mongo 1.3.1 → 1.4.0

data/test/replica_sets/refresh_test.rb

@@ -0,0 +1,123 @@
+$:.unshift(File.join(File.dirname(__FILE__), '..', 'lib'))
+require './test/replica_sets/rs_test_helper'
+require 'benchmark'
+
+# on ports TEST_PORT, RS.ports[1], and TEST + 2.
+class ReplicaSetRefreshTest < Test::Unit::TestCase
+  include Mongo
+
+  def setup
+    @conn = nil
+  end
+
+  def teardown
+    RS.restart_killed_nodes
+    @conn.close if @conn
+  end
+
+  def test_connect_speed
+    Benchmark.bm do |x|
+      x.report("Connect") do
+        10.times do
+          ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+            [RS.host, RS.ports[2]], :background_refresh => false)
+        end
+      end
+
+          @con = ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+            [RS.host, RS.ports[2]], :background_refresh => false)
+
+      x.report("manager") do
+        man = Mongo::PoolManager.new(@con, @con.seeds)
+        10.times do
+          man.connect
+        end
+      end
+    end
+  end
+
+  def test_connect_and_manual_refresh_with_secondaries_down
+    RS.kill_all_secondaries
+
+    rescue_connection_failure do
+      @conn = ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+        [RS.host, RS.ports[2]], :background_refresh => false)
+    end
+
+    assert_equal [], @conn.secondaries
+    assert @conn.connected?
+    assert_equal @conn.read_pool, @conn.primary_pool
+
+    # Refresh with no change to set
+    @conn.refresh
+    assert_equal [], @conn.secondaries
+    assert @conn.connected?
+    assert_equal @conn.read_pool, @conn.primary_pool
+
+    RS.restart_killed_nodes
+    assert_equal [], @conn.secondaries
+    assert @conn.connected?
+    assert_equal @conn.read_pool, @conn.primary_pool
+
+    # Refresh with everything up
+    @conn.refresh
+    assert @conn.read_pool
+    assert @conn.secondaries.length > 0
+  end
+
+  def test_automated_refresh_with_secondaries_down
+    RS.kill_all_secondaries
+
+    rescue_connection_failure do
+      @conn = ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+        [RS.host, RS.ports[2]], :refresh_interval => 2, :background_refresh => true)
+    end
+
+    assert_equal [], @conn.secondaries
+    assert @conn.connected?
+    assert_equal @conn.read_pool, @conn.primary_pool
+
+    RS.restart_killed_nodes
+
+    sleep(3)
+
+    assert @conn.read_pool != @conn.primary_pool, "Read pool and primary pool are identical."
+    assert @conn.secondaries.length > 0, "No secondaries have been added."
+  end
+
+  def test_automated_refresh_with_removed_node
+    @conn = ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+      [RS.host, RS.ports[2]], :refresh_interval => 2, :background_refresh => true)
+
+    assert_equal 2, @conn.secondary_pools.length
+    assert_equal 2, @conn.secondaries.length
+
+    n = RS.remove_secondary_node
+    sleep(4)
+
+    assert_equal 1, @conn.secondaries.length
+    assert_equal 1, @conn.secondary_pools.length
+
+    RS.add_node(n)
+  end
+
+  def test_adding_and_removing_nodes
+    @conn = ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+      [RS.host, RS.ports[2]], :refresh_interval => 2, :background_refresh => true)
+
+    RS.add_node
+    sleep(5)
+
+    @conn2 = ReplSetConnection.new([RS.host, RS.ports[0]], [RS.host, RS.ports[1]],
+      [RS.host, RS.ports[2]], :refresh_interval => 2, :background_refresh => true)
+
+    assert @conn2.secondaries == @conn.secondaries
+    assert_equal 3, @conn.secondary_pools.length
+    assert_equal 3, @conn.secondaries.length
+
+    RS.remove_secondary_node
+    sleep(4)
+    assert_equal 2, @conn.secondary_pools.length
+    assert_equal 2, @conn.secondaries.length
+  end
+end

data/test/replica_sets/replication_ack_test.rb

@@ -20,6 +20,11 @@ class ReplicaSetAckTest < Test::Unit::TestCase
     @col = @db.collection("test-sets")
   end
 
