Skip to content

路由指标按客户端模型与上游模型别名匹配 #560

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mojie126 wants to merge 2 commits into
base: beta
Choose a base branch
from
+191 −4
Open

路由指标按客户端模型与上游模型别名匹配 #560

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4fa1da6
路由指标按客户端模型与上游模型别名匹配
mojie126 Nov 10, 2026
c90cf06
修正路由回放按客户端别名与上游模型匹配
mojie126 Nov 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
195 changes: 191 additions & 4 deletions desktop-tauri/src-tauri/server/src/server/request_stats/routing.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,17 @@
//! 路由页的轻量指标与历史回放；按请求模型隔离，不把映射目标的其它请求混进来。
//!
//! ── 模型匹配口径 ─────────────────────────────────────────────
//! 指标与回放都按「这个模型」过滤，匹配三个模型名里的任意一个：
//! 客户端点的名（`client_model`，可能带 `(low)` 这类后缀）、网关解析出的名
//! （`model`）与上游真名（`upstream_model`）。
//!
//! 三个都要比的原因：客户端可以点带后缀的别名（`DeepSeek-V4.1-Flash(low)`），
//! 网关把它映射到自己的模型名再映射到上游真名 —— 三个名字两两不同。
//! 只比 `client_model` 的话，路由页按 `model` 作用域查指标就查不到这一行：
//! 在途连线于是不高亮、提供商卡片上的请求数也永远是 0。
//!
//! 口径与「模型请求日志」一致（见 `trace` 模块的同一处匹配），否则同一屏里
//! 两处对「这个模型有没有流量」给出相反的答案。

use rusqlite::params;
use serde_json::{json, Value};
Expand All @@ -19,13 +32,16 @@ impl RequestStats {
COALESCE(SUM(ts >= ?1 AND (status <> 0 OR error IS NOT NULL) AND is_test = 0), 0),
COALESCE(SUM(ts >= ?1 AND status >= 200 AND status < 300 AND error IS NULL AND is_test = 0), 0),
COALESCE(SUM(CASE WHEN ts >= ?1 AND status >= 200 AND status < 300 AND error IS NULL AND is_test = 0 THEN duration_ms ELSE 0 END), 0)
FROM requests WHERE (?2 = '' OR COALESCE(NULLIF(client_model, ''), model) = ?2 COLLATE NOCASE)
FROM requests WHERE (?2 = '' OR model = ?2 COLLATE NOCASE
OR client_model = ?2 COLLATE NOCASE OR upstream_model = ?2 COLLATE NOCASE)
AND (ts >= ?1 OR (status = 0 AND error IS NULL AND ts >= ?3))",
params![start, model, crate::server::logging::process_started_at()], |row| Ok((row.get(0)?, row.get(1)?, row.get(2)?, row.get(3)?)),
)?;
let mut stmt = conn.prepare(
"SELECT account_id, COUNT(*) FROM requests WHERE status = 0 AND error IS NULL AND ts >= ?2
AND (?1 = '' OR COALESCE(NULLIF(client_model, ''), model) = ?1 COLLATE NOCASE) AND account_id <> '' GROUP BY account_id",
AND (?1 = '' OR model = ?1 COLLATE NOCASE
OR client_model = ?1 COLLATE NOCASE OR upstream_model = ?1 COLLATE NOCASE)
AND account_id <> '' GROUP BY account_id",
)?;
let mut counts = serde_json::Map::new();
let mut rows = stmt.query(params![model, crate::server::logging::process_started_at()])?;
Expand Down Expand Up @@ -57,12 +73,14 @@ impl RequestStats {
self.with_conn(&guard, |conn| {
let matched: i64 = conn.query_row(
&format!("SELECT COUNT(*) FROM requests WHERE route IS NOT NULL
AND (?1 = '' OR model = ?1 COLLATE NOCASE){status_clause}"),
AND (?1 = '' OR model = ?1 COLLATE NOCASE
OR client_model = ?1 COLLATE NOCASE OR upstream_model = ?1 COLLATE NOCASE){status_clause}"),
params![model], |row| row.get(0),
)?;
let query = format!(
"SELECT {} FROM requests WHERE route IS NOT NULL
AND (?1 = '' OR model = ?1 COLLATE NOCASE){status_clause}
AND (?1 = '' OR model = ?1 COLLATE NOCASE
OR client_model = ?1 COLLATE NOCASE OR upstream_model = ?1 COLLATE NOCASE){status_clause}
ORDER BY ts DESC, row_id DESC LIMIT ?2 OFFSET ?3", sql::REQUEST_COLUMNS,
);
let mut stmt = conn.prepare(&query)?;
Expand All @@ -75,3 +93,172 @@ impl RequestStats {
loaded.unwrap_or_else(|| json!({ "entries": [], "matched": 0, "available": false }))
}
}

#[cfg(test)]
mod tests {
use super::*;
use crate::server::db::Db;
use crate::server::request_stats::{RequestSource, Retention, RunningProgress};
use std::path::PathBuf;
use std::sync::Arc;

