roma 0.8.12 → 0.8.13p1

data/CHANGELOG

@@ -1,3 +1,36 @@
+*0.8.13-p1 (Jan 15 2014)*
+
+* add timeout parameter for a vnode copy [junji torii] 0522750
+* remove some debug log [hiroaki-iwase] 94d96dc
+* Add Gemfile. [Hiroki Matsue] 9029bf8
+* Ignore test related files. [Hiroki Matsue] 556a931
+* fix:add rescue in timer_event_1sec [junji torii] 11cc4ea
+
+*0.8.13 (Sep 12 2013)*
+
+* change target ruby version about encoding problem (upper 1.9.2 => upper 1.9.1) [hiroaki-iwase] e56d2f4
+* modify end message of join. [hiroaki-iwase] a649dfb
+* modify typo [hiroaki-iwase] c47802b
+* fix:for a problem of encording on 1.9.2 [junji torii] 3501622
+* modify ongoing setting changing cmd. [hiroaki-iwase] 190db49
+* Add new async process for calculate latency average & add some commands for change setting. [hiroaki-iwase] 49dc9cb
+* add new function "calculate latency average" [hiroaki-iwase] 40b6266
+* Add command of changing CONNECTION_POOL_EXPTIME & DEFAULT_LOST_ACTION & CONNECTION_DESCRIPTOR_TABLE_SIZE & WRITEBEHIND_SHIFT_SIZE [hiroaki-iwase] de757af
+* add vnode balancing logic [junji torii] bfa16cd
+* add *_rttable_sub_nid commands [junji torii] 8bd4a31
+* add auto_recover function [hiroaki-iwase] 473df72
+* modify message when execute "balse" [hiroaki-iwase] a3d61b1
+* add node-id substitution feature in routingdump command. [junji torii] a3693e8
+* fixed recoverlost_lib's value error [hiroaki-iwase] deb9430
+* add log output(info level) when new wb file was created. [hiroaki-iwase] dcec49a
+* add new function "calculate latency average" [hiroaki-iwase] 4158ccd
+* Add DNS caching function. [Hiroki Matsue] abf4464
+* Modify some default values. Change plugin selecting(plugin_storage.rb is set without question). [hiroaki-iwase] d2542f0
+* add command of changing CONNECTION_EMPOOL_EXPTIME. [hiroaki-iwase] 1d3df93
+* improve mkconfig.rb [hiroaki-iwase] 74be88d
+* remove utf-8 manifesto # -*- coding: utf-8 -*-" # encoding: utf-8 [hiroaki-iwase] d54e30c
+* translate comment JP to ENG [hiroaki-iwase] 1b091fd
+
 *0.8.12 (Feb 12 2013)*
 
 * refactor: for protocol [Hiroki Matsue] a6f9152
@@ -171,4 +204,4 @@
 * fixed bug in t_rclient.rb. [junji torii] 68880b5
 * support running test with Ruby not named as 'ruby'. [Kouhei Sutou] 6fd1ec2
 
-*0.8.0 (Oct 23 2009)* 
+*0.8.0 (Oct 23 2009)* 

data/Gemfile

@@ -0,0 +1,17 @@
+source "https://rubygems.org"
+
+gem 'eventmachine'
+
+# Back-end storages
+group :tokyocabinet do
+  gem 'tokyocabinet', :git => 'https://github.com/roma/tokyocabinet-ruby.git'
+end
+
+group :gdbm do
+  gem 'ffi'
+  gem 'gdbm'
+end
+
+group :sqlite3 do
+  gem 'sqlite3'
+end

data/Rakefile

@@ -1,4 +1,3 @@
-# -*- coding: utf-8 -*-
 require 'rubygems'
 require 'rake'
 
@@ -40,8 +39,8 @@ EXEC_TABLE = Dir.entries(base + 'bin').reject{ |d| d =~ /^\.+$/ || d =~ /^sample
 require File.expand_path(File.join('ruby', 'server', 'lib', 'roma', 'version'), File.dirname(__FILE__))
 VER_NUM = Roma::VERSION
 
