Create profile.yml (#42)

* Create profile.yml

* update profile

* Update profile.yml

* comment out set

* Update profile.yml

* try simplest

* Update profile.yml

* local dalli

* Update profile.yml

* ok

* profile both get and set

* improvements on bench and add multi to profiles

* raw bench

* Update profile.yml

.github/workflows/profile.yml

@@ -0,0 +1,38 @@
+name: Profiles
+
+on: [push, pull_request]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v4
+    - name: Install Memcached 1.6.23
+      working-directory: scripts
+      env:
+        MEMCACHED_VERSION: 1.6.23
+      run: |
+        chmod +x ./install_memcached.sh
+        ./install_memcached.sh
+        memcached -d
+    - name: Set up Ruby
+      uses: ruby/setup-ruby@v1
+      with:
+        ruby-version: 3.4
+        bundler-cache: true # 'bundle install' and cache
+    - name: Run Profiles
+      run: RUBY_YJIT_ENABLE=1 BENCH_TARGET=all bundle exec bin/profile
+    - name: Upload profile results
+      uses: actions/upload-artifact@v4
+      with:
+        name: profile-results
+        path: |
+          client_get_profile.json
+          socket_get_profile.json
+          client_set_profile.json
+          socket_set_profile.json
+          client_get_multi_profile.json
+          socket_get_multi_profile.json
+          client_set_multi_profile.json
+          socket_set_multi_profile.json

bin/benchmark

@@ -42,12 +42,13 @@ bench_target = ENV['BENCH_TARGET'] || 'set'
 bench_time = (ENV['BENCH_TIME'] || 10).to_i
 bench_warmup = (ENV['BENCH_WARMUP'] || 3).to_i
 bench_payload_size = (ENV['BENCH_PAYLOAD_SIZE'] || 700_000).to_i
+payload = 'B' * bench_payload_size
 TERMINATOR = "\r\n"
 puts "yjit: #{RubyVM::YJIT.enabled?}"
 
-client = Dalli::Client.new('localhost', serializer: StringSerializer, compress: false)
-multi_client = Dalli::Client.new('localhost:112111,localhost:11222', serializer: StringSerializer, compress: false)
-string_client = Dalli::Client.new('localhost', serializer: StringSerializer, compress: false)
+client = Dalli::Client.new('localhost', serializer: StringSerializer, compress: false, raw: true)
+multi_client = Dalli::Client.new('localhost:11211,localhost:11222', serializer: StringSerializer, compress: false, raw: true)
+
 # The raw socket implementation is used to benchmark the performance of dalli & the overhead of the various abstractions
 # in the library.
 sock = TCPSocket.new('127.0.0.1', '11211', connect_timeout: 1)
@@ -58,12 +59,8 @@ sock.setsockopt(Socket::SOL_SOCKET, Socket::SO_KEEPALIVE, true)
 # Benchamrks did see an improvement in performance when increasing the SO_SNDBUF buffer size
 sock.setsockopt(Socket::SOL_SOCKET, Socket::SO_SNDBUF, 1024 * 1024 * 8)
 
-TERMINATOR = "\r\n"
-payload = 'B' * bench_payload_size
-
 # ensure the clients are all connected and working
 client.set('key', payload)
-string_client.set('string_key', payload)
 sock.write("set sock_key 0 3600 #{payload.bytesize}\r\n")
 sock.write(payload)
 sock.write(TERMINATOR)

bin/profile

@@ -11,13 +11,12 @@ require 'json'
 
 gemfile do
   source 'https://rubygems.org'
-  gem 'dalli'
   gem 'benchmark-ips'
   gem 'vernier'
   gem 'logger'
 end
 
-require 'dalli'
+require_relative '../lib/dalli'
 require 'benchmark/ips'
 require 'vernier'
 
@@ -46,7 +45,6 @@ TERMINATOR = "\r\n"
 puts "yjit: #{RubyVM::YJIT.enabled?}"
 
 client = Dalli::Client.new('localhost', serializer: StringSerializer, compress: false)
-multi_client = Dalli::Client.new('localhost:112111,localhost:11222', serializer: StringSerializer, compress: false)
 
 # The raw socket implementation is used to benchmark the performance of dalli & the overhead of the various abstractions
 # in the library.
@@ -96,15 +94,31 @@ def sock_get_multi(sock, pairs)
     next unless line.start_with?('VA ') || last_result
 
     _, value_length, _flags, key = line.split
-    value = sock.read(value_length.to_i + TERMINATOR.length)
-    results[key[1..]] = value.chomp!(0..-TERMINATOR.length)
+    results[key[1..]] = sock.read(value_length.to_i)
+    sock.read(TERMINATOR.length)
     break if results.size == pairs.size
     break if last_result
   end
   results
 end
 
-if bench_target == 'get'
+def sock_set_multi(sock, pairs)
+  count = pairs.length
+  tail = ''
+  ttl = 3600
+
+  pairs.each do |key, value|
+    count -= 1
+    tail = count.zero? ? '' : 'q'
+    sock.write(String.new("ms #{key} #{value.bytesize} c F0 T#{ttl} MS #{tail}\r\n", capacity: key.size + value.bytesize + 40))
+    sock.write(value)
+    sock.write(TERMINATOR)
+  end
+  sock.flush
+  sock.gets(TERMINATOR) # clear the buffer
+end
+
+if %w[all get].include?(bench_target)
   Vernier.profile(out: 'client_get_profile.json') do
     start_time = Time.now
     client.get(dalli_key) while Time.now - start_time < bench_time
@@ -118,5 +132,46 @@ if bench_target == 'get'
       sock.read(payload.bytesize)
     end
   end
-  exit 0
+end
+
+if %w[all set].include?(bench_target)
+  Vernier.profile(out: 'client_set_profile.json') do
+    start_time = Time.now
+    client.set(dalli_key, payload, 3600, raw: true) while Time.now - start_time < bench_time
+  end
+
+  Vernier.profile(out: 'socket_set_profile.json') do
+    start_time = Time.now
+    while Time.now - start_time < bench_time
+      sock.write("ms sock_key #{payload.bytesize} T3600 MS\r\n")
+      sock.write(payload)
+      sock.write("\r\n")
+      sock.flush
+      sock.readline # clear the buffer
+    end
+  end
+end
+
+if %w[all get_multi].include?(bench_target)
+  Vernier.profile(out: 'client_get_multi_profile.json') do
+    start_time = Time.now
+    client.get_multi(pairs.keys) while Time.now - start_time < bench_time
+  end
+
+  Vernier.profile(out: 'socket_get_multi_profile.json') do
+    start_time = Time.now
+    sock_get_multi(sock, pairs) while Time.now - start_time < bench_time
+  end
+end
+
+if %w[all set_multi].include?(bench_target)
+  Vernier.profile(out: 'client_set_multi_profile.json') do
+    start_time = Time.now
+    client.set_multi(pairs, 3600, raw: true) while Time.now - start_time < bench_time
+  end
+
+  Vernier.profile(out: 'socket_set_multi_profile.json') do
+    start_time = Time.now
+    sock_set_multi(sock, pairs) while Time.now - start_time < bench_time
+  end
 end