No Callbacks, No Threads - RailsConf 2010

No Callbacks, No Threads & Ruby 1.9async & co-operative web-serversIlya Grigorik@igrigorik

The state of art is not good enough.(we’ve been stuck in the same local minima for several years)

require"active_record”ActiveRecord::Base.establish_connection( :adapter => "mysql", :username => "root", :database => "database", :pool => 5)threads = []10.times do |n| threads <<Thread.new {ActiveRecord::Base.connection_pool.with_connectiondo |conn| res =conn.execute("select sleep(1)")end }endthreads.each { |t| t.join }The “Experiment”vanilla everything…

require"active_record”ActiveRecord::Base.establish_connection( :adapter => "mysql", :username => "root", :database => "database", :pool => 5)threads = []10.times do |n| threads <<Thread.new {ActiveRecord::Base.connection_pool.with_connectiondo |conn| res =conn.execute("select sleep(1)")end }endthreads.each { |t| t.join }5 shared connections# time ruby activerecord-pool.rb## real 0m10.663s# user 0m0.405s# sys 0m0.201s

MongoCouchMySQLPSQL…DriversThreadsRuby VMGILFibers…NetworkMongrelUnicornPassenger…

2MongoCouchMySQLPSQL…DriversThreadsRuby VM14GILFibers…NetworkMongrelUnicorn3We’re as fast as the slowest componentPassenger…

Global Interpreter Lock is a mutual exclusion lock held by a programming language interpreter thread to avoid sharing code that is not thread-safe with other threads. There is always one GIL for one interpreter process.Concurrency is a myth in Ruby(with a few caveats, of course)http://bit.ly/ruby-gil

N-M thread pool in Ruby 1.9…Better but still the same problem!Concurrency is a myth in Rubystill no concurrency in Ruby 1.9http://bit.ly/ruby-gil

Nick – tomorrow @ 11:45amConcurrency is a myth in Rubystill no concurrency in Ruby 1.9http://bit.ly/ruby-gil

Blocks entireRuby VMNot as bad, butavoid it still..Avoid locking interpreter threads at all costslet’s say you’re writing an extension…

Will fetch_xyz() block the VM?when was the last time you asked yourself this question?

require 'rubygems’require 'sequel'DB = Sequel.connect('mysql://root@localhost/test')while trueDB['select sleep(1)'].select.firstendBlocking 1s call!ltrace –ttTg -xmysql_real_query –p ./example.rbmysql.gem under the hood22:10:00.218438 mysql_real_query(0x02740000, "select sleep(1)", 15) = 0 <1.001100>22:10:01.241679 mysql_real_query(0x02740000, "select sleep(1)", 15) = 0 <1.000812>http://bit.ly/c3Pt3f

Blocking calls to mysql_real_querymysql_real_query requires an OS threadBlocking on mysql_real_query blocks the Ruby VMAka, “select sleep(1)” blocks the entire Ruby runtime for 1s(ouch)gem install mysqlwhat you didn’t know…

static VALUE async_query(intargc, VALUE* argv, VALUE obj) { ...send_query( obj, sql ); ...schedule_query( obj, timeout); ...returnget_result(obj); }staticvoidschedule_query(VALUEobj, VALUE timeout) { ...structtimevaltv = { tv_sec: timeout, tv_usec: 0 };for(;;){FD_ZERO(&read);FD_SET(m->net.fd, &read); ret = rb_thread_select(m->net.fd + 1, &read, NULL, NULL, &tv); ...if (m->status == MYSQL_STATUS_READY)break; }}send query and blockRuby: select() = C: rb_thread_select()mysqlplus.gem under the hood

spinning in selectmysqlplus.gem + ruby select

Step 1: Fix the driversMany of our DB drivers don’t respect the underlying Ruby VM. Don’t blame the VM.

DriversRuby VMNetworkWEBRickMongrel made Rails viablestill powers a lot of Rails apps today

Rails rediscovers Apachethe worker/forker model…

*nix IPC is fast! Woo!…Full Ruby VMAn exclusive Ruby VM for EACH requestam I the only one who thinks this is terrible?

“Does not care if your application is thread-safe or not, workers all run within their own isolated address space and only serve one client at a time for maximum robustness.”Robustness? That sounds like a bug.An exclusive Ruby VM for EACH requestam I the only one who thinks this is terrible?

Step 2: consider entire stackThe driver, the web-server, and the networkmust all work together.