-if VER_NUM =~ /([0-9.]+)$/
-  CURRENT_VERSION = $1
+if VER_NUM =~ /([0-9.p-]+)$/
+  CURRENT_VERSION = $1.delete("-")
 else
   CURRENT_VERSION = "0.0.0"
 end

data/ruby/server/lib/roma/async_process.rb

@@ -1,9 +1,9 @@
-# -*- coding: utf-8 -*-
 require 'thread'
 require 'digest/sha1'
+require "timeout"
 
 module Roma
-  
+
   class AsyncMessage
     attr_accessor :event
     attr_accessor :args
@@ -34,16 +34,26 @@ module Roma
   module AsyncProcess
 
     @@async_queue = Queue.new
+    @@async_queue_latency = Queue.new
 
     def self.queue
       @@async_queue
     end
 
+    def self.queue_latency
+      @@async_queue_latency
+    end
+
     def start_async_process
       @async_thread = Thread.new{
         async_process_loop
       }
       @async_thread[:name] = __method__
+
+      @async_thread_latency = Thread.new{
+        async_process_loop_for_latency
+      }
+      @async_thread_latency[:name] = __method__
     rescue =>e
       @log.error("#{e}\n#{$@}")
     end
@@ -57,6 +67,13 @@ module Roma
         sleep 0.1
       end
       @async_thread.exit
+
+      count = 0
+      while @@async_queue_latency.empty? == false && count < 100
+        count += 1
+        sleep 0.1
+      end
+      @async_thread_latency.exit
     end
 
     def async_process_loop
@@ -84,6 +101,31 @@ module Roma
       retry
     end
 
+    def async_process_loop_for_latency
+      loop {
+        while msg = @@async_queue_latency.pop
+          if send("asyncev_#{msg.event}",msg.args)
+            msg.callback.call(msg,true) if msg.callback
+          else
+            if msg.retry?
+              t = Thread.new{
+                msg.wait
+                msg.incr_count
+                @@async_queue_latency.push(msg)
+              }
+              t[:name] = __method__
+            else
+              @log.error("async process retry out:#{msg.inspect}")
+              msg.callback.call(msg,false) if msg.callback
+            end
+          end
+        end
+      }
+    rescue =>e
+      @log.error("#{e}\n#{$@}")
+      retry
+    end
+
     def asyncev_broadcast_cmd(args)
       @log.debug("#{__method__} #{args.inspect}")
       cmd, nids, tout = args
@@ -120,7 +162,7 @@ module Roma
         end
       end
       t[:name] = __method__
-      true      
+      true
     end
 
     def asyncev_start_balance_process(args)
@@ -148,7 +190,7 @@ module Roma
         end
       end
       t[:name] = __method__
-      true      
+      true
     end
 
     def asyncev_redundant(args)
@@ -193,7 +235,7 @@ module Roma
         @log.warn("async redundant failed:#{k}\e#{hname} #{clk} -> #{nid}")
         return false # retry
       end
-      true      
+      true
     end
 
     def asyncev_reqpushv(args)
@@ -243,6 +285,57 @@ module Roma
       t[:name] = __method__
     end
 
+    def asyncev_start_auto_recover_process(args)
+      @log.debug("#{__method__} #{args.inspect}")
+      ###run_join don't have possibility to be true in this case.
+      #if @stats.run_join
+      #  @log.error("#{__method__}:join process running")
+      #  return true
+      #end
+      if @stats.run_recover
+        @log.error("#{__method__}:recover process running.")
+        return false
+      end
+      if @stats.run_balance
+        @log.error("#{__method__}:balance process running")
+        return true
+      end
+
+      @rttable.auto_recover_status = "preparing"
+      t = Thread::new do
+        begin
+          timeout(@rttable.auto_recover_time){
+            loop{ 
+              sleep 1
+              break if @rttable.auto_recover_status != "preparing"
+              #break if @stats.run_join #run_join don't have possibility to be true in this case.
+              break if @stats.run_recover
+              break if @stats.run_balance
+            }
+          }
+          @log.debug("inactivated AUTO_RECOVER")
+        rescue
+          case @rttable.lost_action
+            when :auto_assign, :shutdown
+              @stats.run_recover = true
+              @rttable.auto_recover_status = "executing"
+                begin
+                  @log.debug("auto recover start")
+                  acquired_recover_process
+                rescue => e
+                  @log.error("#{__method__}:#{e.inspect} #{$@}")
+                ensure
+                  @stats.run_recover = false
+                  @rttable.auto_recover_status = "waiting"
+                end
+            when :no_action
+              @log.debug("auto recover NOT start. Because lost action is [no_action]")
+          end
+        end      
+      end
+      t[:name] = __method__
+    end
+
     def asyncev_start_release_process(args)
       @log.debug("#{__method__} #{args}")
       if @stats.run_iterate_storage
@@ -269,7 +362,7 @@ module Roma
       @log.info("#{__method__}:start")
 
