JRuby: Past, Present, and Future

JRuby on Rails is 20 years old! This talk reviews the last 20 years of JRuby development, with a focus on the challenges of supporting Ruby and Rails year after year.

We'll look at the state-of-the-art features that make JRuby attractive and talk about the next generation of JRuby capabilities and applications. JRuby brings a whole new world of opportunities to Ruby — come learn how to use it to do more with Ruby.

Presented at Euruko 2026 in Brno, Czechia.

Transcript

1. JRuby: Past, Present, and Future
2. @headius headius.com Dobrý den! • Charles Oliver Nutter ("headius") • My rst EuRuKo was 2008 (Prague!) • 20 years full-time JRuby dev • Stickers and t-shirts! • Co-founder of Headius Enterprises • fi • "The JRuby Company" I love Ruby, karaoke, and beer!
3. fi How did a Java guy nd Ruby?
7. Professional Career • 1996-1998: University of Minnesota • Java applets, web-based registration system • 1998-2002: Senior Java Developer at various jobs • 2002-2006: Java Enterprise Architect • Converting mainframe apps for US government
9. RubyConf 2004 • My rst Ruby conference • • Hallway hacking: Rake, RubyGems • Pickaxe 2 released by PragProg • A total of 65 attendees! • fi I didn't even know Ruby! Including some names you know...
11. Ko1 に聞け YARV Ko1 に聞け YARV RubyVM Yet Another Yet Another RubyVM ささだ こういち ささだ日本Rubyの会 こういち 東京農工大学大学院 東京農工大学大学院 Ko1@atdot.net 日本Rubyの会 Ko1@atdot.net RubyConf2004 Oct. 2 3 RubyConf2004 Oct. 2 3
12. A. Self Introduction (cont.)  Software • Rava: A JavaVM in Ruby • My best joke software Ear ly in for spirat JRu ion by? ! • Rucheme: A Scheme interpreter in Ruby • Compile to instruction set which I designed • I like Scheme • Nadoka: IRC Client Server software • IRC proxy/bouncer • Ruby’s killer application (… nobody else may agree) 9
15. JRuby is Reborn
16. JRuby is Reborn • 2001-2004: Early work on JRuby • • 2004: low activity, few developers • • Parser, interpreter, basic core classes (Ruby 1.6) Tom Enebo, my university co-worker! It seemed like destiny • But what are we going to do with it?
18. Ruby on Rails • Rails was changing the web dev world • JVM is amazing for web apps • The obvious path: JRuby on Rails! • Could it be done? 🤔
19. The Road to Rails • JRuby in 2005: Nothing worked! • Started with basic tools • Ran every test we could nd, wrote our own • Ruby on the JVM is really hard In other words... challenge accepted! fi •
20. IRB • First step: run the REPL • Language compatibility work • Implement "eval" • Console work: stdio and readline support • One day in early 2006, it worked! $ irb irb(main):001:0> 1 + 2 => 3 irb(main):002:1* class Foo irb(main):003:2* def foo irb(main):004:2* puts 1 irb(main):005:1* end irb(main):006:0> end => :foo irb(main):007:0> Foo.new.foo 1 => nil
21. [] jruby-0.8.3 $ ls bin COPYING.CPL COPYING.LGPL lib samples COPYING COPYING.GPL docs README [] jruby-0.8.3 $ bin/jruby -v ruby 1.8.2 (0) [java] [] jruby-0.8.3 $ bin/jruby -e "puts 'hello, world'" hello, world [] jruby-0.8.3 $ ls samples beer.rb eval.rb java2.rb swing2.rb xslt.rb error.rb fib.rb swing.rb thread.rb [] jruby-0.8.3 $ bin/jruby samples/swing2.rb [] jruby-0.8.3 $ bin/jruby -I ../ruby-1.8.2/lib ../ruby-1.8.2/bin/irb irb(main):001:0> RUBY_VERSION => "1.8.2" irb(main):002:0> PLATFORM => "java" irb(main):003:0>
22. Rake and RubyGems • After RubyConfs 2004 and 2005, tools started to solidify • Rake for build tooling • RubyGems for packaging • Sockets, zlib, YAML fi • File/IO APIs, Dir, lesystem access
23. Spring 2006...
24. It All Comes Together • • JRuby on Rails works! • But it's SUPER SLOW! 😂 • Still in development mode! 🤦 JavaOne 2006 • • 1800 people in the room! Sun Microsystems hires us • Full-time JRuby!
