修复 API Key app-server 认证重试循环
Project CI / Backend tests (pull_request) Failing after 10s
Project CI / Repository checks (pull_request) Failing after 10s
Project CI / Frontend tests (pull_request) Successful in 3m52s
Project CI / Native shell tests (pull_request) Failing after 9m47s

API Key/provider-proxy 握手后主动关闭 remote-control

无 ChatGPT 登录态时将子进程日志级别收敛到 warn

补充 app-server 协议 fixture 回归覆盖与技术方案说明
This commit is contained in:
2026-08-27 18:09:36 +08:00
parent 5942ff25e7
commit 8f15b83aba
2 changed files with 83 additions and 13 deletions
@@ -69,6 +69,19 @@ impl CodexAppServerCredential {
matches!(self, Self::AppDataKey { .. })
}
fn remote_control_disable_reason(
&self,
bridge_through_provider_proxy: bool,
) -> Option<&'static str> {
if bridge_through_provider_proxy {
Some("provider-proxy-auth")
} else if self.uses_app_data_key() {
Some("api-key-auth")
} else {
None
}
}
fn direct_provider_route<'a>(
&'a self,
llm: &'a GameCreatorLlmConfig,
@@ -1444,6 +1457,8 @@ impl CodexAppServerConnection {
.then(|| credential.direct_provider_route(llm))
.flatten()
.map(|(base_url, api_key)| (base_url.to_string(), api_key.to_string()));
let remote_control_disable_reason =
credential.remote_control_disable_reason(direct_provider_route.is_some());
let isolated_codex_home = prepare_isolated_game_creator_codex_home(
working_dir.path(),
credential,
@@ -1549,6 +1564,12 @@ impl CodexAppServerConnection {
.stderr(Stdio::piped())
.kill_on_drop(true);
game_creator_codex_cli_minimal_environment(&mut command);
if remote_control_disable_reason.is_some() {
// API-key and provider-proxy sessions have no ChatGPT auth.json.
// Keep expected remote-control startup diagnostics out of the
// child log after the protocol-level disable below succeeds.
command.env("RUST_LOG", "warn");
}
if let Some(tool_bridge) = tool_bridge.as_ref() {
command.env(DIRECT_TOOL_BRIDGE_URL_ENV, tool_bridge.url());
}
@@ -1634,6 +1655,13 @@ impl CodexAppServerConnection {
.notify("initialized", serde_json::json!({}))
.await
.map_err(platform_llm::LlmError::Transport)?;
if let Some(reason) = remote_control_disable_reason {
connection
.request("remoteControl/disable", serde_json::json!({}))
.await
.map_err(platform_llm::LlmError::Transport)?;
eprintln!("agent.codex_app_server.remote_control disabled reason={reason}");
}
if let Some(skill_root) = connection.inner._skill_root.as_ref() {
connection
.request(
@@ -3213,8 +3241,8 @@ mod tests {
std::fs::create_dir_all(&project_root).expect("project root");
std::fs::create_dir(&assets).expect("assets directory");
std::fs::create_dir(&agent).expect("agent directory");
let workspace = resolve_direct_codex_game_workspace(&project_root)
.expect("resolve project workspace");
let workspace =
resolve_direct_codex_game_workspace(&project_root).expect("resolve project workspace");
assert_eq!(
workspace,
project_root.canonicalize().expect("canonical project root")
@@ -3703,12 +3731,15 @@ case "$initialize" in *'"method":"initialize"'*) ;; *) exit 85 ;; esac
