pwrake 2.1.2 → 2.1.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: c955ecbb671677c131979dcc11b35985fee3aa58
-  data.tar.gz: aa0ebbcc4161b6e78a07656534e97528487b82d2
+  metadata.gz: db880d8483f79d6812bd1920557350a7b4311378
+  data.tar.gz: a1d2779a59eb21dd4df2b8ee5e9a00054a50654e
 SHA512:
-  metadata.gz: a45cd6523db363a562225cb79438de77ca57f96a6cdaba59ddf67cf15147919b814167d2b07e0169b3164704b8ce548de49f67ca7befd5cff07b90e4de870b2f
-  data.tar.gz: 7752375bc140f07f00cae72333d1378375358543646fd42f16e9c5835177b16be21049aa3c1fc52e147dbd5d54df0aa8003e688bb1410938fea2b7fb3ebfe818
+  metadata.gz: 4bc21c0d7a5ef317d5d03883ca918e2c2218e0ec28e13167079127820e0c4800923f4d23c6891e4eff5dcdfb010db4b2d6214a6d9eedf5c6c5597a1353467d7a
+  data.tar.gz: b851a5ffb98ddccf77b52bbd5c4339a900d617ce8018e9c68ee665999ac9a70a8b5295be21c88832a8d9f20cd257da81b9b4c91715f487fde0711a60db4e9931

data/README.md

@@ -86,7 +86,8 @@ In this case, you need the rehash of command paths:
     -d, --debug                      [Pw] Output Debug messages
         --pwrake-conf [FILE]         [Pw] Pwrake configuration file in YAML
         --show-conf, --show-config   [Pw] Show Pwrake configuration options
-        --report LOGDIR              [Pw] Report workflow statistics from LOGDIR to HTML and exit.
+        --report LOGDIR              [Pw] Generate `report.html' (Report of workflow statistics) in LOGDIR and exit.
+        --report-image IMAGE_TYPE    [Pw] Gnuplot output format (png,jpg,svg etc.) in report.html.
         --clear-gfarm2fs             [Pw] Clear gfarm2fs mountpoints left after failure.
 
 ### pwrake_conf.yaml
@@ -125,6 +126,7 @@ In this case, you need the rehash of command paths:
         NOACTION_QUEUE_PRIORITY FIFO(default)|LIFO|RAND
         SHELL_START_INTERVAL    default=0.012 (sec)
         GRAPH_PARTITION         false(default)|true
+        REPORT_IMAGE            default=png
 
 * Options for Gfarm system:
 

data/lib/pwrake/branch/communicator.rb

@@ -108,7 +108,6 @@ class Communicator
     case s
     when /^heartbeat$/
       Log.debug "#{x}: #{s.inspect}"
-      @selector.heartbeat(@reader.io)
     when /^exited$/
       Log.debug "#{x}: #{s.inspect}"
       return false

data/lib/pwrake/branch/shell.rb

@@ -193,12 +193,15 @@ module Pwrake
           msg = "Shell#io_read_loop: exit"
           $stderr.puts(msg)
           Log.error(msg)
+          @exited = true
           @chan.halt
           return "exit"
         when IOError
+          @exited = true
           @chan.halt
           return "ioerror"
         when NBIO::TimeoutError
+          @exited = true
           @chan.halt
           return "timeout"
         end
@@ -236,6 +239,7 @@ module Pwrake
               Rake.application.display_error_message(e)
               Log.error e
               result = "taskfail:#{@id}:#{task.name}"
+              break if @exited
             ensure
               master_w.put_line result
             end

data/lib/pwrake/master/master.rb

@@ -143,6 +143,7 @@ module Pwrake
       @selector.run
 
       Log.info "num_cores=#{sum_ncore}"
+      @option.total_cores = sum_ncore
       @hostinfo_by_id.each do |id,host|
         if ncore = @hostinfo_by_id[id].idle_cores
           Log.info "#{host} id=#{id} ncore=#{ncore}"
@@ -436,10 +437,10 @@ module Pwrake
       Log.debug "Master#finish @selector.run end"
       if !@exited
         @exited = true
-        Log.debug "Master#finish Hander.exit begin"
+        Log.debug "Master#finish Handler.exit begin"
         NBIO::Handler.exit(@hdl_set)
         @selector.run(60)
-        Log.debug "Master#finish Hander.exit end"
+        Log.debug "Master#finish Handler.exit end"
       end
       TaskWrapper.close_task_logger
       Log.debug "Master#finish end"

