Merge remote-tracking branch 'origin/master'
This commit is contained in:
@@ -31,6 +31,11 @@ Host workspace、k3s、Windows、GitHub issue/PR route 见 [references/routes.md
|
||||
|
||||
## P0 边界
|
||||
|
||||
- `trans --help` 是高频直接阅读入口:
|
||||
- 顶层只展示 route 语法、operation 分组和 `trans --help <operation>` 下钻索引。
|
||||
- 顶层与 scoped help 都必须有界且直接返回;出现 `outputTruncated` 或 dump 即视为 CLI 缺陷。
|
||||
- 修复帮助过长时必须收敛信息结构,禁止提高 stdout 阈值。
|
||||
- dump 只允许用于显式 `--full`、`--raw` 或确实无法预先界定的一次性异常证据。
|
||||
- 远端文本修改优先 `trans <route> apply-patch`;不要 download/upload/sed 拼临时 diff 代替 patch。
|
||||
- `upload` / `download` 仅用于必要的二进制文件或生成物:
|
||||
- 远端文本读取优先使用 `cat` / `rg`;
|
||||
|
||||
@@ -42,6 +42,9 @@ description: UniDesk Web 开发与浏览器验证技能。用户处理 UniDesk/H
|
||||
- Workbench Kafka 实时 fanout 的重复验收:
|
||||
- 使用 `observe command --type validateRealtimeFanout --profile <yaml-profile> --provider <provider> --text <first> --second-text <second>`;
|
||||
- connected、refresh/reconnect 和 capability 期望来自 owning YAML,可独立组合,禁止在命令中硬编码 live-only 或 refresh mode;
|
||||
- 每轮只把 `user`、`assistant` 和 `terminal` 对应的 AgentRun/HWLAB 事件族作为必需核心事件;
|
||||
- `tool_call`/`command_output` 与 HWLAB `tool`/`status` 是条件事件对,仅在实际出现时校验数量和 lineage;
|
||||
- 直接答复不得为了通过探针而合成工具事件,超时证据必须披露已经观测和仍然缺失的事件族;
|
||||
- refresh 启用时必须证明 retention 到 live handoff、正式 `user` event、`hwlab.event.v1` 同 envelope,以及主页面单一 EventSource 和同一 UI reducer 终态;
|
||||
- observer 后台维持两个独立 product subscriber 和三条 debug SSE;
|
||||
- 命令提交后按 `status <observer> --command-id <command>` 轮询,不得回退为长连接 `web-probe script`。
|
||||
|
||||
@@ -602,16 +602,19 @@ templates:
|
||||
requiredEventTypes:
|
||||
agentrun:
|
||||
- user_message
|
||||
- tool_call
|
||||
- command_output
|
||||
- assistant_message
|
||||
- terminal_status
|
||||
hwlab:
|
||||
- user
|
||||
- tool
|
||||
- status
|
||||
- assistant
|
||||
- terminal
|
||||
conditionalEventTypePairs:
|
||||
tool-call:
|
||||
agentrun: tool_call
|
||||
hwlab: tool
|
||||
command-output:
|
||||
agentrun: command_output
|
||||
hwlab: status
|
||||
workbenchKafkaDebugReplay:
|
||||
enabled: true
|
||||
topic: hwlab.event.debug.v1
|
||||
|
||||
@@ -2,6 +2,12 @@
|
||||
|
||||
UniDesk 的统一 CLI 实现入口是根目录 `scripts/cli.ts`,运行方式固定为 `bun scripts/cli.ts <command>`;普通根 CLI 子命令仍使用该入口。`trans <route> ...` 是 SSH/WSL/k3s 透传专用入口,wrapper 委托轻量 `scripts/ssh-cli.ts` 启动链路,避免被无关根 CLI 子命令模块的解析或语法错误拖垮;长期参考文档、AGENTS 索引、CLI help 和人工远端操作示例都必须优先写 `trans ...`,不得再把 `bun scripts/cli.ts ssh ...` 作为默认透传入口。面向人工的默认输出采用 Kubernetes 风格的简洁表格、短摘要和 drill-down 命令;JSON 只用于 `--json`、`--raw`、`--full`、`-o json` 或其他明确机器消费模式。所有成功和失败路径仍必须给出可判定状态,避免无输出造成不可观测。
|
||||
|
||||
- 高频 CLI 默认输出与帮助输出是直接阅读入口:
|
||||
- 必须保持有界、低噪声并直接展示,不得触发 `outputTruncated` 或 `/tmp/unidesk-cli-output` dump。
|
||||
- 顶层帮助只展示分组索引和可执行的下钻命令,详细合同通过 `<command> --help <operation>` 或等价 scoped help 获取。
|
||||
- 默认输出或帮助输出触发 dump 视为 CLI 缺陷,必须收敛信息结构,禁止用提高 stdout 阈值掩盖。
|
||||
- dump 只允许用于用户显式请求的 `--full`、`--raw`、机器完整输出,或确实无法预先界定的一次性异常证据。
|
||||
|
||||
主 server 必须在 PATH 上提供 `/root/.local/bin/trans` 可执行 wrapper,内容委托 repo 内版本化 `scripts/trans` 并执行 `bun scripts/ssh-cli.ts ssh "$@"`;交互 shell 可额外提供 alias,但非交互 Codex `exec` 和脚本不能依赖 alias 展开。
|
||||
|
||||
主 server 固定提供 PATH 可见的 `/usr/local/bin/unidesk`,委托版本化 `scripts/unidesk` 执行根 CLI;`/root/.local/bin/unidesk` 可以保留为用户级兼容链接,但不能作为非交互入口的唯一安装位置。高频人工入口优先写 `unidesk <command>`;仓库内验证和未安装 wrapper 的 checkout 使用等价的 `bun scripts/cli.ts <command>`。
|
||||
@@ -274,7 +280,7 @@ GitHub issue/PR 正文局部修补必须优先使用 `trans gh:/owner/repo/issue
|
||||
|
||||
`trans <providerId> [ssh-like args...]` 是面向人的终端透传入口,不包装 JSON 输出,默认由 `scripts/ssh-cli.ts` 只加载 SSH/route/远程前端转发相关模块。主 server 本地执行时会在宿主机启动 `docker exec -i unidesk-backend-core backend-core --ssh-broker ...`,broker 只连接 backend-core 的 Docker 内网 `/ws/ssh`;core 使用 provider WebSocket 下发 open/dispatch 控制消息,但 stdin/stdout/stderr 数据面必须走 provider 主动连接 main server 的 `host.ssh.tcp-pool` TCP warm pool,provider-gateway 最终执行维护用 SSH 连接宿主或 WSL sshd。TTY 策略固定为交互登录 shell 使用 `ssh -tt`,带远端命令的会话使用 `ssh -T`;`apply-patch`、脚本 stdin、`py` 和旧 `apply-patch-v1` fallback 这类命令模式不得被伪终端回显或注入控制字符。该入口不暴露 database,也不改变 frontend/dev frontend/provider ingress 之外的业务边界;provider data TCP port 是 provider 主动连入的数据面端口,不是计算节点入站要求。
|
||||
|
||||
`trans --help` 和 `trans <providerId> --help` 是本地 JSON 帮助命令,必须快速返回;不能把 `--help` 解析成 Provider ID,不能打开交互 shell,也不能等待 provider 会话。
|
||||
`trans --help` 和 `trans <providerId> --help` 是本地 JSON 帮助命令,必须快速返回;不能把 `--help` 解析成 Provider ID,不能打开交互 shell,也不能等待 provider 会话。顶层只展示 route 语法、operation 分组和 scoped help 索引;operation 细节通过 `trans --help <operation>` 获取。任一默认帮助触发 `outputTruncated` 或 dump 都是缺陷,必须缩减顶层内容,禁止提高 stdout 阈值。
|
||||
|
||||
主 server 固定提供 `trans` 缩写,等价于 `bun scripts/ssh-cli.ts ssh "$@"` 的受控 UniDesk SSH 透传入口。这里必须同时保留两层入口:交互式 shell 可额外配置 alias;Codex `exec`、脚本和其他非交互 shell 不会自动展开 alias,所以还必须有 `/root/.local/bin/trans` 可执行 wrapper,内容固定为委托 repo 内版本化脚本:
|
||||
|
||||
|
||||
@@ -226,6 +226,30 @@ HWLAB 与 AgentRun 协同修复必须按 `docs/reference/agentrun.md` 的职责
|
||||
|
||||
HWLAB v0.3 订阅 AgentRun Kafka event 的权威输入是 `agentrun.event.v1`,内存 direct mapper 的权威输出是 `hwlab.event.v1`;三 topic 的 ownership 统一见 `docs/reference/platform-infra.md#kafka-event-bus-boundary`。`directPublish` 使用固定独立 group 消费 AgentRun event、调用生产 mapper 并直接发布完整 HWLAB envelope;`liveKafkaSse` 使用另一个固定 group 的进程级 shared consumer,再向当前 SSE subscriber fanout,同一 Kafka record 不为每个浏览器重复消费。两者不读取数据库、projection inbox/checkpoint/outbox 或 snapshot,不得回写 `agentrun.event.v1`,也不得把 `codex-stdio.raw.v1` 混入 HWLAB product schema。transactional projector、projection outbox relay 和 projection realtime 是三个可独立组合的 capability,关闭时必须保持不启动、不就绪门禁、不参与产品链路。
|
||||
|
||||
### 纯 Kafka 页面刷新与实时交接合同
|
||||
|
||||
- Workbench 的产品权威链固定为:
|
||||
- `codex-stdio.raw.v1 -> agentrun.event.v1 -> HWLAB direct mapper -> hwlab.event.v1 retention -> bounded refresh bootstrap -> shared live fanout -> SSE -> Web ingress/reducer`;
|
||||
- 用户输入必须先成为 `agentrun.event.v1` 的正式 `user_message`,再一对一映射为 `hwlab.event.v1` 的正式 user event;
|
||||
- 用户消息、assistant、tool、terminal 和 final 不得另建数据库、HTTP history 或浏览器本地事实源。
|
||||
- 页面刷新只允许从 retained `hwlab.event.v1` 恢复已授权的 session/trace:
|
||||
- 服务端先建立 shared live handoff 保护,再读取目标分区 end offset 作为 barrier;
|
||||
- barrier 内由有界 retention bootstrap 输出,barrier 后由 shared live fanout 输出;
|
||||
- overlap 只按 Kafka topic/partition/offset 与稳定 event identity 收敛;
|
||||
- messageId 必须由正式 trace/item identity 派生,禁止按正文、时间戳、数组位置或 DOM 内容去重与重排。
|
||||
- refresh bootstrap 是短生命周期读取器:
|
||||
- 到达 barrier 后必须退出,不得把每个浏览器连接变成新的长期 Kafka consumer;
|
||||
- 实时阶段继续复用进程级 shared consumer/fanout;
|
||||
- 同一 Workbench session 只建立一个产品 EventSource,刷新与重连不得复制业务 subscriber。
|
||||
- `kafkaRefreshReplay` 与 `liveKafkaSse` 是可组合、可独立开关的 capability:
|
||||
- topic、group、扫描预算、等待预算、live buffer 上限和失败策略只从选中 node/lane 的 owning YAML/runtime config 读取;
|
||||
- replay 与 live frame 必须进入同一 decode、queue、reducer、row model 和 card;
|
||||
- barrier 不可获得、扫描或 buffer 超限、授权 scope 不完整时必须 typed fail-closed,不得回退 projector/read model/snapshot/sync/gap-fill/finalizer、polling 或 HTTP history。
|
||||
- 关闭此类问题时先用 repo-owned CLI 复用生产 bootstrap/codec/reducer,披露 scanned、matched、applied、barrier、buffered、deduplicated、terminal 与 stable identity:
|
||||
- CLI 通过后,使用 semantic internal WebProbe 在同一已有 session 上执行只刷新、不提交新输入的验收;
|
||||
- 刷新前后 user/agent identity 和可见顺序必须一致,retention 与产品 SSE envelope 多重集必须精确相等;
|
||||
- OTel 应只出现 direct publish、shared fanout receive 和 product SSE write,projector/projection write span 必须为 0。
|
||||
|
||||
纯 Kafka 调试分成三个可复用入口,不以临时脚本作为长期路径。AgentRun 使用 `./scripts/agentrun kafka regenerate agentrun --session-id <ses_agentrun_...> --trace-id <trc_...>`,从 stdio Kafka/JSONL 调用生产 reducer,只输出 `agentrun.event.debug.v1` 并明示 partial reconstruction;HWLAB 使用 `hwlab-cli kafka regenerate hwlab --from kafka|jsonl --session-id <id> --trace-id <id>` 调用生产 mapper,只输出 `hwlab.event.debug.v1`。原 Workbench 页面的管理员 YAML 调试开关只在显式点击后按当前 traceId 建立唯一 debug group,清空隔离面板并复用生产 reducer/card;正式浏览器验收使用 `web-probe observe command <observer> --type validateWorkbenchKafkaDebugReplay`。typed command 继承 observer 的 YAML `internal|public` origin,不接受 URL/IP 来选择内外网。
|
||||
|
||||
产品实时链路的最小关闭证据必须来自一个真实 Workbench turn,并同时证明:AgentRun→HWLAB direct publish 数与 HWLAB shared fanout receive 数一致;产品 SSE 收到 terminal/final response;OTel 覆盖 `hwlab-cloud-api`、`agentrun-manager`、`agentrun-runner` 且 error=0;projection write/projector span 为 0。单步 debug 关闭证据另需证明 stdio frame→AgentRun debug→HWLAB debug→Workbench applied 的逐级 count、顺序、lineage 和 terminal,不得把 debug topic/group 冒充产品 topic/group。KafkaJS 2.2.4 `TimeoutNegativeWarning` 已由 `pikasTech/unidesk#1704` 跟踪空 pending queue 的 1ms 自循环;它不是数据丢失证据,也不得通过 suppress warning、全局 timer patch 或退回产品 polling 路径掩盖。
|
||||
|
||||
@@ -3,6 +3,7 @@ import { createHash } from "node:crypto";
|
||||
import { resolve } from "node:path";
|
||||
import { type UniDeskConfig, repoRoot } from "./config";
|
||||
import { runCommand } from "./command";
|
||||
import { resolveCliChildJsonCommandResult } from "./cli-child-json-recovery";
|
||||
import { coreInternalFetch } from "./microservices";
|
||||
import { capture, compactCapture, parseJsonOutput, shQuote } from "./platform-infra-ops-library";
|
||||
|
||||
@@ -659,11 +660,24 @@ console.log(output);
|
||||
stderrTail: result.stderr.slice(-1000),
|
||||
};
|
||||
}
|
||||
const jsonLine = result.stdout.trim().split("\n").reverse().find((line) => line.trim().startsWith("{"));
|
||||
if (jsonLine === undefined) {
|
||||
return { ok: false, status: 502, error: "NC01 decision-center k8s proxy returned no JSON", stdoutTail: result.stdout.slice(-1000), stderrTail: result.stderr.slice(-1000) };