+  def teardown
+    RS.restart_killed_nodes
+    @conn.close if @conn
+  end
+
   def test_safe_mode_with_w_failure
     assert_raise_error OperationFailure, "timeout" do
       @col.insert({:foo => 1}, :safe => {:w => 4, :wtimeout => 1, :fsync => true})
@@ -33,15 +38,15 @@ class ReplicaSetAckTest < Test::Unit::TestCase
   end
 
   def test_safe_mode_replication_ack
-    @col.insert({:baz => "bar"}, :safe => {:w => 2, :wtimeout => 5000})
+    @col.insert({:baz => "bar"}, :safe => {:w => 3, :wtimeout => 5000})
 
-    assert @col.insert({:foo => "0" * 5000}, :safe => {:w => 2, :wtimeout => 5000})
+    assert @col.insert({:foo => "0" * 5000}, :safe => {:w => 3, :wtimeout => 5000})
     assert_equal 2, @slave1[MONGO_TEST_DB]["test-sets"].count
 
-    assert @col.update({:baz => "bar"}, {:baz => "foo"}, :safe => {:w => 2, :wtimeout => 5000})
+    assert @col.update({:baz => "bar"}, {:baz => "foo"}, :safe => {:w => 3, :wtimeout => 5000})
     assert @slave1[MONGO_TEST_DB]["test-sets"].find_one({:baz => "foo"})
 
-    assert @col.remove({}, :safe => {:w => 2, :wtimeout => 5000})
+    assert @col.remove({}, :safe => {:w => 3, :wtimeout => 5000})
     assert_equal 0, @slave1[MONGO_TEST_DB]["test-sets"].count
   end
 

data/test/replica_sets/rs_test_helper.rb

@@ -11,7 +11,7 @@ class Test::Unit::TestCase
 
   # Generic code for rescuing connection failures and retrying operations.
   # This could be combined with some timeout functionality.
-  def rescue_connection_failure(max_retries=60)
+  def rescue_connection_failure(max_retries=30)
     retries = 0
     begin
       yield
@@ -19,7 +19,7 @@ class Test::Unit::TestCase
       puts "Rescue attempt #{retries}: from #{ex}"
       retries += 1
       raise ex if retries > max_retries
-      sleep(1)
+      sleep(2)
       retry
     end
   end

data/test/timeout_test.rb

@@ -0,0 +1,14 @@
+require './test/test_helper'
+
+class TestTimeout < Test::Unit::TestCase
+
+  def test_timeout
+    @conn = standard_connection(:op_timeout => 2)
+    assert @conn[MONGO_TEST_DB]['test'].save({:a => 1})
+    assert @conn[MONGO_TEST_DB]['test'].find.next
+    assert_raise OperationTimeout do
+      @conn[MONGO_TEST_DB]['test'].find({'$where' => 'function() { while(true) { this.a == 1 } }'}).next
+    end
+  end
+
+end

data/test/tools/repl_set_manager.rb

@@ -10,17 +10,21 @@ end
 
 class ReplSetManager
 
-  attr_accessor :host, :start_port, :ports, :name, :mongods
+  attr_accessor :host, :start_port, :ports, :name, :mongods, :tags, :version
 
   def initialize(opts={})
     @start_port = opts[:start_port] || 30000
     @ports      = []
     @name       = opts[:name] || 'replica-set-foo'
     @host       = opts[:host]  || 'localhost'
-    @retries    = opts[:retries] || 60
+    @retries    = opts[:retries] || 30
     @config     = {"_id" => @name, "members" => []}
     @durable    = opts.fetch(:durable, false)
     @path       = File.join(File.expand_path(File.dirname(__FILE__)), "data")
+    @oplog_size = opts.fetch(:oplog_size, 32)
+    @tags = [{"dc" => "ny", "rack" => "a", "db" => "main"},
+             {"dc" => "ny", "rack" => "b", "db" => "main"},
+             {"dc" => "sf", "rack" => "a", "db" => "main"}]
 
