unified cache metrics (#5002)

* unified cache metrics * lint, use IntCounter * CR comments --------- Co-authored-by: Paul Masurel <[email protected]>
quickwit-oss · Jun 12, 2024 · c228fee · c228fee
1 parent de67160
commit c228fee
Show file tree

Hide file tree

Showing 2 changed files with 41 additions and 20 deletions.
diff --git a/quickwit/quickwit-common/src/metrics.rs b/quickwit/quickwit-common/src/metrics.rs
@@ -90,6 +90,25 @@ pub fn new_counter(
     counter
 }
 
+pub fn new_counter_with_labels(
+    name: &str,
+    help: &str,
+    subsystem: &str,
+    const_labels: &[(&str, &str)],
+) -> IntCounter {
+    let owned_const_labels: HashMap<String, String> = const_labels
+        .iter()
+        .map(|(label_name, label_value)| (label_name.to_string(), label_value.to_string()))
+        .collect();
+    let counter_opts = Opts::new(name, help)
+        .namespace("quickwit")
+        .subsystem(subsystem)
+        .const_labels(owned_const_labels);
+    let counter = IntCounter::with_opts(counter_opts).expect("failed to create counter");
+    prometheus::register(Box::new(counter.clone())).expect("failed to register counter");
+    counter
+}
+
 pub fn new_counter_vec<const N: usize>(
     name: &str,
     help: &str,

diff --git a/quickwit/quickwit-storage/src/metrics.rs b/quickwit/quickwit-storage/src/metrics.rs
@@ -20,7 +20,9 @@
 // See https://prometheus.io/docs/practices/naming/
 
 use once_cell::sync::Lazy;
-use quickwit_common::metrics::{new_counter, new_gauge, IntCounter, IntGauge};
+use quickwit_common::metrics::{
+    new_counter, new_counter_with_labels, new_gauge, IntCounter, IntGauge,
+};
 
 /// Counters associated to storage operations.
 pub struct StorageMetrics {
@@ -95,38 +97,38 @@ pub struct CacheMetrics {
 
 impl CacheMetrics {
     pub fn for_component(component_name: &str) -> Self {
-        let namespace = format!("cache_{component_name}");
+        const CACHE_METRICS_NAMESPACE: &str = "cache";
         CacheMetrics {
             component_name: component_name.to_string(),
             in_cache_count: new_gauge(
                 "in_cache_count",
-                "Count of {component_name} in cache",
-                &namespace,
-                &[],
+                "Count of in cache by component",
+                CACHE_METRICS_NAMESPACE,
+                &[("component_name", component_name)],
             ),
             in_cache_num_bytes: new_gauge(
                 "in_cache_num_bytes",
-                "Number of {component_name} bytes in cache",
-                &namespace,
-                &[],
+                "Number of bytes in cache by component",
+                CACHE_METRICS_NAMESPACE,
+                &[("component_name", component_name)],
             ),
-            hits_num_items: new_counter(
+            hits_num_items: new_counter_with_labels(
                 "cache_hits_total",
-                "Number of {component_name} cache hits",
-                &namespace,
-                &[],
+                "Number of cache hits by component",
+                CACHE_METRICS_NAMESPACE,
+                &[("component_name", component_name)],
             ),
-            hits_num_bytes: new_counter(
+            hits_num_bytes: new_counter_with_labels(
                 "cache_hits_bytes",
-                "Number of {component_name} cache hits in bytes",
-                &namespace,
-                &[],
+                "Number of cache hits in bytes by component",
+                CACHE_METRICS_NAMESPACE,
+                &[("component_name", component_name)],
             ),
-            misses_num_items: new_counter(
+            misses_num_items: new_counter_with_labels(
                 "cache_misses_total",
-                "Number of {component_name} cache misses",
-                &namespace,
-                &[],
+                "Number of cache misses by component",
+                CACHE_METRICS_NAMESPACE,
+                &[("component_name", component_name)],
             ),
         }
     }