_at = now(),
|
|
updated_at = now()
|
|
WHERE instance_id = $1`, strings.TrimSpace(instanceID))
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if result.RowsAffected() == 0 {
|
|
return pgx.ErrNoRows
|
|
}
|
|
return nil
|
|
}
|