From 2a3035ca9babe5a0dd1aa385ce7f32526a163053 Mon Sep 17 00:00:00 2001
From: CMGS <ilskdw@gmail.com>
Date: Tue, 27 Oct 2026 10:06:51 +0800
Subject: [PATCH] perf: check an account's models concurrently in
 /internal/accounts

Per-model health turned the view's one lookup per account into one per
served model, awaited in sequence: 25 accounts and 102 models took 17.7 ms
against a local Redis, 7.3 ms with each account's lookups joined. Also
rewraps the HealthStore paragraph in architecture.md.
---
 crates/views/src/lib.rs | 32 ++++++++++++++++++++------------
 docs/architecture.md    |  6 +++---
 2 files changed, 23 insertions(+), 15 deletions(-)

diff --git a/crates/views/src/lib.rs b/crates/views/src/lib.rs
index 31f030a..8ab4b9e 100644
--- a/crates/views/src/lib.rs
+++ b/crates/views/src/lib.rs
@@ -1321,18 +1321,26 @@ async fn accounts(State(s): State<AppState>, _: GlobalAdmin) -> Response {
     let health = &s.handler.state().health;
     let mut data: Vec<Value> = Vec::with_capacity(cfg.accounts.len());
     for a in &cfg.accounts {
-        let mut cooling = Vec::new();
-        for m in cfg.models.iter().filter(|m| {
-            a.protocols.contains(&m.protocol)
-                && m.provider.as_deref().is_none_or(|p| p == a.provider)
-        }) {
-            if !health
-                .available(&gw_state::health_key(&a.name, &m.name))
-                .await
-            {
-                cooling.push(m.name.as_str());
-            }
-        }
+        let served: Vec<&str> = cfg
+            .models
+            .iter()
+            .filter(|m| {
+                a.protocols.contains(&m.protocol)
+                    && m.provider.as_deref().is_none_or(|p| p == a.provider)
+            })
+            .map(|m| m.name.as_str())
+            .collect();
+        let checks = futures::future::join_all(
+            served
+                .iter()
+                .map(|m| async move { health.available(&gw_state::health_key(&a.name, m)).await }),
+        )
+        .await;
+        let cooling: Vec<&str> = served
+            .into_iter()
+            .zip(checks)
+            .filter_map(|(m, ok)| (!ok).then_some(m))
+            .collect();
         data.push(json!({
             "name": a.name,
             "provider": a.provider,
diff --git a/docs/architecture.md b/docs/architecture.md
index ab4ec71..ce85d35 100644
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -78,9 +78,9 @@ default, so the whole pipeline is testable offline:
 - **`HealthStore`** — account cooldown/recovery, per account and served
   model, so one model's refusals never bench the account for the others.
   In-process breaker by default; `RedisHealth` with `storage.redis_url` — a
-  tripped account is skipped for that model by every instance. Only upstream faults count (5xx, timeouts,
-  broken streams, a 401/402/403 credential or billing refusal), never a
-  client-caused or rate-limit error.
+  tripped account is skipped for that model by every instance. Only upstream
+  faults count (5xx, timeouts, broken streams, a 401/402/403 credential or
+  billing refusal), never a client-caused or rate-limit error.
 - **`Governance`** — rate/quota/TPM counters. In-process by default;
   `RedisGovernance` shares them (including pooled tenant QPS) fleet-wide.
 - **`Moderator`** — the external content review behind `security.moderate`;