|
||||
}
|
||||
return JSON.parse(jsonLine) as unknown;
|
||||
const resolved = resolveCliChildJsonCommandResult({
|
||||
result,
|
||||
requestedStdoutType: "NC01 decision-center k8s proxy JSON",
|
||||
acceptParsed: (value) => typeof value.ok === "boolean" && typeof value.status === "number",
|
||||
});
|
||||
if (resolved.parsed !== null) return resolved.parsed;
|
||||
return {
|
||||
ok: false,
|
||||
status: 502,
|
||||
error: "NC01 decision-center k8s proxy returned unparseable JSON",
|
||||
transport: {
|
||||
exitCode: result.exitCode,
|
||||
timedOut: result.timedOut,
|
||||
stdoutBytes: Buffer.byteLength(result.stdout),
|
||||
stderrBytes: Buffer.byteLength(result.stderr),
|
||||
},
|
||||
diagnostics: resolved.diagnostics,
|
||||
};
|
||||
}
|
||||
|
||||
async function decisionProxyAsync(
|
||||
|
||||
+166
-89
@@ -169,111 +169,188 @@ export function serverHelp(action: string | undefined = undefined): unknown {
|
||||
};
|
||||
}
|
||||
|
||||
export type SshHelpScope = "download";
|
||||
export type SshHelpScope =
|
||||
| "argv" | "exec" | "git" | "node" | "kubectl" | "logs"
|
||||
| "sh" | "bash" | "py" | "ps" | "cmd"
|
||||
| "pwd" | "ls" | "cat" | "head" | "tail" | "stat" | "wc" | "rg" | "find" | "glob" | "skills"
|
||||
| "apply-patch" | "apply-patch-v1" | "upload" | "download";
|
||||
|
||||
type SshOperationHelp = {
|
||||
group: "process" | "shell" | "filesystem" | "patch" | "transfer";
|
||||
description: string;
|
||||
usage: string[];
|
||||
notes?: string[];
|
||||
};
|
||||
|
||||
const SSH_OPERATION_HELP: Record<SshHelpScope, SshOperationHelp> = {
|
||||
argv: {
|
||||
group: "process",
|
||||
description: "以原始 argv 边界运行一个远端进程;需要管道、变量或重定向时改用 sh/bash。",
|
||||
usage: ["trans <route> argv <command> [args...]", "trans D601:/workspace argv git status --short --branch"],
|
||||
},
|
||||
exec: {
|
||||
group: "process",
|
||||
description: "在 k3s workload route 中运行进程,并可显式透传 stdin 或指定容器内 cwd。",
|
||||
usage: ["trans <provider>:k3s:<namespace>:<workload>[:<container>] exec [--cwd /path] [--stdin] -- <command> [args...]"],
|
||||
},
|
||||
git: {
|
||||
group: "process",
|
||||
description: "在 host/workspace 或 Windows workspace route 中运行 Git convenience wrapper。",
|
||||
usage: ["trans <provider>:/workspace git status --short --branch", "trans <provider>:win/<drive>/<path> git diff --check"],
|
||||
},
|
||||
node: {
|
||||
group: "process",
|
||||
description: "在目标 route 中直接运行 Node.js argv。",
|
||||
usage: ["trans <route> node -e 'console.log(process.version)'"],
|
||||
},
|
||||
kubectl: {
|
||||
group: "process",
|
||||
description: "在 k3s control-plane route 上执行有界 kubectl 诊断。",
|
||||
usage: ["trans <provider>:k3s kubectl get pods -n <namespace>"],
|
||||
},
|
||||
logs: {
|
||||
group: "process",
|
||||
description: "从 k3s control-plane 或 workload route 读取有界日志。",
|
||||
usage: ["trans <provider>:k3s:<namespace>:<workload> logs --tail 80", "trans <provider>:k3s logs -n <namespace> -l <selector> --tail 120"],
|
||||
},
|
||||
sh: {
|
||||
group: "shell",
|
||||
description: "从 stdin 或单个命令字符串执行远端 POSIX /bin/sh。",
|
||||
usage: ["trans <route> sh <<'SH'", "trans <route> sh -- '<command && command>'"],
|
||||
notes: ["sh -- 只接受一个 shell command string;单进程多 argv 使用 argv。"],
|
||||
},
|
||||
bash: {
|
||||
group: "shell",
|
||||
description: "仅在需要 pipefail、数组、[[ ]] 等 Bash 语法时执行远端 Bash。",
|
||||
usage: ["trans <route> bash <<'BASH'", "trans <route> bash -- '<bash command>'"],
|
||||
},
|
||||
py: {
|
||||
group: "shell",
|
||||
description: "从本地 stdin 执行远端 Python 源码,并保留 script argv。",
|
||||
usage: ["trans <route> py [script-args...] < script.py"],
|
||||
},
|
||||
ps: {
|
||||
group: "shell",
|
||||
description: "在 Windows route 中执行 PowerShell 源码或 argv;route 已包含 win。",
|
||||
usage: ["trans <provider>:win[/<drive>/<path>] ps <<'PS'", "trans <provider>:win ps '<PowerShell source>'"],
|
||||
},
|
||||
cmd: {
|
||||
group: "shell",
|
||||
description: "在 Windows route 中以 UTF-8 环境执行 cmd.exe/batch。",
|
||||
usage: ["trans <provider>:win[/<drive>/<path>] cmd <<'CMD'", "trans <provider>:win cmd ver"],
|
||||
},
|
||||
pwd: { group: "filesystem", description: "读取 Windows workspace 当前目录。", usage: ["trans <provider>:win/<drive>/<path> pwd"] },
|
||||
ls: { group: "filesystem", description: "有界列出 Windows workspace 文件。", usage: ["trans <provider>:win/<drive>/<path> ls --limit 50"] },
|
||||
cat: { group: "filesystem", description: "读取远端文本文件;二进制文件使用 download。", usage: ["trans <route> cat <path>"] },
|
||||
head: { group: "filesystem", description: "按行读取远端文本文件开头。", usage: ["trans <route> head -n 40 <path>"] },
|
||||
tail: { group: "filesystem", description: "按行读取远端文本文件结尾。", usage: ["trans <route> tail -n 40 <path>"] },
|
||||
stat: { group: "filesystem", description: "读取远端文件元数据。", usage: ["trans <route> stat <path>"] },
|
||||
wc: { group: "filesystem", description: "统计远端文本的行、词、字符或字节。", usage: ["trans <route> wc [-l|-w|-m|-c] <path>"] },
|
||||
rg: { group: "filesystem", description: "执行有界远端 ripgrep;Windows route 使用受控 UTF-8 子集。", usage: ["trans <route> rg [options] <pattern> [path...]"] },
|
||||
find: { group: "filesystem", description: "不依赖 shell quoting 的结构化远端 find。", usage: ["trans <route> find <path...> [--contains TEXT] [--limit N]"] },
|
||||
glob: { group: "filesystem", description: "通过远端 helper 执行有界 glob 匹配。", usage: ["trans <route> glob [--root DIR] [--pattern PATTERN] [--limit N]"] },
|
||||
skills: { group: "filesystem", description: "发现 WSL/Linux 及可选 Windows skill 目录。", usage: ["trans <provider> skills [--scope all|wsl|windows] [--limit N]"] },
|
||||
"apply-patch": {
|
||||
group: "patch",
|
||||
description: "默认远端文本修改入口;本地 v2 engine 计算变更,route 仅负责读写。",
|
||||
usage: ["trans <route> apply-patch [--cwd /path] < patch.diff"],
|
||||
notes: ["失败后重读当前目标块并缩小 hunk;不得自动回退 apply-patch-v1 或整文件 upload。"],
|
||||
},
|
||||
"apply-patch-v1": {
|
||||
group: "patch",
|
||||
description: "显式调用 legacy remote apply_patch helper;只作为人工确认后的兼容入口。",
|
||||
usage: ["trans <route> apply-patch-v1 [--allow-loose] < patch.diff"],
|
||||
},
|
||||
upload: {
|
||||
group: "transfer",
|
||||
description: "上传必要的二进制文件或生成物并自动核对字节数与 SHA-256。",
|
||||
usage: ["trans <route> upload <local-file> <remote-file>"],
|
||||
notes: ["远端文本读取使用 cat/rg,修改使用 apply-patch。"],
|
||||
},
|
||||
download: {
|
||||
group: "transfer",
|
||||
description: "下载必要的二进制文件或生成物并自动核对字节数与 SHA-256。",
|
||||
usage: ["trans <route> download <remote-file> <local-file>"],
|
||||
},
|
||||
};
|
||||
|
||||
const SSH_OPERATION_GROUPS = {
|
||||
process: ["argv", "exec", "git", "node", "kubectl", "logs"],
|
||||
shell: ["sh", "bash", "py", "ps", "cmd"],
|
||||
filesystem: ["pwd", "ls", "cat", "head", "tail", "stat", "wc", "rg", "find", "glob", "skills"],
|
||||
patch: ["apply-patch", "apply-patch-v1"],
|
||||
transfer: ["upload", "download"],
|
||||
} as const satisfies Record<SshOperationHelp["group"], readonly SshHelpScope[]>;
|
||||
|
||||
const SSH_HELP_SCOPES = Object.keys(SSH_OPERATION_HELP) as SshHelpScope[];
|
||||
const SSH_HELP_SCOPE_SET = new Set<string>(SSH_HELP_SCOPES);
|
||||
|
||||
function sshEntrypoint(): "trans" | "tran" | "ssh" {
|
||||
const rawEntrypoint = process.env.UNIDESK_SSH_ENTRYPOINT?.trim();
|
||||
return rawEntrypoint === "trans" || rawEntrypoint === "tran" || rawEntrypoint === "ssh" ? rawEntrypoint : "ssh";
|
||||
}
|
||||
|
||||
export function sshHelpScope(args: string[]): SshHelpScope | undefined {
|
||||
const [top, helpToken, scope] = args;
|
||||
if (top !== "ssh" || args.length !== 3 || !isHelpToken(helpToken)) return undefined;
|
||||
return scope === "download" ? scope : undefined;
|
||||
if (top !== "ssh" || args.length !== 3 || !isHelpToken(helpToken) || !SSH_HELP_SCOPE_SET.has(scope ?? "")) return undefined;
|
||||
return scope as SshHelpScope;
|
||||
}
|
||||
|
||||
export function sshHelp(scope: SshHelpScope | undefined = undefined): unknown {
|
||||
if (scope === "download") return sshDownloadHelp();
|
||||
if (scope !== undefined) return sshOperationHelp(scope);
|
||||
const entrypoint = sshEntrypoint();
|
||||
const operationGroups = Object.fromEntries(
|
||||
Object.entries(SSH_OPERATION_GROUPS).map(([group, operations]) => [group, operations.join(" | ")]),
|
||||
);
|
||||
return {
|
||||
command: "ssh",
|
||||
command: `${entrypoint} --help`,
|
||||
output: "json",
|
||||
description: "Open a Host SSH / WSL SSH maintenance session through the provider-gateway bridge.",
|
||||
usage: [
|
||||
"trans <route>",
|
||||
"trans <providerId> argv <command> [args...]",
|
||||
"trans <providerId>:/absolute/workspace apply-patch < patch.diff",
|
||||
"trans <route> upload <local-file> <remote-file>",
|
||||
"trans <route> download <remote-file> <local-file>",
|
||||
"trans <providerId> apply-patch-v1 [--allow-loose] < patch.diff",
|
||||
"trans <providerId> py [script-args...] < script.py",
|
||||
"trans <providerId> sh [arg...] <<'SH'",
|
||||
"trans <providerId> bash [arg...] <<'BASH'",
|
||||
"trans <providerId> skills [--scope all|wsl|windows] [--limit N]",
|
||||
"trans <providerId> find <path...> [--contains TEXT] [--limit N]",
|
||||
"trans <providerId> glob [--root DIR] [--pattern PATTERN]",
|
||||
"trans NC01:/root/hwlab-v03 git status --short --branch",
|
||||
"trans D601:win ps <<'PS'",
|
||||
"trans D601:win/c/test ps <<'PS'",
|
||||
"trans D601:win/c/test cmd <<'CMD'",
|
||||
"trans D601:win cmd ver",
|
||||
"trans D601:win/c/test cmd cd",
|
||||
"trans D601:win/c/test pwd",
|
||||
"trans D601:win/c/test ls --limit 50",
|
||||
