diff --git a/api/database/v1alpha1/postgresqldatabase_types.go b/api/database/v1alpha1/postgresqldatabase_types.go index faef6b5..5a10305 100644 --- a/api/database/v1alpha1/postgresqldatabase_types.go +++ b/api/database/v1alpha1/postgresqldatabase_types.go @@ -38,6 +38,16 @@ type PostgreSQLDatabaseStatus struct { // +kubebuilder:validation:Minimum=1 // +optional CredentialVersion int64 `json:"credentialVersion,omitempty"` + // RoleOID 只在角色创建提交并回读成功后保存,后续不按同名认领。 + // +kubebuilder:validation:Minimum=1 + // +kubebuilder:validation:Maximum=4294967295 + // +optional + RoleOID int64 `json:"roleOID,omitempty"` + // DatabaseOID 只在数据库创建并回读成功后保存;缺失不代表允许认领已有对象。 + // +kubebuilder:validation:Minimum=1 + // +kubebuilder:validation:Maximum=4294967295 + // +optional + DatabaseOID int64 `json:"databaseOID,omitempty"` // Phase 暂不冻结供应子阶段枚举;它不是操作授权或绑定的替代记录。 // +optional Phase string `json:"phase,omitempty"` @@ -50,6 +60,10 @@ type PostgreSQLDatabaseStatus struct { // +kubebuilder:object:root=true // +kubebuilder:subresource:status // +kubebuilder:resource:scope=Cluster +// +kubebuilder:validation:XValidation:rule="!has(self.status) || !has(self.status.roleOID) || (has(self.status.credentialVersion) && has(self.status.instanceUID))",message="roleOID requires confirmed credentials and instanceUID" +// +kubebuilder:validation:XValidation:rule="!has(self.status) || !has(self.status.databaseOID) || has(self.status.roleOID)",message="databaseOID requires roleOID" +// +kubebuilder:validation:XValidation:rule="!(has(oldSelf.status) && has(oldSelf.status.roleOID)) || (has(self.status) && has(self.status.roleOID) && self.status.roleOID == oldSelf.status.roleOID)",message="confirmed roleOID cannot change or be removed" +// +kubebuilder:validation:XValidation:rule="!(has(oldSelf.status) && has(oldSelf.status.databaseOID)) || (has(self.status) && has(self.status.databaseOID) && self.status.databaseOID == oldSelf.status.databaseOID)",message="confirmed databaseOID cannot change or be removed" // +kubebuilder:validation:XValidation:rule="!has(self.status) || !has(self.status.credentialVersion) || has(self.status.credentialRef)",message="credentialVersion requires credentialRef" // +kubebuilder:validation:XValidation:rule="!(has(oldSelf.status) && has(oldSelf.status.credentialRef)) || (has(self.status) && has(self.status.credentialRef) && self.status.credentialRef == oldSelf.status.credentialRef)",message="recorded credentialRef cannot change or be removed" // +kubebuilder:validation:XValidation:rule="!(has(oldSelf.status) && has(oldSelf.status.credentialVersion)) || (has(self.status) && has(self.status.credentialVersion) && self.status.credentialVersion == oldSelf.status.credentialVersion)",message="confirmed credentialVersion cannot change or be removed" diff --git a/api/database/v1alpha1/resource_status_test.go b/api/database/v1alpha1/resource_status_test.go new file mode 100644 index 0000000..1caca5d --- /dev/null +++ b/api/database/v1alpha1/resource_status_test.go @@ -0,0 +1,40 @@ +package v1alpha1_test + +import ( + "testing" + + databasev1alpha1 "git.ddupan.top/panxiao81/ayatori/api/database/v1alpha1" + apierrors "k8s.io/apimachinery/pkg/api/errors" + "sigs.k8s.io/controller-runtime/pkg/client" +) + +func testResourceStatus(t *testing.T, api client.Client) { + object := validDatabase("resource-status") + if err := api.Create(t.Context(), object); err != nil { + t.Fatal(err) + } + invalid := object.DeepCopy() + invalid.Status.RoleOID = 11 + if err := api.Status().Update(t.Context(), invalid); !apierrors.IsInvalid(err) { + t.Fatal("无凭据不能确认角色", err) + } + object.Status.InstanceUID = "instance-uid" + object.Status.CredentialRef = &databasev1alpha1.CredentialReference{Mount: "application-secrets", Path: "applications/test"} + object.Status.CredentialVersion = 1 + object.Status.RoleOID, object.Status.DatabaseOID = 11, 22 + if err := api.Status().Update(t.Context(), object); err != nil { + t.Fatal(err) + } + for _, change := range []func(*databasev1alpha1.PostgreSQLDatabase){ + func(d *databasev1alpha1.PostgreSQLDatabase) { d.Status.RoleOID = 0 }, + func(d *databasev1alpha1.PostgreSQLDatabase) { d.Status.RoleOID = 12 }, + func(d *databasev1alpha1.PostgreSQLDatabase) { d.Status.DatabaseOID = 0 }, + func(d *databasev1alpha1.PostgreSQLDatabase) { d.Status.DatabaseOID = 23 }, + } { + invalid = object.DeepCopy() + change(invalid) + if err := api.Status().Update(t.Context(), invalid); !apierrors.IsInvalid(err) { + t.Fatal("已确认 OID 不得覆盖或移除", err) + } + } +} diff --git a/api/database/v1alpha1/validation_test.go b/api/database/v1alpha1/validation_test.go index 388f592..3ab5e72 100644 --- a/api/database/v1alpha1/validation_test.go +++ b/api/database/v1alpha1/validation_test.go @@ -62,6 +62,7 @@ func TestDatabaseAPI(t *testing.T) { t.Run("拒绝非法声明", func(t *testing.T) { testInvalidDeclarations(t, client) }) t.Run("status隔离和绑定并发", func(t *testing.T) { testBindingWrites(t, client) }) t.Run("凭据位置与确认版本", func(t *testing.T) { testCredentialStatus(t, client) }) + t.Run("资源创建确认", func(t *testing.T) { testResourceStatus(t, client) }) t.Run("仓库示例", func(t *testing.T) { testSamples(t, client, scheme) }) } diff --git a/config/crd/bases/database.ayatori.ddupan.top_postgresqldatabases.yaml b/config/crd/bases/database.ayatori.ddupan.top_postgresqldatabases.yaml index 8ff07a1..ae55970 100644 --- a/config/crd/bases/database.ayatori.ddupan.top_postgresqldatabases.yaml +++ b/config/crd/bases/database.ayatori.ddupan.top_postgresqldatabases.yaml @@ -221,6 +221,12 @@ spec: format: int64 minimum: 1 type: integer + databaseOID: + description: DatabaseOID 只在数据库创建并回读成功后保存;缺失不代表允许认领已有对象。 + format: int64 + maximum: 4294967295 + minimum: 1 + type: integer instanceUID: description: InstanceUID 记录观察时的实例身份,不把同名新实例视为原目标。 type: string @@ -230,11 +236,28 @@ spec: phase: description: Phase 暂不冻结供应子阶段枚举;它不是操作授权或绑定的替代记录。 type: string + roleOID: + description: RoleOID 只在角色创建提交并回读成功后保存,后续不按同名认领。 + format: int64 + maximum: 4294967295 + minimum: 1 + type: integer type: object required: - spec type: object x-kubernetes-validations: + - message: roleOID requires confirmed credentials and instanceUID + rule: '!has(self.status) || !has(self.status.roleOID) || (has(self.status.credentialVersion) + && has(self.status.instanceUID))' + - message: databaseOID requires roleOID + rule: '!has(self.status) || !has(self.status.databaseOID) || has(self.status.roleOID)' + - message: confirmed roleOID cannot change or be removed + rule: '!(has(oldSelf.status) && has(oldSelf.status.roleOID)) || (has(self.status) + && has(self.status.roleOID) && self.status.roleOID == oldSelf.status.roleOID)' + - message: confirmed databaseOID cannot change or be removed + rule: '!(has(oldSelf.status) && has(oldSelf.status.databaseOID)) || (has(self.status) + && has(self.status.databaseOID) && self.status.databaseOID == oldSelf.status.databaseOID)' - message: credentialVersion requires credentialRef rule: '!has(self.status) || !has(self.status.credentialVersion) || has(self.status.credentialRef)' - message: recorded credentialRef cannot change or be removed diff --git a/docs/database/README.md b/docs/database/README.md index 547b91f..95fc5cf 100644 --- a/docs/database/README.md +++ b/docs/database/README.md @@ -1,7 +1,7 @@ # Database 模块 Database 是 Ayatori 首批实际产品领域之一。当前已包含三资源 API、分层绑定与 Instance 原生 -管理能力观测;尚未完成 Database 供应/导入、Tenant 凭据交付与资源回收链路。 +管理能力观测、凭据准备及显式启用的角色/数据库创建;扩展、导入、Tenant 交付与回收尚未完成。 ## 当前设计(2026-09-24) @@ -12,7 +12,7 @@ Retain 后人工重新绑定与资源侧 Delete。撤销 PostgreSQL ownership re 依据 [ADR-0009](../decisions/0009-database-resource-and-claim.md),当前合同见 [系统规格](specification.md)。下面的迁移来源与已存在代码不反向约束新设计。 registry adapter、专属迁移/测试及 Instance 的 registry 判定现已撤除;Instance 根据完整管理 -能力观察直接判定 Ready。Database 资源与绑定已接入,导入、角色/凭据供应及回收仍未完成。wiki 同步位置见 +能力观察直接判定 Ready。Database 资源与绑定、角色/凭据供应已接入,导入及回收仍未完成。wiki 同步位置见 `homelab-wiki/services/postgresql-tenant-operator.md`,跨仓库发布状态由 wiki 的同步记录维护。 ## 来源基线 @@ -201,14 +201,57 @@ application 保留 I/O 顺序、消费方接口及并发快照,不再复用绑 同样可能保守地要求人工处理,不承诺无损接续;多副本部署应启用既有 leader election。 成功只设置 `CredentialsReady=True`,Database Ready 仍为 False/ProvisioningIncomplete, -Tenant 仍未完成交付。本切片没有 PostgreSQL role/database 创建、扩展安装、ESO 投射、 -Retain 释放或 Delete 清理,也不会解除 finalizer。不要作为完整 DBaaS 部署。 +Tenant 仍未完成交付。凭据准备本身不创建 PostgreSQL 资源;后续创建需另行启用下述用例。 +扩展安装、ESO 投射、Retain 释放或 Delete 清理尚未接入,也不会解除 finalizer。不要作为完整 DBaaS 部署。 单元测试穷举前置条件;真实 API server + 隔离 Bao 验证创建、状态确认、幂等、重启、并发、 依赖恢复、固定位置、不确定结果、确认保存失败和删除边界;实际 manager 验证 watch 驱动 及重启。该 fixture 只声明 Instance 前置 Ready,真实 PostgreSQL 管理能力由既有 Instance 集成测试覆盖,不把凭据准备验收当成实际建库或应用登录验收。 +## PostgreSQL 资源创建闭环 + +在管理 Secret、OpenBao 认证和凭据准备配置之上,显式设置 `--database-provision-resources` +才启用外部角色/数据库写入;默认关闭。升级时先安装新 CRD,供应字段见 +[API 合同](api-reference.md#当前-api-切片)。只处理 Provision 来源,导入资源不进入创建流程。 + +`DatabaseProvisioning` 在每轮验证双向绑定、UID、删除/Released 状态、保护和当前 Instance +Ready,读取固定位置的已确认凭据。所有 SQL 操作复用 InstanceService 已有的管理连接及 +Secret 刷新机制,不再开一个管理连接池。执行前重新查询非 superuser 管理能力;外部操作 +前后直接回读 API 快照,发生并发变更则停止确认,不盲目覆盖 status。 +启用资源创建时,`DatabaseReconciliation` 在同一条 reconcile 中先运行凭据准备/检查, +再运行资源供应用例,不同时注册独立凭据 controller,避免两个 worker 争写同一资源状态。 + +创建分为三个可观察步骤: + +1. 保存 CreatingRole,事务内创建无管理特权的 LOGIN owner 并授予管理账号 SET 权限; + 提交并回读成功后保存 `status.roleOID`。 +2. 保存 CreatingDatabase,创建以已确认角色为 owner 且 `ALLOW_CONNECTIONS false` 的数据库; + 回读成功后保存 `status.databaseOID`。 +3. 在已确认对象上以 owner 收紧 ACL:撤销 PUBLIC CONNECT、授予 owner CONNECT,再开放 + 连接入口。ACL 收敛可幂等重试,不重置密码或更换 owner。 + +OID 只保存在 CR,表示本次成功创建的回读身份,不是 registry 或认领机制。未知同名对象、 +已确认对象消失/被重建、owner/角色特权漂移均报 Conflict。CreatingRole/CreatingDatabase +未留下对应 OID 时,下轮保守停止;即使请求可能尚未发出,也不尝试推断或自动补记。 +网络结果不确定、创建成功后确认写入失败同样交给人工处理。诊断包含 Database、Instance、 +实际 database/role、失败步骤;确认 OID 独立保留。依赖或明确权限拒绝可等待恢复。 +OID 不是跨集群/备份恢复的稳定身份,恢复后须人工核对,不能靠匹配 OID 推导管理权。 + +`ResourcesReady=True` 仅表示角色、数据库及连接 ACL 已确认,不代表扩展或 Tenant 交付完成; +Database/Tenant 总体 Ready 仍为 False。当前不轮换密码、不做运行期应用登录健康检查,也不 +撤销其他数据库的 PUBLIC 权限。管理员须保证共享实例中其他数据库的接入策略满足隔离要求。 + +真实 API server + PostgreSQL + OpenBao 测试覆盖实际密码登录、owner 建表、应用管理权限拒绝、 +无关账号连接拒绝、重试/重启、权限恢复、同名冲突、角色重建、并发授权、创建结果丢失、 +确认持久化失败、删除停止与实际 manager 的观察链路。保留外部操作/API 写入间的非原子边界, +不承诺控制面与数据库之间的事务或对任意管理员并发 DDL 的无损恢复。 + +遵循 PostgreSQL 官方的 +[CREATE DATABASE 非事务与 owner 权限合同](https://www.postgresql.org/docs/17/sql-createdatabase.html) +和 [CREATEROLE 的成员授权](https://www.postgresql.org/docs/17/role-attributes.html)。 +不新增 controller 通用生命周期框架,继续采用既有直接观察、resourceVersion 保护和条件报告。 + ## OpenBao Kubernetes 认证会话 公共 `internal/infra/openbao.KubernetesSession` 复用官方 Kubernetes auth helper 和 `LifetimeWatcher` diff --git a/docs/database/api-reference.md b/docs/database/api-reference.md index befafea..60c415b 100644 --- a/docs/database/api-reference.md +++ b/docs/database/api-reference.md @@ -2,9 +2,9 @@ | 项目 | 内容 | | --- | --- | -| 状态 | API schema 与绑定 controller 已实现;供应、交付与删除清理未接入 | +| 状态 | API、绑定、凭据准备及角色/数据库创建已实现;扩展、导入、交付与删除清理未接入 | | API group/version | `database.ayatori.ddupan.top/v1alpha1` | -| 最后更新 | 2026-09-27 | +| 最后更新 | 2026-09-29 | 以 [系统规格](specification.md) 与 [ADR-0009](../decisions/0009-database-resource-and-claim.md) 为准。类型与生成的 CRD 已纳入源码, @@ -26,6 +26,7 @@ Go 类型位于 `api/database/v1alpha1`,CRD 随 `config/crd` 发布;manager | Database | `status.instanceUID` | 观察时的 Instance 身份 | | Database | `status.credentialRef.mount/path` | 首次写入前固定的 KV v2 位置,不随部署配置迁移 | | Database | `status.credentialVersion` | 创建并回读成功后确认的正整数版本;省略表示未确认 | +| Database | `status.roleOID`、`status.databaseOID` | 分别创建并回读成功后确认的 PostgreSQL OID;不能凭同名补记 | | Tenant | `spec.provision.instanceRef.name` | 动态申请来源,与 `spec.databaseRef` 互斥且必须二选一 | | Tenant | `spec.provision.database/loginRole` | 可省略,语义默认值由 controller 解析,不由 CRD 推导 | | Tenant | `spec.databaseRef.name` | 显式申请已有 Database,不额外指定 Instance | @@ -47,6 +48,13 @@ KV 版本,删除或版本漂移均需人工处理,不回退旧版本或生 `CreationStarted` 且无确认版本表示创建未完成确认,重入时停在 Conflict;不尝试推断 进程中断前请求是否发出。完整执行和测试边界见[凭据准备闭环](README.md#凭据准备闭环)。 +角色/数据库 OID 为 1–4294967295,写入后不可修改或移除。角色确认要求已有凭据确认与 +Instance UID;数据库确认要求先有角色确认。`ResourcesReady` 的 CreatingRole/CreatingDatabase +没有相应 OID 时,下轮转 Conflict;依赖故障不清空确认。旧 CRD 裁剪 OID 会阻止继续下一步, +但不能回滚已发出的外部创建,因此必须先升级 CRD 再启用供应。 +`ResourcesReady=True/Available` 只表示角色、数据库和 ACL 已确认,总体 Ready 仍不放行。 +详见[资源创建闭环](README.md#postgresql-资源创建闭环)。 + 示例:[Instance](../../config/samples/database_v1alpha1_postgresqlinstance.yaml)、 [导入 Database](../../config/samples/database_v1alpha1_postgresqldatabase.yaml)、 [动态/已有资源申请](../../config/samples/database_v1alpha1_postgresqltenant.yaml)。 @@ -131,7 +139,7 @@ Database 是平台管理的集群级资源,不归属于应用 namespace,也 普通申请者通过 Tenant 申请使用,不能自行修改 Database 回收策略或将 Released 资源重新开放; 这些资源管理操作由平台管理员授权。controller 的绑定协调权限与用户申请权限分别配置。 -以下是行为合同,具体 schema 见当前 API 切片与生成的 CRD;后端行为尚未实现: +以下是行为合同,具体 schema 见当前 API 切片与生成的 CRD;导入、交付与回收尚未实现: | 内容 | 合同 | | --- | --- | diff --git a/docs/database/deployment.md b/docs/database/deployment.md index e8d25f3..0ebd18d 100644 --- a/docs/database/deployment.md +++ b/docs/database/deployment.md @@ -1,6 +1,6 @@ # 部署与配置 -> 本页区分已实现的 Instance 观测、认证和凭据准备配置,与尚未接入的 PostgreSQL 供应/交付合同。 +> 本页区分已实现的 Instance 观测、认证、凭据准备和资源创建配置,与尚未接入的完整供应/交付合同。 > 完整 Database 服务仍不可部署使用;当前可执行入口见 [模块说明](README.md)。 | 项目 | 内容 | @@ -46,6 +46,7 @@ Instance 观测)与 `--database-root-cert`(公开 PostgreSQL CA PEM 路径 | `--openbao-auth-role` | 已实现,启用时必填 | OpenBao 登录 role | | `--openbao-ca-cert` | 已实现,默认系统信任根 | OpenBao 公开 CA PEM 路径 | | `--database-credential-mount` | 已实现,默认空 | 显式设置后启用应用凭据准备,要求已配置 OpenBao 认证 | +| `--database-provision-resources` | 已实现,默认 false | 启用角色/数据库及 ACL 创建;要求管理 Secret namespace、凭据 mount 和认证配置,先升级 CRD | | `--openbao-service-account-namespace` | 已实现,启用时必填 | TokenRequest 目标 SA 的固定 namespace | | `--openbao-service-account-name` | 已实现,启用时必填 | TokenRequest 目标 SA 名称 | | `--openbao-token-audience` | 已实现,`openbao` | SA JWT audience,必须匹配 OpenBao role | diff --git a/internal/bootstrap/database.go b/internal/bootstrap/database.go index d94af8f..b23d59c 100644 --- a/internal/bootstrap/database.go +++ b/internal/bootstrap/database.go @@ -17,10 +17,11 @@ import ( ) type databaseOptions struct { - secretNamespace string - rootCert string - credentialMount string - credentialPrefix string + secretNamespace string + rootCert string + credentialMount string + credentialPrefix string + provisionResources bool } func (o *databaseOptions) bindFlags(flags *flag.FlagSet) { @@ -29,6 +30,7 @@ func (o *databaseOptions) bindFlags(flags *flag.FlagSet) { flags.StringVar(&o.rootCert, "database-root-cert", "", "PostgreSQL 管理连接信任的公开 CA bundle 路径") flags.StringVar(&o.credentialMount, "database-credential-mount", "", "应用凭据 KV v2 mount;为空时不启用凭据准备") flags.StringVar(&o.credentialPrefix, "database-credential-prefix", "applications", "应用凭据路径前缀;已有固定位置不随配置变化迁移") + flags.BoolVar(&o.provisionResources, "database-provision-resources", false, "启用 PostgreSQL 角色和数据库创建;需要 Instance 管理凭据及 OpenBao 凭据准备") } func (o databaseOptions) configureManager(options *ctrl.Options) { @@ -40,25 +42,30 @@ func (o databaseOptions) configureManager(options *ctrl.Options) { // registerDatabaseControllers 封装 Database 的内部装配,并返回在 manager 停止后执行的清理。 func registerDatabaseControllers(ctx context.Context, manager ctrl.Manager, options databaseOptions, baoClient *bao.Client) (func(), error) { closeDatabaseConnections := func() {} + var backend *application.InstanceService if options.secretNamespace != "" { service, err := setupInstanceObservation(manager, options.secretNamespace, options.rootCert) if err != nil { return nil, fmt.Errorf("set up Instance observation: %w", err) } closeDatabaseConnections = service.Close + backend = service } if err := wireBindingController(manager.GetClient(), manager.GetAPIReader()).SetupWithManager(ctx, manager); err != nil { closeDatabaseConnections() return nil, fmt.Errorf("set up Database binding controller: %w", err) } - if err := setupCredentialPreparation(manager, options, baoClient); err != nil { + if err := registerDatabaseSupplyController(manager, options, baoClient, backend); err != nil { closeDatabaseConnections() - return nil, fmt.Errorf("set up Database credential preparation: %w", err) + return nil, fmt.Errorf("register Database supply controller: %w", err) } return closeDatabaseConnections, nil } -func setupCredentialPreparation(manager ctrl.Manager, options databaseOptions, baoClient *bao.Client) error { +func registerDatabaseSupplyController(manager ctrl.Manager, options databaseOptions, baoClient *bao.Client, backend *application.InstanceService) error { + if options.provisionResources && (backend == nil || options.credentialMount == "") { + return fmt.Errorf("database resource provisioning requires management Secret namespace and credential mount") + } if options.credentialMount == "" { return nil } @@ -69,6 +76,9 @@ func setupCredentialPreparation(manager ctrl.Manager, options databaseOptions, b if err != nil { return err } + if options.provisionResources { + return wireProvisioningController(manager.GetClient(), manager.GetAPIReader(), store, backend).SetupWithManager(manager) + } return wireCredentialController(manager.GetClient(), manager.GetAPIReader(), store).SetupWithManager(manager) } diff --git a/internal/bootstrap/database_wiring.go b/internal/bootstrap/database_wiring.go index e445ba9..edc215e 100644 --- a/internal/bootstrap/database_wiring.go +++ b/internal/bootstrap/database_wiring.go @@ -26,3 +26,13 @@ func wireInstanceController(writer client.Client, reader client.Reader, observer service := &application.InstanceReconciliation{Resources: resources, Observer: observer} return databasecontroller.NewInstanceReconciler(writer, service, resources, namespace) } + +func wireProvisioningController(writer client.Client, reader client.Reader, store application.CredentialStore, backend application.ProvisioningBackend) *databasecontroller.ProvisioningReconciler { + resources := &kubernetes.ProvisioningResources{Client: writer, Reader: reader} + credentialResources := &kubernetes.CredentialResources{Client: writer, Reader: reader} + service := &application.DatabaseReconciliation{ + Credentials: &application.CredentialPreparation{Resources: credentialResources, Store: store}, + Provisioning: &application.DatabaseProvisioning{Resources: resources, Credentials: store, Backend: backend}, + } + return databasecontroller.NewProvisioningReconciler(writer, service) +} diff --git a/internal/bootstrap/database_wiring_test.go b/internal/bootstrap/database_wiring_test.go index 05ab9ac..58be259 100644 --- a/internal/bootstrap/database_wiring_test.go +++ b/internal/bootstrap/database_wiring_test.go @@ -39,6 +39,11 @@ func TestDatabaseExplicitWiring(t *testing.T) { if !ok || resources.Reader != directReader || resources.Client != writer || reconciler.Service.Observer != observer || reconciler.Presenter != resources { t.Fatal("Instance 的服务或读取边界未按依赖图注入") } + provisioner := wireProvisioningController(writer, directReader, nil, observer) + provisioningResources, ok := provisioner.Service.Provisioning.Resources.(*kubernetes.ProvisioningResources) + if !ok || provisioningResources.Client != writer || provisioningResources.Reader != directReader || provisioner.Service.Provisioning.Backend != observer { + t.Fatal("资源供应必须复用已有 Instance 管理服务和 API 客户端") + } } // 防止领域重新依赖用例/存储,也防止 controller 再次私自构造具体 adapter。 