     @arbiter_count   = opts[:arbiter_count]   || 2
     @secondary_count = opts[:secondary_count] || 2
@@ -33,35 +37,59 @@ class ReplSetManager
     end
 
     @mongods   = {}
+    version_string = `mongod --version`
+    version_string =~ /(\d\.\d\.\d)/
+    @version = $1.split(".").map {|d| d.to_i }
   end
 
   def start_set
-    puts "** Starting a replica set with #{@count} nodes"
+    begin
+    con = Mongo::Connection.new(@host, @start_port)
+      rescue Mongo::ConnectionFailure
+    end
 
-    system("killall mongod")
+    if con && ensure_up(1, con)
+      should_start = false
+      puts "** Replica set already started."
+    else
+      should_start = true
+      system("killall mongod")
+      puts "** Starting a replica set with #{@count} nodes"
+    end
 
     n = 0
     (@primary_count + @secondary_count).times do
-      init_node(n)
+      init_node(n, should_start) do |attrs|
+        if @version[0] >= 2
+          attrs['tags'] = @tags[n % @tags.size]
+        end
+      end
       n += 1
     end
 
     @passive_count.times do
-      init_node(n) do |attrs|
+      init_node(n, should_start) do |attrs|
         attrs['priority'] = 0
       end
       n += 1
     end
 
     @arbiter_count.times do
-      init_node(n) do |attrs|
+      init_node(n, should_start) do |attrs|
         attrs['arbiterOnly'] = true
       end
       n += 1
     end
 
-    initiate
-    ensure_up
+    if con && ensure_up(1, con)
+      @mongods.each do |k, v|
+        v['up'] = true
+        v['pid'] = File.open(File.join(v['db_path'], 'mongod.lock')).read.strip
+      end
+    else
+      initiate
+      ensure_up
+    end
   end
 
   def cleanup_set
@@ -71,18 +99,20 @@ class ReplSetManager
     end
   end
 
-  def init_node(n)
+  def init_node(n, should_start=true)
     @mongods[n] ||= {}
     port = @start_port + n
     @ports << port
     @mongods[n]['port'] = port
     @mongods[n]['db_path'] = get_path("rs-#{port}")
     @mongods[n]['log_path'] = get_path("log-#{port}")
-    system("rm -rf #{@mongods[n]['db_path']}")
-    system("mkdir -p #{@mongods[n]['db_path']}")
-
     @mongods[n]['start'] = start_cmd(n)
-    start(n)
+
+    if should_start
+      system("rm -rf #{@mongods[n]['db_path']}")
+      system("mkdir -p #{@mongods[n]['db_path']}")
+      start(n)
+    end
 
     member = {'_id' => n, 'host' => "#{@host}:#{@mongods[n]['port']}"}
 
@@ -96,13 +126,64 @@ class ReplSetManager
     @config['members'] << member
   end
 
+  def journal_switch
+    if @version[0] >= 2
+      if @durable
+        "--journal"
+      else
+        "--nojournal"
+      end
+    elsif @durable
+      "--journal"
+    end
+  end
+
   def start_cmd(n)
     @mongods[n]['start'] = "mongod --replSet #{@name} --logpath '#{@mongods[n]['log_path']}' " +
-     " --dbpath #{@mongods[n]['db_path']} --port #{@mongods[n]['port']} --fork"
+     "--oplogSize #{@oplog_size} #{journal_switch} --dbpath #{@mongods[n]['db_path']} --port #{@mongods[n]['port']} --fork"
     @mongods[n]['start'] += " --dur" if @durable
     @mongods[n]['start']
   end
 
