Keep microsandbox checkpoints as native snapshots

docs/getting-started/operations.md

@@ -165,7 +165,7 @@ An installation with independent workspace storage or nodes needs the full proce
 - For a [quota mode](../workspace-provider.md#quota-mode) namespace, the entire XFS filesystem holding the export as one block-level copy: a storage or LVM snapshot, or an image of the frozen or unmounted device. Only such a copy keeps project IDs, quota limits, inode identities, the filesystem UUID and the `oac-workspace-quota` journal consistent. A file-level copy, `tar`, `rsync` or `xfsdump` loses them; the command then refuses the restored objects instead of serving them unbounded.
 - Each node's state directory, `/var/lib/oac-node/.oac/nodes/<installation-id>/`, and its provider storage: Docker volumes or microsandbox's store. Include all compute and snapshot resources still referenced by the database. See [when a node host fails](./nodes.md#when-a-node-host-fails).
 
-- Each private `checkpoint_root` in its entirety, including `.oac-checkpoint-store`, archives, operation journals, lock files and tombstones, together with the matching Core database. It is separate from the native SDK store and independent workspace namespace. Preserve its identity and private ownership; the [checkpoint transfer contract](../sandbox-provider.md#checkpoint-transfer) owns its lifecycle.
+- Each private `checkpoint_root` in its entirety, including `.oac-checkpoint-store`, checkpoints, operation journals, lock files and tombstones, together with the matching Core database. It is separate from the native SDK store and independent workspace namespace. Preserve its identity and private ownership; the [checkpoint transfer contract](../sandbox-provider.md#checkpoint-transfer) owns its lifecycle.
 
 ### Establish a stopped-write window
 

docs/sandbox-provider.md

@@ -256,11 +256,11 @@ A Provider relies on these rules:
 
 A checkpoint-capable generation reports `CheckpointCompatibility`: opaque `ArtifactDomain` and `ExecutionClass` tokens. Every verified `SnapshotIdentity` carries the same qualification. Core compares these values without interpreting CPU features, filesystem paths or storage implementations. A restore destination must be online and ready for the allocation's immutable deployment generation, match both tokens, and have active capacity plus a retained slot when moving from another node. The allocation, Device and Session identities remain unchanged. The Session and deployment transaction commits the destination route, placement and exact restore intent before target-side native work. No capacity or compatible destination leaves the retained snapshot owned until its configured deadline. The target must already hold that exact generation; Core does not automatically prepare historical generations on new nodes. A node keeps a superseded generation's provider until its checkpoint retention obligations end.
 
-`Suspend` may report `SourceStopped` only after publishing a verified full archive durably, stopping the exact source compute and settling source-local capture and cleanup obligations. Core persists the snapshot before marking it suspended; native absence alone is insufficient. `Suspend.ObserveOnly` may finish archive publication and source cleanup for an already verified snapshot of the same operation, but cannot capture another snapshot or stop a source that has resumed running. If a previously published archive is missing or damaged, `ObserveOnly` may return its exact durable ownership receipt with `Status: unknown` and `SourceStopped: false` for cleanup; that receipt proves neither current archive usability nor source absence. `Resume` still verifies the archive before execution. After the source-settlement barrier, source-node unavailability does not prevent restoration. Deletion and ordinary expiry can transfer an idle suspended allocation's cleanup route to an online node in the same artifact domain; deletion does not require execution-class compatibility. The cleanup destination must already serve the same immutable generation and have retained capacity. Without such a node, cleanup stays pending and ownership is not released; operators must restore an eligible node or its capacity. An unknown running or restoring writer never moves merely because its node is unreachable.
+`Suspend` may report `SourceStopped` only after publishing a verified full checkpoint durably, stopping the exact source compute and settling source-local capture and cleanup obligations. Core persists the snapshot before marking it suspended; native absence alone is insufficient. `Suspend.ObserveOnly` may finish checkpoint publication and source cleanup for an already verified snapshot of the same operation, but cannot capture another snapshot or stop a source that has resumed running. If a previously published checkpoint is missing or damaged, `ObserveOnly` may return its exact durable ownership receipt with `Status: unknown` and `SourceStopped: false` for cleanup; that receipt proves neither current checkpoint usability nor source absence. `Resume` still verifies the checkpoint before execution. After the source-settlement barrier, source-node unavailability does not prevent restoration. Deletion and ordinary expiry can transfer an idle suspended allocation's cleanup route to an online node in the same artifact domain; deletion does not require execution-class compatibility. The cleanup destination must already serve the same immutable generation and have retained capacity. Without such a node, cleanup stays pending and ownership is not released; operators must restore an eligible node or its capacity. An unknown running or restoring writer never moves merely because its node is unreachable.
 
