Cache Flipper feature flags in L1 and L2 caches

In https://gitlab.com/gitlab-com/gl-infra/production/issues/928, we saw
a significant amount of network traffic and CPU usage due to Redis
checking feature flags via Flipper. Since these flags are hit with every
request, the overhead becomes significant. To alleviate Redis overhead,
we now cache the data in the following way:

* L1: A thread-local memory store for 1 minute
* L2: Redis for 1 hour

changelogs/unreleased/sh-cache-flipper-checks-in-memory.yml

+---
+title: Cache Flipper feature flags in L1 and L2 caches
+merge_request: 30276
+author:
+type: performance

lib/feature.rb

@@ -103,10 +103,27 @@ class Feature
         feature_class: FlipperFeature,
         gate_class: FlipperGate)
 
+      # Redis L2 cache
+      redis_cache_adapter =
+        Flipper::Adapters::ActiveSupportCacheStore.new(
+          active_record_adapter,
+          l2_cache_backend,
+          expires_in: 1.hour)
+
+      # Thread-local L1 cache: use a short timeout since we don't have a
+      # way to expire this cache all at once
       Flipper::Adapters::ActiveSupportCacheStore.new(
-        active_record_adapter,
-        Rails.cache,
-        expires_in: 1.hour)
+        redis_cache_adapter,
+        l1_cache_backend,
+        expires_in: 1.minute)
+    end
+
+    def l1_cache_backend
+      Gitlab::ThreadMemoryCache.cache_backend
+    end
+
+    def l2_cache_backend
+      Rails.cache
     end
   end
 

spec/lib/feature_spec.rb

@@ -144,6 +144,68 @@ describe Feature do
       expect(described_class.enabled?(:enabled_feature_flag)).to be_truthy
     end
 
+    it { expect(described_class.l1_cache_backend).to eq(Gitlab::ThreadMemoryCache.cache_backend) }
+    it { expect(described_class.l2_cache_backend).to eq(Rails.cache) }
+
+    it 'caches the status in L1 and L2 caches',
+       :request_store, :use_clean_rails_memory_store_caching do
+      described_class.enable(:enabled_feature_flag)
+      flipper_key = "flipper/v1/feature/enabled_feature_flag"
+
+      expect(described_class.l2_cache_backend)
+        .to receive(:fetch)
+        .once
+        .with(flipper_key, expires_in: 1.hour)
+        .and_call_original
+
+      expect(described_class.l1_cache_backend)
+        .to receive(:fetch)
+        .once
+        .with(flipper_key, expires_in: 1.minute)
+        .and_call_original
+
+      2.times do
+        expect(described_class.enabled?(:enabled_feature_flag)).to be_truthy
+      end
+    end
+
+    context 'cached feature flag', :request_store do
+      let(:flag) { :some_feature_flag }
+
+      before do
+        described_class.flipper.memoize = false
+        described_class.enabled?(flag)
+      end
+
+      it 'caches the status in L1 cache for the first minute' do
+        expect do
+          expect(described_class.l1_cache_backend).to receive(:fetch).once.and_call_original
+          expect(described_class.l2_cache_backend).not_to receive(:fetch)
+          expect(described_class.enabled?(flag)).to be_truthy
+        end.not_to exceed_query_limit(0)
+      end
+
+      it 'caches the status in L2 cache after 2 minutes' do
+        Timecop.travel 2.minutes do
+          expect do
+            expect(described_class.l1_cache_backend).to receive(:fetch).once.and_call_original
+            expect(described_class.l2_cache_backend).to receive(:fetch).once.and_call_original
+            expect(described_class.enabled?(flag)).to be_truthy
+          end.not_to exceed_query_limit(0)
+        end
+      end
+
+      it 'fetches the status after an hour' do
+        Timecop.travel 61.minutes do
+          expect do
+            expect(described_class.l1_cache_backend).to receive(:fetch).once.and_call_original
+            expect(described_class.l2_cache_backend).to receive(:fetch).once.and_call_original
+            expect(described_class.enabled?(flag)).to be_truthy
+          end.not_to exceed_query_limit(1)
+        end
+      end
+    end
+
     context 'with an individual actor' do
       CustomActor = Struct.new(:flipper_id)