26. JRuby Today
27. @headius headius.com JRuby = Opportunity • • Ruby is amazing, but has many challenges • C extensions break optimization • Real world needs parallelism, big data • Ruby ecosystem missing functionality JRuby is enterprise-ready! • JVM is trusted, optimized, parallel, scalable
29. CRuby JRuby High Concurrency CLI Tools Low Memory Data Pipelines Legacy C Exts Web Apps Forking API Servers POSIX Ecosystem Big Data JVM Ecosystem Packaged Deployment GUIs and Games
30. JRuby Worldwide • Thousands of applications running JRuby • Production users since 2007 • Most users could not use Ruby without JRuby • JRuby opens up new opportunities for Ruby • I'll highlight some users along the way
31. @headius headius.com User: Quantum Inventions • Software for intelligent transport systems, focusing on smart city mobility and eet mobility as a service. • Rails-based GUI apps deployed onpremises with Oracle DB and WebLogic • User-acceptance testing using packaged JRuby apps within customer environments • Ruby-based data processing packaged as JRuby apps run by customers https://quantuminventions.com/ fl • "We use [JRuby] for running Ruby in environments without Ruby!"
32. @headius headius.com JVM Language • JRuby is a JVM language • Libraries, ecosystem, extensions written in JVM languages • JVM can optimize all of this code • No C extensions to prevent optimization • True concurrency everywhere
33. opt_new for JRuby
34. Class#new Optimization • CRuby 4 added opt_new to optimize Class#new • • Using C code to allocate + initialize is hard to optimize JRuby implemented in 2016 • Normal Class#new: allocate and call #initialize directly • Everything can inline and optimize
35. Object.new allocation rates 1.46x 1.5 1.125 1x 0.75 0.375 0 CRuby 3.4 CRuby 4.0 JRuby
36. Object.new allocation rates 30 23.6x 22.5 15 7.5 0 1x 1.46x CRuby 3.4 CRuby 4.0 JRuby
37. JIT Compilers • • HotSpot JIT introduced in 1999 • 27 years of improvements • Thousands of development years of work JRuby compiles to JVM bytecode • JVM JIT compiles that to native code • First native JIT for Ruby!
38. Garbage Collectors • • JVM has the best GCs! • Scale up to enormous heap sizes • High throughput, low pause times GC won't be your problem • Cleans up faster than you allocate
39. @headius headius.com Concurrency • • Ractors bring concurrency to Ruby • Many VM challenges • Dif cult* to use (immutability, C extensions) JRuby Threads are already 100% parallel • Low overhead for threads and shared memory • Easy* to use! fi * Concurrency is never easy.