/// 临时库的 RAII 守卫：目录只在用例结束时删一次。
struct TempDir(PathBuf);

impl Drop for TempDir {
fn drop(&mut self) {
let _ = std::fs::remove_dir_all(&self.0);
}
}

fn temp_stats(tag: &str) -> (Arc<RequestStats>, TempDir) {
static SEQ: std::sync::atomic::AtomicUsize = std::sync::atomic::AtomicUsize::new(0);
let id = SEQ.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
let dir = std::env::temp_dir()
.join(format!("request-stats-routing-{}-{}-{id}", std::process::id(), tag));
let _ = std::fs::remove_dir_all(&dir);
let db = Db::open(&dir.join("agent2api.db")).expect("临时库应当能建起来");
(Arc::new(RequestStats::with_db(Some(db), Retention::default)), TempDir(dir))
}

/// 插一条在途行并回写承载它的账号。
fn insert_running(
stats: &RequestStats,
id: &str,
model: &str,
client_model: &str,
account_id: &str,
) {
// 行的时间戳要明确晚于「进程启动时刻」。
//
// 陈旧清理按 `ts < 进程启动时刻` 判僵尸行，而那个时刻是首次调用时
// 锁定的全局值：并行跑用例时，它可能在某个用例取完时间戳之后才被锁定，
// 那一行的 ts 就早于它，行会被当成上一轮遗留补成 408（账号与上游
// 模型一并清空）。生产环境里进程启动时刻早于任何请求，不会有这个竞态。
//
// 这里先取一次全局值把它锁住，再在此之上留出一秒余量：无论谁先锁定，
// 插入行的 ts 都晚于它。
let _ = crate::server::logging::process_started_at();
let ts = crate::server::logging::process_started_at() + 1000;
stats.record_started(
id,
ts,
model,
client_model,
"",
&RequestSource::default(),
false,
);
stats.update_running(id, &RunningProgress {
provider: "trae".to_string(),
account_id: account_id.to_string(),
account_name: "mojie126".to_string(),
upstream_model: model.to_string(),
upstream_reasoning: String::new(),
attempts: 1,
first_response_ms: None,
phase: "streaming".to_string(),
phase_started_at: None,
attempt_details: Vec::new(),
route: None,
sensitive_hits: Vec::new(),
});
}

/// 客户端点的是带后缀的别名，网关按自己的模型名建独立配置时，
/// 在途请求也要算进这个模型的指标 —— 否则路由页的在途连线不会高亮。
#[test]
fn a_running_request_is_counted_by_the_gateway_model_name() {
let (stats, _dir) = temp_stats("gateway-model");
insert_running(
&stats,
"req-1",
"deepseek-v4.1-flash",
"DeepSeek-V4.1-Flash(low)",
"acct-trae",
);
let metrics = stats.routing_metrics(Some("deepseek-v4.1-flash"));
assert_eq!(
metrics["accountRunning"].get("acct-trae").and_then(Value::as_i64),
Some(1),
"在途请求应按网关模型名计入账号在途数：{metrics}"
);
}

/// 反过来，客户端点的别名也要能查到同一批在途请求。
#[test]
fn a_running_request_is_counted_by_the_client_model_alias() {
let (stats, _dir) = temp_stats("client-model");
insert_running(
&stats,
"req-1",
"deepseek-v4.1-flash",
"DeepSeek-V4.1-Flash(low)",
"acct-trae",
);
let metrics = stats.routing_metrics(Some("DeepSeek-V4.1-Flash(low)"));
assert_eq!(
metrics["accountRunning"].get("acct-trae").and_then(Value::as_i64),
Some(1),
"在途请求应按客户端别名计入账号在途数：{metrics}"
);
}

/// 别的模型的请求不能算进来。
#[test]
fn a_running_request_of_another_model_is_not_counted() {
let (stats, _dir) = temp_stats("other-model");
insert_running(
&stats,
"req-1",
"glm-5.3",
"glm-5.3",
"acct-trae",
);
let metrics = stats.routing_metrics(Some("deepseek-v4.1-flash"));
assert_eq!(
metrics["accountRunning"].get("acct-trae").and_then(Value::as_i64),
None,
"别的模型的请求不该计入本模型的账号在途数：{metrics}"
);
}

/// 历史回放与指标必须使用同一套模型名匹配，客户端别名不能只出现在一侧。
#[test]
fn replays_match_the_same_client_alias_as_metrics() {
let (stats, _dir) = temp_stats("replay-client-model");
insert_running(
&stats,
"req-1",
"deepseek-v4.1-flash",
"DeepSeek-V4.1-Flash(low)",
"acct-trae",
);
stats.update_running(
"req-1",
&RunningProgress {
provider: "trae".to_string(),
account_id: "acct-trae".to_string(),
account_name: "mojie126".to_string(),
upstream_model: "deepseek-v4.1-flash".to_string(),
upstream_reasoning: String::new(),
attempts: 1,
first_response_ms: None,
phase: "streaming".to_string(),
phase_started_at: None,
attempt_details: Vec::new(),
route: Some(json!({ "accountId": "acct-trae" })),
sensitive_hits: Vec::new(),
},
);
for model in ["DeepSeek-V4.1-Flash(low)", "deepseek-v4.1-flash"] {
let replays = stats.routing_replays(Some(model), 0, 20, "all");
assert_eq!(
replays["matched"].as_i64(),
Some(1),
"历史回放应按客户端别名或上游模型名命中：{model}，{replays}"
);
assert_eq!(replays["entries"].as_array().map(Vec::len), Some(1));
}
}
}
Loading