"trans D601:win/c/test cat README.md",
|
||||
"trans D601:win/c/test head -n 40 README.md",
|
||||
"trans D601:win/c/test tail -n 40 README.md",
|
||||
"trans D601:win/c/test stat README.md",
|
||||
"trans D601:win/c/test wc README.md",
|
||||
"trans D601:win/c/test rg -i needle .",
|
||||
"trans D601:win/c/test git status --short --branch",
|
||||
"trans D601:win/c/test git diff --check",
|
||||
"trans D601:win/c/test git commit -m 'fix: update docs'",
|
||||
"trans D601:win/c/test git exact-commit status",
|
||||
"trans D601:win/c/test git exact-commit plan --path src/main.c",
|
||||
"trans D601:win/c/test git exact-commit run --path src/main.c --message 'fix: exact change' --confirm",
|
||||
"trans D601:win skills [--scope agents|codex|all] [--limit N]",
|
||||
"trans NC01:k3s",
|
||||
"trans NC01:k3s kubectl get pods -n hwlab-dev",
|
||||
"trans NC01:k3s",
|
||||
"trans NC01:k3s kubectl get pipelineruns -n hwlab-ci",
|
||||
"trans NC01:k3s sh <<'SH'",
|
||||
"trans NC01:k3s:hwlab-dev:hwlab-cloud-api:hwlab-cloud-api exec --cwd /app -- pwd",
|
||||
"trans NC01:k3s:hwlab-dev:hwlab-cloud-api:hwlab-cloud-api apply-patch --cwd /app <<'PATCH'",
|
||||
"trans NC01:k3s:hwlab-dev:hwlab-cloud-api apply-patch-v1 <<'PATCH'",
|
||||
"tar -C /path/to/files -cf - . | trans NC01:k3s:unidesk:code-queue exec --cwd /root/unidesk --stdin -- tar -xf - -C /root/unidesk",
|
||||
"trans D601:win/c/test apply-patch <<'PATCH'",
|
||||
"trans D601:win upload ./tool.mjs C:\\Temp\\tool.mjs",
|
||||
"trans D601:win download C:\\Temp\\tool.mjs ./tool.mjs",
|
||||
"trans NC01:k3s:hwlab-dev:hwlab-cloud-api node -e 'console.log(process.version)'",
|
||||
"trans NC01:k3s:hwlab-dev:hwlab-cloud-api sh <<'SH'",
|
||||
"trans NC01:k3s:hwlab-dev:hwlab-cloud-api logs --tail 80",
|
||||
"trans NC01:k3s logs -n agentrun-ci -l tekton.dev/pipelineRun=<run> --tail 120",
|
||||
],
|
||||
notes: [
|
||||
"trans --help and trans <route> --help print this JSON help and never open an interactive session; the underlying ssh subcommand keeps the same help behavior.",
|
||||
"For non-interactive remote commands, prefer argv for a single process and explicit sh/bash stdin for shell logic.",
|
||||
"Windows routes have explicit Windows operations, not POSIX shell aliases: `ps` runs Windows PowerShell from stdin or one inline command, `cmd` runs cmd.exe/batch from stdin or one command line, and `skills` discovers Windows skill directories.",
|
||||
"On Windows routes, the route already contains `win`; write `trans D601:win/c/test ps`, not `trans D601:win/c/test win ps`. Repository commands can use the git convenience wrapper, including `git status`, `git diff`, and non-interactive `git commit -m ...`; commands that need shell review should use `ps` or `cmd`.",
|
||||
"Windows routes include read-only filesystem convenience operations `pwd`, `ls`, `cat`, `head`, `tail`, `stat`, `wc`, and a bounded UTF-8 `rg` subset. These are implemented through a Windows fs backend with UTF-8/binary checks and bounded output; they do not imply POSIX `sh`/`bash` availability.",
|
||||
"For Windows PowerShell, use `trans <provider>:win ps <<'PS'`; the PowerShell body is written to a temporary .ps1 with UTF-8 settings and executed by powershell.exe. Do not use POSIX `sh` or `bash` for Windows PowerShell.",
|
||||
"For Windows cmd.exe, use `trans <provider>:win/<drive>/<path> cmd <<'CMD'`; `cmd` with no command-line arguments reads the UTF-8 batch body from stdin, injects UTF-8/Python encoding defaults, runs it from a temp .cmd file, and deletes the temp file.",
|
||||
"`argv` executes direct argv tokens only: `trans <route> argv ls -la` is valid, but `trans <route> argv 'ls -la'` is rejected because the single string would be treated as an executable path; use `sh -- 'ls -la'` or `bash -- 'ls -la'` for one-line shell logic.",
|
||||
"For one-line remote shell logic without a heredoc, use `sh -- '<command && command>'` for POSIX syntax or `bash -- '<command && command>'` for Bash syntax. Outer shell operators written outside trans, such as `trans NC01:/repo sed ... && sed ...`, are parsed by the local shell before UniDesk starts and therefore cannot be redirected by the CLI.",
|
||||
"`sh --` and `bash --` accept exactly one shell command string. For direct argv commands with multiple tokens, use `argv`, `exec`, or a known direct subcommand such as `git`, `rg`, `sed`, or `cat`.",
|
||||
"The removed `script` and `shell` operations intentionally fail. The operation name must declare the shell dialect: `sh` maps to target `/bin/sh`, while `bash` maps to target `bash`.",
|
||||
"sh and bash helper modes inject a tiny POSIX-compatible printf wrapper before user shell text, so portable printf headings such as `printf \"--- section ---\\n\"` work consistently under dash/sh and bash. Direct argv commands are unchanged.",
|
||||
"For arbitrary stdin streams into a workload command, use a workload route plus `exec --stdin -- <command> ...`; this keeps the route as location-only and avoids heredoc/base64/tar shell wrapping.",
|
||||
"`apply-patch` is the default remote text patch entry and uses the v2 local line-based patch engine with remote read/write operations, including Windows routes such as `D601:win/c/test`, so long Unicode/Chinese lines and pure insertion hunks avoid the legacy remote shell hunk parser. Plain multi-file Update File patches on POSIX host/k3s and Windows workspace routes use bulk read/write operations to avoid per-file SSH round trips. Its stdout follows Codex apply_patch text output rather than UniDesk JSON output; stderr keeps Codex-style failure text and appends one `UNIDESK_APPLY_PATCH_TIMING` JSON summary with durationMs, patchBytes, fileCount, hunkCount, changedCount, remoteOperationCount, remoteOperationCounts and remoteElapsedMs so slow patch runs can be attributed without changing success stdout.",
|
||||
"`upload` 和 `download` 只用于必要的二进制文件或生成物;远端文本应避免反复传输,读取使用 `cat`/`rg`,修改使用 `apply-patch`。成功传输的 JSON 会在 `hint` 中重复该提示,通过远端临时文件写入,自动核对两端字节数和 SHA-256,并返回 `verification.automatic=true`、`verification.verified=true` 和 `verification.match.{bytes,sha256}=true`;该 JSON 已是完整性证据,无需额外执行 `sha256sum`。下载通过 `host.ssh.tcp-pool` 流式传输并输出进度 JSON;受控异步产物任务可传入 `--inactivity-timeout-ms`,只要数据持续流动就不受普通短命令预算影响。Windows route 下载(例如 `trans D601:win download C:\\Temp\\tool.mjs ./tool.mjs`)会自动把盘符路径映射到同一 provider 的 WSL `/mnt/<drive>` host route,同时仍返回字节数和 SHA-256 校验证据。",
|
||||
"`apply-patch-v1` is the only legacy fallback entry: it rejects low-context update hunks by default, reports the matched file:line for each hunk on stderr, and only accepts --allow-loose when the caller has manually reviewed an intentionally ambiguous insertion.",
|
||||
"`sh` inherits provider proxy variables such as HTTP_PROXY/HTTPS_PROXY/ALL_PROXY/NO_PROXY and runs target `/bin/sh`; use `bash` only for Bash syntax such as `pipefail`, arrays, substring expansion, or `[[ ... ]]`, not as a proxy workaround.",
|
||||
"Route syntax is `{provider}:{plane}[:{scope...}] {operation} [operation-args...]`: the first argv token locates a distributed target only, and every following token belongs to the operation parser. Host workspace routes use `<provider>:/absolute/workspace`; WSL providers can use `<provider>:win ps` for Windows PowerShell and `<provider>:win cmd` for Windows cmd.exe, with `<provider>:win/c/test ...` mapping the Windows cwd to `C:\\test`; native k3s providers such as D601 and G14 use `<provider>:k3s` for the control plane and `<provider>:k3s:<namespace>:<workload>[:<container>]` for a workload/container. In k3s routes, `:` is the distributed route separator; `/...` is only an in-container filesystem cwd and never selects a container. Prefer operation `--cwd /path` when a container is also specified.",
|
||||
"Use `win`, not `win32`; the win route is a Windows operation plane. `ps` and `cmd` both set UTF-8/Python encoding defaults, while `cmd` also sets `chcp 65001`.",
|
||||
"`<provider>:win skills` discovers the current Windows user's `%USERPROFILE%\\.agents\\skills` by default; use `--scope all` to include `%USERPROFILE%\\.codex\\skills`.",
|
||||
"Do not put operation names in any colon route segment, including nested k3s namespace/workload/container segments.",
|
||||
"Do not use post-provider shorthand such as `trans NC01 k3s ...`; write `trans NC01:k3s ...` so location and operation stay separated.",
|
||||
"If an ssh-like remote command fails with timeout/kex/exit-255 friction, stderr includes one low-noise UNIDESK_SSH_HINT JSON line with the argv retry command.",
|
||||
"Ordinary non-interactive ssh/trans/tran remote commands have a hard top-level runtime timeout capped at 60s. Timeout writes UNIDESK_SSH_RUNTIME_TIMEOUT or UNIDESK_TRAN_TIMEOUT_HINT and disconnects the broker; long CI/CD, trace, logs, build, or hardware work must use submit-and-poll / short query loops instead of keeping trans open. Whole-file `download` is the narrow exception: controlled async callers can pass `--inactivity-timeout-ms` for a verified progress-emitting tcp-pool transfer.",
|
||||
"Only slow ssh/trans/tran runtime writes UNIDESK_SSH_TIMING JSON to stderr; operations over 10s are marked level=warning even when they succeed, because slow successful calls are a distributed performance monitoring signal. Check provider latency, remote command cost, helper bootstrap, or remote patch optimization before repeating high-frequency work. Routine short calls do not emit timing noise.",
|
||||
"The local trans/tran wrapper must not add provider/plane directory locks; rely on k8s/Tekton/Argo/Lease or server-side TTL queues for coordination.",