40. @headius headius.com fi User: OpenVox Server • Unpredictable performance of con g compiler on CRuby • Scaling did not linearly improve throughput, diminishing returns with more nodes • JRuby brought near-linear scaling with the same code • Codebase combines Ruby and Clojure • https://github.com/OpenVoxProject "Using JRuby to put the compiler into the JVM alleviated that [scaling] problem"
41. @headius headius.com puts "single-threaded version" 10.times do t = Time.now 1000.times.map do JSON.parse File.read("1MB.json") end puts Time.now - t end
42. @headius headius.com puts "Thread version" 10.times do t = Time.now 100.times.map do Thread.new do 10.times.map do JSON.parse File.read("1MB.json") end end end.each(&:join) puts Time.now - t end
43. @headius headius.com puts "Ractor version" 10.times do t = Time.now 100.times.map do Ractor.new do 10.times.map do JSON.parse File.read("1MB.json") end end end.each(&:join) puts Time.now - t end
44. @headius headius.com Threads vs Ractors (~6 cores) CRuby threads CRuby ractors 6x 6 4.5 3.7x 3 1.5 0 JRuby threads 1.6x
45. @headius headius.com Threads vs Ractors (20 cores) CRuby threads CRuby ractors 20x 20 15 10 5 1.7x 0 JRuby threads 3.8x
46. @headius headius.com User Focus: Poshmark • Leading social commerce marketplace with millions of users • High concurrency and heavy transaction volumes daily • Leveraging JVM ecosystem for resilience, growth, and scaling • Migrating to MongoDB Java driver reduced p50 by 30% and p95 by 35% • https://poshmark.com/ "JRuby has been integral to Poshmark's stack for over a decade and remains a cornerstone of scalability and innovation"
47. @headius headius.com Libraries • • JVM ecosystem has hundreds of thousands of libraries • Graphics, GUIs, servers, document formats, ML/AI/LLM • One of the largest collections in the dev world All available to Ruby! • Easy integration into Ruby apps and code • "Magic" to make Java APIs look like Ruby
48. Example: Charts with JFreeChart
49. @headius headius.com Dependencies Jar le • Jar le for JVM deps • jar 'org.jfree:jfreechart:1.5.5' jar 'org.jfree:org.jfree.chart3d:2.1.0' Like Bundler's Gem le $ lock_jars • lock_jars command • Like `bundle install` -- jar root dependencies -org.jfree:jfreechart:1.5.5:compile org.jfree:org.jfree.chart3d:2.1.0:compile org.jfree:org.jfree.svg:5.0.6:compile org.jfree:org.jfree.pdf:2.0:compile fi fi fi Jars.lock updated
50. Import Classes java_import org.jfree.chart3d.Chart3DFactory java_import org.jfree.chart3d.data.DefaultKeyedValues java_import org.jfree.chart3d.legend.LegendAnchor java_import org.jfree.chart3d.marker.CategoryMarker ...
51. Load Data into Dataset require 'json' dataset = StandardCategoryDataset3D.new data = JSON.load(File.read("revenue.json")) data.each do |name, subset| values = DefaultKeyedValues.new subset.each { values.put(_1, _2) } dataset.add_series_as_row name, values end
52. Create a 3D Bar Chart chart = Chart3DFactory.create_bar_chart( "Quarterly Revenues", "Application Monitoring Companies", dataset, nil, "Quarter", "$million Revenues") chart.chart_box_color = Color.new(255, 255, 255, 127) chart.legend_anchor = LegendAnchor::BOTTOM_RIGHT chart.plot.gridline_paint_for_values = Color::BLACK
53. @headius headius.com Generate PNG width, height = 600, 500 image = BufferedImage.new(width, height, BufferedImage::TYPE_INT_RGB) graphics = image.create_graphics chart.draw(graphics, Rectangle.new(width, height)) png_file = File.open("category_chart.png", "w") ImageIO.write(image, "PNG", png_file.to_outputstream)
54. @headius headius.com