+  def remove_secondary_node
+    primary = get_node_with_state(1)
+    con = get_connection(primary)
+    config = con['local']['system.replset'].find_one
+    secondary = get_node_with_state(2)
+    host_port = "#{@host}:#{@mongods[secondary]['port']}"
+    kill(secondary)
+    @mongods.delete(secondary)
+    @config['members'].reject! {|m| m['host'] == host_port}
+    @config['version'] = config['version'] + 1
+
+    begin
+    con['admin'].command({'replSetReconfig' => @config})
+    rescue Mongo::ConnectionFailure
+    end
+
+    con.close
+
+    return secondary
+  end
+
+  def add_node(n=nil)
+    primary = get_node_with_state(1)
+    con = get_connection(primary)
+    init_node(n || @mongods.length)
+
+    config = con['local']['system.replset'].find_one
+    @config['version'] = config['version'] + 1
+
+    # We expect a connection failure on reconfigure here.
+    begin
+      con['admin'].command({'replSetReconfig' => @config})
+    rescue Mongo::ConnectionFailure
+    end
+
+    con.close
+    ensure_up
+  end
+
   def kill(node, signal=2)
     pid = @mongods[node]['pid']
     puts "** Killing node with pid #{pid} at port #{@mongods[node]['port']}"
@@ -127,6 +208,7 @@ class ReplSetManager
       con['admin'].command({'replSetStepDown' => 90})
     rescue Mongo::ConnectionFailure
     end
+    con.close
   end
 
   def kill_secondary
@@ -135,6 +217,15 @@ class ReplSetManager
     return node
   end
 
+  def kill_all_secondaries
+    nodes = get_all_nodes_with_state(2)
+    if nodes
+      nodes.each do |n|
+        kill(n)
+      end
+    end
+  end
+
   def restart_killed_nodes
     nodes = @mongods.keys.select do |key|
       @mongods[key]['up'] == false
@@ -159,21 +250,26 @@ class ReplSetManager
   end
   alias :restart :start
 
-  def ensure_up
+  def ensure_up(n=nil, connection=nil)
     print "** Ensuring members are up..."
 
-    attempt do
-      con = get_connection
+    attempt(n) do
+      con = connection || get_connection
       status = con['admin'].command({'replSetGetStatus' => 1})
       print "."
-      if status['members'].all? { |m| m['health'] == 1 && [1, 2, 7].include?(m['state']) } &&
+      if status['members'].all? { |m| m['health'] == 1 &&
+         [1, 2, 7].include?(m['state']) } &&
          status['members'].any? { |m| m['state'] == 1 }
         print "all members up!\n\n"
+        con.close
         return status
       else
+        con.close
         raise Mongo::OperationFailure
       end
     end
+
+    return false
   end
 
   def primary
@@ -207,6 +303,20 @@ class ReplSetManager
     attempt do
       con['admin'].command({'replSetInitiate' => @config})
     end
+
+    con.close
+  end
+
+  def get_all_nodes_with_state(state)
+    status = ensure_up
+    nodes = status['members'].select {|m| m['state'] == state}
+    nodes = nodes.map do |node|
+      host_port = node['name'].split(':')
+      port = host_port[1] ? host_port[1].to_i : 27017
+      @mongods.keys.detect {|key| @mongods[key]['port'] == port}
+    end
+
+    nodes == [] ? false : nodes
   end
 
   def get_node_with_state(state)
@@ -215,7 +325,7 @@ class ReplSetManager
     if node
       host_port = node['name'].split(':')
       port = host_port[1] ? host_port[1].to_i : 27017
-      key = @mongods.keys.detect {|key| @mongods[key]['port'] == port}
+      key = @mongods.keys.detect {|n| @mongods[n]['port'] == port}
       return key
     else
       return false
@@ -247,19 +357,20 @@ class ReplSetManager
     File.join(@path, name)
   end
 
-  def attempt
+  def attempt(retries=nil)
     raise "No block given!" unless block_given?
     count = 0
 
-    while count < @retries do
+    while count < (retries || @retries) do
       begin
         return yield
         rescue Mongo::OperationFailure, Mongo::ConnectionFailure => ex
-          sleep(1)
+          sleep(2)
           count += 1
       end
     end
 
+    puts "NO MORE ATTEMPTS"
     raise ex
   end