       exclude_nodes = @rttable.exclude_nodes_for_recover(@stats.ap_str, @stats.rep_host)
-      
+
       @do_acquired_recover_process = true
       loop do
         break unless @do_acquired_recover_process
@@ -293,7 +386,7 @@ module Roma
     ensure
       @do_acquired_recover_process = false
     end
- 
+
     def join_process
       @log.info("#{__method__}:start")
       count = 0
@@ -309,7 +402,7 @@ module Roma
           @log.warn("#{__method__}:vnode dose not found")
           return false
         end
-        ret = req_push_a_vnode(vn, nodes[0], is_primary)      
+        ret = req_push_a_vnode(vn, nodes[0], is_primary)
         if ret == :rejected
           sleep 5
         else
@@ -317,10 +410,10 @@ module Roma
           count += 1
         end
       end
-      @log.info("#{__method__} has done.")
     rescue => e
       @log.error("#{e.inspect} #{$@}")
     ensure
+      @log.info("#{__method__} has done.")
       @do_join_process = false
     end
 
@@ -339,7 +432,7 @@ module Roma
           @log.warn("#{__method__}:vnode dose not found")
           return false
         end
-        ret = req_push_a_vnode(vn, nodes[0], is_primary)      
+        ret = req_push_a_vnode(vn, nodes[0], is_primary)
         if ret == :rejected
           sleep 5
         else
@@ -369,11 +462,11 @@ module Roma
       Roma::Messaging::ConPool.instance.return_connection(src_nid,con)
       # waiting for pushv
       count = 0
-      while @rttable.search_nodes(vn).include?(@stats.ap_str)==false && count < 300
+      while @rttable.search_nodes(vn).include?(@stats.ap_str)==false && count < @stats.reqpushv_timeout_count
         sleep 0.1
         count += 1
       end
-      if count >= 300
+      if count >= @stats.reqpushv_timeout_count
         @log.warn("#{__method__}:request has been time-out.vn=#{vn} nid=#{src_nid}")
         return :timeout
       end
@@ -383,7 +476,7 @@ module Roma
       @rttable.proc_failed(src_nid)
       false
     end
- 
+
     def release_process
       @log.info("#{__method__}:start.")
 
@@ -397,7 +490,7 @@ module Roma
         @rttable.each_vnode do |vn, nids|
           break unless @do_release_process
           if nids.include?(@stats.ap_str)
-            
+
             to_nid, new_nids = @rttable.select_node_for_release(@stats.ap_str, @stats.rep_host, nids)
             unless sync_a_vnode_for_release(vn, to_nid, new_nids)
               @log.warn("#{__method__}:error at vn=#{vn} to_nid=#{to_nid} new_nid=#{new_nids}")
@@ -413,10 +506,10 @@ module Roma
       @do_release_process = false
       Roma::Messaging::ConPool.instance.close_all
     end
- 
+
     def sync_a_vnode_for_release(vn, to_nid, new_nids)
       nids = @rttable.search_nodes(vn)
-      
+
       if nids.include?(to_nid)==false || (is_primary && nids[0]!=to_nid)
         @log.debug("#{__method__}:#{vn} #{to_nid}")
         # change routing data at the vnode and synchronize a data
@@ -444,7 +537,7 @@ module Roma
         if clk.is_a?(Integer) == false
           clk,new_nids = @rttable.search_nodes_with_clk(vn)
         end
-        
+
         cmd = "setroute #{vn} #{clk - 1}"
         new_nids.each{ |nn| cmd << " #{nn}"}
         res = async_broadcast_cmd("#{cmd}\r\n")
@@ -459,7 +552,7 @@ module Roma
 