55. @headius headius.com User Example: eazyBI World-class reporting and business intelligence, integrated into the Atlassian stack • JRuby user since 2011 • Integration with Java libraries like Mondrian OLAP, Apache POI, Batik • Deployment as a Jira plugin .jar le • Real multithreading without global lock for heavy parallel worker pools and data import jobs • Deploy on Windows more easily than CRuby • https://eazybi.com/ fi •
56. Performance
57. Concurrency
58. Libraries
59. • Open-world exploration, construction, and combat game available on Steam • DragonRuby used for front-end graphics • World management handled by JRuby backend •
61. Part 5: The Future of JRuby
62. Experimental Ruby Features
63. Ractor? • • Ractor = Thread? • ractor_shim gem just uses Thread internally • json example is same code for Ractor and Thread Don't really need Ractor on JRuby • Everything can run in parallel already • Immutable state remains just as safe
64. Box? • • JRuby has always supported isolated runtimes • org.jruby.Ruby.new_instance => new JRuby runtime • Deploy fully isolated apps in one process Experiment: Box on top of org.jruby.Ruby • Fully parallel as usual • Fully isolated is better than monkey-patching
65. funky.rb module Funky def self.call "hello from Funky" end end
66. example.rb require './box' box = JRuby::Box.new box.require("./funky.rb") obj = box::Funky.call puts obj puts Funky rescue puts $! $ jruby example.rb hello from Funky uninitialized constant Funky
67. box.rb module JRuby class Box def self.new Class.new(Box) do @ruby = org.jruby.Ruby.new_instance end end def self.const_missing(const) @ruby.object.const_get(const) end def self.require(feature) @ruby.load_service.require(feature) end end end
68. Stable and Edge • JRuby 10.0.x is the "stable" line • Ruby 3.4 compatible • Mostly bug and security xes JRuby 10.1.x is the "edge" line • Ruby 4.0 compatible (released only 3 months after CRuby 4.0!) • Active performance experimentation • More invasive refactoring and improvements fi •
69. Optimizations in 10.1
70. Shrinking Objects • • Reduced object size in JRuby 10.1 • Smaller objects across the system • Better throughput, less GC overhead Much smaller JRuby processes!
71. fl JRuby 10.0 • 32-byte objects • JVM object header: 12 bytes • Flags: 4 bytes (shared by all objects) • Class pointer: 4 bytes • Instance variable array: 4 bytes padding lock ivars ags 32 bytes 12 4 4 4 • Instance variable table lock: 4 bytes • Padding for object alignment: 4 bytes 4 4 class header
72. fl JRuby 10.1 • padding lock ivars ags 24-byte objects • JVM object header: 12 bytes • Flags: 1 byte (other objects have their own) • Class pointer: 4 bytes 12 • Instance variable array: 4 bytes 1 1 • Instance variable table lock: 1 byte 4 Padding for object alignment: 2 bytes 4 2 • 24 bytes class header
73. Object.new memory size 40 bytes 32 bytes 24 bytes CRuby 4 JRuby 10 JRuby 10.1
74. JVM Improvements • • Compressed Pointers (Compressed OOPs, 2006) • Use 32 bits instead of 64 bits for object references • GC manages indirection Compact Object Headers (Project Lilliput, 2023) • Shrink the size of all JVM objects
75. fl Project Lilliput • padding lock ivars ags class header Object headers 12 bytes to 4 bytes 24 bytes • • • • Reduced class pointer size Lazy identity hash code (object ID) Expand other data as needed 16-byte objects! 12 16 bytes 1 1 4 1 1 4 4 4 2 4 2 Normal Lilliput
76. Object.new memory size 40 bytes 32 bytes 24 bytes 16 bytes CRuby 4 JRuby 10 JRuby 10.1 JRuby 10.1 (L)
77. Array.new memory size 48 bytes 40 bytes CRuby 4 40 bytes JRuby 10 JRuby 10.1
78. Array.new(1000) memory size 8,040 bytes CRuby 4 4,064 bytes 4,056 bytes JRuby 10 JRuby 10.1
79. Fixnum Splitting • JRuby always boxes integers and oats • JRuby 10.0 integers: 40 bytes • JRuby 10.1 integers: 32 or 24 bytes • • short version goes down to 24 bytes But this is an object so there's also a 4-byte pointer reference fl • 32 byte object + 8 byte long