diff --git a/internal/bootstrap/manager_integration_test.go b/internal/bootstrap/manager_integration_test.go index 337afe2..8ab475c 100644 --- a/internal/bootstrap/manager_integration_test.go +++ b/internal/bootstrap/manager_integration_test.go @@ -67,7 +67,7 @@ func TestBootstrapWithRealAPIServer(t *testing.T) { t.Fatal(err) } options.database.credentialMount = "secret" - if err := setupCredentialPreparation(manager, options.database, fixtureClient); err != nil { + if err := registerDatabaseSupplyController(manager, options.database, fixtureClient, nil); err != nil { t.Fatal(err) } done := make(chan error, 1) diff --git a/internal/bootstrap/options_test.go b/internal/bootstrap/options_test.go index 2ddb410..2499041 100644 --- a/internal/bootstrap/options_test.go +++ b/internal/bootstrap/options_test.go @@ -63,12 +63,15 @@ func TestCredentialPreparationOptions(t *testing.T) { if options.database.credentialMount != "applications-kv" || options.database.credentialPrefix != "database" { t.Fatal("凭据准备参数未传入领域装配") } - if err := setupCredentialPreparation(nil, options.database, nil); err == nil { + if err := registerDatabaseSupplyController(nil, options.database, nil, nil); err == nil { t.Fatal("启用凭据准备必须有显式配置的认证 client") } - if err := setupCredentialPreparation(nil, databaseOptions{}, nil); err != nil { + if err := registerDatabaseSupplyController(nil, databaseOptions{}, nil, nil); err != nil { t.Fatal("默认停用凭据准备不应要求后端") } + if err := registerDatabaseSupplyController(nil, databaseOptions{provisionResources: true}, nil, nil); err == nil { + t.Fatal("启用资源供应必须有管理连接及凭据准备") + } } func TestManagerFlagOverrides(t *testing.T) { diff --git a/internal/database/adapter/kubernetes/provisioning_resources.go b/internal/database/adapter/kubernetes/provisioning_resources.go new file mode 100644 index 0000000..de20f16 --- /dev/null +++ b/internal/database/adapter/kubernetes/provisioning_resources.go @@ -0,0 +1,111 @@ +package kubernetes + +import ( + "context" + "fmt" + "reflect" + + databasev1alpha1 "git.ddupan.top/panxiao81/ayatori/api/database/v1alpha1" + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" + "k8s.io/apimachinery/pkg/api/equality" + apierrors "k8s.io/apimachinery/pkg/api/errors" + "k8s.io/apimachinery/pkg/api/meta" + metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" + "k8s.io/apimachinery/pkg/types" + "sigs.k8s.io/controller-runtime/pkg/client" +) + +const resourcesReadyCondition = "ResourcesReady" + +type ProvisioningResources struct { + Client client.Client + Reader client.Reader +} + +func (r *ProvisioningResources) Load(ctx context.Context, name string) (*application.ProvisioningRecord, error) { + // 同一组 API 事实复用映射;供应用例不持有凭据准备用例的快照或 repository。 + facts, err := (&CredentialResources{Client: r.Client, Reader: r.Reader}).Load(ctx, name) + if err != nil || facts == nil { + return nil, err + } + object := &databasev1alpha1.PostgreSQLDatabase{} + if err := r.Reader.Get(ctx, types.NamespacedName{Name: name}, object); err != nil { + return nil, err + } + if object.ResourceVersion != facts.Revision { + return nil, provisioningSnapshotConflict(name) + } + record := &application.ProvisioningRecord{Target: facts.Target, Revision: facts.Revision, TenantGeneration: facts.TenantGeneration, + InstanceGeneration: facts.InstanceGeneration, Credentials: facts.Status, State: provisioningStatus(object)} + if facts.Instance == nil { + return record, nil + } + instance := &databasev1alpha1.PostgreSQLInstance{} + if err := r.Reader.Get(ctx, types.NamespacedName{Name: facts.Database.Instance}, instance); err != nil { + return nil, err + } + if string(instance.UID) != facts.Instance.Identity.UID || instance.Generation != facts.InstanceGeneration || + (!instance.DeletionTimestamp.IsZero()) != facts.Instance.Deleting || currentReady(instance.Generation, instance.Status.Conditions) != facts.Instance.Ready { + return nil, provisioningSnapshotConflict(name) + } + mapped, err := instanceRecord(instance) + if err != nil { + return nil, err + } + record.InstanceTarget = mapped.Target + return record, nil +} + +func provisioningStatus(object *databasev1alpha1.PostgreSQLDatabase) provisioning.State { + state := provisioning.State{RoleOID: uint32(object.Status.RoleOID), DatabaseOID: uint32(object.Status.DatabaseOID), Phase: provisioning.Pending} + if condition := meta.FindStatusCondition(object.Status.Conditions, resourcesReadyCondition); condition != nil { + state.Phase, state.Message = provisioning.Phase(condition.Reason), condition.Message + } + return state +} + +func (r *ProvisioningResources) Save(ctx context.Context, record *application.ProvisioningRecord, state provisioning.State) (*application.ProvisioningRecord, error) { + object := &databasev1alpha1.PostgreSQLDatabase{} + if err := r.Reader.Get(ctx, types.NamespacedName{Name: record.Database.Identity.Name}, object); err != nil { + return nil, err + } + if object.ResourceVersion != record.Revision || string(object.UID) != record.Database.Identity.UID { + return nil, provisioningSnapshotConflict(object.Name) + } + previous := object.Status.DeepCopy() + object.Status.RoleOID, object.Status.DatabaseOID = int64(state.RoleOID), int64(state.DatabaseOID) + status := metav1.ConditionFalse + if state.Phase == provisioning.Available { + status = metav1.ConditionTrue + } + meta.SetStatusCondition(&object.Status.Conditions, metav1.Condition{Type: resourcesReadyCondition, Status: status, + Reason: string(state.Phase), Message: state.Message, ObservedGeneration: object.Generation}) + if !equality.Semantic.DeepEqual(*previous, object.Status) { + if err := r.Client.Status().Update(ctx, object); err != nil { + return nil, err + } + } + if object.Status.RoleOID != int64(state.RoleOID) || object.Status.DatabaseOID != int64(state.DatabaseOID) { + return nil, fmt.Errorf("资源 OID 未被 API 保留;请先升级 Database CRD,未继续供应") + } + updated := *record + updated.Revision, updated.State = object.ResourceVersion, provisioningStatus(object) + return &updated, nil +} + +func (r *ProvisioningResources) CheckCurrent(ctx context.Context, record *application.ProvisioningRecord) error { + current, err := r.Load(ctx, record.Database.Identity.Name) + if err != nil { + return err + } + if current == nil || !reflect.DeepEqual(current, record) { + return provisioningSnapshotConflict(record.Database.Identity.Name) + } + return nil +} + +func provisioningSnapshotConflict(name string) error { + return apierrors.NewConflict(databasev1alpha1.GroupVersion.WithResource("postgresqldatabases").GroupResource(), name, + fmt.Errorf("供应快照发生变化;停止本轮并重新观察")) +} diff --git a/internal/database/adapter/postgresql/provisioning_fixture_integration_test.go b/internal/database/adapter/postgresql/provisioning_fixture_integration_test.go new file mode 100644 index 0000000..3141203 --- /dev/null +++ b/internal/database/adapter/postgresql/provisioning_fixture_integration_test.go @@ -0,0 +1,184 @@ +//go:build integration + +package postgresql_test + +import ( + "context" + "os/exec" + "regexp" + "strings" + "testing" + "time" + + databasev1alpha1 "git.ddupan.top/panxiao81/ayatori/api/database/v1alpha1" + "git.ddupan.top/panxiao81/ayatori/internal/database/adapter/kubernetes" + "git.ddupan.top/panxiao81/ayatori/internal/database/adapter/openbao" + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + databasecontroller "git.ddupan.top/panxiao81/ayatori/internal/database/controller" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" + bao "github.com/openbao/openbao/api/v2" + "k8s.io/apimachinery/pkg/api/meta" + "k8s.io/apimachinery/pkg/runtime" + ctrl "sigs.k8s.io/controller-runtime" + "sigs.k8s.io/controller-runtime/pkg/client" + "sigs.k8s.io/controller-runtime/pkg/envtest" +) + +type provisioningFixture struct { + *credentialFixture + api client.Client + scheme *runtime.Scheme + store *openbao.Credentials + resources *kubernetes.ProvisioningResources + usecase application.DatabaseProvisioning +} + +func newProvisioningFixture(t *testing.T) *provisioningFixture { + t.Helper() + f := newCredentialFixture(t) + useNativeManager(t, f) + if _, err := envtest.InstallCRDs(f.config, envtest.CRDInstallOptions{Paths: []string{"../../../../config/crd/bases"}, ErrorIfPathMissing: true}); err != nil { + t.Fatal(err) + } + scheme := runtime.NewScheme() + if err := databasev1alpha1.AddToScheme(scheme); err != nil { + t.Fatal(err) + } + api, err := client.New(f.config, client.Options{Scheme: scheme}) + if err != nil { + t.Fatal(err) + } + object := &databasev1alpha1.PostgreSQLInstance{Name: "supply-instance", Spec: databasev1alpha1.PostgreSQLInstanceSpec{ + Endpoint: databasev1alpha1.PostgreSQLEndpoint{Host: fixtureHost, HostAddr: fixtureAddress, Port: int32(f.port), SSLMode: "disable"}, + AdminCredentialRef: databasev1alpha1.AdminCredentialReference{Name: secretName, UsernameKey: managementUsernameKey, PasswordKey: managementPasswordKey}, + }} + if err := api.Create(f.ctx, object); err != nil { + t.Fatal(err) + } + instanceResources := &kubernetes.InstanceResources{Client: api, Reader: api} + instanceService := &application.InstanceReconciliation{Resources: instanceResources, Observer: f.service} + result, err := instanceService.Reconcile(f.ctx, object.Name) + if err != nil { + t.Fatal(err) + } + if err := instanceResources.PresentInstance(f.ctx, result); err != nil { + t.Fatal(err) + } + if err := api.Get(f.ctx, client.ObjectKeyFromObject(object), object); err != nil { + t.Fatal(err) + } + if !meta.IsStatusConditionTrue(object.Status.Conditions, "Ready") { + t.Fatal("实际管理账号未就绪") + } + store := provisioningBao(t) + resources := &kubernetes.ProvisioningResources{Client: api, Reader: api} + return &provisioningFixture{credentialFixture: f, api: api, scheme: scheme, store: store, resources: resources, + usecase: application.DatabaseProvisioning{Resources: resources, Credentials: store, Backend: f.service}} +} + +// 三后端组合验收只启动自己的 dev Bao;不读取环境 token 或外部地址。 +func provisioningBao(t *testing.T) *openbao.Credentials { + t.Helper() + const image = "openbao/openbao@sha256:5b2486ab0fb90bbc788cc345b0a08616dfb375873ee8be5df3a2fd4d378a67e0" + const token = "AYATORI-TEST-ONLY-supply-token" + prepare, cancel := context.WithTimeout(t.Context(), 5*time.Minute) + defer cancel() + if exec.CommandContext(prepare, "docker", "image", "inspect", image).Run() != nil { + if output, err := exec.CommandContext(prepare, "docker", "pull", image).CombinedOutput(); err != nil { + t.Fatalf("隔离 Bao 镜像准备失败:%s", output) + } + } + ctx, stop := context.WithTimeout(t.Context(), time.Minute) + defer stop() + output, err := exec.CommandContext(ctx, "docker", "run", "--pull=never", "--rm", "-d", "-p", "127.0.0.1::8200", image, + "server", "-dev", "-dev-root-token-id="+token, "-dev-listen-address=0.0.0.0:8200").Output() + if err != nil { + t.Fatal("隔离 Bao 启动失败") + } + id := strings.TrimSpace(string(output)) + if !regexp.MustCompile(`^[a-f0-9]{64}$`).MatchString(id) { + t.Fatal("无效容器 ID") + } + t.Cleanup(func() { + cleanup, cancel := context.WithTimeout(context.Background(), 30*time.Second) + defer cancel() + if exec.CommandContext(cleanup, "docker", "rm", "-f", id).Run() != nil { + t.Error("隔离 Bao 清理失败") + } + }) + output, err = exec.CommandContext(ctx, "docker", "inspect", "--format", `{{(index (index .NetworkSettings.Ports "8200/tcp") 0).HostPort}}`, id).Output() + if err != nil { + t.Fatal("隔离 Bao 端口不可读") + } + backend, err := bao.NewClient(&bao.Config{Address: "http://127.0.0.1:" + strings.TrimSpace(string(output)), Timeout: 5 * time.Second}) + if err != nil { + t.Fatal("隔离 Bao client 构造失败") + } + backend.SetToken(token) + for { + if _, err := backend.Sys().HealthWithContext(ctx); err == nil { + break + } + select { + case <-ctx.Done(): + t.Fatal("隔离 Bao 健康检查超时") + case <-time.After(100 * time.Millisecond): + } + } + store, err := openbao.NewCredentials(backend, "secret", "applications") + if err != nil { + t.Fatal(err) + } + return store +} + +func (f *provisioningFixture) bound(t *testing.T, name string) *databasev1alpha1.PostgreSQLDatabase { + t.Helper() + tenant := &databasev1alpha1.PostgreSQLTenant{Name: strings.ReplaceAll(name, "_", "-"), Namespace: controllerNamespace, + Spec: databasev1alpha1.PostgreSQLTenantSpec{Provision: &databasev1alpha1.DatabaseProvisionRequest{ + InstanceRef: databasev1alpha1.InstanceReference{Name: "supply-instance"}, + Database: databasev1alpha1.PostgreSQLIdentifier(name), LoginRole: databasev1alpha1.PostgreSQLIdentifier(name), + }}} + if err := f.api.Create(f.ctx, tenant); err != nil { + t.Fatal(err) + } + resources := &kubernetes.BindingResources{Client: f.api, Reader: f.api} + binder := databasecontroller.NewBindingReconciler(f.api, &application.BindingService{Resources: resources}, resources) + if _, err := binder.Reconcile(f.ctx, ctrl.Request{NamespacedName: client.ObjectKeyFromObject(tenant)}); err != nil { + t.Fatal(err) + } + if err := f.api.Get(f.ctx, client.ObjectKeyFromObject(tenant), tenant); err != nil { + t.Fatal(err) + } + database := &databasev1alpha1.PostgreSQLDatabase{} + if err := f.api.Get(f.ctx, client.ObjectKey{Name: string(tenant.Status.DatabaseRef.Name)}, database); err != nil { + t.Fatal(err) + } + prepare := application.CredentialPreparation{Resources: &kubernetes.CredentialResources{Client: f.api, Reader: f.api}, Store: f.store} + if err := prepare.Reconcile(f.ctx, database.Name); err != nil { + t.Fatal(err) + } + return database +} + +func (f *provisioningFixture) reconcile(t *testing.T, database *databasev1alpha1.PostgreSQLDatabase, phase provisioning.Phase) { + t.Helper() + if err := f.usecase.Reconcile(f.ctx, database.Name); err != nil { + t.Fatal(err) + } + f.status(t, database, phase) +} + +func (f *provisioningFixture) status(t *testing.T, database *databasev1alpha1.PostgreSQLDatabase, phase provisioning.Phase) { + t.Helper() + if err := f.api.Get(f.ctx, client.ObjectKeyFromObject(database), database); err != nil { + t.Fatal(err) + } + condition := meta.FindStatusCondition(database.Status.Conditions, "ResourcesReady") + if condition == nil || condition.Reason != string(phase) { + t.Fatalf("期望资源阶段 %s,实际 %v", phase, condition) + } + if meta.IsStatusConditionTrue(database.Status.Conditions, "Ready") { + t.Fatal("角色建库不代表扩展与交付完成") + } +} diff --git a/internal/database/adapter/postgresql/provisioning_integration_test.go b/internal/database/adapter/postgresql/provisioning_integration_test.go new file mode 100644 index 0000000..8089bc9 --- /dev/null +++ b/internal/database/adapter/postgresql/provisioning_integration_test.go @@ -0,0 +1,243 @@ +//go:build integration + +package postgresql_test + +import ( + "context" + "errors" + "sync" + "sync/atomic" + "testing" + + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/credential" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/instance" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" + "github.com/jackc/pgx/v5" + apierrors "k8s.io/apimachinery/pkg/api/errors" +) + +func testProvisioningLogin(t *testing.T, f *provisioningFixture) { + t.Helper() + database := f.bound(t, "supplied") + f.reconcile(t, database, provisioning.Pending) + if database.Status.RoleOID == 0 || database.Status.DatabaseOID != 0 { + t.Fatal("角色确认未独立保存") + } + f.service.Forget("supply-instance") + f.reconcile(t, database, provisioning.Pending) + if database.Status.DatabaseOID == 0 { + t.Fatal("数据库未独立确认") + } + if got := f.queryPostgres(t, "SELECT datallowconn FROM pg_database WHERE datname='supplied'"); got != "f" { + t.Fatal("ACL 收紧前连接入口应关闭") + } + f.reconcile(t, database, provisioning.Available) + beforeRole, beforeDB := database.Status.RoleOID, database.Status.DatabaseOID + for range 2 { + f.reconcile(t, database, provisioning.Available) + } + if beforeRole != database.Status.RoleOID || beforeDB != database.Status.DatabaseOID { + t.Fatal("幂等协调替换了对象") + } + value, err := f.store.ReadCredential(f.ctx, credential.Location{Mount: database.Status.CredentialRef.Mount, Path: database.Status.CredentialRef.Path}, database.Status.CredentialVersion) + if err != nil { + t.Fatal(err) + } + config, err := pgx.ParseConfig("") + if err != nil { + t.Fatal("fixture 配置失败") + } + config.Host, config.Port = fixtureAddress, uint16(f.port) + config.User, config.Database, config.Password = "supplied", "supplied", value.SecretData()["password"].(string) + config.TLSConfig, config.Fallbacks = nil, nil + connection, err := pgx.ConnectConfig(f.ctx, config) + if err != nil { + t.Fatal("实际应用密码不能登录已供应数据库") + } + defer func() { _ = connection.Close(context.Background()) }() + if _, err := connection.Exec(f.ctx, "CREATE TABLE app_data (id integer)"); err != nil { + t.Fatal("应用 owner 不能创建表") + } + if _, err := connection.Exec(f.ctx, "CREATE ROLE should_be_denied"); err == nil { + t.Fatal("应用具有 CREATEROLE") + } + if _, err := connection.Exec(f.ctx, "CREATE DATABASE should_be_denied"); err == nil { + t.Fatal("应用具有 CREATEDB") + } + f.queryPostgres(t, "CREATE ROLE unrelated_login LOGIN PASSWORD '"+fixturePassword+"'") + config.User, config.Password = "unrelated_login", fixturePassword + if outsider, err := pgx.ConnectConfig(f.ctx, config); err == nil { + _ = outsider.Close(f.ctx) + t.Fatal("其他账号可连接受管数据库") + } +} + +func TestDatabaseProvisioningWithRealBackends(t *testing.T) { + f := newProvisioningFixture(t) + t.Run("创建重启幂等与实际登录", func(t *testing.T) { testProvisioningLogin(t, f) }) + t.Run("未知同名不认领", func(t *testing.T) { + database := f.bound(t, "occupied") + f.queryPostgres(t, "CREATE ROLE occupied LOGIN") + before := f.queryPostgres(t, "SELECT oid FROM pg_roles WHERE rolname='occupied'") + f.reconcile(t, database, provisioning.Conflict) + f.reconcile(t, database, provisioning.Conflict) + if database.Status.RoleOID != 0 || f.queryPostgres(t, "SELECT oid FROM pg_roles WHERE rolname='occupied'") != before { + t.Fatal("未知角色被认领或修改") + } + }) + t.Run("未知同名数据库不认领", func(t *testing.T) { + database := f.bound(t, "occupied_database") + f.queryPostgres(t, "CREATE DATABASE occupied_database") + before := f.queryPostgres(t, "SELECT oid FROM pg_database WHERE datname='occupied_database'") + f.reconcile(t, database, provisioning.Conflict) + if database.Status.RoleOID != 0 || database.Status.DatabaseOID != 0 || + f.queryPostgres(t, "SELECT oid FROM pg_database WHERE datname='occupied_database'") != before || + f.queryPostgres(t, "SELECT count(*) FROM pg_roles WHERE rolname='occupied_database'") != "0" { + t.Fatal("未知数据库被认领、修改或继续创建了角色") + } + }) + t.Run("已确认角色后权限恢复", func(t *testing.T) { + database := f.bound(t, "permission_restore") + f.reconcile(t, database, provisioning.Pending) + f.queryPostgres(t, "ALTER ROLE native_manager NOCREATEDB") + f.reconcile(t, database, provisioning.Unavailable) + f.queryPostgres(t, "ALTER ROLE native_manager CREATEDB") + f.reconcile(t, database, provisioning.Pending) + f.reconcile(t, database, provisioning.Available) + }) + t.Run("删除前置阻止写入", func(t *testing.T) { + database := f.bound(t, "deleting_supply") + if err := f.api.Delete(f.ctx, database); err != nil { + t.Fatal(err) + } + f.reconcile(t, database, provisioning.Stopped) + if len(database.Finalizers) == 0 || database.Status.RoleOID != 0 { + t.Fatal("删除边界被供应绕过") + } + }) + t.Run("确认保存失败转人工冲突", func(t *testing.T) { + database := f.bound(t, "lost_confirmation") + failed := f.usecase + failed.Resources = &failRoleConfirmation{ProvisioningResources: f.resources} + if err := failed.Reconcile(f.ctx, database.Name); err == nil { + t.Fatal("未注入确认写入故障") + } + f.reconcile(t, database, provisioning.Conflict) + if database.Status.RoleOID != 0 || f.queryPostgres(t, "SELECT count(*) FROM pg_roles WHERE rolname='lost_confirmation'") != "1" { + t.Fatal("失败恢复认领或清理了残留角色") + } + }) + t.Run("数据库确认失败保留关闭入口", func(t *testing.T) { + database := f.bound(t, "lost_database_confirmation") + f.reconcile(t, database, provisioning.Pending) + failed := f.usecase + failed.Resources = &failDatabaseConfirmation{ProvisioningResources: f.resources} + if err := failed.Reconcile(f.ctx, database.Name); err == nil { + t.Fatal("未注入数据库确认写入故障") + } + f.reconcile(t, database, provisioning.Conflict) + if database.Status.RoleOID == 0 || database.Status.DatabaseOID != 0 || f.queryPostgres(t, "SELECT datallowconn FROM pg_database WHERE datname='lost_database_confirmation'") != "f" { + t.Fatal("未确认数据库被认领或开放连接") + } + }) + t.Run("创建成功但调用方丢失结果", func(t *testing.T) { + database := f.bound(t, "lost_role_response") + failed := f.usecase + failed.Backend = &lostRoleResponse{ProvisioningBackend: f.service} + if err := failed.Reconcile(f.ctx, database.Name); err != nil { + t.Fatal(err) + } + f.reconcile(t, database, provisioning.Conflict) + if database.Status.RoleOID != 0 || f.queryPostgres(t, "SELECT count(*) FROM pg_roles WHERE