-Recovery uses `Resume.ObserveOnly` for the persisted target and operation. `RestoreAttemptClosed` is an exact operation ID, not a generic absence flag: the adapter may return it only after durably closing an attempt that never entered native execution and fencing every late request for that ID. An adapter may also close a durably admitted attempt when it can prove that its current invocation failed before dispatching native restoration. The microsandbox adapter does this only when the fresh request's wire deadline expires before native dispatch; archive corruption, qualification errors and opaque SDK failures do not authorize an automatic retry. Core then persists a new operation ID before attempting execution. Once native restoration has been dispatched, an unknown outcome retains the same target, ownership and original retention deadline; a missing native listing, helper death or timeout does not authorize replay. Such a restore can remain unavailable until explicit deletion or ordinary retention expiry, and pending inputs can reach their public deadline while waiting. User deletion still requires exact native settlement before resources and ownership can be released.
+Recovery uses `Resume.ObserveOnly` for the persisted target and operation. `RestoreAttemptClosed` is an exact operation ID, not a generic absence flag: the adapter may return it only after durably closing an attempt that never entered native execution and fencing every late request for that ID. An adapter may also close a durably admitted attempt when it can prove that its current invocation failed before dispatching native restoration. The microsandbox adapter does this only when the fresh request's wire deadline expires before native dispatch; checkpoint corruption, qualification errors and opaque SDK failures do not authorize an automatic retry. Core then persists a new operation ID before attempting execution. Once native restoration has been dispatched, an unknown outcome retains the same target, ownership and original retention deadline; a missing native listing, helper death or timeout does not authorize replay. Such a restore can remain unavailable until explicit deletion or ordinary retention expiry, and pending inputs can reach their public deadline while waiting. User deletion still requires exact native settlement before resources and ownership can be released.
 
-The microsandbox adapter requires an explicit private `checkpoint_root`, outside every guest-accessible filesystem. Its directory is mode `0700`; a private namespace marker identifies the actual archive store. The default installation creates a node-local private store. Cross-node restoration requires operators to mount the same private store on each participating node; mount paths may differ. Runtime state directories and workspace bindings are never used to infer this root. The adapter verifies the archive, external filesystem identity and native execution class again before restoration. Restored RAM and file descriptors do not promise continuity of external TCP connections or replay safety for application side effects.
+The microsandbox adapter requires an explicit private `checkpoint_root`, outside every guest-accessible filesystem. Its directory is mode `0700`; a private namespace marker identifies the actual checkpoint store. The default installation creates a node-local private store. Cross-node restoration requires operators to mount the same private store on each participating node; mount paths may differ. Each retained checkpoint is an uncompressed native snapshot of the guest's memory and root-disk layers, so a shared store carries that volume for every suspension and cross-node restore. Runtime state directories and workspace bindings are never used to infer this root. The adapter verifies the checkpoint, external filesystem identity and native execution class again before restoration. Restored RAM and file descriptors do not promise continuity of external TCP connections or replay safety for application side effects.
 
 ### Cold replacement after checkpoint retention
 

docs/zh/getting-started/operations.md

@@ -1,7 +1,7 @@
 ---
 title: "运维"
 source: docs/getting-started/operations.md
-source_hash: 2f6435641f47d00db1ae4556594d99367223270013dfdf282c16648f8a6cd275
+source_hash: c61c4d11097be8e0718c6c095c55b10a47020e5a2f83f24c504eb6753ca00b5c
 ---
 
 Core 安装运维人员的日常任务：`oac` 命令、服务健康状态、Core 密钥、备份和卸载。节点有[专门的指南](./nodes.md)，[故障排查](./troubleshooting.md)列出了安装程序和访问方面的提示信息。设置见[配置参考](../configuration.md)。
@@ -167,7 +167,7 @@ docker run --rm -v "${project}_data:/data" -v ~/oac-backup:/backup:ro --entrypoi
 - 对于[配额模式](../workspace-provider.md#quota-mode)命名空间，将承载导出的整个 XFS 文件系统作为一份块级副本备份：存储或 LVM 快照，或已冻结或已卸载设备的镜像。只有这样的副本能保持 project ID、配额限制、inode 标识、文件系统 UUID 和 `oac-workspace-quota` 日志库的一致。文件级复制、`tar`、`rsync` 或 `xfsdump` 会丢失这些信息；此时该命令会拒绝恢复后的对象，而不是在无上限状态下提供它们。
 - 各节点状态目录 `/var/lib/oac-node/.oac/nodes/<installation-id>/` 及其提供商存储：Docker 卷或 microsandbox 存储。包含数据库仍然引用的全部计算资源和快照。参见[节点主机故障时](./nodes.md#when-a-node-host-fails)。
 