|
||||
"Use -- before a remote command that intentionally starts with a dash.",
|
||||
description: "通过 provider-gateway bridge 访问 Host SSH、WSL、Windows 或 k3s 目标。",
|
||||
routeSyntax: {
|
||||
general: `${entrypoint} <route> <operation> [operation-args...]`,
|
||||
hostWorkspace: "<provider>:/absolute/workspace",
|
||||
k3sControlPlane: "<provider>:k3s",
|
||||
k3sWorkload: "<provider>:k3s:<namespace>:<workload>[:<container>]",
|
||||
windowsWorkspace: "<provider>:win[/<drive>/<path>]",
|
||||
githubContent: "gh:/<owner>/<repo>[/issue|/pr]/<number>",
|
||||
},
|
||||
operationGroups,
|
||||
scopedHelp: {
|
||||
command: `${entrypoint} --help <operation>`,
|
||||
availableOperationCount: SSH_HELP_SCOPES.length,
|
||||
examples: [`${entrypoint} --help apply-patch`, `${entrypoint} --help download`, `${entrypoint} --help ps`],
|
||||
},
|
||||
boundaries: [
|
||||
"route 只定位目标,route 后的 token 全部交给 operation parser。",
|
||||
"单进程优先 argv;shell 逻辑显式使用 sh/bash,Windows 使用 ps/cmd。",
|
||||
"远端文本读取优先 cat/rg,修改优先 apply-patch;upload/download 仅用于必要的二进制或生成物。",
|
||||
"普通 trans 短连接上限保持 60s;长流程使用受控 submit-and-poll 入口。",
|
||||
],
|
||||
};
|
||||
}
|
||||
|
||||
function sshOperationHelp(scope: Exclude<SshHelpScope, "download">): unknown {
|
||||
const entrypoint = sshEntrypoint();
|
||||
const detail = SSH_OPERATION_HELP[scope];
|
||||
return {
|
||||
command: `${entrypoint} --help ${scope}`,
|
||||
output: "json",
|
||||
scope,
|
||||
group: detail.group,
|
||||
description: detail.description,
|
||||
routeSyntax: `${entrypoint} <route> ${scope} [operation-args...]`,
|
||||
usage: detail.usage.map((value) => value.replace(/^trans\b/u, entrypoint)),
|
||||
notes: detail.notes ?? [],
|
||||
next: [`${entrypoint} --help`, `${entrypoint} --help ${scope}`],
|
||||
};
|
||||
}
|
||||
|
||||
function sshDownloadHelp(): unknown {
|
||||
const configuredRepoRoot = process.env.UNIDESK_TRANS_REPO_ROOT?.trim() || null;
|
||||
const rawEntrypoint = process.env.UNIDESK_SSH_ENTRYPOINT?.trim();
|
||||
const entrypoint = rawEntrypoint === "trans" || rawEntrypoint === "tran" || rawEntrypoint === "ssh"
|
||||
? rawEntrypoint
|
||||
: "ssh";
|
||||
const entrypoint = sshEntrypoint();
|
||||
return {
|
||||
command: `${entrypoint} download`,
|
||||
command: `${entrypoint} --help download`,
|
||||
output: "json",
|
||||
scope: "download",
|
||||
description: "下载一个必要的二进制文件或生成物,并自动核对远端/本地字节数与 SHA-256;远端文本改用 cat/rg/apply-patch。",
|
||||
|
||||
@@ -206,6 +206,10 @@ export interface HwlabRuntimeWebProbeRealtimeFanoutProfileSpec {
|
||||
agentrun: readonly string[];
|
||||
hwlab: readonly string[];
|
||||
}>;
|
||||
readonly conditionalEventTypePairs: Readonly<Record<string, Readonly<{
|
||||
agentrun: string;
|
||||
hwlab: string;
|
||||
}>>>;
|
||||
}
|
||||
|
||||
export interface HwlabRuntimeWebProbeAuthLoginSpec {
|
||||
@@ -1514,6 +1518,24 @@ function webProbeRealtimeFanoutProfileConfig(value: unknown, path: string): Hwla
|
||||
const hwlabEventTypes = nonEmptyStringArrayField(requiredEventTypes, "hwlab", `${path}.requiredEventTypes`);
|
||||
if (new Set(agentrunEventTypes).size !== agentrunEventTypes.length) throw new Error(`${path}.requiredEventTypes.agentrun must not contain duplicates`);
|
||||
if (new Set(hwlabEventTypes).size !== hwlabEventTypes.length) throw new Error(`${path}.requiredEventTypes.hwlab must not contain duplicates`);
|
||||
const conditionalEventTypePairsRaw = asRecord(raw.conditionalEventTypePairs, `${path}.conditionalEventTypePairs`);
|
||||
if (Object.keys(conditionalEventTypePairsRaw).length === 0) throw new Error(`${path}.conditionalEventTypePairs must contain at least one pair`);
|
||||
const conditionalEventTypePairs = Object.fromEntries(Object.entries(conditionalEventTypePairsRaw).map(([id, value]) => {
|
||||
if (!/^[a-z][a-z0-9-]*$/u.test(id)) throw new Error(`${path}.conditionalEventTypePairs.${id} must use a stable lowercase id`);
|
||||
const pairPath = `${path}.conditionalEventTypePairs.${id}`;
|
||||
const pair = asRecord(value, pairPath);
|
||||
const unknownFields = Object.keys(pair).filter((field) => field !== "agentrun" && field !== "hwlab");
|
||||
if (unknownFields.length > 0) throw new Error(`${pairPath}.${unknownFields[0]} is not allowed`);
|
||||
const agentrun = stringField(pair, "agentrun", pairPath);
|
||||
const hwlab = stringField(pair, "hwlab", pairPath);
|
||||
if (agentrunEventTypes.includes(agentrun)) throw new Error(`${pairPath}.agentrun duplicates a required event type`);
|
||||
if (hwlabEventTypes.includes(hwlab)) throw new Error(`${pairPath}.hwlab duplicates a required event type`);
|
||||
return [id, { agentrun, hwlab }];
|
||||
}));
|
||||
const conditionalAgentrunTypes = Object.values(conditionalEventTypePairs).map((pair) => pair.agentrun);
|
||||
const conditionalHwlabTypes = Object.values(conditionalEventTypePairs).map((pair) => pair.hwlab);
|
||||
if (new Set(conditionalAgentrunTypes).size !== conditionalAgentrunTypes.length) throw new Error(`${path}.conditionalEventTypePairs agentrun types must not contain duplicates`);
|
||||
if (new Set(conditionalHwlabTypes).size !== conditionalHwlabTypes.length) throw new Error(`${path}.conditionalEventTypePairs hwlab types must not contain duplicates`);
|
||||
return {
|
||||
subscriberCount: 2,
|
||||
debugStreams: debugStreams as ("stdio" | "agentrun" | "hwlab")[],
|
||||
@@ -1535,6 +1557,7 @@ function webProbeRealtimeFanoutProfileConfig(value: unknown, path: string): Hwla
|
||||
agentrun: agentrunEventTypes,
|
||||
hwlab: hwlabEventTypes,
|
||||
},
|
||||
conditionalEventTypePairs,
|
||||
};
|
||||
}
|
||||
|
||||
|
||||
@@ -36,6 +36,7 @@ test("realtime fanout keeps the YAML terminal timeout when durationMs is absent"
|
||||
},
|
||||
forbiddenRequestPaths: [],
|
||||
requiredEventTypes: { agentrun: ["terminal_status"], hwlab: ["terminal"] },
|
||||
conditionalEventTypePairs: { "tool-call": { agentrun: "tool_call", hwlab: "tool" } },
|
||||
expectedKafka: {
|
||||
topics: { stdio: "codex-stdio.raw.v1", agentrun: "agentrun.event.v1", hwlab: "hwlab.event.v1" },
|
||||
groups: { directPublish: "direct", transactionalProjector: "projector", liveSse: "live" },
|
||||
@@ -46,6 +47,91 @@ test("realtime fanout keeps the YAML terminal timeout when durationMs is absent"
|
||||
assert.equal(effectiveProfile(profile, null).terminalTimeoutMs, 480_000);
|
||||
assert.equal(effectiveProfile(profile, undefined).terminalTimeoutMs, 480_000);
|
||||
assert.equal(effectiveProfile(profile, 5_000).terminalTimeoutMs, 5_000);
|
||||
assert.deepEqual(effectiveProfile(profile, null).conditionalEventTypePairs, profile.conditionalEventTypePairs);
|
||||
});
|
||||
|
||||
test("realtime fanout terminal completion does not require absent optional tool families", () => {
|
||||
const source = nodeWebObserveRunnerRealtimeSource();
|
||||
const build = new Function(`${source}\nreturn realtimeDebugTurnCompletion;`) as () => (
|
||||
snapshot: Record<string, any>, traceId: string, profile: Record<string, any>,
|
||||
) => Record<string, any>;
|
||||
const completion = build();
|
||||
const traceId = "trc_direct_answer";
|
||||
const profile = {
|
||||
debugStreams: ["stdio", "agentrun", "hwlab"],
|
||||
requiredEventTypes: {
|
||||
agentrun: ["user_message", "assistant_message", "terminal_status"],
|
||||
hwlab: ["user", "assistant", "terminal"],
|
||||
},
|
||||
conditionalEventTypePairs: {
|
||||
"tool-call": { agentrun: "tool_call", hwlab: "tool" },
|
||||
"command-output": { agentrun: "command_output", hwlab: "status" },
|
||||
},
|
||||
};
|
||||
const snapshot = {
|
||||
records: {
|
||||
stdio: [{ traceId, eventType: "codex.stdio.stdout" }],
|
||||
agentrun: ["user_message", "backend_status", "assistant_message", "terminal_status"].map((eventType) => ({ traceId, eventType })),
|
||||
hwlab: ["user", "backend", "assistant", "terminal"].map((eventType) => ({ traceId, eventType })),
|
||||
},
|
||||
};
|
||||
|
||||
const directAnswer = completion(snapshot, traceId, profile);
|
||||
assert.equal(directAnswer.complete, true);
|
||||
assert.deepEqual(directAnswer.missingEventTypes, { agentrun: [], hwlab: [] });
|
||||
|
||||
const missingTerminal = completion({
|
||||
records: {
|
||||
...snapshot.records,
|
||||
agentrun: snapshot.records.agentrun.filter((row: Record<string, string>) => row.eventType !== "terminal_status"),
|
||||
hwlab: snapshot.records.hwlab.filter((row: Record<string, string>) => row.eventType !== "terminal"),
|
||||
},
|
||||
}, traceId, profile);
|
||||
assert.equal(missingTerminal.complete, false);
|
||||
assert.deepEqual(missingTerminal.missingEventTypes, { agentrun: ["terminal_status"], hwlab: ["terminal"] });
|
||||
assert.ok(missingTerminal.observedEventTypes.agentrun.includes("assistant_message"));
|
||||
});
|
||||
|
||||
test("realtime fanout validates optional tool events only when they occur", () => {
|
||||
const source = nodeWebObserveRunnerRealtimeSource();
|
||||
const build = new Function(`${source}\nreturn realtimeValidateConditionalEventTypePairs;`) as () => (
|
||||
input: Record<string, any>,
|
||||
) => void;
|
||||
const validate = build();
|
||||
const traceId = "trc_optional_tool";
|
||||
const base = {
|
||||
turn: 1,
|
||||
traceId,
|
||||
profile: {
|
||||
conditionalEventTypePairs: {
|
||||
"tool-call": { agentrun: "tool_call", hwlab: "tool" },
|
||||
"command-output": { agentrun: "command_output", hwlab: "status" },
|
||||
},
|
||||
},
|
||||
};
|
||||
|
||||
assert.doesNotThrow(() => validate({ ...base, debug: { records: { agentrun: [], hwlab: [] } } }));