printf '%s\n' '{"id":1,"result":{"codexHome":"/tmp","platformFamily":"unix","platformOs":"linux","userAgent":"fixture"}}'
IFS= read -r initialized
case "$initialized" in *'"method":"initialized"'*) ;; *) exit 86 ;; esac
IFS= read -r remote_control_disable
case "$remote_control_disable" in *'"method":"remoteControl/disable"'*) ;; *) exit 89 ;; esac
printf '%s\n' '{"id":2,"result":{}}'
IFS= read -r extra_roots
case "$extra_roots" in *'"method":"skills/extraRoots/set"'*) ;; *) exit 87 ;; esac
printf '%s\n' '{"id":2,"result":{}}'
printf '%s\n' '{"id":3,"result":{}}'
IFS= read -r skills_list
case "$skills_list" in *'"method":"skills/list"'*) ;; *) exit 88 ;; esac
printf '%s\n' '{"id":3,"result":{"data":[{"skills":[{"name":"agc-browser-playtest"},{"name":"agc-client-projection"},{"name":"agc-project-structure"},{"name":"agc-web-game-development"},{"name":"taonier-art-assets"}],"errors":[]}]}}'
printf '%s\n' '{"id":4,"result":{"data":[{"skills":[{"name":"agc-browser-playtest"},{"name":"agc-client-projection"},{"name":"agc-project-structure"},{"name":"agc-web-game-development"},{"name":"taonier-art-assets"}],"errors":[]}]}}'
while IFS= read -r line; do :; done
"#,
)
@@ -4016,6 +4047,32 @@ while IFS= read -r line; do :; done
);
}
#[test]
fn remote_control_is_disabled_when_the_isolated_home_has_no_chatgpt_auth() {
let api_key = CodexAppServerCredential::AppDataKey {
fingerprint: "api-key".to_string(),
};
assert_eq!(
api_key.remote_control_disable_reason(false),
Some("api-key-auth")
);
assert_eq!(
api_key.remote_control_disable_reason(true),
Some("provider-proxy-auth")
);
let auth_bridge = CodexAppServerCredential::AuthBridge {
fingerprint: "auth-bridge".to_string(),
auth_json: br#"{"tokens":{"access_token":"fixture"}}"#.to_vec(),
api_key: None,
};
assert_eq!(auth_bridge.remote_control_disable_reason(false), None);
assert_eq!(
auth_bridge.remote_control_disable_reason(true),
Some("provider-proxy-auth")
);
}
#[test]
fn codex_app_server_auth_bridge_snapshot_drives_pool_and_isolated_home() {
let temp = tempfile::tempdir().expect("temp dir");
@@ -4146,12 +4203,15 @@ case "$initialize" in *'"method":"initialize"'*) ;; *) exit 43 ;; esac
printf '%s\n' '{"id":1,"result":{"codexHome":"/tmp","platformFamily":"unix","platformOs":"linux","userAgent":"fixture"}}'
IFS= read -r initialized
case "$initialized" in *'"method":"initialized"'*) ;; *) exit 44 ;; esac
IFS= read -r remote_control_disable
case "$remote_control_disable" in *'"method":"remoteControl/disable"'*) ;; *) exit 47 ;; esac
printf '%s\n' '{"id":2,"result":{}}'
IFS= read -r thread_start
case "$thread_start" in *'"method":"thread/start"'*'"modelProvider":"genarrative_agc"'*) ;; *) exit 45 ;; esac
printf '%s\n' '{"id":2,"result":{"thread":{"id":"thread-1"}}}'
printf '%s\n' '{"id":3,"result":{"thread":{"id":"thread-1"}}}'
IFS= read -r turn_start
case "$turn_start" in *'"method":"turn/start"'*'"outputSchema"'*) ;; *) exit 46 ;; esac
