agent-dx/platform · openJiuwen/agent-dx

neatlinefix(platform): preserve cross-node recovery identity and retries

Files	Last commit	Last update
adxlet	fix(platform): preserve cross-node recovery identity and retries Signed-off-by: neatline <fuyipeng3@huawei.com>	1 day ago
api	!176 merge codex/cross-node-recovery into master fix(platform): preserve cross-node recovery identity and retries Created-by: neatline Commit-by: neatline Merged-by: robbluo Description: Paired: [GitHub #226](https://github.com/openJiuwen-ai/agent-dx/pull/226) ↔ [GitCode !176](https://gitcode.com/openJiuwen/agent-dx/merge_requests/176) /kind bug Pod 替换触发跨节点恢复时，调度器为 checkpoint 存储和源节点排除追加的条件会留在临时 spec 中，与 Redis 的权威 Environment spec 不一致。托管卷执行授权因此校验冲突，无卷路径也会留下不一致的调度状态。恢复 Assignment 持久化后，将调度请求和放置快照恢复为权威 spec，保持 Assignment、generation、设备与资源预留；其他规格或归属变化仍被拒绝。恢复按失败阶段决定重试：本地容量不足、后端 restore 前的临时执行准入失败可重试；checkpoint 读取、后端 restore/初始化/readiness/路由失败，在确认清理后进入 Failed。明确的控制 RPC Unauthenticated、PermissionDenied 复用已有 BackendFailure 错误码，保留原因并停止准入重试；Internal 保留诊断信息，继续按原阶段策略处理。不新增错误类型。清理不确定时保留必要占用，Running 结果发布失败仍只补交结果。授权已落库但回复失败时，保留本次 runtime id；确认清理后的 Paused 提交携带 stopped_runtime_ids，Coordinator 释放独占 Occupancy，下一次授权使用该 runtime 作为 previous。新增回归收敛为 8 个：Coordinator 规格与预留不变量 2 个，自动/按需跨节点卷恢复及授权后错误处理 3 个，adxlet 可重试/不可重试边界的表驱动测试 2 个，协议错误映射 1 个。合并共享 fixture，移除重复的阶段测试；HTTP 超时语义复用现有错误码矩阵断言。同步恢复和错误契约文档。 ## 验证基于 `upstream/master`（`6a4c6d87`）验证： - 237 个不同测试通过（包含既有用例）：Protocol 33、adxlet pause/resume 81、Coordinator 恢复单元测试 2、RPC 集成 65、按需启动/恢复 24、调度 17、现有 HTTP 错误契约 15。 - 4 个 shared-checkpoint 恢复场景额外通过默认 2 MiB 测试线程栈复验。 - `cargo fmt --all -- --check` 通过。 - `cargo clippy --workspace --all-targets --all-features --locked --offline -j2 -- -D warnings` 通过。 - `python3 build/docs/check.py`、`git diff --check` 通过。本地集成使用真实 Redis 和 mTLS，RuntimeDriver、readiness 与共享对象存储为测试夹具。授权错误注入发生在真实 RPC 完成 Redis 授权后的回复边界。不代表真实 sandboxd、共享文件系统或 Kubernetes 滚动升级验收。TLS/连接失败若没有明确的认证拒绝状态，仍按依赖不可用处理；本补丁不会自动复活历史 Failed 记录。 <!-- bot4-pr-sync-meta schema_version: 1 source: gitcode gitcode_repo: openJiuwen/agent-dx gitcode_mr: 176 github_repo: openJiuwen-ai/agent-dx github_pr: 226 github_branch: sync/pr-176 --> See merge request: openJiuwen/agent-dx!176	1 day ago
coordinator	fix(platform): preserve cross-node recovery identity and retries Signed-off-by: neatline <fuyipeng3@huawei.com>	1 day ago
crates	fix(platform): preserve cross-node recovery identity and retries Signed-off-by: neatline <fuyipeng3@huawei.com>	1 day ago
deployment	feat(trace): support core flow tracing Signed-off-by: yanzidi <yanzidi@huawei.com>	3 days ago
runtime	fix(runtime): address command observation review findings Copy exit-state mutex values without overlapping guards, preserve structured observation failures, and validate HTTP wait results by stable command identity. Cover error metadata, legacy contracts and lock ordering with regressions and update the runtime data contract. Signed-off-by: 月可 <hukeyang.hky@antgroup.com>	1 day ago
sdk	fix(sdk): preserve command observation HTTP error metadata Enrich flat Execd observation errors without replacing their HTTP exception, status, payload or request identity. Cover all command read paths and the real HTTP transport with offline regressions. Signed-off-by: 月可 <hukeyang.hky@antgroup.com>	1 day ago