Node imposes the full-stack requirementsNode imposes async driversNode imposes async frameworksSurprise: Node is “fast”

We can ignore the performanceissues at our own perilor, we can just fix the problem

>I’ll take Ruby over JSgem install eventmachine

p "Starting"EM.rundop"Running in EM reactor"endp”won’t get here"whiletruedo timersnetwork_ioother_ioendEventMachine Reactorconcurrency without threadEventMachine: The Speed DemonWednesday @ 11:45am – Aman Gupta

Non-blocking IO requires non-blocking drivers:AMQP http://github.com/tmm1/amqpMySQLPlushttp://github.com/igrigorik/em-mysqlplusMemcachedhttp://github.com/astro/remcachedDNS http://github.com/astro/em-dnsRedishttp://github.com/madsimian/em-redisMongoDBhttp://github.com/tmm1/rmongoHTTPRequesthttp://github.com/igrigorik/em-http-requestWebSockethttp://github.com/igrigorik/em-websocketAmazon S3 http://github.com/peritor/happeningAnd many others: http://wiki.github.com/eventmachine/eventmachine/protocol-implementations

gem install em-mysqlplusEventMachine.rundoconn=EventMachine::MySQL.new(:host => 'localhost') query =conn.query("select sleep(1)")query.callback { |res| pres.all_hashes }query.errback { |res| pres.all_hashes } puts ”executing…”end# > ruby em-mysql-test.rb## executing…# [{"sleep(1)"=>"0"}]callback fired 1s after “executing”em-mysqlplus: exampleasyncMySQL driver

non-blocking driverrequire'mysqlplus'defconnect(opts)conn=connect_socket(opts)EM.watch(conn.socket, EventMachine::MySQLConnection, conn, opts, self)enddefconnect_socket(opts)conn=Mysql.initconn.real_connect(host, user, pass, db, port, socket, ...)conn.reconnect=falseconnendEM.watch: reactor will poll & notifyem-mysqlplus: under the hoodmysqlplus + reactor loop

Features: Maintains C-based mysql gem API

Deferrables for every query with callback & errback

Connection query queue - pile 'em up!

Auto-retry on deadlockshttp://github.com/igrigorik/em-mysqlplusem-mysqlplusmysqlplus + reactor loop

We can do better than node.jsall the benefits of evented code without the drawbacks

Ruby 1.9 Fibers are a means of creating code blocks which can be paused and resumed by our application (think lightweight threads, minus the thread scheduler and less overhead). f=Fiber.new {whiletruedoFiber.yield"Hi”end}pf.resume# => Hipf.resume# => Hipf.resume# => HiManual / cooperative scheduling!Ruby 1.9 Fibersand cooperative schedulinghttp://bit.ly/d2hYw0

Fibers vs Threads: creation time much lowerFibers vs Threads: memory usage is much lowerRuby 1.9 Fibersand cooperative schedulinghttp://bit.ly/aesXy5

defquery(sql)f = Fiber.currentconn=EventMachine::MySQL.new(:host => 'localhost')q = conn.query(sql)c.callback { f.resume(conn) }c.errback { f.resume(conn) }return Fiber.yieldendEventMachine.rundoFiber.new { res =query('select sleep(1)') puts "Results: #{res.fetch_row.first}"}.resumeendException, async!Untangling Evented Code with Fibershttp://bit.ly/d2hYw0

defquery(sql)f = Fiber.currentconn=EventMachine::MySQL.new(:host => 'localhost')q = conn.query(sql)c.callback { f.resume(conn) }c.errback { f.resume(conn) } return Fiber.yieldendEventMachine.rundoFiber.new{ res =query('select sleep(1)') puts "Results: #{res.fetch_row.first}" }.resumeend1. Wrap into a continuationUntangling Evented Code with Fibershttp://bit.ly/d2hYw0

defquery(sql)f=Fiber.currentconn=EventMachine::MySQL.new(:host => 'localhost')q = conn.query(sql)c.callback { f.resume(conn) }c.errback { f.resume(conn) }returnFiber.yieldendEventMachine.rundoFiber.new{ res =query('select sleep(1)') puts "Results: #{res.fetch_row.first}" }.resumeend2. Pause the continuationUntangling Evented Code with Fibershttp://bit.ly/d2hYw0

defquery(sql)f=Fiber.currentconn=EventMachine::MySQL.new(:host => 'localhost')q = conn.query(sql)c.callback { f.resume(conn) }c.errback { f.resume(conn) }returnFiber.yieldendEventMachine.rundoFiber.new{ res =query('select sleep(1)') puts "Results: #{res.fetch_row.first}" }.resumeend3. Resume the continuationUntangling Evented Code with Fibershttp://bit.ly/d2hYw0

Good news, you don’t even have to muck around with Fibers!gem install em-synchronyhttp://github.com/igrigorik/em-synchrony Fiber aware connection pool with sync/async query support

Multi request interface which accepts any callback enabled client

Fibered iterator to allow concurrency control & mixing of sync / async

em-http-request: .get, etc are synchronous, while .aget, etc are async

em-mysqlplus: .query is synchronous, while .aquery is async

No Callbacks, No Threads - RailsConf 2010

More Related Content

What's hot

Viewers also liked

Similar to No Callbacks, No Threads - RailsConf 2010

More from Ilya Grigorik

Recently uploaded

No Callbacks, No Threads - RailsConf 2010

Editor's Notes