Add experimental caching metrics (#3558)

Fix #3554 This creates a new metric recorded only if we set the configuration option `telemetry.metrics.common.experimental_cache_metrics.enabled` to `true`. * `apollo.router.operations.entity` (histogram): cache hit ratio per subgraph and entity type This simulates an entity cache to find out if it would be useful. Each time we do a subgraph query, we use as cache key: - subgraph name - entity type - query - vary headers - entity key We record if we have seen this entity before (using a bloom filter) and calculate the cache hit ratio for that query, per subgraph and entity type. --------- Signed-off-by: Benjamin Coenen <[email protected]> Co-authored-by: Benjamin Coenen <[email protected]>
apollographql · Sep 12, 2023 · 961fff4 · 961fff4
1 parent 0cceca7
commit 961fff4
Show file tree

Hide file tree

Showing 9 changed files with 393 additions and 31 deletions.
diff --git a/.changesets/maint_bnjjj_caching_metrics.md b/.changesets/maint_bnjjj_caching_metrics.md
@@ -0,0 +1,38 @@
+### Add experimental caching metrics ([PR #3532](https://github.com/apollographql/router/pull/3532))
+
+It adds a metric only if you configure `telemetry.metrics.common.experimental_cache_metrics.enabled` to `true`. It will generate metrics to evaluate which entities would benefit from caching. It simulates a cache with a TTL, configurable at `telemetry.metrics.common.experimental_cache_metrics.ttl` (default: 5 seconds), and measures the cache hit rate per entity type and subgraph.
+
+example
+
+```
+# HELP apollo.router.operations.entity.cache_hit
+# TYPE apollo_router_operations_entity.cache_hit histogram
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="0.05"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="0.1"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="0.25"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="0.5"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="1"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="2.5"} 3
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="5"} 4
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="10"} 4
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="20"} 4
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="1000"} 4
+apollo_router_operations_entity_cache_hitbucket{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version="",le="+Inf"} 4
+apollo_router_operations_entity_cache_hitsum{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version=""} 7
+apollo_router_operations_entity_cache_hitcount{entity_type="Product",service_name="apollo-router",subgraph="products",otel_scope_name="apollo/router",otel_scope_version=""} 4
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="0.05"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="0.1"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="0.25"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="0.5"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="1"} 0
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="2.5"} 1
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="5"} 1
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="10"} 1
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="20"} 1
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="1000"} 1
+apollo_router_operations_entity_cache_hitbucket{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version="",le="+Inf"} 1
+apollo_router_operations_entity_cache_hitsum{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version=""} 1
+apollo_router_operations_entity_cache_hitcount{entity_type="User",service_name="apollo-router",subgraph="users",otel_scope_name="apollo/router",otel_scope_version=""} 1
+```
+
+By [@bnjjj](https://github.com/bnjjj) [@Geal](https://github.com/geal) in https://github.com/apollographql/router/pull/3532
diff --git a/Cargo.lock b/Cargo.lock
@@ -281,6 +281,7 @@ dependencies = [
  "aws-types",
  "axum",
  "base64 0.21.2",
+ "bloomfilter",
  "brotli",
  "buildstructor 0.5.3",
  "bytes",
@@ -1098,6 +1099,17 @@ dependencies = [
  "generic-array 0.14.7",
 ]
 
+[[package]]
+name = "bloomfilter"
+version = "1.0.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b92db7965d438b8b4b1c1d0aedd188440a1084593c9eb7f6657e3df7e906d934"
+dependencies = [
+ "bit-vec",
+ "getrandom 0.2.10",
+ "siphasher",
+]
+
 [[package]]
 name = "brotli"
 version = "3.3.4"
@@ -5793,6 +5805,12 @@ dependencies = [
  "time",
 ]
 
+[[package]]
+name = "siphasher"
+version = "1.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "54ac45299ccbd390721be55b412d41931911f654fa99e2cb8bfb57184b2061fe"
+
 [[package]]
 name = "slab"
 version = "0.4.8"

diff --git a/apollo-router/Cargo.toml b/apollo-router/Cargo.toml
@@ -67,6 +67,7 @@ async-trait = "0.1.73"
 atty = "0.2.14"
 axum = { version = "0.6.20", features = ["headers", "json", "original-uri"] }
 base64 = "0.21.2"
+bloomfilter = "1.0.12"
 buildstructor = "0.5.3"
 bytes = "1.4.0"
 clap = { version = "4.4.2", default-features = false, features = [
@@ -163,6 +164,7 @@ prost = "0.11.9"
 prost-types = "0.11.9"
 proteus = "0.5.0"
 rand = "0.8.5"
+rand_core = "0.6.4"
 rhai = { version = "1.15.1", features = ["sync", "serde", "internals"] }
 regex = "1.9.5"
 reqwest = { version = "0.11.19", default-features = false, features = [
@@ -236,7 +238,6 @@ memchr = "2.6.3"
 brotli = "3.3.4"
 zstd = "0.12.4"
 zstd-safe = "6.0.6"
-rand_core = "0.6.4"
 # note: AWS dependencies should always use the same version
 aws-sigv4 = "0.56.0"
 aws-credential-types = "0.56.0"

diff --git a/...r/src/configuration/snapshots/apollo_router__configuration__tests__schema_generation.snap b/...r/src/configuration/snapshots/apollo_router__configuration__tests__schema_generation.snap
@@ -4314,6 +4314,23 @@ expression: "&schema"
                     "format": "double"
                   }
                 },
+                "experimental_cache_metrics": {
+                  "description": "Experimental metrics to know more about caching strategies",
+                  "type": "object",
+                  "properties": {
+                    "enabled": {
+                      "description": "Enable experimental metrics",
+                      "default": false,
+                      "type": "boolean"
+                    },
+                    "ttl": {
+                      "description": "Potential TTL for a cache if we had one (default: 5secs)",
+                      "default": "5s",
+                      "type": "string"
+                    }
+                  },
+                  "additionalProperties": false
+                },
                 "resources": {
                   "description": "Resources",
                   "default": {},

diff --git a/apollo-router/src/plugins/telemetry/config.rs b/apollo-router/src/plugins/telemetry/config.rs
@@ -94,6 +94,28 @@ pub(crate) struct MetricsCommon {
     /// Custom buckets for histograms
     #[serde(default = "default_buckets")]
     pub(crate) buckets: Vec<f64>,
+    /// Experimental metrics to know more about caching strategies
+    pub(crate) experimental_cache_metrics: ExperimentalCacheMetricsConf,
+}
+
+#[derive(Clone, Debug, Deserialize, JsonSchema)]
+#[serde(deny_unknown_fields, rename_all = "snake_case", default)]
+pub(crate) struct ExperimentalCacheMetricsConf {
+    /// Enable experimental metrics
+    pub(crate) enabled: bool,
+    #[serde(with = "humantime_serde")]
+    #[schemars(with = "String")]
+    /// Potential TTL for a cache if we had one (default: 5secs)
+    pub(crate) ttl: Duration,
+}
+
+impl Default for ExperimentalCacheMetricsConf {
+    fn default() -> Self {
+        Self {
+            enabled: false,
+            ttl: Duration::from_secs(5),
+        }
+    }
 }
 
 fn default_buckets() -> Vec<f64> {
@@ -110,6 +132,7 @@ impl Default for MetricsCommon {
             service_namespace: None,
             resources: HashMap::new(),
             buckets: default_buckets(),
+            experimental_cache_metrics: ExperimentalCacheMetricsConf::default(),
         }
     }
 }