Introduce ractor and ractor-only categories

This commit modifies the benchmark runner to add 2 new categories:

--category=ractor-only Runs only those benchmarks in benchmarks-ractor
using the Ractor harness.

--category=ractor In addition to the ractor-only benchmarks, this will
run benchmarks specified in benchmarks.yml with `ractor: true` using the
Ractor harness.

In addition to this we've modified the burn in and run once scripts to
use the new harnesses and categories too

Co-Authored-By: Luke Gruber <[email protected]>

Author: eightbitraptor

benchmarks.yml

@@ -31,6 +31,7 @@ mail:
 psych-load:
   desc: psych-load repeatedly loads a small selection of YAML files taken from various OSS projects.
   category: headline
+  ractor: true
 railsbench:
   desc: railsbench is a read-only tiny SQLite-backed Rails app, querying a small selection of .html.erb routes and JSON routes.
   category: headline
@@ -54,16 +55,20 @@ binarytrees:
   desc: binarytrees from the Computer Language Benchmarks Game.
 blurhash:
   desc: blurhash (blurred preview image) calculation
+  ractor: true
 erubi:
   desc: erubi compiles a simple Erb template into a method with erubi, then times evaluating that method.
+  ractor: true
 etanni:
   desc: etanni is an older, extremely simple template-lang format that basically turns your template into an "eval" with a lot of heredocs.
 fannkuchredux:
   desc: fannkuchredux from the Computer Language Benchmarks Game.
+  ractor: true
 fluentd:
   desc: fluentd is a log collector, which parses logs in a server and forwards them to various destinations.
 graphql:
   desc: GraphQL gem parsing a large file, uses Racc which has some Ruby->native->Ruby calls
+  ractor: true
 graphql-native:
   desc: GraphQL gem parsing a large file, but using a native parser
 knucleotide:
@@ -72,26 +77,33 @@ lee:
   desc: lee is a circuit-board layout solver, deployed in a plausibly reality-like way
 matmul:
   desc: matrix multiplication benchmark
+  ractor: true
 nbody:
   desc: nbody from the Computer Language Benchmarks Game.
 nqueens:
   desc: solver for the N-Queens problem
+  ractor: true
 optcarrot:
   desc: optcarrot is a functional headless NES emulator, run on a specific game cartridge for a specific number of frames.
 protoboeuf:
   desc: protoboeuf (pure-Ruby protobuf) message decoding
+  ractor: true
 protoboeuf-encode:
   desc: protoboeuf (pure-Ruby protobuf) message encoding
+  ractor: true
 rack:
   desc: test the performance of the Rack framework with barely any routing.
 ruby-json:
   desc: an optimized version of the json_pure gem's pure Ruby JSON parser.
+  ractor: true
 rubyboy:
   desc: Rubyboy is a functional headless GameBoy emulator, run on a specific game cartridge for a specific number of frames.
+  ractor: true
 rubykon:
   desc: Ruby solver for Go (the boardgame.) Runs many iterations forward from an initial starting board.
 sudoku:
   desc: sudoku solver
+  ractor: true
 tinygql:
   desc: TinyGQL gem parsing a large file in pure Ruby
 
@@ -102,14 +114,17 @@ tinygql:
   desc: 30k_ifelse tests thousands of nested methods containing simple if/else statements.
   category: micro
   single_file: true
+  ractor: true
 30k_methods:
   desc: 30k_methods tests thousands of nested method calls that mostly just call out to other single-call methods.
   category: micro
   single_file: true
+  ractor: true
 attr_accessor:
   desc: attr_accessor tests the performance of getting instance variable values via an attr_accessor imemo.
   category: micro
   single_file: true
+  ractor: true
 cfunc_itself:
   desc: cfunc_itself just calls the 'itself' method many, many times.
   category: micro
@@ -130,10 +145,12 @@ fib:
   desc: Fib is a simple exponential-time recursive Fibonacci number generator.
   category: micro
   single_file: true
+  ractor: true
 getivar:
   desc: getivar tests the performance of getting instance variable values.
   category: micro
   single_file: true
+  ractor: true
 keyword_args:
   desc: keyword_args tests the performance of method calls with keyword arguments.
   category: micro
@@ -142,10 +159,12 @@ loops-times:
   desc: nested loop Integer#times and array access microbenchmark
   category: micro
   single_file: true
+  ractor: true
 object-new:
   desc: instantiate a new object in a loop to test allocation performance
   category: micro
   single_file: true
+  ractor: true
 respond_to:
   desc: respond_to tests the performance of the respond_to? method.
   category: micro
@@ -154,6 +173,7 @@ ruby-xor:
   desc: pure-Ruby string XOR microbenchmark, analogous to xorcist C extension.
   category: micro
   single_file: true
+  ractor: true
 setivar:
   desc: setivar tests the performance of setting instance variable values.
   category: micro
@@ -166,14 +186,18 @@ setivar_young:
   desc: setivar_object tests the performance of setting instance variables to an object, to test write barrier speed on young objects.
   category: micro
   single_file: true
+  ractor: true
 str_concat:
   desc: str_concat tests the performance of string concatenation in multiple different encodings.
   category: micro
   single_file: true
+  ractor: true
 throw:
   desc: microbenchmark for the throw instruction and stack unwinding.
   category: micro
   single_file: true
+  ractor: true
+
 #
 # Ractor-only benchmarks
 #

