Notes from RubyConf 2017 AU talks April 24, 2017 Reinvesting in Ruby by Tim Riley Video
Rails was defining the Ruby ecosystem, time to explore other alternatives Introduces dry-rb gems on a blog post example Minimalistic, single-responsibility, composable gems for common tasks dry-validation , dry-monads , dry-matcher - pattern matchingActors in Ruby! Why let Elixir have all the fun by Marcos Matos Video
Threads vs. forks (forks heavy - even with CoW optimisations). Forking does not have built-in way to communicate - some form of IPC is required, e.g. message queues. GIL is not a problem for IO workloads - it is not held during IO (network calls or file reads). Core issue is sharing mutable state and managing threads Introduction to Celluloidasync programming, actors -> OO + message passing message passing, actors mutate their own, isolated state, process messages from mailbox actors - thread based workers managed worker pool supervision of workers, composable to hierarchies linking -> ‘running/dead’ dependency Dcell (distributed actors, ZeroMQ) Defragging Ruby by Aaron Patterson Video
Ruby heap layout (pages, slots) Memory alignment - objects on 40 byte boundary, pages 16k (407, 408 objects on a page) Certain objects do not get allocated on the heap (pointer tagging - alignment frees lower bits for tagging) Idea: defragment heap, to free pages after they get fragmented after GC - improve locality, performance. Run GC.compact after app startup, before forking new Unicorn workers. ObjectSpace.dump_all + determining page address => heap fragmentation map An algorithm to compact heap (two finger compaction, or Cheneys GC algoithm ) Obstacles (memory address dependent code - e.g. object_id, C extensions). Certain pages still cannot get freed, only compacted. Solution - keep a hash table of un-relocatable objects. Sounds hacky, maybe no other way to keep backwards-compat? Code in a public repo Video
Have a benchmarking suite - benchmark-ips , method_profiler Profile methods, find out the bottleneck - memory, CPU, IO bound workload? Upgrade MRI Ruby - constant perf improvements in newer versions Try multithreading Try alternative implementations JRuby, Rubinius - might be better for certain workloads, benchmarks required Ruby, HTTP2 by Hiro Asari Video
Intro to how H2 works There’s no proper support in Ruby/Rack yet, apart from a few H2 clients. Rack is solely request/response oriented.