|
||||
assert.doesNotThrow(() => validate({
|
||||
...base,
|
||||
debug: {
|
||||
records: {
|
||||
agentrun: [{ traceId, eventType: "tool_call", eventId: "evt_tool" }],
|
||||
hwlab: [{ traceId, eventType: "tool", sourceEventId: "evt_tool" }],
|
||||
},
|
||||
},
|
||||
}));
|
||||
assert.throws(
|
||||
() => validate({
|
||||
...base,
|
||||
debug: {
|
||||
records: {
|
||||
agentrun: [{ traceId, eventType: "command_output", eventId: "evt_output" }],
|
||||
hwlab: [],
|
||||
},
|
||||
},
|
||||
}),
|
||||
/conditional event pair command-output count differs/u,
|
||||
);
|
||||
});
|
||||
|
||||
test("realtime fanout retries a YAML-bounded transient health navigation", async () => {
|
||||
@@ -321,3 +407,174 @@ test("realtime fanout proves one formal user event through AgentRun, HWLAB Kafka
|
||||
/product subscriber lacks pre-terminal or terminal event|product SSE lacks the formal user event/u,
|
||||
);
|
||||
});
|
||||
|
||||
function replayEnvelope(eventId: string, sourceSeq: number, terminal = false) {
|
||||
return {
|
||||
envelopeFingerprint: `fingerprint:${eventId}`,
|
||||
eventId,
|
||||
sourceEventId: `source:${eventId}`,
|
||||
sourceSeq,
|
||||
eventType: terminal ? "terminal" : "assistant",
|
||||
traceId: "trc_joint_quiet",
|
||||
hwlabSessionId: "ses_joint_quiet",
|
||||
runId: "run_joint_quiet",
|
||||
commandId: "cmd_joint_quiet",
|
||||
userMessageId: null,
|
||||
terminal,
|
||||
};
|
||||
}
|
||||
|
||||
test("realtime replay joint quiet waits for a late formal Kafka envelope before accepting 32/32", async () => {
|
||||
const source = nodeWebObserveRunnerRealtimeSource();
|
||||
let now = 0;
|
||||
let sleepCount = 0;
|
||||
const first = replayEnvelope("evt_first", 1);
|
||||
const late = replayEnvelope("evt_late", 2, true);
|
||||
let kafkaRows = [first];
|
||||
const productRows = [first, late];
|
||||
const build = new Function(
|
||||
"sleep",
|
||||
"Date",
|
||||
`${source}\nreturn realtimeWaitReplayedTraceJointQuiet;`,
|
||||
) as (
|
||||
sleep: () => Promise<void>,
|
||||
date: { now: () => number },
|
||||
) => (
|
||||
debug: Record<string, any>,
|
||||
key: string,
|
||||
subscriber: Record<string, any>,
|
||||
traceId: string,
|
||||
label: string,
|
||||
profile: Record<string, number>,
|
||||
) => Promise<Record<string, any>>;
|
||||
const waitForJointQuiet = build(
|
||||
async () => {
|
||||
sleepCount += 1;
|
||||
now += 250;
|
||||
if (sleepCount === 1) kafkaRows = [first, late];
|
||||
},
|
||||
{ now: () => now },
|
||||
);
|
||||
const debug = {
|
||||
page: {
|
||||
async evaluate() {
|
||||
return {
|
||||
connected: {}, connectedEventCount: { hwlab: 1 }, openCount: { hwlab: 1 }, errors: { hwlab: [] },
|
||||
records: { stdio: [], agentrun: [], hwlab: structuredClone(kafkaRows) },
|
||||
};
|
||||
},
|
||||
},
|
||||
};
|
||||
const subscriber = {
|
||||
page: {
|
||||
async evaluate() {
|
||||
return { connected: {}, connectedEventCount: 1, openCount: 1, errors: [], records: structuredClone(productRows) };
|
||||
},
|
||||
},
|
||||
};
|
||||
|
||||
const result = await waitForJointQuiet(
|
||||
debug,
|
||||
"turn-1-trc_joint_quiet",
|
||||
subscriber,
|
||||
"trc_joint_quiet",
|
||||
"reconnected",
|
||||
{ barrierTimeoutMs: 2_000, terminalQuietMs: 500 },
|
||||
);
|
||||
|
||||
assert.equal(result.comparison.equal, true);
|
||||
assert.equal(result.comparison.sourceCount, 2);
|
||||
assert.equal(result.comparison.productCount, 2);
|
||||
assert.equal(result.quietForMs, 500);
|
||||
assert.ok(sleepCount >= 3);
|
||||
});
|
||||
|
||||
test("realtime replay joint quiet rejects a permanent 31/32 split with bounded identity diagnostics", async () => {
|
||||
const source = nodeWebObserveRunnerRealtimeSource();
|
||||
let now = 0;
|
||||
const first = replayEnvelope("evt_first", 1);
|
||||
const late = replayEnvelope("evt_late", 2, true);
|
||||
const build = new Function(
|
||||
"sleep",
|
||||
"Date",
|
||||
`${source}\nreturn realtimeWaitReplayedTraceJointQuiet;`,
|
||||
) as (
|
||||
sleep: () => Promise<void>,
|
||||
date: { now: () => number },
|
||||
) => (
|
||||
debug: Record<string, any>,
|
||||
key: string,
|
||||
subscriber: Record<string, any>,
|
||||
traceId: string,
|
||||
label: string,
|
||||
profile: Record<string, number>,
|
||||
) => Promise<Record<string, any>>;
|
||||
const waitForJointQuiet = build(async () => { now += 250; }, { now: () => now });
|
||||
const debug = {
|
||||
page: {
|
||||
async evaluate() {
|
||||
return { connected: {}, connectedEventCount: { hwlab: 1 }, openCount: { hwlab: 1 }, errors: { hwlab: [] }, records: { stdio: [], agentrun: [], hwlab: [first] } };
|
||||
},
|
||||
},
|
||||
};
|
||||
const subscriber = {
|
||||
page: {
|
||||
async evaluate() {
|
||||
return { connected: {}, connectedEventCount: 1, openCount: 1, errors: [], records: [first, late] };
|
||||
},
|
||||
},
|
||||
};
|
||||
|
||||
await assert.rejects(
|
||||
() => waitForJointQuiet(
|
||||
debug,
|
||||
"turn-1-trc_joint_quiet",
|
||||
subscriber,
|
||||
"trc_joint_quiet",
|
||||
"reconnected",
|
||||
{ barrierTimeoutMs: 750, terminalQuietMs: 250 },
|
||||
),
|
||||
(error: Error & { details?: Record<string, any> }) => {
|
||||
assert.match(error.message, /did not converge before the joint quiet timeout/u);
|
||||
assert.equal(error.details?.sourceCount, 1);
|
||||
assert.equal(error.details?.productCount, 2);
|
||||
assert.equal(error.details?.productOnly?.[0]?.identity?.eventId, "evt_late");
|
||||
assert.equal(error.details?.productOnly?.[0]?.count, 1);
|
||||
return true;
|
||||
},
|
||||
);
|
||||
});
|
||||
|
||||
test("realtime envelope equality is an exact multiset even when counts match", () => {
|
||||
const source = nodeWebObserveRunnerRealtimeSource();
|
||||
const build = new Function(`${source}\nreturn { compare: realtimeEnvelopeMultisetComparison, assertEqual: realtimeAssertEnvelopePassthrough };`) as () => {
|
||||
compare: (kafkaRows: Record<string, any>[], productRows: Record<string, any>[]) => Record<string, any>;
|
||||
assertEqual: (kafkaRows: Record<string, any>[], productRows: Record<string, any>[], requireAll: boolean, label: string) => void;
|
||||
};
|
||||
const helpers = build();
|
||||
const first = replayEnvelope("evt_first", 1);
|
||||
const sourceTerminal = replayEnvelope("evt_terminal", 2, true);
|
||||
const changedTerminal = { ...sourceTerminal, envelopeFingerprint: "fingerprint:changed", sourceEventId: "source:changed" };
|
||||
const comparison = helpers.compare([first, sourceTerminal], [first, changedTerminal]);
|
||||
|
||||
assert.equal(comparison.equal, false);
|
||||
assert.equal(comparison.sourceCount, 2);
|
||||
assert.equal(comparison.productCount, 2);
|
||||
assert.equal(comparison.sourceOnly[0].identity.eventId, "evt_terminal");
|
||||
assert.equal(comparison.productOnly[0].identity.sourceEventId, "source:changed");
|
||||
assert.throws(
|
||||
() => helpers.assertEqual([first, sourceTerminal], [first, changedTerminal], true, "reconnected"),
|
||||
/product SSE envelope absent from hwlab\.event\.v1/u,
|
||||
);
|
||||
});
|
||||
|
||||
test("realtime fanout closes the first debug stream only after the final joint quiet gate", () => {
|
||||
const source = nodeWebObserveRunnerRealtimeSource();
|
||||
const afterSecond = source.indexOf("const firstReplayAfterSecond = await realtimeWaitReplayedTraceJointQuiet");
|
||||
const finalGate = source.indexOf("const firstReplayFinal = await realtimeWaitReplayedTraceJointQuiet");
|
||||
const close = source.indexOf("await realtimeCloseDebugStreams(debug, first.debugKey);");
|
||||
|
||||
assert.ok(afterSecond > 0);
|
||||
assert.ok(finalGate > afterSecond);
|
||||
assert.ok(close > finalGate);
|
||||
});
|
||||
|
||||
@@ -112,7 +112,7 @@ async function validateRealtimeFanout(command) {
|
||||
const firstTerminal = await realtimeWaitTraceEvent(subscriberA, first.traceId, true, effective.terminalTimeoutMs);
|
||||
await realtimeWaitUiTerminal(first.traceId, effective.terminalTimeoutMs);
|
||||
const firstStable = await realtimeWaitTurnStable(debug, first.debugKey, first.traceId, [subscriberA], effective);
|
||||
const firstDebug = firstStable.debug;
|
||||
let firstDebug = firstStable.debug;
|
||||
realtimeEnrichTurnExecutionIds(first, firstDebug);
|
||||
const firstProductA = firstStable.products[0];
|
||||
realtimeAssertStableProductTransport(firstProductA, "subscriber-a-first-terminal");
|
||||
@@ -127,13 +127,25 @@ async function validateRealtimeFanout(command) {
|
||||
preTerminalSnapshots: [bBeforeDisconnect],
|
||||
profile: effective,
|
||||
});
|
||||
await realtimeCloseDebugStreams(debug, first.debugKey);
|
||||
|
||||
subscriberB2 = await realtimeNewSubscriber(storageState, sessionId, "subscriber-b-reconnected", effective);
|
||||
const bReconnected = await realtimeWaitProductConnected(subscriberB2, effective.barrierTimeoutMs);
|
||||
realtimeAssertConnectedContract(bReconnected, sessionId, "subscriber-b-reconnected", effective, { requireRetainedEvents: effective.expectedProductSse.replayPriorTraceOnReconnect });
|
||||
await sleep(effective.reconnectQuietMs);
|
||||
const reconnectBeforeSecond = await realtimeProductSnapshot(subscriberB2);
|
||||
let reconnectBeforeSecond;
|
||||
if (effective.expectedProductSse.replayPriorTraceOnReconnect) {
|
||||
const firstReplayBeforeSecond = await realtimeWaitReplayedTraceJointQuiet(
|
||||
debug,
|
||||
first.debugKey,
|
||||
subscriberB2,
|
||||
first.traceId,
|
||||
"subscriber-b-reconnected-before-second-turn",