rolname='lost_role_response'") != "1" { + t.Fatal("丢失结果后应保留未认领角色") + } + }) + t.Run("同名角色被重建", func(t *testing.T) { + database := f.bound(t, "replaced_role") + f.reconcile(t, database, provisioning.Pending) + f.queryPostgres(t, "DROP ROLE replaced_role; CREATE ROLE replaced_role LOGIN") + f.reconcile(t, database, provisioning.Conflict) + if database.Status.DatabaseOID != 0 { + t.Fatal("重建角色被当成原 owner 继续供应") + } + }) + t.Run("创建步骤并发只有一个获准", func(t *testing.T) { + database := f.bound(t, "concurrent_supply") + service := f.usecase + resources := &concurrentProvisioningResources{ProvisioningResources: f.resources} + resources.loaded.Add(2) + service.Resources = resources + results := make(chan error, 2) + for range 2 { + go func() { results <- service.Reconcile(f.ctx, database.Name) }() + } + success, conflict := 0, 0 + for range 2 { + err := <-results + if err == nil { + success++ + } else if apierrors.IsConflict(err) { + conflict++ + } else { + t.Fatal(err) + } + } + if success != 1 || conflict != 1 { + t.Fatal("同一快照不得重复授权创建") + } + f.status(t, database, provisioning.Pending) + if database.Status.RoleOID == 0 { + t.Fatal("胜方未确认角色") + } + }) + t.Run("实际manager观察与重启", func(t *testing.T) { testProvisioningManager(t, f) }) +} + +type failRoleConfirmation struct { + application.ProvisioningResources +} + +func (r *failRoleConfirmation) Save(ctx context.Context, record *application.ProvisioningRecord, state provisioning.State) (*application.ProvisioningRecord, error) { + if state.RoleOID != 0 { + return nil, errors.New("injected confirmation persistence failure") + } + return r.ProvisioningResources.Save(ctx, record, state) +} + +type failDatabaseConfirmation struct { + application.ProvisioningResources +} + +func (r *failDatabaseConfirmation) Save(ctx context.Context, record *application.ProvisioningRecord, state provisioning.State) (*application.ProvisioningRecord, error) { + if state.DatabaseOID != 0 { + return nil, errors.New("injected database confirmation persistence failure") + } + return r.ProvisioningResources.Save(ctx, record, state) +} + +type lostRoleResponse struct { + application.ProvisioningBackend +} + +func (b *lostRoleResponse) CreateLoginRole(ctx context.Context, target instance.ObservationTarget, value credential.ApplicationCredential) (uint32, error) { + if _, err := b.ProvisioningBackend.CreateLoginRole(ctx, target, value); err != nil { + return 0, err + } + return 0, application.ErrResourceUncertain +} + +type concurrentProvisioningResources struct { + application.ProvisioningResources + readers atomic.Int32 + loaded sync.WaitGroup +} + +func (r *concurrentProvisioningResources) Load(ctx context.Context, name string) (*application.ProvisioningRecord, error) { + record, err := r.ProvisioningResources.Load(ctx, name) + if r.readers.Add(1) <= 2 { + r.loaded.Done() + r.loaded.Wait() + } + return record, err +} diff --git a/internal/database/adapter/postgresql/provisioning_watch_integration_test.go b/internal/database/adapter/postgresql/provisioning_watch_integration_test.go new file mode 100644 index 0000000..a16deb1 --- /dev/null +++ b/internal/database/adapter/postgresql/provisioning_watch_integration_test.go @@ -0,0 +1,97 @@ +//go:build integration + +package postgresql_test + +import ( + "context" + "sync/atomic" + "testing" + "time" + + "git.ddupan.top/panxiao81/ayatori/internal/database/adapter/kubernetes" + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + databasecontroller "git.ddupan.top/panxiao81/ayatori/internal/database/controller" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/instance" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" + "k8s.io/apimachinery/pkg/api/meta" + ctrl "sigs.k8s.io/controller-runtime" + "sigs.k8s.io/controller-runtime/pkg/client" + controllerconfig "sigs.k8s.io/controller-runtime/pkg/config" + metricsserver "sigs.k8s.io/controller-runtime/pkg/metrics/server" +) + +func testProvisioningManager(t *testing.T, f *provisioningFixture) { + t.Helper() + database := f.bound(t, "watch_supply") + // 首次 list/watch 自动供应;第二个新 manager 只观察确认记录,不重新建库。 + var roleOID, databaseOID int64 + for range 2 { + skipRepeatedName := true + manager, err := ctrl.NewManager(f.config, ctrl.Options{Scheme: f.scheme, Metrics: metricsserver.Options{BindAddress: "0"}, HealthProbeBindAddress: "0", + Controller: controllerconfig.Controller{SkipNameValidation: &skipRepeatedName}}) + if err != nil { + t.Fatal(err) + } + resources := &kubernetes.ProvisioningResources{Client: manager.GetClient(), Reader: manager.GetAPIReader()} + backend := &countProvisioningObservations{ProvisioningBackend: f.service} + service := &application.DatabaseReconciliation{ + Credentials: &application.CredentialPreparation{Resources: &kubernetes.CredentialResources{Client: manager.GetClient(), Reader: manager.GetAPIReader()}, Store: f.store}, + Provisioning: &application.DatabaseProvisioning{Resources: resources, Credentials: f.store, Backend: backend}, + } + if err := databasecontroller.NewProvisioningReconciler(manager.GetClient(), service).SetupWithManager(manager); err != nil { + t.Fatal(err) + } + ctx, stop := context.WithCancel(f.ctx) + done := make(chan error, 1) + go func() { done <- manager.Start(ctx) }() + // 使用 scope 确保任何失败都先停止 worker,再由 fixture 关闭共享连接。 + func() { + defer func() { + stop() + select { + case err := <-done: + if err != nil { + t.Error(err) + } + case <-time.After(10 * time.Second): + t.Error("供应 manager 未停止") + } + }() + if !manager.GetCache().WaitForCacheSync(ctx) { + t.Fatal("供应 cache 未同步") + } + deadline := time.After(15 * time.Second) + for { + if err := f.api.Get(f.ctx, client.ObjectKeyFromObject(database), database); err != nil { + t.Fatal(err) + } + if meta.IsStatusConditionTrue(database.Status.Conditions, "ResourcesReady") && backend.reads.Load() > 0 { + break + } + select { + case <-deadline: + t.Fatal("watch 未推动资源创建") + case <-time.After(100 * time.Millisecond): + } + } + if roleOID != 0 && (roleOID != database.Status.RoleOID || databaseOID != database.Status.DatabaseOID) { + t.Fatal("manager 重启替换已确认资源") + } + roleOID, databaseOID = database.Status.RoleOID, database.Status.DatabaseOID + }() + } + f.status(t, database, provisioning.Available) +} + +type countProvisioningObservations struct { + application.ProvisioningBackend + reads atomic.Int32 +} + +func (b *countProvisioningObservations) InspectResources(ctx context.Context, target instance.ObservationTarget, name, role string) (provisioning.Observation, error) { + observation, err := b.ProvisioningBackend.InspectResources(ctx, target, name, role) + if err == nil && name == "watch_supply" { + b.reads.Add(1) + } + return observation, err +} diff --git a/internal/database/adapter/postgresql/resources.go b/internal/database/adapter/postgresql/resources.go new file mode 100644 index 0000000..07d7a8f --- /dev/null +++ b/internal/database/adapter/postgresql/resources.go @@ -0,0 +1,159 @@ +package postgresql + +import ( + "context" + "errors" + "regexp" + + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/credential" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" + "github.com/jackc/pgx/v5" + "github.com/jackc/pgx/v5/pgconn" +) + +var resourceIdentifier = regexp.MustCompile(`^[a-z][a-z0-9_]{0,62}$`) + +const inspectResourcesSQL = ` +SELECT COALESCE(r.oid, 0), COALESCE(d.oid, 0), + COALESCE(r.rolcanlogin AND NOT (r.rolsuper OR r.rolcreatedb OR r.rolcreaterole OR r.rolreplication OR r.rolbypassrls) + AND NOT EXISTS (SELECT FROM pg_catalog.pg_auth_members m WHERE m.member = r.oid), false), + COALESCE(d.datdba, 0), COALESCE(d.datallowconn, false), + EXISTS (SELECT FROM pg_catalog.aclexplode(COALESCE(d.datacl, pg_catalog.acldefault('d', d.datdba))) a + WHERE a.grantee = 0 AND a.privilege_type = 'CONNECT') +FROM (SELECT 1) seed +LEFT JOIN pg_catalog.pg_roles r ON r.rolname = $1 +LEFT JOIN pg_catalog.pg_database d ON d.datname = $2` + +type resourceReader interface { + QueryRow(context.Context, string, ...any) pgx.Row +} + +func inspectResources(ctx context.Context, reader resourceReader, name, role string) (provisioning.Observation, error) { + var result provisioning.Observation + if !resourceIdentifier.MatchString(name) || !resourceIdentifier.MatchString(role) { + return result, application.ErrResourceConflict + } + err := reader.QueryRow(ctx, inspectResourcesSQL, role, name).Scan( + &result.RoleOID, &result.DatabaseOID, &result.RoleSafe, &result.OwnerOID, &result.AllowConnections, &result.PublicConnect, + ) + if err != nil { + return provisioning.Observation{}, application.ErrResourceUnavailable + } + return result, nil +} + +func (d *database) InspectResources(ctx context.Context, name, role string) (provisioning.Observation, error) { + return inspectResources(ctx, d.pool, name, role) +} + +// CREATE ROLE 与 membership 在一个原生事务提交;不修改任何已有角色。 +// simple protocol 使用 pgx 的参数转义,避免自行拼接密码字面量;错误不带 SQL 或驱动响应。 +func (d *database) CreateLoginRole(ctx context.Context, value credential.ApplicationCredential) (uint32, error) { + if value.Validate() != nil { + return 0, application.ErrResourceConflict + } + data := value.SecretData() + role := data["username"].(string) + transaction, err := d.pool.Begin(ctx) + if err != nil { + return 0, application.ErrResourceUnavailable + } + defer func() { _ = transaction.Rollback(ctx) }() + statement := "CREATE ROLE " + pgx.Identifier{role}.Sanitize() + " LOGIN NOSUPERUSER NOCREATEDB NOCREATEROLE NOREPLICATION NOBYPASSRLS NOINHERIT PASSWORD $1" + if _, err := transaction.Exec(ctx, statement, pgx.QueryExecModeSimpleProtocol, data["password"]); err != nil { + return 0, creationError(err) + } + if _, err := transaction.Exec(ctx, "GRANT "+pgx.Identifier{role}.Sanitize()+" TO CURRENT_USER WITH SET TRUE, INHERIT FALSE"); err != nil { + return 0, application.ErrResourceUncertain + } + var oid uint32 + if err := transaction.QueryRow(ctx, "SELECT oid FROM pg_catalog.pg_roles WHERE rolname=$1", role).Scan(&oid); err != nil { + return 0, application.ErrResourceUncertain + } + if err := transaction.Commit(ctx); err != nil { + return 0, application.ErrResourceUncertain + } + observed, err := d.InspectResources(ctx, data["database"].