     def sync_a_vnode(vn, to_nid, is_primary=nil)
       nids = @rttable.search_nodes(vn)
-      
+
       if nids.include?(to_nid)==false || (is_primary && nids[0]!=to_nid)
         @log.debug("#{__method__}:#{vn} #{to_nid} #{is_primary}")
         # change routing data at the vnode and synchronize a data
@@ -488,7 +581,7 @@ module Roma
         if clk.is_a?(Integer) == false
           clk,nids = @rttable.search_nodes_with_clk(vn)
         end
-        
+
         cmd = "setroute #{vn} #{clk - 1}"
         nids.each{ |nn| cmd << " #{nn}"}
         res = async_broadcast_cmd("#{cmd}\r\n")
@@ -534,7 +627,7 @@ module Roma
       con = Roma::Messaging::ConPool.instance.get_connection(nid)
 
       @do_push_a_vnode_stream = true
-      
+
       con.write("spushv #{hname} #{vn}\r\n")
 
       res = con.gets # READY\r\n or error string
@@ -631,6 +724,51 @@ module Roma
       @log.info("#{__method__}:stop")
     end
 
+    def asyncev_calc_latency_average(args)
+      latency,cmd = args
+      #@log.debug(__method__)
+
+      if !@stats.latency_data.key?(cmd) #only first execute target cmd
+        @stats.latency_data[cmd].store("latency", Array.new())
+        @stats.latency_data[cmd].store("latency_max", Hash.new())
+        @stats.latency_data[cmd]["latency_max"].store("current", 0)
+        @stats.latency_data[cmd].store("latency_min", Hash.new())
+        @stats.latency_data[cmd]["latency_min"].store("current", 99999)
+        @stats.latency_data[cmd].store("time", Time.now.to_i)
+      end
+
+      begin
+        @stats.latency_data[cmd]["latency"].delete_at(0) if @stats.latency_data[cmd]["latency"].length >= 10
+        @stats.latency_data[cmd]["latency"].push(latency)
+
+        @stats.latency_data[cmd]["latency_max"]["current"] = latency if latency > @stats.latency_data[cmd]["latency_max"]["current"]
+        @stats.latency_data[cmd]["latency_min"]["current"] = latency if latency < @stats.latency_data[cmd]["latency_min"]["current"]
+
+      rescue =>e
+        @log.error("#{__method__}:#{e.inspect} #{$@}")
+
+      ensure
+        if @stats.latency_check_time_count != nil && Time.now.to_i - @stats.latency_data[cmd]["time"] > @stats.latency_check_time_count
+          average = @stats.latency_data[cmd]["latency"].inject(0.0){|r,i| r+=i }/@stats.latency_data[cmd]["latency"].size
+          max = @stats.latency_data[cmd]["latency_max"]["current"]
+          min = @stats.latency_data[cmd]["latency_min"]["current"]
+          @log.debug("Latency average[#{cmd}]: #{sprintf("%.8f", average)}"+
+                     "(denominator=#{@stats.latency_data[cmd]["latency"].length}"+
+                     " max=#{sprintf("%.8f", max)}"+
+                     " min=#{sprintf("%.8f", min)})"
+                    )
+
+          @stats.latency_data[cmd]["time"] =  Time.now.to_i
+          @stats.latency_data[cmd]["latency_past"] = @stats.latency_data[cmd]["latency"]
+          @stats.latency_data[cmd]["latency"] = []
+          @stats.latency_data[cmd]["latency_max"]["past"] = @stats.latency_data[cmd]["latency_max"]["current"]
+          @stats.latency_data[cmd]["latency_max"]["current"] = 0
+          @stats.latency_data[cmd]["latency_min"]["past"] = @stats.latency_data[cmd]["latency_min"]["current"]
+          @stats.latency_data[cmd]["latency_min"]["current"] = 99999
+        end
+      end
+      true
+    end
 
   end # module AsyncProcess
 

data/ruby/server/lib/roma/command/bg_command_receiver.rb

@@ -1,4 +1,3 @@
-# -*- coding: utf-8 -*-
 require 'roma/async_process'
 require 'roma/messaging/con_pool'
 require 'roma/command/vn_command_receiver'
@@ -12,7 +11,7 @@ module Roma
       def ev_balance(s)
         res = broadcast_cmd("rbalance\r\n")
         if @stats.run_join == false &&
-            @stats.run_recover == false && 
+            @stats.run_recover == false &&
             @stats.run_balance == false &&
             @rttable.vnode_balance(@stats.ap_str)==:less
           Roma::AsyncProcess::queue.push(Roma::AsyncMessage.new('start_balance_process'))
@@ -25,7 +24,7 @@ module Roma
 