80. 28 bytes 28 bytes y ub JR y ub JR y ub JR y ub JR y ub JR by Ru C .1 10 .1 10 .1 10 .1 10 .0 10 4 L) ) (L t( or sh ng lo t or sh ng lo fi Pointer + xnum size 44 bytes 36 bytes 20 bytes 8 bytes
81. 32k Fixnum Array Size 1,431,616 bytes 911,416 bytes 649,272 bytes 262,184 bytes y ub JR y ub JR y ub JR by Ru C .1 10 .1 10 .0 10 4 L) t( or sh t or sh
82. while i < 100_000_000 7 6.34x 5.25 3.08x 3.5 1.75 0 1x 1.06x CRuby 3.4 CRuby 4.0 JRuby 10.0 Times faster JRuby 10.1
83. Marshal Load Performance medium_data = { :id => 1, :name => "test" * 100, :values => (1..1000).to_a, :nested => { :a => 1, :b => 2, :c => (1..100).to_a } } # Load only benchmark dumped = Marshal.dump(medium_data) loop { t = Time.now 100000.times { Marshal.load(dumped) } puts Time.now - t }
84. Marshal.load 100k medium data 6 5.23s 4.5 3 2.03s 1.83s 1.5 0 CRuby 4.0 JRuby 10.0 JRuby 10.1
85. Beyond JRuby 10.1
86. ! N O SO C O M IN G Advanced Shaping • Guessed at rst allocate • Never change after that 10.1+: Allow object shape to evolve • Expand as more instance variables discovered • Shape selection at allocation site fi • 10.0: Object shapes in JRuby are xed fi •
87. ! N O SO C O M IN G Specialized Strings • Codepoint (immutable) • • Compact • • Use 2-byte padding for UTF-16 char Embed small string in 8-byte long Java String wrapper (immutable) • Reduce cost of calling JVM APIs
88. ! N O SO G IN O M C Specialized Array • Exact size using Java elds (no Object[]) • Direct numeric values (packed integers or doubles) • Mixed (both numerics and objects) • Any JVM List type Including thread-safe, trees, native memory fi •
89. ! N O SO C O M IN G Specialized Hash • Open addressing • Direct key/value elds (2*N elds instead of buckets or Object[]) • Mutexed fallback where needed • Any JVM Map fi Concurrent, trees, database result set, ... fi •
90. ! N O SO G IN O M C Direct Argument Passing • All arguments passed directly on stack • Even to core class methods • Optional values, keyword layout in call site metadata • Zero-allocation for all forms up to N wide • JVM inlining optimizes register use
91. Foo.new(1) vs Foo.new(a:1) allocation rates 28.2x 30 23.8x 22.5 15 WE CAN FIX THIS 7.5 1x 5.37x 2.8x 1x 2.65x 0 Foo.new(1) CRuby 3.4 Foo.new(a:1) CRuby 4.0 JRuby JRuby (L) 3.29x
92. ! N O SO G IN O M C Value Type Numerics • • Boxed numeric objects are slow and heavy • All Integer and Float are objects in JRuby • JVM is good, but not that good yet Project Valhalla brings true value types (stack-allocated) • JRuby will move numerics to value types • Zero-alloc maths and compact collections for free!
93. ! N O SO G IN O M C Fast Native Access • • FFI is the future of native integration • Avoid C extensions and problems they cause • Portable to all Ruby implementations Project Panama brings fast Foreign Functions and Memory • Optimized native invocation and memory management • JIT, GC understand and optimize native access
95. What's Next?
96. JRuby in 2026 • JRuby remains 100% Open Source Software • One self-funded developer, many community members • We need your support! • Try JRuby and help us improve • Present JRuby to your companies • Sponsor a JRuby developer to keep working
97. @headius headius.com Commercial Support • Expert support for your JRuby apps • New dev, migrations, upgrades • Bug prioritization, security SLAs • Help with your code under NDA • headius.com/welcome • Site by
98. Closing
99. I have spent 20 years on JRuby
100. Because I love Ruby
101. I want to help Ruby grow
102. Let's work together
103. To make Ruby and JRuby great!
104. @headius headius.com Thank You! • • Please get in touch! • @headius on socials • headius@headius.com • JRuby Matrix chat JRuby support • headius.com/welcome