|
||||
effective,
|
||||
);
|
||||
firstDebug = firstReplayBeforeSecond.debug;
|
||||
reconnectBeforeSecond = firstReplayBeforeSecond.product;
|
||||
} else {
|
||||
reconnectBeforeSecond = await realtimeProductSnapshot(subscriberB2);
|
||||
}
|
||||
realtimeAssertStableProductTransport(reconnectBeforeSecond, "subscriber-b-reconnected-before-second-turn");
|
||||
if (effective.expectedProductSse.replayPriorTraceOnReconnect) {
|
||||
realtimeAssertReplayedTrace(firstDebug, reconnectBeforeSecond, first.traceId, "subscriber-b-reconnected-before-second-turn");
|
||||
@@ -161,7 +173,17 @@ async function validateRealtimeFanout(command) {
|
||||
realtimeAssertStableProductTransport(secondProductA, "subscriber-a-second-terminal");
|
||||
realtimeAssertStableProductTransport(secondProductB, "subscriber-b-reconnected-second-terminal");
|
||||
if (effective.expectedProductSse.replayPriorTraceOnReconnect) {
|
||||
realtimeAssertReplayedTrace(firstDebug, secondProductB, first.traceId, "subscriber-b-reconnected-after-second-turn");
|
||||
const firstReplayAfterSecond = await realtimeWaitReplayedTraceJointQuiet(
|
||||
debug,
|
||||
first.debugKey,
|
||||
subscriberB2,
|
||||
first.traceId,
|
||||
"subscriber-b-reconnected-after-second-turn",
|
||||
effective,
|
||||
);
|
||||
firstDebug = firstReplayAfterSecond.debug;
|
||||
realtimeAssertStableProductTransport(firstReplayAfterSecond.product, "subscriber-b-reconnected-after-second-turn");
|
||||
realtimeAssertReplayedTrace(firstDebug, firstReplayAfterSecond.product, first.traceId, "subscriber-b-reconnected-after-second-turn");
|
||||
} else if (secondProductB.records.some((item) => item.traceId === first.traceId)) {
|
||||
throw new Error("reconnected subscriber received first-turn events contrary to the owning YAML");
|
||||
}
|
||||
@@ -179,12 +201,25 @@ async function validateRealtimeFanout(command) {
|
||||
await realtimeCloseDebugStreams(debug, second.debugKey);
|
||||
|
||||
const productA = await realtimeProductSnapshot(subscriberA);
|
||||
const productB2 = await realtimeProductSnapshot(subscriberB2);
|
||||
let productB2 = await realtimeProductSnapshot(subscriberB2);
|
||||
if (effective.expectedProductSse.replayPriorTraceOnReconnect) {
|
||||
const firstReplayFinal = await realtimeWaitReplayedTraceJointQuiet(
|
||||
debug,
|
||||
first.debugKey,
|
||||
subscriberB2,
|
||||
first.traceId,
|
||||
"subscriber-b-reconnected-final",
|
||||
effective,
|
||||
);
|
||||
firstDebug = firstReplayFinal.debug;
|
||||
productB2 = firstReplayFinal.product;
|
||||
}
|
||||
realtimeAssertStableProductTransport(productA, "subscriber-a-final");
|
||||
realtimeAssertStableProductTransport(productB2, "subscriber-b-reconnected-final");
|
||||
if (effective.expectedProductSse.replayPriorTraceOnReconnect) {
|
||||
realtimeAssertReplayedTrace(firstDebug, productB2, first.traceId, "subscriber-b-reconnected-final");
|
||||
}
|
||||
await realtimeCloseDebugStreams(debug, first.debugKey);
|
||||
const mainStreams = network.filter((item) => item.path === effective.productEventsPath && item.sessionId === sessionId);
|
||||
const forbidden = network.filter((item) => realtimeForbiddenRequest(item, effective.forbiddenRequestPaths));
|
||||
if (mainStreams.length !== 1) throw new Error("main Workbench session SSE connection count is " + mainStreams.length + ", expected 1");
|
||||
@@ -747,6 +782,7 @@ function realtimeFanoutEffectiveProfile(profile, durationMs) {
|
||||
reconnectQuietMs: Number(profile.reconnectQuietMs),
|
||||
forbiddenRequestPaths: Array.isArray(profile.forbiddenRequestPaths) ? profile.forbiddenRequestPaths.map(String) : [],
|
||||
requiredEventTypes: profile.requiredEventTypes && typeof profile.requiredEventTypes === "object" ? profile.requiredEventTypes : {},
|
||||
conditionalEventTypePairs: profile.conditionalEventTypePairs && typeof profile.conditionalEventTypePairs === "object" ? sanitize(profile.conditionalEventTypePairs) : {},
|
||||
expectedProductSse: sanitize(expectedProductSse),
|
||||
expectedKafka: sanitize(expectedKafka),
|
||||
};
|
||||
@@ -1043,15 +1079,44 @@ async function realtimeWaitTurnIds(debug, key, traceId, timeoutMs) {
|
||||
}
|
||||
|
||||
async function realtimeWaitDebugTurnComplete(debug, key, traceId, profile) {
|
||||
return realtimePoll(async () => {
|
||||
const snapshot = await realtimeDebugSnapshot(debug, key);
|
||||
if (!profile.debugStreams.every((stream) => (snapshot.records[stream] || []).some((item) => item.traceId === traceId))) return null;
|
||||
const agentrunTypes = new Set((snapshot.records.agentrun || []).filter((item) => item.traceId === traceId).map((item) => item.eventType));
|
||||
const hwlabTypes = new Set((snapshot.records.hwlab || []).filter((item) => item.traceId === traceId).map((item) => item.eventType));
|
||||
if (!(profile.requiredEventTypes.agentrun || []).every((type) => agentrunTypes.has(type))) return null;
|
||||
if (!(profile.requiredEventTypes.hwlab || []).every((type) => hwlabTypes.has(type))) return null;
|
||||
return snapshot;
|
||||
}, profile.terminalTimeoutMs, "realtime debug terminal event families");
|
||||
let latest = null;
|
||||
try {
|
||||
return await realtimePoll(async () => {
|
||||
const snapshot = await realtimeDebugSnapshot(debug, key);
|
||||
latest = realtimeDebugTurnCompletion(snapshot, traceId, profile);
|
||||
return latest.complete ? snapshot : null;
|
||||
}, profile.terminalTimeoutMs, "realtime debug terminal event families");
|
||||
} catch (error) {
|
||||
const wrapped = error instanceof Error ? error : new Error(String(error));
|
||||
const missing = latest?.missingEventTypes || { agentrun: [], hwlab: [] };
|
||||
wrapped.message = "realtime debug terminal event families missing agentrun=[" + missing.agentrun.join(",") + "] hwlab=[" + missing.hwlab.join(",") + "] after " + profile.terminalTimeoutMs + "ms";
|
||||
wrapped.details = { ...(wrapped.details || {}), traceId, completion: latest, valuesRedacted: true };
|
||||
throw wrapped;
|
||||
}
|
||||
}
|
||||
|
||||
function realtimeDebugTurnCompletion(snapshot, traceId, profile) {
|
||||
const observedEventTypes = {};
|
||||
const recordCounts = {};
|
||||
for (const stream of profile.debugStreams || []) {
|
||||
const rows = (snapshot.records?.[stream] || []).filter((item) => item.traceId === traceId);
|
||||
observedEventTypes[stream] = [...new Set(rows.map((item) => item.eventType).filter(Boolean))];
|
||||
recordCounts[stream] = rows.length;
|
||||
}
|
||||
const missingEventTypes = {
|
||||
agentrun: (profile.requiredEventTypes?.agentrun || []).filter((type) => !observedEventTypes.agentrun?.includes(type)),
|
||||
hwlab: (profile.requiredEventTypes?.hwlab || []).filter((type) => !observedEventTypes.hwlab?.includes(type)),
|
||||
};
|
||||
const missingStreams = (profile.debugStreams || []).filter((stream) => !recordCounts[stream]);
|
||||
return {
|
||||
complete: missingStreams.length === 0 && missingEventTypes.agentrun.length === 0 && missingEventTypes.hwlab.length === 0,
|
||||
traceId,
|
||||
observedEventTypes,
|
||||
missingEventTypes,
|
||||
missingStreams,
|
||||
recordCounts,
|
||||
valuesRedacted: true,
|
||||
};
|
||||
}
|
||||
|
||||
async function realtimeWaitTurnStable(debug, key, traceId, subscribers, profile) {
|
||||
@@ -1184,6 +1249,61 @@ function realtimeAssertReplayedTrace(debug, product, traceId, label) {
|
||||
realtimeAssertEnvelopePassthrough(kafkaRows, productRows, true, label + " retained trace");
|
||||
}
|
||||
|
||||
async function realtimeWaitReplayedTraceJointQuiet(debug, key, subscriber, traceId, label, profile) {
|
||||
const timeoutMs = profile.barrierTimeoutMs;
|
||||
const deadline = Date.now() + timeoutMs;
|
||||
let previousSignature = null;
|
||||
let quietSince = null;
|
||||
let latest = null;
|
||||
while (Date.now() < deadline) {
|
||||
const debugSnapshot = await realtimeDebugSnapshot(debug, key);
|
||||
const productSnapshot = await realtimeProductSnapshot(subscriber);
|
||||
const kafkaRows = (debugSnapshot.records?.hwlab || []).filter((item) => item.traceId === traceId);
|
||||
const productRows = (productSnapshot.records || []).filter((item) => item.traceId === traceId);
|
||||
const comparison = realtimeEnvelopeMultisetComparison(kafkaRows, productRows);
|
||||
const signature = JSON.stringify({
|
||||
kafka: realtimeEnvelopeMultisetSignature(kafkaRows),
|
||||
product: realtimeEnvelopeMultisetSignature(productRows),
|
||||
});
|
||||
const now = Date.now();
|
||||
if (signature !== previousSignature) {
|
||||
previousSignature = signature;
|
||||
quietSince = now;
|
||||
}
|
||||
const quietForMs = quietSince === null ? 0 : now - quietSince;
|
||||
const terminalSeen = productRows.some((item) => item.terminal === true);
|
||||
const sourceTransport = {
|
||||
openCount: debugSnapshot.openCount?.hwlab || 0,
|
||||
connectedEventCount: debugSnapshot.connectedEventCount?.hwlab || 0,
|
||||
errorCount: (debugSnapshot.errors?.hwlab || []).length,
|
||||
};
|
||||
const sourceTransportStable = sourceTransport.openCount === 1 && sourceTransport.connectedEventCount === 1 && sourceTransport.errorCount === 0;
|
||||
latest = { comparison, terminalSeen, quietForMs, sourceTransport, sourceTransportStable };
|
||||
if (comparison.equal && terminalSeen && sourceTransportStable && quietForMs >= profile.terminalQuietMs) {
|
||||
return { debug: debugSnapshot, product: productSnapshot, comparison, quietForMs, valuesRedacted: true };
|
||||
}
|
||||
await sleep(250);
|
||||
}
|
||||