data/lib/pwrake/master/master_application.rb

@@ -26,10 +26,10 @@ module Pwrake
       standard_exception_handling do
         init("pwrake")  # <- parse options here
         @role = @master = Master.new
-        load_rakefile
         t = Time.now
         @master.init
         @master.setup_branches
+        load_rakefile
         begin
           Log.debug "init: #{Time.now-t} sec"
           t = Time.now
@@ -62,7 +62,11 @@ module Pwrake
       opts.each_with_index do |a,i|
         if a[0] == '--version'
           a[3] = lambda { |value|
-            puts "rake, version #{RAKEVERSION}"
+            if defined? RAKEVERSION
+              puts "rake, version #{RAKEVERSION}"
+            elsif defined? Rake::VERSION
+              puts "rake, version #{Rake::VERSION}"
+            end
             puts "pwrake, version #{Pwrake::VERSION}"
             exit
           }
@@ -139,9 +143,12 @@ module Pwrake
         "[Pw] Show Pwrake configuration options",
         lambda {|value| options.show_conf = true }
        ],
-       ['--report LOGDIR',"[Pw] Report workflow statistics from LOGDIR to HTML and exit.",
+       ['--report LOGDIR',"[Pw] Generate `report.html' (Report of workflow statistics) in LOGDIR and exit.",
         lambda {|value| options.report_dir = value }
        ],
+       ['--report-image IMAGE_TYPE',"[Pw] Gnuplot output format (png,jpg,svg etc.) in report.html.",
+        lambda {|value| options.report_image = value }
+       ],
        ['--clear-gfarm2fs',"[Pw] Clear gfarm2fs mountpoints left after failure.",
          lambda { |value|
            Option.new.clear_gfarm2fs

data/lib/pwrake/nbio.rb

@@ -3,7 +3,7 @@ require "fiber"
 module Pwrake
 module NBIO
 
-  class TimeoutError < IOError
+  class TimeoutError < StandardError
   end
 
   class Selector
@@ -11,7 +11,6 @@ module NBIO
     def initialize
       @reader = {}
       @writer = {}
-      @hb_time = {}
       @running = false
     end
 
@@ -19,12 +18,10 @@ module NBIO
 
     def add_reader(hdl)
       @reader[hdl.io] = hdl
-      heartbeat(hdl.io) if hdl.timeout
     end
 
     def delete_reader(hdl)
       @reader.delete(hdl.io)
-      delete_heartbeat(hdl.io)
     end
 
     def add_writer(hdl)
@@ -51,19 +48,9 @@ module NBIO
       hdl.respond_to?(:host) ? hdl.host : nil
     end
 
-    # called when IO start and receive heartbeat
-    def heartbeat(io)
-      @hb_time[io] = Time.now
-      @hb_earliest = @hb_time.values.min
-    end
-
-    def delete_heartbeat(io)
-      @hb_time.delete(io)
-      @hb_earliest = @hb_time.values.min
-    end
-
     def run(timeout=nil)
       @running = true
+      init_heartbeat if timeout
       while @running && !empty?
         if $debug
           Log.debug "Selector#run: "+caller[0..1].join(", ")+
@@ -74,21 +61,16 @@ module NBIO
       end
     ensure
       @running = false
+      @hb_time = nil
     end
 
     private
     def run_select(timeout)
-      r, w = IO.select(@reader.keys,@writer.keys,[],timeout)
+      to = (timeout) ? timeout*0.75 : nil
+      r, w, = IO.select(@reader.keys,@writer.keys,[],to)
+      check_heartbeat(r,timeout) if timeout
       r.each{|io| @reader[io].call} if r
       w.each{|io| @writer[io].call} if w
-      while timeout && @hb_earliest && Time.now - @hb_earliest > timeout
-        io = @hb_time.key(@hb_earliest)
-        if hdl = @reader[io]
-          e = TimeoutError.new("HB Timeout (#{timeout}s) #<Reader:%x> #{io.inspect}"%hdl.__id__)
-          hdl.error(e)
-        end
-        delete_heartbeat(io)
-      end
     rescue IOError => e
       em = "#{e.class.name}: #{e.message}"
       @reader.keys.each do |io|
@@ -112,6 +94,33 @@ module NBIO
       #raise e
     end
 