-- 每个私有 `checkpoint_root` 的全部内容，包括 `.oac-checkpoint-store`、归档、operation journal、lock 文件与 tombstone，并与匹配的 Core 数据库一起保留。它独立于原生 SDK store 和工作区 namespace。保留其身份和私有所有权；[检查点转移契约](../sandbox-provider.md#checkpoint-transfer)定义其生命周期。
+- 每个私有 `checkpoint_root` 的全部内容，包括 `.oac-checkpoint-store`、检查点、operation journal、lock 文件与 tombstone，并与匹配的 Core 数据库一起保留。它独立于原生 SDK store 和工作区 namespace。保留其身份和私有所有权；[检查点转移契约](../sandbox-provider.md#checkpoint-transfer)定义其生命周期。
 
 ### 建立停止写入窗口 {#establish-a-stopped-write-window}
 

docs/zh/sandbox-provider.md

@@ -1,7 +1,7 @@
 ---
 title: "沙箱提供商协议"
 source: docs/sandbox-provider.md
-source_hash: 9d36546c75d10805c104092536d8bb13a87b887bef1573f2f4fd5ee94d1673c7
+source_hash: 6476bd4c551ee002c2006eba645cfd6ebe3f107b9369570d2fa386fb9c0ef132
 ---
 
 沙箱提供商协议连接 Core 与为 Core 管理的（`openai_hosted`）Environment 提供计算资源的 adapter，并由该 adapter 运行启动 Runtime daemon 的有界引导流程。本页面向 adapter 作者：说明提供商必须实现什么，以及可以依赖 Core 做什么。Docker、microsandbox 和 E2B 实现了该协议。
@@ -258,11 +258,11 @@ Core 在初始化完成、没有 root 或 Subagent Turn 处于排队、进行中
 
 支持检查点的 generation 报告 `CheckpointCompatibility`，包含不透明的 `ArtifactDomain` 与 `ExecutionClass` token。每个已验证 `SnapshotIdentity` 携带同样的资格。Core 只比较这些值，不解释 CPU 特性、文件系统路径或存储实现。恢复目标必须在线、对 allocation 的不可变 deployment generation 已就绪、两个 token 均匹配，并有 active 容量；跨 node 时还需 retained slot。allocation、Device 和 Session 身份保持不变。Session 与 deployment 事务先提交目标路由、placement 和精确 restore intent，再执行目标侧原生操作。没有容量或兼容目标时，保留快照的所有权一直持续到配置期限。目标必须已持有该精确 generation；Core 不会在新节点自动准备历史 generation。节点会保留被取代的 generation 的提供商，直到检查点保留义务结束。
 
-`Suspend` 只有在持久发布已验证的完整归档、停止精确 source compute，并结清 source 本地 capture 与 cleanup 义务后，才能报告 `SourceStopped`。Core 先持久化快照，再标记 suspended；仅原生资源不存在并不足够。`Suspend.ObserveOnly` 可以为同一 operation 已验证的快照完成归档发布和源清理，但不能重新捕获快照，也不能停止已经恢复 running 的源。此前已发布的归档缺失或损坏时，`ObserveOnly` 可以返回其精确持久 ownership receipt，标记 `Status: unknown` 和 `SourceStopped: false`，供清理使用；该 receipt 不证明归档当前可用，也不证明源不存在。`Resume` 仍在执行前验证归档。完成源结清屏障后，source node 不可用不再阻止恢复。删除和正常到期可以将空闲 suspended allocation 的清理路由转给同一 artifact domain 内的在线 node；删除不要求 execution class 匹配。清理目标必须已提供同一不可变 generation，且有 retained 容量。没有这样的 node 时，清理保持 pending，不释放所有权；运维需恢复符合条件的 node 或其容量。未知 running 或 restoring writer 不会仅因 node 不可达而迁移。
+`Suspend` 只有在持久发布已验证的完整检查点、停止精确 source compute，并结清 source 本地 capture 与 cleanup 义务后，才能报告 `SourceStopped`。Core 先持久化快照，再标记 suspended；仅原生资源不存在并不足够。`Suspend.ObserveOnly` 可以为同一 operation 已验证的快照完成检查点发布和源清理，但不能重新捕获快照，也不能停止已经恢复 running 的源。此前已发布的检查点缺失或损坏时，`ObserveOnly` 可以返回其精确持久 ownership receipt，标记 `Status: unknown` 和 `SourceStopped: false`，供清理使用；该 receipt 不证明检查点当前可用，也不证明源不存在。`Resume` 仍在执行前验证检查点。完成源结清屏障后，source node 不可用不再阻止恢复。删除和正常到期可以将空闲 suspended allocation 的清理路由转给同一 artifact domain 内的在线 node；删除不要求 execution class 匹配。清理目标必须已提供同一不可变 generation，且有 retained 容量。没有这样的 node 时，清理保持 pending，不释放所有权；运维需恢复符合条件的 node 或其容量。未知 running 或 restoring writer 不会仅因 node 不可达而迁移。
 