(string), role) + if err != nil || observed.RoleOID != oid || !observed.RoleSafe { + return 0, application.ErrResourceUncertain + } + return oid, nil +} + +func (d *database) CreateOwnedDatabase(ctx context.Context, name, role string, roleOID uint32) (uint32, error) { + observed, err := d.InspectResources(ctx, name, role) + if err != nil { + return 0, err + } + if roleOID == 0 || observed.RoleOID != roleOID || !observed.RoleSafe || observed.DatabaseOID != 0 { + return 0, application.ErrResourceConflict + } + // 不能放入事务。先关闭连接入口,避免默认 PUBLIC CONNECT 暴露未收紧的数据库。 + statement := "CREATE DATABASE " + pgx.Identifier{name}.Sanitize() + " OWNER " + pgx.Identifier{role}.Sanitize() + " ALLOW_CONNECTIONS false" + if _, err := d.pool.Exec(ctx, statement); err != nil { + return 0, creationError(err) + } + observed, err = d.InspectResources(ctx, name, role) + if err != nil || observed.DatabaseOID == 0 || observed.OwnerOID != roleOID || observed.RoleOID != roleOID { + return 0, application.ErrResourceUncertain + } + return observed.DatabaseOID, nil +} + +func (d *database) ConfigureAccess(ctx context.Context, name, role string, state provisioning.State) error { + transaction, err := d.pool.Begin(ctx) + if err != nil { + return application.ErrResourceUnavailable + } + defer func() { _ = transaction.Rollback(ctx) }() + observed, err := inspectResources(ctx, transaction, name, role) + if err != nil { + return err + } + if state.RoleOID == 0 || state.DatabaseOID == 0 || state.Check(observed) != nil { + return application.ErrResourceConflict + } + if _, err := transaction.Exec(ctx, "SET LOCAL ROLE "+pgx.Identifier{role}.Sanitize()); err != nil { + return application.ErrResourceUnavailable + } + for _, statement := range []string{ + "REVOKE CONNECT ON DATABASE " + pgx.Identifier{name}.Sanitize() + " FROM PUBLIC", + "GRANT CONNECT ON DATABASE " + pgx.Identifier{name}.Sanitize() + " TO " + pgx.Identifier{role}.Sanitize(), + "ALTER DATABASE " + pgx.Identifier{name}.Sanitize() + " ALLOW_CONNECTIONS true", + } { + if _, err := transaction.Exec(ctx, statement); err != nil { + return application.ErrResourceUnavailable + } + } + // SET LOCAL 不污染池内连接;ACL 失败可在已确认对象上幂等重试,不改密码或 owner。 + if err := transaction.Commit(ctx); err != nil { + return application.ErrResourceUnavailable + } + observed, err = d.InspectResources(ctx, name, role) + if err != nil { + return err + } + if state.Check(observed) != nil { + return application.ErrResourceConflict + } + if observed.PublicConnect || !observed.AllowConnections { + return application.ErrResourceUnavailable + } + return nil +} + +func creationError(err error) error { + if serverError, ok := errors.AsType[*pgconn.PgError](err); ok { + switch serverError.Code { + case "42710", "42P04", "23505": + return application.ErrResourceConflict + case "42501", "25006", "28P01", "28000": + return application.ErrResourceUnavailable + } + } + if pgconn.SafeToRetry(err) { + return application.ErrResourceUnavailable + } + return application.ErrResourceUncertain +} diff --git a/internal/database/adapter/postgresql/resources_test.go b/internal/database/adapter/postgresql/resources_test.go new file mode 100644 index 0000000..4ed30f2 --- /dev/null +++ b/internal/database/adapter/postgresql/resources_test.go @@ -0,0 +1,35 @@ +package postgresql + +import ( + "errors" + "testing" + + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + "github.com/jackc/pgx/v5/pgconn" +) + +func TestCreationErrorsDoNotExposeServerDetails(t *testing.T) { + for _, test := range []struct { + code string + want error + }{ + {"42710", application.ErrResourceConflict}, + {"42P04", application.ErrResourceConflict}, + {"23505", application.ErrResourceConflict}, + {"42501", application.ErrResourceUnavailable}, + {"25006", application.ErrResourceUnavailable}, + {"57014", application.ErrResourceUncertain}, + {"XX000", application.ErrResourceUncertain}, + } { + actual := creationError(&pgconn.PgError{Code: test.code, Message: "unsafe SQL and credential detail"}) + if !errors.Is(actual, test.want) { + t.Fatalf("SQLSTATE %s 分类错误", test.code) + } + if actual.Error() != test.want.Error() { + t.Fatal("后端错误携带原始响应") + } + } + if !errors.Is(creationError(errors.New("connection lost after send")), application.ErrResourceUncertain) { + t.Fatal("未知网络结果不允许自动重试创建") + } +} diff --git a/internal/database/application/database_provisioning.go b/internal/database/application/database_provisioning.go new file mode 100644 index 0000000..ea2049b --- /dev/null +++ b/internal/database/application/database_provisioning.go @@ -0,0 +1,152 @@ +package application + +import ( + "context" + "errors" + "fmt" + + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/credential" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/instance" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" +) + +type ProvisioningRecord struct { + credential.Target + Revision string + TenantGeneration int64 + InstanceGeneration int64 + InstanceTarget instance.ObservationTarget + Credentials credential.State + State provisioning.State +} + +type ProvisioningResources interface { + Load(context.Context, string) (*ProvisioningRecord, error) + Save(context.Context, *ProvisioningRecord, provisioning.State) (*ProvisioningRecord, error) + CheckCurrent(context.Context, *ProvisioningRecord) error +} + +type DatabaseProvisioning struct { + Resources ProvisioningResources + Credentials CredentialStore + Backend ProvisioningBackend +} + +func (s DatabaseProvisioning) Reconcile(ctx context.Context, name string) error { + record, err := s.Resources.Load(ctx, name) + if err != nil || record == nil || !record.RequiresPreparation() { + return err + } + if state, resume := record.State.Resume(); !resume { + if state == record.State { + return nil + } + _, err := s.Resources.Save(ctx, record, state) + return err + } + if issue := record.Check(); issue != nil { + phase := provisioning.Unavailable + if issue.Phase == credential.Conflict { + phase = provisioning.Conflict + } + if issue.Phase == credential.Stopped { + phase = provisioning.Stopped + } + return s.report(ctx, record, phase, issue.Message) + } + if !record.Credentials.Confirmed() || record.Credentials.Location == nil || record.Credentials.Phase != credential.Prepared { + return s.report(ctx, record, provisioning.Unavailable, "等待已确认且当前可用的应用凭据") + } + value, err := s.Credentials.ReadCredential(ctx, *record.Credentials.Location, record.Credentials.Version) + if err != nil { + return s.report(ctx, record, provisioning.Unavailable, "无法读取已确认凭据;未执行 PostgreSQL 写入") + } + if issue := record.CheckCredential(value); issue != nil { + return s.report(ctx, record, provisioning.Conflict, issue.Message) + } + observed, err := s.Backend.InspectResources(ctx, record.InstanceTarget, record.Database.Name, record.Database.LoginRole) + if err != nil { + return s.report(ctx, record, provisioning.Unavailable, "无法验证当前 PostgreSQL 资源和管理权限") + } + if err := record.State.Check(observed); err != nil { + return s.report(ctx, record, provisioning.Conflict, err.Error()+";请人工核对,未认领或覆盖") + } + if err := s.Resources.CheckCurrent(ctx, record); err != nil { + return err + } + if record.State.RoleOID == 0 { + return s.createRole(ctx, record, value) + } + if record.State.DatabaseOID == 0 { + return s.createDatabase(ctx, record) + } + if err := s.Backend.ConfigureAccess(ctx, record.InstanceTarget, record.Database.Name, record.Database.LoginRole, record.State); err != nil { + return s.backendFailure(ctx, record, err, "数据库访问权限收敛") + } + if err := s.Resources.CheckCurrent(ctx, record); err != nil { + return err + } + return s.report(ctx, record, provisioning.Available, "角色和数据库已确认,PUBLIC CONNECT 已撤销;扩展与 Tenant 交付尚未完成") +} + +func (s DatabaseProvisioning) createRole(ctx context.Context, record *ProvisioningRecord, value credential.ApplicationCredential) error { + record, err := s.Resources.Save(ctx, record, record.State.WithPhase(provisioning.CreatingRole, "开始创建登录角色;尚未持久确认")) + if err != nil { + return err + } + if err := s.Resources.CheckCurrent(ctx, record); err != nil { + return err + } + oid, err := s.Backend.CreateLoginRole(ctx, record.InstanceTarget, value) + if err != nil { + return s.backendFailure(ctx, record, err, "登录角色创建") + } + if oid == 0 { + return s.backendFailure(ctx, record, ErrResourceUncertain, "登录角色创建") + } + if err := s.Resources.CheckCurrent(ctx, record); err != nil { + return err + } + state := record.State.WithPhase(provisioning.Pending, "登录角色已创建并回读,等待创建数据库") + state.RoleOID = oid + _, err = s.Resources.Save(ctx, record, state) + return err +} + +func (s DatabaseProvisioning) createDatabase(ctx context.Context, record *ProvisioningRecord) error { + record, err := s.Resources.Save(ctx, record, record.State.WithPhase(provisioning.CreatingDatabase, "开始创建数据库;尚未持久确认")) + if err != nil { + return err + } + if err := s.Resources.CheckCurrent(ctx, record); err != nil { + return err + } + oid, err := s.Backend.CreateOwnedDatabase(ctx, record.InstanceTarget, record.Database.Name, record.Database.LoginRole, record.State.RoleOID) + if err != nil { + return s.backendFailure(ctx, record, err, "数据库创建") + } + if oid == 0 { + return s.backendFailure(ctx, record, ErrResourceUncertain, "数据库创建") + } + if err := s.Resources.CheckCurrent(ctx, record); err != nil { + return err + } + state := record.State.WithPhase(provisioning.Pending, "数据库已创建并回读,等待收紧访问权限;连接入口仍关闭") + state.DatabaseOID = oid + _, err = s.Resources.Save(ctx, record, state) + return err +} + +func (s DatabaseProvisioning) backendFailure(ctx context.Context, record *ProvisioningRecord, err error, step string) error { + if errors.Is(err, ErrResourceUnavailable) { + return s.report(ctx, record, provisioning.Unavailable, step+"被拒绝或暂不可用,保留已确认步骤并等待依赖恢复") + } + return s.report(ctx, record, provisioning.Conflict, step+"冲突或结果不确定;请核对目标与已确认 OID,未自动认领、改密或清理") +} + +func (s DatabaseProvisioning) report(ctx context.Context, record *ProvisioningRecord, phase provisioning.Phase, message string) error { + message = fmt.Sprintf("Database %s / Instance %s / database %s / role %s:%s", record.Database.Identity.Name, + record.Database.Instance, record.Database.Name, record.Database.LoginRole, message) + _, err := s.Resources.Save(ctx, record, record.State.WithPhase(phase, message)) + return err +} diff --git a/internal/database/application/database_reconciliation.go b/internal/database/application/database_reconciliation.go new file mode 100644 index 0000000..142704f --- /dev/null +++ b/internal/database/application/database_reconciliation.go @@ -0,0 +1,17 @@ +package application + +import "context" + +// DatabaseReconciliation 顺序编排两个已有用例,不复制它们的领域判断。 +// 启用资源创建时由同一个 controller 驱动,避免两个 worker 争写同一 Database status。 +type DatabaseReconciliation struct { + Credentials *CredentialPreparation + Provisioning *DatabaseProvisioning +} + +func (s DatabaseReconciliation) Reconcile(ctx context.Context, name string) error { + if err := s.Credentials.Reconcile(ctx, name); err != nil { + return err + } + return s.Provisioning.Reconcile(ctx, name) +} diff --git a/internal/database/application/instance_service.go b/internal/database/application/instance_service.go index b1498e0..fc782c3 100644 --- a/internal/database/application/instance_service.go +++ b/internal/database/application/instance_service.go @@ -35,6 +35,7 @@ var ( // Database 与 Connector 复用原项目 internal/instance/service.go 的能力边界。 // Metadata 只查询版本与可用扩展,不能产生领域 Ready。 type Database interface { + ResourceDatabase InspectMetadata(context.Context) (DatabaseMetadata, error) InspectManagement(context.Context) (DatabaseMetadata, error) Close() @@ -93,16 +94,45 @@ func (s *InstanceService) ObserveManagement(ctx context.Context, target instance } func (s *InstanceService) observe(ctx context.Context, target instance.ObservationTarget, management bool) (InstanceObservation, error) { - if err := target.Validate(); err != nil { - return InstanceObservation{}, err - } s.mu.Lock() defer s.mu.Unlock() + current, err := s.connection(ctx, target) + if err != nil { + return InstanceObservation{}, err + } + name := target.Identity().Name() + var metadata DatabaseMetadata + if management { + metadata, err = current.database.InspectManagement(ctx) + } else { + metadata, err = current.database.InspectMetadata(ctx) + metadata.Management = instance.ManagementChecks{} + } + if err != nil { + s.release(name) + return InstanceObservation{}, err + } + if metadata.Version == "" { + s.release(name) + return InstanceObservation{}, ErrObservation + } + if err := s.checkCredentials(ctx, target, current.credentials); err != nil { + return InstanceObservation{}, err + } + return InstanceObservation{target: target, version: metadata.Version, + extensions: instance.ObserveExtensionSupport(metadata.AvailableExtensions), management: metadata.Management}, nil +} + +// connection 由持有 mu 的调用者使用;不缓存观察结果或管理权限。 +func (s *InstanceService) connection(ctx context.Context, target instance.ObservationTarget) (*entry, error) { + if err := target.Validate(); err != nil { + return nil, err + } if s.closed { - return InstanceObservation{}, ErrClosed + return nil, ErrClosed } if err := ctx.Err(); err != nil { - return InstanceObservation{}, err + return nil, err } // 先读取有效凭据。读取失败时不得继续使用缓存中的旧连接。 @@ -110,11 +140,11 @@ func (s *InstanceService) observe(ctx context.Context, target instance.Observati credentials, err := s.source.Read(ctx, target.Definition().AdminCredential()) if err != nil { s.release(name) - return InstanceObservation{}, credentialError(err) + return nil, credentialError(err) } if credentials.username == "" || credentials.password == "" { s.release(name) - return InstanceObservation{}, ErrCredentialsInvalid + return nil, ErrCredentialsInvalid } // 连接身份与有效值均未变化时复用 pgxpool;generation 本身不要求换池。 @@ -127,7 +157,7 @@ func (s *InstanceService) observe(ctx context.Context, target instance.Observati if current == nil { database, err := s.connector.Connect(ctx, target.Definition().Endpoint(), credentials) if err != nil { - return InstanceObservation{}, err + return nil, err } current = &entry{ target: target, @@ -137,39 +167,22 @@ func (s *InstanceService) observe(ctx context.Context, target instance.Observati s.entries[name] = current } - var metadata DatabaseMetadata - if management { - metadata, err = current.database.InspectManagement(ctx) - } else { - metadata, err = current.database.InspectMetadata(ctx) - // 即使 adapter 误填权限,也不能把只读 metadata 入口升级为 Ready。 - metadata.Management = instance.ManagementChecks{} - } - if err != nil { - s.release(name) - return InstanceObservation{}, err - } - if metadata.Version == "" { - s.release(name) - return InstanceObservation{}, ErrObservation - } + return current, nil +} +func (s *InstanceService) checkCredentials(ctx context.Context, target instance.ObservationTarget, credentials Credentials) error { + name := target.Identity().Name() // 回读后再检查凭据,避免把轮换前取得的结果交给新凭据的调用链。 latest, err := s.source.Read(ctx, target.Definition().AdminCredential()) if err != nil { s.release(name) - return InstanceObservation{}, credentialError(err) + return credentialError(err) } if latest != credentials { s.release(name) - return InstanceObservation{}, ErrCredentialsChanged + return ErrCredentialsChanged } - return InstanceObservation{ - target: target, - version: metadata.Version, - extensions: instance.ObserveExtensionSupport(metadata.AvailableExtensions), - management: metadata.Management, - }, nil + return nil } func credentialError(err error) error { diff --git a/internal/database/application/instance_service_test.go b/internal/database/application/instance_service_test.go index d9c2a68..c1b5341 100644 --- a/internal/database/application/instance_service_test.go +++ b/internal/database/application/instance_service_test.go @@ -37,6 +37,7 @@ func (s *sourceStub) Read(context.Context, instance.CredentialReference) (Creden } type databaseStub struct { + ResourceDatabase closes int err error metadata DatabaseMetadata diff --git a/internal/database/application/provisioning_backend.go b/internal/database/application/provisioning_backend.go new file mode 100644 index 0000000..a97f629 --- /dev/null +++ b/internal/database/application/provisioning_backend.go @@ -0,0 +1,102 @@ +package application + +import ( + "context" + "errors" + + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/credential" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/instance" + "git.ddupan.top/panxiao81/ayatori/internal/database/domain/provisioning" +) + +var ( + ErrResourceConflict = errors.New("PostgreSQL resource conflict; manual resolution required") + ErrResourceUncertain = errors.New("PostgreSQL creation outcome uncertain; manual resolution required") + ErrResourceUnavailable = errors.New("PostgreSQL operation unavailable before creation") +) + +// ResourceDatabase 是供应实际需要的后端能力,不暴露 SQL 或任意回调给用例。 +type ResourceDatabase interface { + InspectResources(context.Context, string, string) (provisioning.Observation, error) + CreateLoginRole(context.Context, credential.ApplicationCredential) (uint32, error) + CreateOwnedDatabase(context.Context, string, string, uint32) (uint32, error) + ConfigureAccess(context.Context, string, string, provisioning.State) error +} + +type ProvisioningBackend interface { + InspectResources(context.Context, instance.ObservationTarget, string, string) (provisioning.Observation, error) + CreateLoginRole(context.Context, instance.ObservationTarget, credential.ApplicationCredential) (uint32, error) + CreateOwnedDatabase(context.Context, instance.ObservationTarget, string, string, uint32) (uint32, error) + ConfigureAccess(context.Context, instance.ObservationTarget, string, string, provisioning.State) error +} + +func (s *InstanceService) InspectResources(ctx context.Context, target instance.ObservationTarget, name, role string) (provisioning.Observation, error) { + var result provisioning.Observation + err := s.withManagementConnection(ctx, target, false, func(database Database) (err error) { + result, err = database.InspectResources(ctx, name, role) + return err + }) + if err != nil { + return provisioning.Observation{}, err + } + return result, nil +} + +func (s *InstanceService) CreateLoginRole(ctx context.Context, target instance.ObservationTarget, value credential.ApplicationCredential) (uint32, error) { + var oid uint32 + err := s.withManagementConnection(ctx, target, true, func(database Database) (err error) { + oid, err = database.CreateLoginRole(ctx, value) + return err + }) + if err != nil { + return 0, err + } + return oid, nil +} + +func (s *InstanceService) CreateOwnedDatabase(ctx context.Context, target instance.ObservationTarget, name, role string, roleOID uint32) (uint32, error) { + var oid uint32 + err := s.withManagementConnection(ctx, target, true, func(database Database) (err error) { + oid, err = database.CreateOwnedDatabase(ctx, name, role, roleOID) + return err + }) + if err != nil { + return 0, err + } + return oid, nil +} + +func (s *InstanceService) ConfigureAccess(ctx context.Context, target instance.ObservationTarget, name, role string, state provisioning.State) error { + return s.withManagementConnection(ctx, target, false, func(database Database) error { + return database.ConfigureAccess(ctx, name, role, state) + }) +} + +// 所有供应操作与观察共用同一个连接登记和 Secret 刷新边界。 +// 创建后凭据回读失败不能冒充明确未执行;用例必须保留不确定诊断。 +func (s *InstanceService) withManagementConnection(ctx context.Context, target instance.ObservationTarget, creating bool, operation func(Database) error) error { + s.mu.Lock() + defer s.mu.Unlock() + current, err := s.connection(ctx, target) + if err != nil { + return ErrResourceUnavailable + } + metadata, err := current.database.InspectManagement(ctx) + checks := metadata.Management + if err != nil || checks.Connection != instance.CheckPassed || checks.Roles != instance.CheckPassed || checks.Databases != instance.CheckPassed || checks.Grants != instance.CheckPassed { + return ErrResourceUnavailable + } + if err := s.checkCredentials(ctx, target, current.credentials); err != nil { + return ErrResourceUnavailable + } + if err := operation(current.database); err != nil { + return err + } + if err := s.checkCredentials(ctx, target, current.credentials); err != nil { + if creating { + return ErrResourceUncertain + } + return ErrResourceUnavailable + } + return nil +} diff --git a/internal/database/controller/credential_controller.go b/internal/database/controller/credential_controller.go index a3d66a7..ae140ae 100644 --- a/internal/database/controller/credential_controller.go +++ b/internal/database/controller/credential_controller.go @@ -35,12 +35,14 @@ func (r *CredentialReconciler) SetupWithManager(manager ctrl.Manager) error { } return ctrl.NewControllerManagedBy(manager). Named("database-credentials").For(&databasev1alpha1.PostgreSQLDatabase{}). - Watches(&databasev1alpha1.PostgreSQLTenant{}, handler.EnqueueRequestsFromMapFunc(r.requestsForTenant)). - Watches(&databasev1alpha1.PostgreSQLInstance{}, handler.EnqueueRequestsFromMapFunc(r.requestsForInstance)). + Watches(&databasev1alpha1.PostgreSQLTenant{}, handler.EnqueueRequestsFromMapFunc(databaseRequestsForTenant)). + Watches(&databasev1alpha1.PostgreSQLInstance{}, handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, object client.Object) []ctrl.Request { + return databaseRequestsForInstance(ctx, r.Client, object) + })). Complete(r) } -func (r *CredentialReconciler) requestsForTenant(_ context.Context, object client.Object) []ctrl.Request { +func databaseRequestsForTenant(_ context.Context, object client.Object) []ctrl.Request { tenant := object.(*databasev1alpha1.PostgreSQLTenant) if tenant.Status.DatabaseRef == nil { return nil @@ -48,9 +50,9 @@ func (r *CredentialReconciler) requestsForTenant(_ context.Context, object clien return []ctrl.Request{{Name: string(tenant.Status.DatabaseRef.Name)}} } -func (r *CredentialReconciler) requestsForInstance(ctx context.Context, object client.Object) []ctrl.Request { +func databaseRequestsForInstance(ctx context.Context, cache client.Client, object client.Object) []ctrl.Request { databases := &databasev1alpha1.PostgreSQLDatabaseList{} - if err := r.Client.List(ctx, databases); err != nil { + if err := cache.List(ctx, databases); err != nil { ctrl.LoggerFrom(ctx).Error(err, "无法映射 Instance 凭据准备事件;等待低频重试") return nil } diff --git a/internal/database/controller/instance_controller_test.go b/internal/database/controller/instance_controller_test.go index 489a0c9..a897c14 100644 --- a/internal/database/controller/instance_controller_test.go +++ b/internal/database/controller/instance_controller_test.go @@ -18,6 +18,7 @@ import ( ) type instanceBackend struct { + application.ResourceDatabase checks instance.ManagementChecks err error inspect func() diff --git a/internal/database/controller/provisioning_controller.go b/internal/database/controller/provisioning_controller.go new file mode 100644 index 0000000..4d9e713 --- /dev/null +++ b/internal/database/controller/provisioning_controller.go @@ -0,0 +1,43 @@ +package controller + +import ( + "context" + "errors" + "time" + + databasev1alpha1 "git.ddupan.top/panxiao81/ayatori/api/database/v1alpha1" + "git.ddupan.top/panxiao81/ayatori/internal/database/application" + ctrl "sigs.k8s.io/controller-runtime" + "sigs.k8s.io/controller-runtime/pkg/client" + "sigs.k8s.io/controller-runtime/pkg/handler" +) + +type ProvisioningReconciler struct { + Client client.Client + Service *application.DatabaseReconciliation +} + +func NewProvisioningReconciler(cache client.Client, service *application.DatabaseReconciliation) *ProvisioningReconciler { + return &ProvisioningReconciler{Client: cache, Service: service} +} + +func (r *ProvisioningReconciler) Reconcile(ctx context.Context, request ctrl.Request) (ctrl.Result, error) { + operationContext, cancel := context.WithTimeout(ctx, 30*time.Second) + defer cancel() + if err := r.Service.Reconcile(operationContext, request.Name); err != nil { + return ctrl.Result{}, err + } + return ctrl.Result{RequeueAfter: dependencyRetry}, nil +} + +func (r *ProvisioningReconciler) SetupWithManager(manager ctrl.Manager) error { + if r.Client == nil || r.Service == nil { + return errors.New("provisioning controller requires injected client and use case") + } + return ctrl.NewControllerManagedBy(manager).Named("database-provisioning"). + For(&databasev1alpha1.PostgreSQLDatabase{}). + Watches(&databasev1alpha1.PostgreSQLTenant{}, handler.EnqueueRequestsFromMapFunc(databaseRequestsForTenant)). + Watches(&databasev1alpha1.PostgreSQLInstance{}, handler.EnqueueRequestsFromMapFunc(func(ctx context.Context, object client.Object) []ctrl.Request { + return databaseRequestsForInstance(ctx, r.Client, object) + })).Complete(r) +} diff --git a/internal/database/domain/credential/preparation.go b/internal/database/domain/credential/preparation.go index 12750a1..e6f3dff 100644 --- a/internal/database/domain/credential/preparation.go +++ b/internal/database/domain/credential/preparation.go @@ -106,14 +106,14 @@ func (t Target) Check() *Issue { return &Issue{Unavailable, "等待 Database 与 Tenant 双向绑定完成"} } if *database.Tenant != t.Tenant.Identity || *t.Tenant.Database != database.Identity { - return &Issue{Conflict, "双向绑定的名称或 UID 不匹配,未创建凭据"} + return &Issue{Conflict, "双向绑定的名称或 UID 不匹配,未继续供应"} } if t.Tenant.Deleting || t.Tenant.Phase != binding.Bound || !t.DatabaseProtected || !t.TenantProtected { - return &Issue{Stopped, "Tenant 未完成绑定、正在删除或缺少 finalizer 保护,未创建凭据"} + return &Issue{Stopped, "Tenant 未完成绑定、正在删除或缺少 finalizer 保护,未继续供应"} } request, err := t.Tenant.Request.Resolve(t.Tenant.Identity) if err != nil || (request.Provision != nil && !database.MatchesProvision(request, t.Tenant.Identity)) || request.Name != database.Identity.Name { - return &Issue{Conflict, "Tenant 申请与 Database 目标不一致,未创建凭据"} + return &Issue{Conflict, "Tenant 申请与 Database 目标不一致,未继续供应"} } if t.Instance == nil || database.InstanceUID == "" { return &Issue{Unavailable, "等待 Instance 与已记录的实例身份"} diff --git a/internal/database/domain/provisioning/state.go b/internal/database/domain/provisioning/state.go new file mode 100644 index 0000000..df09845 --- /dev/null +++ b/internal/database/domain/provisioning/state.go @@ -0,0 +1,66 @@ +// Package provisioning 保存 PostgreSQL 资源供应的确认与恢复规则,不依赖后端或 API。 +package provisioning + +import "fmt" + +type Phase string + +const ( + Pending Phase = "Pending" + CreatingRole Phase = "CreatingRole" + CreatingDatabase Phase = "CreatingDatabase" + Available Phase = "Available" + Conflict Phase = "Conflict" + Unavailable Phase = "Unavailable" + Stopped Phase = "Stopped" +) + +// OID 是已成功创建并回读的对象身份,不是从名称推导出的管理授权。 +// 仅保存在 CR;不建立 PostgreSQL registry,也不承诺备份还原后的自动认领。 +type State struct { + RoleOID uint32 + DatabaseOID uint32 + Phase Phase + Message string +} + +type Observation struct { + RoleOID uint32 + DatabaseOID uint32 + RoleSafe bool + OwnerOID uint32 + PublicConnect bool + AllowConnections bool +} + +func (s State) WithPhase(phase Phase, message string) State { + s.Phase, s.Message = phase, message + return s +} + +func (s State) Resume() (State, bool) { + if s.Phase == Conflict { + return s, false + } + if (s.Phase == CreatingRole && s.RoleOID == 0) || (s.Phase == CreatingDatabase && s.DatabaseOID == 0) { + return s.WithPhase(Conflict, "外部创建未留下成功确认;请核对目标角色和数据库,不自动认领或重复创建"), false + } + return s, true +} + +// Check 既阻止未知同名对象,也拒绝已确认对象消失、被重建或权限漂移。 +func (s State) Check(o Observation) error { + if s.RoleOID != o.RoleOID { + return fmt.Errorf("角色身份不匹配:记录 OID=%d,观察 OID=%d", s.RoleOID, o.RoleOID) + } + if s.DatabaseOID != o.DatabaseOID { + return fmt.Errorf("数据库身份不匹配:记录 OID=%d,观察 OID=%d", s.DatabaseOID, o.DatabaseOID) + } + if o.RoleOID != 0 && !o.RoleSafe { + return fmt.Errorf("已确认角色的登录属性、特权或成员关系发生变化") + } + if o.DatabaseOID != 0 && o.OwnerOID != s.RoleOID { + return fmt.Errorf("数据库 owner 与已确认角色不匹配") + } + return nil +} diff --git a/internal/database/domain/provisioning/state_test.go b/internal/database/domain/provisioning/state_test.go new file mode 100644 index 0000000..4cd9952 --- /dev/null +++ b/internal/database/domain/provisioning/state_test.go @@ -0,0 +1,52 @@ +package provisioning + +import "testing" + +func TestResume(t *testing.T) { + for _, test := range []struct { + name string + state State + resume bool + }{ + {"初次", State{}, true}, + {"角色创建不确定", State{Phase: CreatingRole}, false}, + {"库创建不确定", State{RoleOID: 11, Phase: CreatingDatabase}, false}, + {"已确认角色", State{RoleOID: 11, Phase: Pending}, true}, + {"已确认资源暂不可用", State{RoleOID: 11, DatabaseOID: 22, Phase: Unavailable}, true}, + {"冲突保持", State{RoleOID: 11, Phase: Conflict, Message: "首次诊断"}, false}, + } { + t.Run(test.name, func(t *testing.T) { + state, resume := test.state.Resume() + if resume != test.resume || state.RoleOID != test.state.RoleOID || state.DatabaseOID != test.state.DatabaseOID { + t.Fatal("恢复决策丢失确认或错误放行") + } + if !resume && state.Phase != Conflict { + t.Fatal("不确定创建未转冲突") + } + if test.state.Phase == Conflict && state != test.state { + t.Fatal("冲突诊断被覆盖") + } + }) + } +} + +func TestObservedIdentity(t *testing.T) { + state := State{RoleOID: 11, DatabaseOID: 22} + good := Observation{RoleOID: 11, DatabaseOID: 22, RoleSafe: true, OwnerOID: 11} + if err := state.Check(good); err != nil { + t.Fatal(err) + } + for _, observed := range []Observation{ + {}, {RoleOID: 12, DatabaseOID: 22, RoleSafe: true, OwnerOID: 12}, + {RoleOID: 11, DatabaseOID: 23, RoleSafe: true, OwnerOID: 11}, + {RoleOID: 11, DatabaseOID: 22, RoleSafe: false, OwnerOID: 11}, + {RoleOID: 11, DatabaseOID: 22, RoleSafe: true, OwnerOID: 99}, + } { + if state.Check(observed) == nil { + t.Fatal("对象身份或权限漂移被接受") + } + } + if (State{}).Check(good) == nil { + t.Fatal("未确认同名对象被认领") + } +}