const error = new Error(label + " Kafka source and product replay envelope multisets did not converge before the joint quiet timeout");
|
||||
error.details = {
|
||||
traceId,
|
||||
timeoutMs,
|
||||
terminalQuietMs: profile.terminalQuietMs,
|
||||
terminalSeen: latest?.terminalSeen === true,
|
||||
sourceTransport: latest?.sourceTransport || null,
|
||||
sourceTransportStable: latest?.sourceTransportStable === true,
|
||||
quietForMs: latest?.quietForMs || 0,
|
||||
sourceCount: latest?.comparison?.sourceCount || 0,
|
||||
productCount: latest?.comparison?.productCount || 0,
|
||||
sourceOnly: latest?.comparison?.sourceOnly || [],
|
||||
productOnly: latest?.comparison?.productOnly || [],
|
||||
invalidSource: latest?.comparison?.invalidSource || [],
|
||||
invalidProduct: latest?.comparison?.invalidProduct || [],
|
||||
valuesRedacted: true,
|
||||
};
|
||||
throw error;
|
||||
}
|
||||
|
||||
function realtimeConnectedSummary(connected) {
|
||||
return {
|
||||
realtimeSource: connected?.realtimeSource || null,
|
||||
@@ -1241,6 +1361,7 @@ function realtimeValidateTurnEvidence(input) {
|
||||
const hwlabTypes = new Set(input.debug.records.hwlab.filter((item) => item.traceId === input.traceId).map((item) => item.eventType));
|
||||
for (const type of profile.requiredEventTypes.agentrun || []) if (!agentrunTypes.has(type)) throw new Error("turn " + input.turn + " AgentRun lacks " + type);
|
||||
for (const type of profile.requiredEventTypes.hwlab || []) if (!hwlabTypes.has(type)) throw new Error("turn " + input.turn + " HWLAB lacks " + type);
|
||||
realtimeValidateConditionalEventTypePairs(input);
|
||||
const identityRows = Object.values(input.debug.records).flat().filter((item) => item.traceId === input.traceId);
|
||||
if (!input.runId || new Set(identityRows.map((item) => item.runId)).size !== 1) throw new Error("turn " + input.turn + " runId missing or mismatched across streams");
|
||||
const commandIds = new Set(identityRows.map((item) => item.commandId).filter(Boolean));
|
||||
@@ -1267,6 +1388,30 @@ function realtimeValidateTurnEvidence(input) {
|
||||
}
|
||||
}
|
||||
|
||||
function realtimeValidateConditionalEventTypePairs(input) {
|
||||
const pairs = input.profile.conditionalEventTypePairs || {};
|
||||
const agentrunRows = (input.debug.records.agentrun || []).filter((item) => item.traceId === input.traceId);
|
||||
const hwlabRows = (input.debug.records.hwlab || []).filter((item) => item.traceId === input.traceId);
|
||||
for (const [id, pair] of Object.entries(pairs)) {
|
||||
const sourceRows = agentrunRows.filter((item) => item.eventType === pair.agentrun);
|
||||
const projectedRows = hwlabRows.filter((item) => item.eventType === pair.hwlab);
|
||||
if (sourceRows.length === 0 && projectedRows.length === 0) continue;
|
||||
if (sourceRows.length !== projectedRows.length) {
|
||||
throw new Error("turn " + input.turn + " conditional event pair " + id + " count differs AgentRun=" + sourceRows.length + " HWLAB=" + projectedRows.length);
|
||||
}
|
||||
for (const source of sourceRows) {
|
||||
if (!source.eventId || !projectedRows.some((projected) => projected.sourceEventId === source.eventId)) {
|
||||
throw new Error("turn " + input.turn + " conditional event pair " + id + " lacks HWLAB sourceEventId lineage");
|
||||
}
|
||||
}
|
||||
for (const projected of projectedRows) {
|
||||
if (!projected.sourceEventId || !sourceRows.some((source) => source.eventId === projected.sourceEventId)) {
|
||||
throw new Error("turn " + input.turn + " conditional event pair " + id + " has an orphan HWLAB event");
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
function realtimeExpectedStreamSessionId(stream, hwlabSessionId) {
|
||||
if (stream !== "stdio") return hwlabSessionId;
|
||||
const base = String(hwlabSessionId || "")
|
||||
@@ -1276,26 +1421,97 @@ function realtimeExpectedStreamSessionId(stream, hwlabSessionId) {
|
||||
return base ? "ses_agentrun_" + base : null;
|
||||
}
|
||||
|
||||
function realtimeAssertEnvelopePassthrough(kafkaRows, productRows, requireAllKafkaRows, label) {
|
||||
const sameEnvelope = (left, right) => left.envelopeFingerprint && left.envelopeFingerprint === right.envelopeFingerprint
|
||||
&& left.eventId === right.eventId
|
||||
&& left.sourceEventId === right.sourceEventId
|
||||
&& left.sourceSeq === right.sourceSeq
|
||||
&& left.eventType === right.eventType
|
||||
&& left.traceId === right.traceId
|
||||
&& left.hwlabSessionId === right.hwlabSessionId
|
||||
&& left.runId === right.runId
|
||||
&& left.commandId === right.commandId
|
||||
&& left.userMessageId === right.userMessageId
|
||||
&& left.terminal === right.terminal;
|
||||
for (const productRow of productRows) {
|
||||
if (!kafkaRows.some((kafkaRow) => sameEnvelope(kafkaRow, productRow))) throw new Error(label + " received a product SSE envelope absent from hwlab.event.v1");
|
||||
}
|
||||
if (requireAllKafkaRows) {
|
||||
if (productRows.length !== kafkaRows.length) throw new Error(label + " envelope count differs from hwlab.event.v1");
|
||||
for (const kafkaRow of kafkaRows) {
|
||||
if (!productRows.some((productRow) => sameEnvelope(kafkaRow, productRow))) throw new Error(label + " did not receive an unchanged hwlab.event.v1 envelope");
|
||||
function realtimeEnvelopeIdentity(row) {
|
||||
return {
|
||||
envelopeFingerprint: row?.envelopeFingerprint || null,
|
||||
eventId: row?.eventId || null,
|
||||
sourceEventId: row?.sourceEventId || null,
|
||||
sourceSeq: Number.isFinite(row?.sourceSeq) ? row.sourceSeq : null,
|
||||
eventType: row?.eventType || null,
|
||||
traceId: row?.traceId || null,
|
||||
hwlabSessionId: row?.hwlabSessionId || null,
|
||||
runId: row?.runId || null,
|
||||
commandId: row?.commandId || null,
|
||||
userMessageId: row?.userMessageId || null,
|
||||
terminal: row?.terminal === true,
|
||||
};
|
||||
}
|
||||
|
||||
function realtimeEnvelopeIdentityKey(row) {
|
||||
const identity = realtimeEnvelopeIdentity(row);
|
||||
return identity.envelopeFingerprint ? JSON.stringify(identity) : null;
|
||||
}
|
||||
|
||||
function realtimeEnvelopeMultiset(rows) {
|
||||
const counts = new Map();
|
||||
const invalid = [];
|
||||
for (const row of rows) {
|
||||
const identity = realtimeEnvelopeIdentity(row);
|
||||
const key = realtimeEnvelopeIdentityKey(row);
|
||||
if (!key) {
|
||||
if (invalid.length < 8) invalid.push(identity);
|
||||
continue;
|
||||
}
|
||||
const current = counts.get(key);
|
||||
if (current) current.count += 1;
|
||||
else counts.set(key, { count: 1, identity });
|
||||
}
|
||||
return { counts, invalid };
|
||||
}
|
||||
|
||||
function realtimeEnvelopeMultisetSignature(rows) {
|
||||
const multiset = realtimeEnvelopeMultiset(rows);
|
||||
const counts = [...multiset.counts.entries()].map(([key, entry]) => [key, entry.count]).sort((left, right) => left[0].localeCompare(right[0]));
|
||||
return JSON.stringify({ counts, invalid: multiset.invalid });
|
||||
}
|
||||
|
||||
function realtimeEnvelopeMultisetComparison(kafkaRows, productRows) {
|
||||
const source = realtimeEnvelopeMultiset(kafkaRows);
|
||||
const product = realtimeEnvelopeMultiset(productRows);
|
||||
const sourceOnly = [];
|
||||
const productOnly = [];
|
||||
let differs = source.invalid.length > 0 || product.invalid.length > 0;
|
||||
const keys = new Set([...source.counts.keys(), ...product.counts.keys()]);
|
||||
for (const key of keys) {
|
||||
const sourceEntry = source.counts.get(key);
|
||||
const productEntry = product.counts.get(key);
|
||||
const sourceCount = sourceEntry?.count || 0;
|
||||
const productCount = productEntry?.count || 0;
|
||||
if (sourceCount > productCount) {
|
||||
differs = true;
|
||||
if (sourceOnly.length < 8) sourceOnly.push({ identity: sourceEntry.identity, count: sourceCount - productCount });
|
||||
}
|
||||
if (productCount > sourceCount) {
|
||||
differs = true;
|
||||
if (productOnly.length < 8) productOnly.push({ identity: productEntry.identity, count: productCount - sourceCount });
|
||||
}
|
||||
}
|
||||
return {
|
||||
equal: !differs && kafkaRows.length === productRows.length,
|
||||
sourceCount: kafkaRows.length,
|
||||
productCount: productRows.length,
|
||||
sourceOnly,
|
||||
productOnly,
|
||||
invalidSource: source.invalid,
|
||||
invalidProduct: product.invalid,
|
||||
valuesRedacted: true,
|
||||
};
|
||||
}
|
||||
|
||||
function realtimeAssertEnvelopePassthrough(kafkaRows, productRows, requireAllKafkaRows, label) {
|
||||
const comparison = realtimeEnvelopeMultisetComparison(kafkaRows, productRows);
|
||||
if (comparison.productOnly.length > 0 || comparison.invalidProduct.length > 0) {
|
||||
const error = new Error(label + " received a product SSE envelope absent from hwlab.event.v1");
|
||||
error.details = comparison;
|
||||
throw error;
|
||||
}
|
||||
if (requireAllKafkaRows && !comparison.equal) {
|
||||
const message = comparison.sourceCount !== comparison.productCount
|
||||
? label + " envelope count differs from hwlab.event.v1"
|
||||
: label + " did not receive an unchanged hwlab.event.v1 envelope";
|
||||
const error = new Error(message);
|
||||
error.details = comparison;
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -28,6 +28,14 @@ test("realtime fanout runner contract derives topics, groups, and independent ca
|
||||
projectionOutboxRelay: false,
|
||||
projectionRealtime: false,
|
||||
});
|
||||
assert.deepEqual(profiles["pure-kafka-live"]?.requiredEventTypes, {
|
||||
agentrun: ["user_message", "assistant_message", "terminal_status"],
|
||||
hwlab: ["user", "assistant", "terminal"],
|
||||