+    def init_heartbeat
+      t = Time.now
+      @hb_time = {}
+      @reader.each_key{|io| @hb_time[io] = t}
+    end
+
+    def check_heartbeat(ios,timeout)
+      t = Time.now
+      rds = @reader.dup
+      if ios
+        ios.each do |io|
+          @hb_time[io] = t
+          rds.delete(io)
+        end
+      end
+      rds.each do |io,hdl|
+        if hdl.check_timeout
+          tdif = t - @hb_time[io]
+          if tdif > timeout
+            m = "Heartbeat Timeout: no response during #{tdif}s "+
+              "> timeout #{timeout}s from host=#{get_host(io)}"
+            hdl.error(TimeoutError.new(m))
+          end
+        end
+      end
+    end
+
   end
 
 #------------------------------------------------------------------
@@ -216,7 +225,7 @@ module NBIO
       @chunk_size = 8192
     end
     attr_reader :io
-    attr_accessor :timeout
+    attr_accessor :check_timeout
 
     # call from Selector#run
     def call
@@ -316,10 +325,9 @@ module NBIO
       @n_chan = n_chan
       @queue = @n_chan.times.map{|i| FiberReaderQueue.new(self)}
       @default_queue = FiberReaderQueue.new(self)
-      @timeout = true
+      @check_timeout = true
     end
     attr_reader :queue
-    attr_accessor :timeout
     attr_accessor :default_queue
 
     def [](ch)

data/lib/pwrake/option/option.rb

@@ -39,6 +39,7 @@ module Pwrake
 
     attr_reader :counter
     attr_reader :logger
+    attr_accessor :total_cores
 
     DEFAULT_CONFFILES = ["pwrake_conf.yaml","PwrakeConf.yaml"]
 
@@ -121,6 +122,7 @@ module Pwrake
         'PLOT_PARALLELISM',
         'SHOW_CONF',
         ['REPORT_DIR','REPORT'],
+        'REPORT_IMAGE',
         'FAILED_TARGET', # rename(default), delete, leave
         'FAILURE_TERMINATION', # wait, kill, continue
         'QUEUE_PRIORITY', # RANK(default), FIFO, LIFO, DFS

data/lib/pwrake/option/option_filesystem.rb

@@ -12,7 +12,8 @@ module Pwrake
     def setup_filesystem
 
       @worker_progs = %w[
-        parallel/processor_count
+        parallel/processor_count.rb
+        pwrake/worker/reader
         pwrake/worker/writer
         pwrake/worker/log_executor
         pwrake/worker/executor

data/lib/pwrake/report/parallelism.rb

@@ -64,7 +64,7 @@ module Pwrake
       d
     end
 
-    def plot_parallelism(file)
+    def plot_parallelism(file,fmt)
       a = count_start_end_from_csv(file)
       return if a.size < 4
 
@@ -101,8 +101,8 @@ module Pwrake
 
       IO.popen("gnuplot","r+") do |f|
         f.puts "
-set terminal png
-set output '#{base}.png'
+set terminal #{fmt}
+set output '#{base}.#{fmt}'
 #set rmargin 10
 set title '#{base}'
 set xlabel 'time (sec)'
@@ -115,17 +115,17 @@ plot '#{fpara}' w l axis x1y1 title 'parallelism'
 "
       end
 
-      #puts "Parallelism plot: #{base}.png"
+      #puts "Parallelism plot: #{base}.#{fmt}"
     end
 
 
-    def plot_parallelism2(csvtable, base)
+    def plot_parallelism2(csvtable, base, fmt)
       a = count_start_end_from_csv_table(csvtable)
       return if a.size < 4
 
       density = exec_density(a)
 
-      fimg = base+'/parallelism.png'
+      fimg = base+'/parallelism.'+fmt
 
       n = a.size
       i = 0
@@ -155,7 +155,7 @@ plot '#{fpara}' w l axis x1y1 title 'parallelism'
       if system("which gnuplot >/dev/null 2>&1")
       IO.popen("gnuplot","r+") do |f|
         f.print "
-set terminal png
+set terminal #{fmt}
 set output '#{fimg}'
 #set rmargin 10
 set title '#{base}'
@@ -259,7 +259,7 @@ plot '-' w l axis x1y1 title 'parallelism', '-' w l axis x1y2 title 'exec/sec'
       h
     end
 