       def ev_rbalance(s)
         if @stats.run_join == false &&
-            @stats.run_recover == false && 
+            @stats.run_recover == false &&
             @stats.run_balance == false &&
             @rttable.vnode_balance(@stats.ap_str)==:less
           Roma::AsyncProcess::queue.push(Roma::AsyncMessage.new('start_balance_process'))
@@ -37,7 +36,7 @@ module Roma
 
       def ev_release(s)
         if @stats.run_join == false &&
-            @stats.run_recover == false && 
+            @stats.run_recover == false &&
             @stats.run_balance == false &&
             @stats.run_release == false &&
             @stats.run_iterate_storage == false

data/ruby/server/lib/roma/command/rt_command_receiver.rb

@@ -36,20 +36,24 @@ module Roma
 
       # routingdump [yaml|json|yamlbytes|bin]\r\n
       def ev_routingdump(s)
+        rt = @rttable
+        rd = @rttable.sub_nid_rd(@addr)
+        rt = Roma::Routing::RoutingTable.new(rd) if rd
+
         if s.length == 1
-          dmp = @rttable.dump
+          dmp = rt.dump
           send_data("#{dmp.length}\r\n#{dmp}\r\nEND\r\n")
         elsif s[1] == 'yaml'
-          dmp = @rttable.dump_yaml
+          dmp = rt.dump_yaml
           send_data("#{dmp}\r\nEND\r\n")
         elsif s[1] == 'json'
-          dmp = @rttable.dump_json
+          dmp = rt.dump_json
           send_data("#{dmp}\r\nEND\r\n")
         elsif s[1] == 'yamlbytes'
-          dmp = @rttable.dump_yaml
+          dmp = rt.dump_yaml
           send_data("#{dmp.length + 7}\r\nEND\r\n")
         elsif s[1] == 'bin'
-          dmp = @rttable.dump_binary
+          dmp = rt.dump_binary
           send_data("#{dmp.length}\r\n#{dmp}\r\nEND\r\n")
         else
           send_data("CLIENT_ERROR\r\n")
@@ -107,6 +111,69 @@ module Roma
         send_data("CLIENT_ERROR\r\n")
       end
 
+      # set_auto_recover [true|false] <sec>
+      def ev_set_auto_recover(s)
+        #check argument
+        if /^true$|^false$/ !~ s[1]
+          return send_data("CLIENT_ERROR arguments must be true or false\r\n")
+        elsif s.length != 2 && s.length != 3
+          return send_data("CLIENT_ERROR number of arguments(0 for 1)\r\n")
+        elsif s.length == 3 && s[2].to_i < 1
+          return send_data("CLIENT_ERROR length must be greater than zero\r\n")
+        end
+        res = broadcast_cmd("rset_auto_recover #{s[1]} #{s[2]}\r\n")
+        if s[1] == "true"
+          @rttable.auto_recover = true
+        elsif s[1] == "false"
+          @rttable.auto_recover = false
+        end
+        @rttable.auto_recover_status = "waiting"
+        @rttable.auto_recover_time = s[2].to_i if s[2]
+        res[@stats.ap_str] = "STORED"
+        send_data("#{res}\r\n")
+      end
+
+      def ev_rset_auto_recover(s)
+        if /^true$|^false$/ !~ s[1]
+          return send_data("CLIENT_ERROR arguments must be true or false #{s[1]} #{s[1].class} \r\n")
+        elsif s.length != 2 && s.length != 3
+          return send_data("CLIENT_ERROR number of arguments(0 for 1)\r\n")
+        elsif s.length == 3 && s[2].to_i < 1
+          return send_data("CLIENT_ERROR length must be greater than zero\r\n")
+        end
+        if s[1] == "true"
+          @rttable.auto_recover = true
+        elsif s[1] == "false"
+          @rttable.auto_recover = false
+        end
+        @rttable.auto_recover_status = "waiting"
+        @rttable.auto_recover_time = s[2].to_i if s[2]
+        send_data("STORED\r\n")
+      end
+
+      # set_lost_action [auto_assign|shutdown]
+      def ev_set_lost_action(s)
+        if s.length != 2 || /^auto_assign$|^shutdown$/ !~ s[1]
+          return send_data("CLIENT_ERROR changing lost_action must be auto_assign or shutdown\r\n")
+        elsif /^auto_assign$|^shutdown$/ !~ @rttable.lost_action
+          return send_data("CLIENT_ERROR can use this command only current lost action is auto_assign or shutdwn mode\r\n")
+        end
+        res = broadcast_cmd("rset_lost_action #{s[1]}\r\n")
+        @rttable.lost_action = s[1].to_sym
+        res[@stats.ap_str] = "STORED"
+        send_data("#{res}\r\n")
+      end
+      
+      def ev_rset_lost_action(s)
+        if s.length != 2 || /^auto_assign$|^shutdown$/ !~ s[1]
+          return send_data("CLIENT_ERROR changing lost_action must be auto_assign or shutdown\r\n")
+        elsif /^auto_assign$|^shutdown$/ !~ @rttable.lost_action
+          return send_data("CLIENT_ERROR can use this command only current lost action is auto_assign or shutdwn mode\r\n")
+        end
+        @rttable.lost_action = s[1].to_sym
+        send_data("STORED\r\n")
+      end  
+
       # set_threshold_for_failover <n>
       def ev_set_threshold_for_failover(s)
         if s.length != 2 || s[1].to_i == 0
@@ -147,6 +214,67 @@ module Roma
         send_data("STORED\r\n")        
       end
 