printf '%s\n' '{"id":3,"result":{"turn":{"id":"turn-1","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"id":4,"result":{"turn":{"id":"turn-1","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"method":"turn/started","params":{"threadId":"thread-1","turn":{"id":"turn-1","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"method":"item/fileChange/patchUpdated","params":{"threadId":"thread-1","turnId":"turn-1","itemId":"change-1","patch":"*** SECRET PATCH /private/project"}}'
printf '%s\n' '{"method":"item/commandExecution/outputDelta","params":{"threadId":"thread-1","turnId":"turn-1","itemId":"command-1","delta":"Bearer secret-command-output"}}'
@@ -4267,14 +4327,17 @@ case "$initialize" in *'"method":"initialize"'*) ;; *) exit 70 ;; esac
printf '%s\n' '{"id":1,"result":{"codexHome":"/tmp","platformFamily":"unix","platformOs":"linux","userAgent":"fixture"}}'
IFS= read -r initialized
case "$initialized" in *'"method":"initialized"'*) ;; *) exit 71 ;; esac
IFS= read -r remote_control_disable
case "$remote_control_disable" in *'"method":"remoteControl/disable"'*) ;; *) exit 76 ;; esac
printf '%s\n' '{"id":2,"result":{}}'
IFS= read -r thread_start
case "$thread_start" in *'"method":"thread/start"'*'"approvalPolicy":"never"'*'"sandbox":"read-only"'*) ;; *) exit 72 ;; esac
case "$thread_start" in *'workspace-write'*|*'workspaceWrite'*|*'writableRoots'*) exit 73 ;; esac
printf '%s\n' '{"id":2,"result":{"thread":{"id":"home-thread"}}}'
printf '%s\n' '{"id":3,"result":{"thread":{"id":"home-thread"}}}'
IFS= read -r turn_start
case "$turn_start" in *'"method":"turn/start"'*'"approvalPolicy":"never"'*) ;; *) exit 74 ;; esac
case "$turn_start" in *'"sandboxPolicy"'*|*'workspaceWrite'*|*'writableRoots'*) exit 75 ;; esac
printf '%s\n' '{"id":3,"result":{"turn":{"id":"home-turn","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"id":4,"result":{"turn":{"id":"home-turn","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"method":"item/completed","params":{"threadId":"home-thread","turnId":"home-turn","item":{"id":"item-1","type":"fileChange"}}}'
while IFS= read -r line; do :; done
"#,
@@ -4326,15 +4389,18 @@ while IFS= read -r line; do :; done
IFS= read -r initialize
printf '%s\n' '{"id":1,"result":{"codexHome":"/tmp","platformFamily":"unix","platformOs":"linux","userAgent":"fixture"}}'
IFS= read -r initialized
IFS= read -r remote_control_disable
case "$remote_control_disable" in *'"method":"remoteControl/disable"'*) ;; *) exit 77 ;; esac
printf '%s\n' '{"id":2,"result":{}}'
IFS= read -r thread_start
printf '%s\n' '{"id":2,"result":{"thread":{"id":"thread-cancel"}}}'
printf '%s\n' '{"id":3,"result":{"thread":{"id":"thread-cancel"}}}'
IFS= read -r turn_start
sleep 0.2
printf '%s\n' '{"id":3,"result":{"turn":{"id":"turn-cancel","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"id":4,"result":{"turn":{"id":"turn-cancel","items":[],"status":"inProgress"}}}'
IFS= read -r interrupt
case "$interrupt" in *'"method":"turn/interrupt"'*'"turnId":"turn-cancel"'*) ;; *) exit 51 ;; esac
: > "$HOME/interrupt-seen"
printf '%s\n' '{"id":4,"result":{}}'
