Use process-wide memory cache for feature flags

When we switched from a single-threaded application server (Unicorn) to
a multithreaded one (Puma), we did not realize that Puma often reaps
threads after a request is done and recreates them later. This makes the
thread-local cache ineffective, as the cache does not store anything
beyond the lifetime of the thread.

Since `ActiveSupport::Cache::MemoryStore` is thread-safe, we should be
able to switch the L1 cache for feature flags to use this to reduce load
on Redis.

Since read and write access is synchronized, this does have the side
effect of adding contention when feature flags are accessed.

Discovered in
https://gitlab.com/gitlab-com/gl-infra/infrastructure/-/issues/9414

changelogs/unreleased/sh-use-process-cache-for-feature-flags.yml

+---
+title: Use process-wide memory cache for feature flags
+merge_request: 26935
+author:
+type: performance

lib/feature.rb

@@ -134,7 +134,11 @@ class Feature
     end
 
     def l1_cache_backend
-      Gitlab::ThreadMemoryCache.cache_backend
+      if Gitlab::Utils.to_boolean(ENV['USE_THREAD_MEMORY_CACHE'])
+        Gitlab::ThreadMemoryCache.cache_backend
+      else
+        Gitlab::ProcessMemoryCache.cache_backend
+      end
     end
 
     def l2_cache_backend

lib/gitlab/process_memory_cache.rb

+# frozen_string_literal: true
+
+module Gitlab
+  class ProcessMemoryCache
+    # ActiveSupport::Cache::MemoryStore is thread-safe:
+    # https://github.com/rails/rails/blob/2f1fefe456932a6d7d2b155d27b5315c33f3daa1/activesupport/lib/active_support/cache/memory_store.rb#L19
+    @cache = ActiveSupport::Cache::MemoryStore.new
+
+    def self.cache_backend
+      @cache
+    end
+  end
+end

spec/lib/feature_spec.rb

@@ -146,7 +146,15 @@ 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) }
+    context 'with USE_THREAD_MEMORY_CACHE defined' do
+      before do
+        stub_env('USE_THREAD_MEMORY_CACHE', '1')
+      end
+
+      it { expect(described_class.l1_cache_backend).to eq(Gitlab::ThreadMemoryCache.cache_backend) }
+    end
+
+    it { expect(described_class.l1_cache_backend).to eq(Gitlab::ProcessMemoryCache.cache_backend) }
     it { expect(described_class.l2_cache_backend).to eq(Rails.cache) }
 
     it 'caches the status in L1 and L2 caches',

spec/spec_helper.rb

@@ -193,8 +193,10 @@ RSpec.configure do |config|
     # expect(Gitlab::Git::KeepAround).to receive(:execute).and_call_original
     allow(Gitlab::Git::KeepAround).to receive(:execute)
 
-    # Clear thread cache and Sidekiq queues
-    Gitlab::ThreadMemoryCache.cache_backend.clear
+    [Gitlab::ThreadMemoryCache, Gitlab::ProcessMemoryCache].each do |cache|
+      cache.cache_backend.clear
+    end
+
     Sidekiq::Worker.clear_all
 
     # Temporary patch to force admin mode to be active by default in tests when