burn_in.rb

@@ -41,7 +41,7 @@
     args.num_long_runs = v.to_i
   end
 
-  opts.on("--category=headline,other,micro", "when given, only benchmarks with specified categories will run") do |v|
+  opts.on("--category=headline,other,micro,ractor,ractor-only", "when given, only benchmarks with specified categories will run") do |v|
     args.categories = v.split(",")
   end
 
@@ -59,11 +59,18 @@ def free_file_path(parent_dir, name_prefix)
   end
 end
 
-def run_benchmark(bench_name, no_yjit, logs_path, run_time, ruby_version)
+def run_benchmark(bench_id, no_yjit, logs_path, run_time, ruby_version)
   # Determine the path to the benchmark script
-  script_path = File.join('benchmarks', bench_name, 'benchmark.rb')
+  bench_name = bench_id.sub('ractor/', '')
+  bench_dir, harness = if bench_name == bench_id
+    ['benchmarks', 'harness']
+  else
+    ['benchmarks-ractor', 'harness-ractor']
+  end
+
+  script_path = File.join(bench_dir, bench_name, 'benchmark.rb')
   if not File.exist?(script_path)
-    script_path = File.join('benchmarks', bench_name + '.rb')
+    script_path = File.join(bench_dir, bench_name + '.rb')
   end
 
   # Assemble random environment variable options to test
@@ -101,7 +108,7 @@ def run_benchmark(bench_name, no_yjit, logs_path, run_time, ruby_version)
   cmd = [
     'ruby',
     *test_options,
-    "-Iharness",
+    "-I#{harness}",
     script_path,
   ].compact
   cmd_str = cmd.shelljoin
@@ -133,7 +140,7 @@ def run_benchmark(bench_name, no_yjit, logs_path, run_time, ruby_version)
     puts "ERROR"
 
     # Write command executed and output
-    out_path = free_file_path(logs_path, "error_#{bench_name}")
+    out_path = free_file_path(logs_path, "error_#{bench_name.gsub('/', '_')}")
     puts "writing output file #{out_path}"
     contents = ruby_version + "\n\n" + "pid #{status.pid}\n" + user_cmd_str + "\n\n" + output
     File.write(out_path, contents)
@@ -191,11 +198,38 @@ def test_loop(bench_names, no_yjit, logs_path, run_time, ruby_version)
 
 # Extract the names of benchmarks in the categories we want
 metadata = YAML.load_file('benchmarks.yml')
-metadata = metadata.filter do |bench_name, entry|
-  category = entry.fetch('category', 'other')
-  args.categories.include? category
+bench_names = []
+
+if args.categories.include?('ractor-only')
+  # Only include benchmarks with ractor/ prefix (from benchmarks-ractor directory)
+  bench_names = metadata.keys.select { |name| name.start_with?('ractor/') }
+elsif args.categories.include?('ractor')
+  # Include both ractor/ prefixed benchmarks and those with ractor: true
+  metadata.each do |name, entry|
+    if name.start_with?('ractor/') || entry['ractor']
+      bench_names << name
+    end
+  end
+
+  # Also include regular category benchmarks if other categories are specified
+  if args.categories.any? { |cat| ['headline', 'other', 'micro'].include?(cat) }
+    metadata.each do |name, entry|
+      category = entry.fetch('category', 'other')
+      if args.categories.include?(category) && !bench_names.include?(name)
+        bench_names << name
+      end
+    end
+  end
+else
+  # Regular category filtering
+  metadata.each do |name, entry|
+    category = entry.fetch('category', 'other')
+    if args.categories.include?(category)
+      bench_names << name
+    end
+  end
 end
-bench_names = metadata.map { |name, entry| name }
+
 bench_names.sort!
 
 # Fork the test processes