printf '%s\n' '{"id":5,"result":{}}'
while IFS= read -r line; do :; done
"#,
)
@@ -4388,10 +4454,13 @@ while IFS= read -r line; do :; done
IFS= read -r initialize
printf '%s\n' '{"id":1,"result":{"codexHome":"/tmp","platformFamily":"unix","platformOs":"linux","userAgent":"fixture"}}'
IFS= read -r initialized
IFS= read -r remote_control_disable
case "$remote_control_disable" in *'"method":"remoteControl/disable"'*) ;; *) exit 78 ;; esac
printf '%s\n' '{"id":2,"result":{}}'
IFS= read -r thread_start
printf '%s\n' '{"id":2,"result":{"thread":{"id":"thread-timeout"}}}'
printf '%s\n' '{"id":3,"result":{"thread":{"id":"thread-timeout"}}}'
IFS= read -r turn_start
printf '%s\n' '{"id":3,"result":{"turn":{"id":"turn-timeout","items":[],"status":"inProgress"}}}'
printf '%s\n' '{"id":4,"result":{"turn":{"id":"turn-timeout","items":[],"status":"inProgress"}}}'
while IFS= read -r line; do :; done
"#,
)
@@ -210,6 +210,7 @@ Supervisor 认领该回执后,由父 run 自己为每个原 delivery 逐一创
- 进程与节点:External Runner 按“有效 Agent LLM 凭据/Responses 路由 + `projectId/agentId/sessionId/runId`”隔离长期 `codex app-server --stdio`,即每个权威节点 run 直接持有自己的 Codex CLI 子进程与 ephemeral thread,每次完整权威请求映射 turn。同一节点 turn 串行,节点之间进程级隔离;单节点连接失败不得使其它节点同时失去终态。Codex thread 不写 durable recovery;节点完成、重启、retry、handoff 和 finalization 仍只认 AGC 账本。
- LLM 配置:`apiKind` 始终只接受 `openai_responses`;非空 Key 转换为 app-server model providerbase URL 生效,Key 仅走专用环境变量;空 Key 只桥接用户 Codex `auth.json`,不继承环境 `CODEX_API_KEY`。设置面板在 app-server 模式继续显示并保存 model、effort、stream、全局/逐 Agent Key 与路由配置;`openai_chat / anthropic` 明确提示切 `provider`,不得悄悄忽略。`stream=true` 接入 app-server 文本 delta`webSearchEnabled=true` 只允许 DirectProject 经客户端审核的 `agc_web_search` 使用,不得启用 Codex 原生 webSearch 或任意网络。
- 安全与取消:临时 cwd、隔离 `CODEX_HOME` 与 OS HOME、read-only、network off、never approval,并在启动前关闭 web/multi-agent/shell/browser/plugin/image 等原生能力;取消从 turn-start pending 阶段就跟踪且只 interrupt 当前 turn。已发送 turn 后连接断开或终态丢失进入 reconciliation,只关闭当前节点进程且不重放同一 request slot;明确 failed/interrupted 不按 transport 重试。
- remote-control 认证边界:没有 ChatGPT `auth.json` 的 API Key / provider-proxy app-server 会在 `initialize` 后、其它 RPC 前立即调用 `remoteControl/disable`,避免上游因 `desired_state=Unknown` 进入 1Hz 认证重试;只有实际桥接 ChatGPT 登录态的 AuthBridge 保持 remote-control 可用。禁用成功后 API Key 子进程使用 `RUST_LOG=warn` 收敛预期认证噪音,禁用 RPC 失败则连接创建失败关闭,不伪造 `auth.json` 或静默继续。
- 资源与退出:app-server pool 按实际凭据快照/base URL/API kind/CLI 版本和节点 run 身份隔离并做有界 LRU;空 AppData Key 必须读取同一份有界 `auth.json` 字节来生成池指纹并桥接隔离登录态,继承的 `CODEX_API_KEY` 始终移除,node thread 也只淘汰 inactive LRU。Runner 正常、强制和 watchdog 退出都显式关池,Linux child 绑定 parent-death signal,防止强杀 Runner 后遗留带凭据孤儿进程。stdout NDJSON 与 stderr 无换行记录均有硬上限;stderr 原文不写入诊断,只记录固定分类、总字节数、SHA-256 和可取得的退出状态。
- 旧配置迁移:既有 AppData 若没有 `agentMode`,只有全局和逐 Agent 路由均为 `openai_responses` 时迁移到 `codex_app_server`;存在 `openai_chat / anthropic` 时显式保留 `provider`,避免打开项目自动恢复时把所有节点批量写成 `invalid-config`。用户确认端点支持 Responses 后,可在设置中显式切换并保留原 model/base URL/API Key。
- 验收:fake JSON-RPC fixture、三态 UI/config、配置指纹、unknown-terminal 零重放、旧两种模式回归和显式 ignored 真实 smoke 全部通过后,才可视为模式切换完成。