feat(backend): admin 手动运维接真实 lifecycle/provision
管理后台节点页的「拉黑/恢复(draining/up)」与「一键换机(replace)」此前 注入的是 StubLifecycle/StubProvision(Ready()=false,UI 置灰、操作返回 服务不可用)。本次接真实实现: - RealLifecycle:复用 scheduler 的乐观锁状态流转(UPDATE nodes WHERE status=from + node_events + directory_version bump),手动运维与调度 自动循环共用同一条 canonical 路径;读当前状态作 from 守卫,幂等安全。 - RealProvision:委托 provision.Service.Replace 的 make-before-break 换机 (先拉起新节点再 drain 旧节点,容量不掉)。 - BuildServices:lifecycle 恒为真实;provision 在服务初始化成功时为真实, 否则回退 not-ready stub(无厂商凭证时点击换机给出明确错误)。 补 services_real_test.go:换机委托(空 UUID 起新编排)、错误透传、 非法目标状态拒绝、Ready 标志。server 全量 23 包测试通过。 e2e(真实换机)待机群 + 厂商凭证。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
@@ -2,10 +2,16 @@ package admin
|
||||
|
||||
import (
|
||||
"context"
|
||||
"database/sql"
|
||||
"errors"
|
||||
"fmt"
|
||||
"strconv"
|
||||
"time"
|
||||
|
||||
"github.com/wangjia/pangolin/server/internal/codes"
|
||||
"github.com/wangjia/pangolin/server/internal/provision"
|
||||
"github.com/wangjia/pangolin/server/internal/scheduler"
|
||||
"github.com/wangjia/pangolin/server/internal/scheduler/orchestrate"
|
||||
)
|
||||
|
||||
// ErrServiceUnavailable is returned by stub services whose real implementation
|
||||
@@ -134,6 +140,55 @@ func (StubLifecycle) TransitionStatus(context.Context, int64, string, string) er
|
||||
// Ready reports false.
|
||||
func (StubLifecycle) Ready() bool { return false }
|
||||
|
||||
// RealLifecycle is the SQL-backed lifecycle service. It reuses the scheduler's
|
||||
// canonical optimistic-lock transition (UPDATE nodes ... WHERE status=from, plus
|
||||
// a node_events row and a directory_version bump so clients re-fetch), so manual
|
||||
// admin transitions and the autonomous scheduler loop share one code path.
|
||||
type RealLifecycle struct {
|
||||
db *sql.DB
|
||||
lc orchestrate.LifecycleService
|
||||
}
|
||||
|
||||
// NewRealLifecycle wires the lifecycle adapter over db.
|
||||
func NewRealLifecycle(db *sql.DB) *RealLifecycle {
|
||||
return &RealLifecycle{
|
||||
db: db,
|
||||
lc: scheduler.NewOrchestrateLifecycle(scheduler.NewSQLLifecycle(db, nil)),
|
||||
}
|
||||
}
|
||||
|
||||
// TransitionStatus moves nodeID to target ("draining" | "up"). It reads the
|
||||
// current status as the optimistic-lock guard, so the transition is a safe
|
||||
// no-op if the node already moved.
|
||||
func (l *RealLifecycle) TransitionStatus(ctx context.Context, nodeID int64, target, actor string) error {
|
||||
if target != "draining" && target != "up" {
|
||||
return fmt.Errorf("admin: unsupported target status %q", target)
|
||||
}
|
||||
var from string
|
||||
switch err := l.db.QueryRowContext(ctx, `SELECT status FROM nodes WHERE id = ?`, nodeID).Scan(&from); err {
|
||||
case nil:
|
||||
case sql.ErrNoRows:
|
||||
return fmt.Errorf("admin: node %d not found", nodeID)
|
||||
default:
|
||||
return err
|
||||
}
|
||||
if from == target {
|
||||
return nil // already in target state — idempotent no-op
|
||||
}
|
||||
n, err := l.lc.TransitionStatus(ctx, strconv.FormatInt(nodeID, 10), from, target,
|
||||
map[string]any{"actor": actor, "source": "admin"})
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if n == 0 {
|
||||
return fmt.Errorf("admin: node %d status changed concurrently", nodeID)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// Ready reports true — the lifecycle service only needs the database.
|
||||
func (*RealLifecycle) Ready() bool { return true }
|
||||
|
||||
// --------------------------------------------------------------------------
|
||||
// Node provisioning service (#14) — not yet implemented
|
||||
// --------------------------------------------------------------------------
|
||||
@@ -159,6 +214,34 @@ func (StubProvision) Replace(context.Context, int64, string) error { return ErrS
|
||||
// Ready reports false.
|
||||
func (StubProvision) Ready() bool { return false }
|
||||
|
||||
// provReplacer is the slice of provision.Service the admin replace control needs
|
||||
// (declared as an interface so tests can fake it without a database).
|
||||
type provReplacer interface {
|
||||
Replace(ctx context.Context, nodeID int64, replacementUUID string) (*provision.ReplaceResult, error)
|
||||
}
|
||||
|
||||
// RealProvision drives a one-click make-before-break node replacement via the
|
||||
// real provision.Service (doc/04 §4.1): a fresh node is brought UP before the
|
||||
// old one drains, so capacity never dips.
|
||||
type RealProvision struct {
|
||||
svc provReplacer
|
||||
}
|
||||
|
||||
// NewRealProvision wires the provision adapter over the real service.
|
||||
func NewRealProvision(svc provReplacer) *RealProvision { return &RealProvision{svc: svc} }
|
||||
|
||||
// Replace provisions a replacement for nodeID. An empty replacementUUID starts a
|
||||
// fresh orchestration (the service mints its own idempotency key). The actor is
|
||||
// recorded in the admin audit log by the caller.
|
||||
func (p *RealProvision) Replace(ctx context.Context, nodeID int64, _ string) error {
|
||||
_, err := p.svc.Replace(ctx, nodeID, "")
|
||||
return err
|
||||
}
|
||||
|
||||
// Ready reports true — the provision service is wired. Without vendor
|
||||
// credentials the Replace call fails with a clear error when invoked.
|
||||
func (*RealProvision) Ready() bool { return true }
|
||||
|
||||
// Services bundles the three downstream services the admin UI depends on.
|
||||
type Services struct {
|
||||
Codes CodesService
|
||||
|
||||
Reference in New Issue
Block a user