+      # cleat RTTABLE_SUB_NID map
+      def ev_clear_rttable_sub_nid(s)
+        res = broadcast_cmd("rclear_rttable_sub_nid\r\n")
+        @rttable.sub_nid.clear()
+        res[@stats.ap_str] = "CLEARED"
+        send_data("#{res}\r\n")
+      end
+
+      def ev_rclear_rttable_sub_nid(s)
+        @rttable.sub_nid.clear()
+        send_data("CLEARED\r\n")
+      end
+      
+      # add_rttable_sub_nid <netmask> <regexp> <replace>
+      def ev_add_rttable_sub_nid(s)
+        if s.length != 4
+          return send_data("usage:add_rttable_sub_nid <netmask> <regexp> <replace>\r\n")
+        end
+        res = broadcast_cmd("radd_rttable_sub_nid #{s[1]} #{s[2]} #{s[3]}\r\n")
+        @rttable.sub_nid[s[1]] = {:regexp => "#{s[2]}", :replace => "#{s[3]}"}
+        res[@stats.ap_str] = "ADDED"
+        send_data("#{res}\r\n")
+      end
+
+      # radd_rttable_sub_nid <netmask> <regexp> <replace>
+      def ev_radd_rttable_sub_nid(s)
+        if s.length != 4
+          return send_data("usage:add_rttable_sub_nid <netmask> <regexp> <replace>\r\n")
+        end
+        @rttable.sub_nid[s[1]] = {:regexp => "#{s[2]}", :replace => "#{s[3]}"}
+        send_data("ADDED\r\n")
+      end
+
+      # delete_rttable_sub_nid <netmask>
+      def ev_delete_rttable_sub_nid(s)
+        if s.length != 2
+          return send_data("usage:delete_rttable_sub_nid <netmask>\r\n")
+        end
+
+        res = broadcast_cmd("rdelete_rttable_sub_nid #{s[1]}\r\n")
+        unless @rttable.sub_nid.delete s[1]
+          res[@stats.ap_str] = "NOT_FOUND"
+        else
+          res[@stats.ap_str] = "DELETED"
+        end
+        send_data("#{res}\r\n")
+      end
+
+      # rdelete_rttable_sub_nid <netmask>
+      def ev_rdelete_rttable_sub_nid(s)
+        if s.length != 2
+          return send_data("usage:delete_rttable_sub_nid <netmask>\r\n")
+        end
+
+        unless @rttable.sub_nid.delete s[1]
+          send_data("NOT_FOUND\r\n")
+        else
+          send_data("DELETED\r\n")
+        end
+      end
+
     end # module RoutingCommandReceiver
   end # module Command
 end # module Roma