});
|
||||
assert.deepEqual(profiles["pure-kafka-live"]?.conditionalEventTypePairs, {
|
||||
"tool-call": { agentrun: "tool_call", hwlab: "tool" },
|
||||
"command-output": { agentrun: "command_output", hwlab: "status" },
|
||||
});
|
||||
});
|
||||
|
||||
test("Workbench Kafka debug replay runner contract is derived from owning YAML", () => {
|
||||
|
||||
@@ -41,8 +41,14 @@ test("validateRealtimeFanout parses the YAML profile and two independent prompts
|
||||
refreshHandoff: true,
|
||||
replayPriorTraceOnReconnect: true,
|
||||
});
|
||||
assert.ok(spec.webProbe?.realtimeFanoutProfiles?.["pure-kafka-live"]?.requiredEventTypes.agentrun.includes("user_message"));
|
||||
assert.ok(spec.webProbe?.realtimeFanoutProfiles?.["pure-kafka-live"]?.requiredEventTypes.hwlab.includes("user"));
|
||||
assert.deepEqual(spec.webProbe?.realtimeFanoutProfiles?.["pure-kafka-live"]?.requiredEventTypes, {
|
||||
agentrun: ["user_message", "assistant_message", "terminal_status"],
|
||||
hwlab: ["user", "assistant", "terminal"],
|
||||
});
|
||||
assert.deepEqual(spec.webProbe?.realtimeFanoutProfiles?.["pure-kafka-live"]?.conditionalEventTypePairs, {
|
||||
"tool-call": { agentrun: "tool_call", hwlab: "tool" },
|
||||
"command-output": { agentrun: "command_output", hwlab: "status" },
|
||||
});
|
||||
});
|
||||
|
||||
test("validateRealtimeFanout refuses body as a substitute for the second prompt", () => {
|
||||
|
||||
@@ -3,15 +3,74 @@ import { describe, expect, test } from "bun:test";
|
||||
import { repoRoot } from "./config";
|
||||
import { sshHelp, sshHelpScope } from "./help";
|
||||
|
||||
describe("ssh download scoped help", () => {
|
||||
test("selects only the exact help-first download scope", () => {
|
||||
type RenderedHelp = {
|
||||
ok: boolean;
|
||||
command: string;
|
||||
data: {
|
||||
command?: string;
|
||||
scope?: string;
|
||||
runtime?: { repoRoot: string };
|
||||
routeSyntax?: Record<string, string> | string;
|
||||
operationGroups?: Record<string, string>;
|
||||
scopedHelp?: { availableOperationCount: number };
|
||||
usage?: string[];
|
||||
outputTruncated?: boolean;
|
||||
dump?: unknown;
|
||||
};
|
||||
};
|
||||
|
||||
function runTransHelp(...args: string[]): { stdout: string; rendered: RenderedHelp } {
|
||||
const result = spawnSync("bun", ["scripts/ssh-cli.ts", ...args], {
|
||||
cwd: repoRoot,
|
||||
env: {
|
||||
...process.env,
|
||||
UNIDESK_SSH_ENTRYPOINT: "trans",
|
||||
UNIDESK_TRANS_REPO_ROOT: repoRoot,
|
||||
},
|
||||
encoding: "utf8",
|
||||
});
|
||||
expect(result.status).toBe(0);
|
||||
expect(result.stderr).toBe("");
|
||||
return { stdout: result.stdout, rendered: JSON.parse(result.stdout) as RenderedHelp };
|
||||
}
|
||||
|
||||
describe("ssh bounded progressive help", () => {
|
||||
test("selects exact help-first operation scopes", () => {
|
||||
expect(sshHelpScope(["ssh", "--help", "download"])).toBe("download");
|
||||
expect(sshHelpScope(["ssh", "help", "download"])).toBe("download");
|
||||
expect(sshHelpScope(["ssh", "help", "apply-patch"])).toBe("apply-patch");
|
||||
expect(sshHelpScope(["ssh", "--help", "ps"])).toBe("ps");
|
||||
expect(sshHelpScope(["ssh", "--help"])).toBeUndefined();
|
||||
expect(sshHelpScope(["ssh", "--help", "unknown"])).toBeUndefined();
|
||||
expect(sshHelpScope(["ssh", "NC01:k3s", "--help"])).toBeUndefined();
|
||||
expect(sshHelpScope(["ssh", "--help", "download", "extra"])).toBeUndefined();
|
||||
});
|
||||
|
||||
test("keeps the top-level model compact while indexing routes and every scoped operation", () => {
|
||||
const top = sshHelp() as {
|
||||
routeSyntax: Record<string, string>;
|
||||
operationGroups: Record<string, string>;
|
||||
scopedHelp: { command: string; availableOperationCount: number };
|
||||
outputTruncated?: boolean;
|
||||
};
|
||||
const serialized = JSON.stringify(top);
|
||||
const grouped = Object.values(top.operationGroups).flatMap((value) => value.split(" | "));
|
||||
|
||||
expect(Buffer.byteLength(serialized, "utf8")).toBeLessThan(6_144);
|
||||
expect(top.outputTruncated).toBeUndefined();
|
||||
expect(top.routeSyntax.general).toContain("<route> <operation>");
|
||||
expect(top.routeSyntax.k3sWorkload).toContain(":<namespace>:<workload>");
|
||||
expect(top.routeSyntax.windowsWorkspace).toContain(":win");
|
||||
expect(top.operationGroups.patch).toBe("apply-patch | apply-patch-v1");
|
||||
expect(top.operationGroups.transfer).toBe("upload | download");
|
||||
expect(new Set(grouped).size).toBe(top.scopedHelp.availableOperationCount);
|
||||
expect(top.scopedHelp.command).toContain("--help <operation>");
|
||||
for (const operation of grouped) {
|
||||
const scope = sshHelpScope(["ssh", "--help", operation]);
|
||||
expect(scope).toBe(operation);
|
||||
expect(Buffer.byteLength(JSON.stringify(sshHelp(scope)), "utf8")).toBeLessThan(6_144);
|
||||
}
|
||||
});
|
||||
|
||||
test("keeps download help bounded and discloses the executing source root", () => {
|
||||
const scoped = sshHelp("download") as {
|
||||
scope: string;
|
||||
@@ -20,13 +79,11 @@ describe("ssh download scoped help", () => {
|
||||
examples: Record<string, string>;
|
||||
contract: { pathOrder: string[]; transport: string; verification: { automatic: boolean } };
|
||||
};
|
||||
const full = sshHelp() as { usage: string[] };
|
||||
const serialized = JSON.stringify(scoped);
|
||||
|
||||
expect(scoped.scope).toBe("download");
|
||||
expect(scoped.usage.length).toBeLessThanOrEqual(4);
|
||||
expect(scoped.usage.length).toBeLessThan(full.usage.length);
|
||||
expect(Buffer.byteLength(serialized, "utf8")).toBeLessThan(10_240);
|
||||
expect(Buffer.byteLength(serialized, "utf8")).toBeLessThan(6_144);
|
||||
expect(serialized).not.toContain("git exact-commit");
|
||||
expect(scoped.runtime.repoRoot).toBe(repoRoot);
|
||||
expect(scoped.runtime.cliPath).toBe(`${repoRoot}/scripts/ssh-cli.ts`);
|
||||
@@ -40,46 +97,32 @@ describe("ssh download scoped help", () => {
|
||||
"trans NC01:k3s:hwlab-v03:hwlab-cloud-api:hwlab-cloud-api download /tmp/trace.json ./trace.json",
|
||||
);
|
||||
expect(scoped.examples.windowsDrive).toContain("D:\\tmp\\trace.json");
|
||||
expect(scoped.examples.explicitWorktree).toBe(
|
||||
`UNIDESK_TRANS_REPO_ROOT=${repoRoot} trans NC01:k3s:hwlab-v03:hwlab-cloud-api:hwlab-cloud-api download /tmp/trace.json ./trace.json`,
|
||||
);
|
||||
});
|
||||
|
||||
test("renders the original trans command without stdout dump fallback", () => {
|
||||
const result = spawnSync("bun", ["scripts/ssh-cli.ts", "--help", "download"], {
|
||||
cwd: repoRoot,
|
||||
env: {
|
||||
...process.env,
|
||||
UNIDESK_SSH_ENTRYPOINT: "trans",
|
||||
UNIDESK_TRANS_REPO_ROOT: repoRoot,
|
||||
},
|
||||
encoding: "utf8",
|
||||
});
|
||||
test("renders compact top-level and scoped help without dump fallback", () => {
|
||||
const top = runTransHelp("--help");
|
||||
expect(Buffer.byteLength(top.stdout, "utf8")).toBeLessThan(6_144);
|
||||
expect(top.rendered.ok).toBe(true);
|
||||
expect(top.rendered.command).toBe("trans --help");
|
||||
expect(top.rendered.data.outputTruncated).toBeUndefined();
|
||||
expect(top.rendered.data.dump).toBeUndefined();
|
||||
expect(top.rendered.data.operationGroups?.transfer).toContain("download");
|
||||
|
||||
expect(result.status).toBe(0);
|
||||
expect(result.stderr).toBe("");
|
||||
expect(Buffer.byteLength(result.stdout, "utf8")).toBeLessThan(10_240);
|
||||
const rendered = JSON.parse(result.stdout) as {
|
||||
ok: boolean;
|
||||
command: string;
|
||||
data: { scope: string; runtime: { repoRoot: string }; usage: string[]; outputTruncated?: boolean };
|
||||
};
|
||||
expect(rendered.ok).toBe(true);
|
||||
expect(rendered.command).toBe("trans --help download");
|
||||
expect(rendered.data.scope).toBe("download");
|
||||
expect(rendered.data.runtime.repoRoot).toBe(repoRoot);
|
||||
expect(rendered.data.usage.length).toBeLessThanOrEqual(4);
|
||||
expect(rendered.data.outputTruncated).toBeUndefined();
|
||||
const routeTop = runTransHelp("NC01:k3s", "--help");
|
||||
expect(routeTop.rendered.command).toBe("trans NC01:k3s --help");
|
||||
expect(routeTop.rendered.data.outputTruncated).toBeUndefined();
|
||||
expect(routeTop.rendered.data.operationGroups).toEqual(top.rendered.data.operationGroups);
|
||||
|
||||
const unified = spawnSync("bun", ["scripts/cli.ts", "ssh", "--help", "download"], {
|
||||
cwd: repoRoot,
|
||||
env: process.env,
|
||||
encoding: "utf8",
|
||||
});
|
||||
expect(unified.status).toBe(0);
|
||||
expect(unified.stderr).toBe("");
|
||||
const unifiedRendered = JSON.parse(unified.stdout) as { data: { scope: string; runtime: { repoRoot: string } } };
|
||||
expect(unifiedRendered.data.scope).toBe("download");
|
||||
expect(unifiedRendered.data.runtime.repoRoot).toBe(repoRoot);
|
||||
const applyPatch = runTransHelp("--help", "apply-patch");
|
||||
expect(applyPatch.rendered.command).toBe("trans --help apply-patch");
|
||||
expect(applyPatch.rendered.data.scope).toBe("apply-patch");
|
||||
expect(applyPatch.rendered.data.usage?.[0]).toContain("apply-patch");
|
||||
expect(applyPatch.rendered.data.outputTruncated).toBeUndefined();
|
||||
|
||||
const download = runTransHelp("--help", "download");
|
||||
expect(download.rendered.command).toBe("trans --help download");
|
||||
expect(download.rendered.data.scope).toBe("download");
|
||||
expect(download.rendered.data.runtime?.repoRoot).toBe(repoRoot);
|
||||
expect(download.rendered.data.outputTruncated).toBeUndefined();
|
||||
});
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user