-    def plot_parallelism_by_pattern(csvtable, base, pattern)
+    def plot_parallelism_by_pattern(csvtable, base, pattern, fmt)
       y_max = 0
       t_end = 0
       para = {}
@@ -290,13 +290,13 @@ plot '-' w l axis x1y1 title 'parallelism', '-' w l axis x1y2 title 'exec/sec'
         end
       end
 
-      fimg = base+'/para_cmd.png'
+      fimg = base+'/para_cmd.'+fmt
 
       if system("which gnuplot >/dev/null 2>&1")
       IO.popen("gnuplot","r+") do |f|
         #begin f = $stdout
         f.print "
-set terminal png
+set terminal #{fmt}
 set output '#{fimg}'
 set title '#{base}'
 set xlabel 'time (sec)'
@@ -338,10 +338,10 @@ set ylabel 'parallelism'
       return grid
     end
 
-    def plot_parallelism_by_host(csvtable,base)
-      fpng = base+"/para_host.png"
+    def plot_parallelism_by_host(csvtable,base,fmt)
+      fimg = base+"/para_host."+fmt
       data = read_time_by_host_from_csv(csvtable)
-      return fpng if data.size == 0
+      return fimg if data.size == 0
 
       grid = []
       hosts = data.keys.sort
@@ -353,8 +353,8 @@ set ylabel 'parallelism'
       if system("which gnuplot >/dev/null 2>&1")
       IO.popen("gnuplot","r+") do |f|
         f.puts "
-set terminal png
-set output '#{fpng}'
+set terminal #{fmt}
+set output '#{fimg}'
 #set rmargin 7
 set lmargin 16
 set pm3d map
@@ -385,7 +385,7 @@ set format y ''
         f.printf "e\n"
       end
       end
-      fpng
+      fimg
     end
 
   end

data/lib/pwrake/report/report.rb

@@ -43,6 +43,8 @@ EOL
       @id = @@id
       @base = @dir
 
+      @img_fmt = option['REPORT_IMAGE'] || 'png'
+
       @cmd_file = File.join(@dir,option['COMMAND_CSV_FILE'])
       @task_file = File.join(@dir,option['TASK_CSV_FILE'])
       @html_file = File.join(@dir,'report.html')
@@ -196,15 +198,15 @@ EOL
       end
       html << "</table>\n"
       html << "<h2>Parallelism</h2>\n"
-      fimg = Parallelism.plot_parallelism2(@sh_table,@base)
+      fimg = Parallelism.plot_parallelism2(@sh_table,@base,@img_fmt)
       html << "<img src='./#{File.basename(fimg)}' align='top'/></br>\n"
 
       html << "<h2>Parallelism by command</h2>\n"
-      fimg3 = Parallelism.plot_parallelism_by_pattern(@sh_table,@base,@pattern)
+      fimg3 = Parallelism.plot_parallelism_by_pattern(@sh_table,@base,@pattern,@img_fmt)
       html << "<img src='./#{File.basename(fimg3)}' align='top'/></br>\n"
 
       html << "<h2>Parallelism by host</h2>\n"
-      fimg2 = Parallelism.plot_parallelism_by_host(@sh_table,@base)
+      fimg2 = Parallelism.plot_parallelism_by_host(@sh_table,@base,@img_fmt)
       html << "<img src='./#{File.basename(fimg2)}' align='top'/></br>\n"
 
       html << "<h2>Command statistics</h2>\n"
@@ -285,11 +287,11 @@ EOL
           command_list << cmd
         end
       end
-      hist_image = @base+"/hist.png"
+      hist_image = @base+"/hist."+@img_fmt
       if system("which gnuplot >/dev/null 2>&1")
       IO.popen("gnuplot","r+") do |f|
         f.puts "
-set terminal png # size 480,360
+set terminal #{@img_fmt} # size 480,360
 set output '#{hist_image}'
 set ylabel 'histogram'
 set xlabel 'Execution time (sec)'