-恢复用 `Resume.ObserveOnly` 观察已持久化的 target 与 operation。`RestoreAttemptClosed` 是精确 operation ID，不是通用 absence 标志：adapter 只有在持久关闭从未进入原生执行的 attempt，并隔离该 ID 的所有迟到请求后才能返回。如果 adapter 能证明当前调用在派发原生恢复之前失败，也可以关闭已持久 admitted 的 attempt。microsandbox adapter 仅在 fresh 请求的 wire deadline 于原生派发之前到期时执行这种关闭；归档损坏、资格错误和不透明 SDK 失败都不授权自动重试。Core 随后先持久化新的 operation ID，再尝试执行。一旦已派发原生恢复，结果未知的 attempt 保留相同 target、所有权与原始 retention deadline；原生列表缺失、helper 死亡或超时都不授权重放。这种恢复可能一直不可用，直到显式删除或正常保留期到期；等待中的 input 可能达到其公共 deadline。用户删除仍需精确原生 settlement，之后才能释放资源与所有权。
+恢复用 `Resume.ObserveOnly` 观察已持久化的 target 与 operation。`RestoreAttemptClosed` 是精确 operation ID，不是通用 absence 标志：adapter 只有在持久关闭从未进入原生执行的 attempt，并隔离该 ID 的所有迟到请求后才能返回。如果 adapter 能证明当前调用在派发原生恢复之前失败，也可以关闭已持久 admitted 的 attempt。microsandbox adapter 仅在 fresh 请求的 wire deadline 于原生派发之前到期时执行这种关闭；检查点损坏、资格错误和不透明 SDK 失败都不授权自动重试。Core 随后先持久化新的 operation ID，再尝试执行。一旦已派发原生恢复，结果未知的 attempt 保留相同 target、所有权与原始 retention deadline；原生列表缺失、helper 死亡或超时都不授权重放。这种恢复可能一直不可用，直到显式删除或正常保留期到期；等待中的 input 可能达到其公共 deadline。用户删除仍需精确原生 settlement，之后才能释放资源与所有权。
 
-microsandbox adapter 要求显式的私有 `checkpoint_root`，位于所有 guest 可访问文件系统之外。目录权限为 `0700`，私有 namespace marker 标识真实归档存储。默认安装创建 node 本地私有 store。跨 node 恢复要求运维为参与 node 挂载同一个私有 store，挂载路径可以不同。Runtime state 目录与 workspace binding 都不用于推导该 root。adapter 在恢复前再次验证归档、外部文件系统身份和原生 execution class。恢复 RAM 与文件描述符不保证外部 TCP 连接连续，也不保证应用副作用可安全重放。
+microsandbox adapter 要求显式的私有 `checkpoint_root`，位于所有 guest 可访问文件系统之外。目录权限为 `0700`，私有 namespace marker 标识真实检查点存储。默认安装创建 node 本地私有 store。跨 node 恢复要求运维为参与 node 挂载同一个私有 store，挂载路径可以不同。每个保留的检查点都是 guest 内存与根磁盘层的未压缩原生快照，因此共享 store 要为每次挂起和跨 node 恢复承载这部分数据量。Runtime state 目录与 workspace binding 都不用于推导该 root。adapter 在恢复前再次验证检查点、外部文件系统身份和原生 execution class。恢复 RAM 与文件描述符不保证外部 TCP 连接连续，也不保证应用副作用可安全重放。
 
 ### 检查点保留期后的冷替换 {#cold-replacement-after-checkpoint-retention}
 

services/core/internal/sandbox/microsandbox/checkpoint_linux.go

@@ -16,7 +16,7 @@ import (
 	"github.com/MiniMax-AI/OpenAgentCore/services/core/internal/sandbox"
 )
 
-// CheckpointStore opens only an explicitly configured, private archive root.
+// CheckpointStore opens only an explicitly configured, private checkpoint root.
 // Its marker is an operator-owned identity, never inferred from workspace paths.
 func CheckpointStore(c Config) (string, error) {
 	if !filepath.IsAbs(c.CheckpointRoot) || filepath.Clean(c.CheckpointRoot) != c.CheckpointRoot {