From 2e28f9be470c7c7604d568c53b0762320ad23788 Mon Sep 17 00:00:00 2001
From: Hengfei Yang <hengfei.yang@gmail.com>
Date: Tue, 25 Nov 2025 14:14:46 +0800
Subject: [PATCH] fix: use result cache env for metrics (#9281)

---
 src/config/src/config.rs               | 10 +---------
 src/router/http/mod.rs                 |  4 +---
 src/service/promql/search/cache/mod.rs |  4 ++--
 src/service/promql/search/mod.rs       | 10 ++++++----
 4 files changed, 10 insertions(+), 18 deletions(-)

diff --git a/src/config/src/config.rs b/src/config/src/config.rs
index 5c2f009f6d..197822cb09 100644
--- a/src/config/src/config.rs
+++ b/src/config/src/config.rs
@@ -1170,12 +1170,6 @@ pub struct Common {
         help = "Strategy to use for result cache, default is both, possible value - both, overlap, duration"
     )]
     pub result_cache_selection_strategy: String,
-    #[env_config(
-        name = "ZO_METRICS_CACHE_ENABLED",
-        default = true,
-        help = "Enable result cache for PromQL metrics queries"
-    )]
-    pub metrics_cache_enabled: bool,
     #[env_config(name = "ZO_SWAGGER_ENABLED", default = true)]
     pub swagger_enabled: bool,
     #[env_config(
@@ -2812,16 +2806,14 @@ fn check_disk_cache_config(cfg: &mut Config) -> Result<(), anyhow::Error> {
     // disable disk cache for local disk storage
     if cfg.common.is_local_storage
         && !cfg.common.result_cache_enabled
-        && !cfg.common.metrics_cache_enabled
         && !cfg.common.feature_query_streaming_aggs
     {
         cfg.disk_cache.enabled = false;
     }
 
-    // disable result cache and metrics cache if disk cache is disabled
+    // disable result cache if disk cache is disabled
     if !cfg.disk_cache.enabled {
         cfg.common.result_cache_enabled = false;
-        cfg.common.metrics_cache_enabled = false;
         cfg.common.feature_query_streaming_aggs = false;
     }
 
diff --git a/src/router/http/mod.rs b/src/router/http/mod.rs
index 7212c20911..d2617e96e9 100644
--- a/src/router/http/mod.rs
+++ b/src/router/http/mod.rs
@@ -179,9 +179,7 @@ async fn dispatch(
     }
 
     // check if the request need to be proxied by body
-    if (cfg.common.result_cache_enabled || cfg.common.metrics_cache_enabled)
-        && is_querier_route_by_body(&path)
-    {
+    if cfg.common.result_cache_enabled && is_querier_route_by_body(&path) {
         return proxy_querier_by_body(req, payload, client, new_url, start).await;
     }
 
diff --git a/src/service/promql/search/cache/mod.rs b/src/service/promql/search/cache/mod.rs
index 361135fb13..835f78fc63 100644
--- a/src/service/promql/search/cache/mod.rs
+++ b/src/service/promql/search/cache/mod.rs
@@ -55,7 +55,7 @@ static GLOBAL_CACHE: Lazy<Vec<RwLock<MetricsIndex>>> = Lazy::new(|| {
 
 pub async fn init() -> Result<()> {
     let cfg = get_config();
-    if !cfg.common.metrics_cache_enabled {
+    if !cfg.common.result_cache_enabled {
         return Ok(());
     }
 
@@ -352,7 +352,7 @@ pub async fn set(
 /// load the cache item from the secondary storage
 pub async fn load(cache_key: &str) -> Result<()> {
     let cfg = get_config();
-    if !cfg.common.metrics_cache_enabled {
+    if !cfg.common.result_cache_enabled {
         return Ok(());
     }
     let Some((key, start, end)) = parse_cache_item_key(cache_key) else {
diff --git a/src/service/promql/search/mod.rs b/src/service/promql/search/mod.rs
index 2b1968441b..557f7efba4 100644
--- a/src/service/promql/search/mod.rs
+++ b/src/service/promql/search/mod.rs
@@ -100,7 +100,7 @@ async fn search_in_cluster(
 
     // cache disabled if result cache is disabled or use_cache is false or start == end or step == 0
     let cache_disabled =
-        !cfg.common.metrics_cache_enabled || !req.use_cache || start == end || step == 0;
+        !cfg.common.result_cache_enabled || !req.use_cache || start == end || step == 0;
     // adjust start and end time
     let (start, end) = adjust_start_end(start, end, step);
 
@@ -144,11 +144,13 @@ async fn search_in_cluster(
         match cache::get(query, start, end, step).await {
             Ok(Some((new_start, values))) => {
                 let took = start_time.elapsed().as_millis() as i32;
+                let cache_ratio = (new_start - start) as f64 / (end - start) as f64;
                 config::metrics::QUERY_METRICS_CACHE_RATIO
                     .with_label_values(&[&req.org_id])
-                    .observe((new_start - start) as f64 / (end - start) as f64);
+                    .observe(cache_ratio);
                 log::info!(
-                    "[trace_id {trace_id}] promql->search->cache: hit cache, took: {took} ms"
+                    "[trace_id {trace_id}] promql->search->cache: hit cache, cache ratio: {:.2}%, took: {took} ms",
+                    cache_ratio * 100.0,
                 );
                 (new_start, values)
             }
@@ -348,7 +350,7 @@ async fn search_in_cluster(
     .await;
 
     // cache the result
-    if cfg.common.metrics_cache_enabled
+    if cfg.common.result_cache_enabled
         && let Some(matrix) = values.get_ref_matrix_values()
         && let Err(err) = cache::set(
             trace_id,