data/lib/pwrake/report/report_multi.rb

@@ -2,14 +2,15 @@ module Pwrake
 
   class ReportMulti
 
-    def initialize(list,pattern)
+    def initialize(list,pattern,options={})
       @reports = list.map do |base|
         r = Report.new(base,pattern)
         puts r.base+" elap=#{r.elap}"
         r
       end
       @pattern = pattern
-      @elap_png = 'elap.png'
+      @img_fmt = options['REPORT_IMAGE'] || 'png'
+      @elap_img = 'elap.#{@img_fmt}'
     end
 
     def report(stat_html)
@@ -29,7 +30,7 @@ module Pwrake
       end
       html << "</table>\n"
       html << "<h2>Elapsed time</h2>\n"
-      html << "<img src='./#{File.basename(@elap_png)}'  align='top'/></br>\n"
+      html << "<img src='./#{File.basename(@elap_img)}'  align='top'/></br>\n"
 
       html << "<h2>Histogram of Execution time</h2>\n"
       html << report_histogram()
@@ -51,8 +52,8 @@ module Pwrake
       ymax = 10**(mid+wid)
       IO.popen("gnuplot","r+") do |f|
         f.puts "
-set terminal png size 640,480
-set output '#{@elap_png}'
+set terminal #{@img_fmt} size 640,480
+set output '#{@elap_img}'
 set xlabel 'ncore'
 set ylabel 'time (sec)'
 set yrange [#{ymin}:#{ymax}]
@@ -64,7 +65,7 @@ plot #{a}/x,'-' w lp lw 2 ps 2 title 'elapsed time'
         end
         f.puts "e"
       end
-      puts "Ncore-time plot: "+@elap_png
+      puts "Ncore-time plot: "+@elap_img
     end
 
     def report_histogram
@@ -81,7 +82,7 @@ plot #{a}/x,'-' w lp lw 2 ps 2 title 'elapsed time'
       end
 
       @cmd_rep.each_key do |cmd|
-        @images[cmd] = 'hist_'+cmd.gsub(/[\/.]/,'_')+'.png'
+        @images[cmd] = 'hist_'+cmd.gsub(/[\/.]/,'_')+'.'+@img_fmt
       end
       histogram_plot
       histogram_html
@@ -106,7 +107,7 @@ plot #{a}/x,'-' w lp lw 2 ps 2 title 'elapsed time'
       @cmd_rep.each do |cmd,cmd_rep|
         IO.popen("gnuplot","r+") do |f|
           f.puts "
-set terminal png # size 480,360
+set terminal #{@img_fmt} # size 480,360
 set output '#{@images[cmd]}'
 set ylabel 'histogram'
 set xlabel 'Execution time (sec)'
@@ -139,7 +140,7 @@ set title '#{cmd}'"
       @cmd_rep.each do |cmd,cmd_rep|
         IO.popen("gnuplot","r+") do |f|
           f.puts "
-set terminal png # size 480,360
+set terminal #{@img_fmt} # size 480,360
 set output '#{@images[cmd]}'
 set nohidden3d
 set palette rgb 33,13,10
@@ -193,4 +194,3 @@ set title '#{cmd}'"
 
   end
 end
-

data/lib/pwrake/version.rb

@@ -1,3 +1,3 @@
 module Pwrake
-  VERSION = "2.1.2"
+  VERSION = "2.1.3"
 end

data/lib/pwrake/worker/executor.rb

@@ -2,168 +2,147 @@ module Pwrake
 
   class Executor
 
-    LIST = {}
-    CHARS='0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
-    TLEN=32
-
-    def initialize(dir_class,id,shell_cmd,shell_rc)
+    def initialize(selector,dir_class,id)
+      @selector = selector
       @id = id
-      @shell_rc = shell_rc
-      @shell_cmd = shell_cmd || ENV['SHELL'] || '/bin/sh'
-      @terminator = ""
-      TLEN.times{ @terminator << CHARS[rand(CHARS.length)] }
       @out = Writer.instance
       @log = LogExecutor.instance
-      @queue = Queue.new
+      @queue = []
+      @rd_list = []
       @dir = dir_class.new
-      @spawn_in, @sh_in = IO.pipe
-      @sh_out, @spawn_out = IO.pipe
-      @sh_err, @spawn_err = IO.pipe
-      LIST[@id] = self
-      @exec_thread = start_exec_thread
+      @dir.open
+      @dir.open_messages.each{|m| @log.info(m)}
+      @out.puts "#{@id}:open"
     end
 
-    def execute(cmd)
-      @queue.enq(cmd)
+    def stop
+      @stopped = true
+      @queue.clear
     end
 
-    def start_exec_thread
-      Thread.new do
-        begin
-          @dir.open
-          @dir.open_messages.each{|m| @log.info(m)}
-          @pid = Kernel.spawn(@shell_cmd,
-                              :out=>@spawn_out,
-                              :err=>@spawn_err,
-                              :in=>@spawn_in,
-                              :chdir=>@dir.current)
-          begin
-            @out.puts "#{@id}:open"
-            @shell_rc.each do |cmd|
-              run_rc(cmd)
-            end
-            while cmd = @queue.deq
-              run(cmd)
-            end
-            @sh_in.puts("exit")
-            @sh_in.flush
-          ensure
-            status = nil
-            begin
-              Timeout.timeout(5){
-                pid,status = Process.waitpid2(@pid)
-              }
-            rescue => exc
-              @log.error(([exc.to_s]+exc.backtrace).join("\n"))
-              @log.info("#{@id}:kill INT sh @pid=#{@pid}")
-              Process.kill("INT",@pid)
-              pid,status = Process.waitpid2(@pid)
-            end
-            @log.info("shell exit status: "+status.inspect)
-          end
-        rescue => exc
-          @out.puts "#{@id}:exc:#{exc}"
-          @log.error(([exc.to_s]+exc.backtrace).join("\n"))
-        ensure
-          @dir.close_messages.each{|m| @log.info(m)}
-          @dir.close
-        end
+    def close
+      if @thread
+        @thread.join(15)
+        sleep 0.1
       end
+      @thread = Thread.new do
+        @dir.close_messages.each{|m| @log.info(m)}
+        @dir.close
+      end
+    rescue => exc
+      @log.error(([exc.to_s]+exc.backtrace).join("\n"))
     end
 
-    def run(cmd)
-      case cmd
-      when Proc
-        cmd.call
-      when "cd"
-        @dir.cd
-        run_command("cd "+@dir.current)
-        #
-      when /^cd\s+(.*)$/
-        @dir.cd($1)
-        run_command("cd "+@dir.current)
-        #
-      when /^exit\b/
-        close
-        @out.puts "#{@id}:exit"
-        #
-      when String
-        run_command(cmd)
-        #
-      else
-        raise RuntimeError,"invalid cmd: #{cmd.inspect}"
+    def join
+      if @thread
+        @thread.join(15)
       end
+    rescue => exc
+      @log.error(([exc.to_s]+exc.backtrace).join("\n"))
     end
 
-    def run_rc(cmd)
-      run_command_main(cmd){|s| @log.info "<"+s if @log}
+    def execute(cmd)
+      return if @stopped
+      @queue.push(cmd)
+      start_process
     end
 
-    def run_command(cmd)
-      run_command_main(cmd){|s| @out.puts s}
+    def start_process
+      return if @thread      # running
+      command = @queue.shift
+      return if !command     # empty queue
+      @spawn_in, @sh_in = IO.pipe
+      @sh_out, @spawn_out = IO.pipe
+      @sh_err, @spawn_err = IO.pipe
+
+      @pid = Kernel.spawn(command,
+                          :in=>@spawn_in,
+                          :out=>@spawn_out,
+                          :err=>@spawn_err,
+                          :chdir=>@dir.current,
+                          :pgroup=>true
+                         )
+      @log.info "pid=#{@pid} started. command=#{command.inspect}"
+
+      @thread = Thread.new do
+        @pid2,@status = Process.waitpid2(@pid)
+        @spawn_in.close
+        @spawn_out.close
+        @spawn_err.close
+      end
+
+      @rd_out = Reader.new(@sh_out,"o")
+      @rd_err = Reader.new(@sh_err,"e")
+      @rd_list = [@rd_out,@rd_err]
+
+      @selector.add_reader(@sh_out){callback(@rd_out)}
+      @selector.add_reader(@sh_err){callback(@rd_err)}
     end
 
-    def run_command_main(cmd)
-      if /\\$/ =~ cmd  # command line continues
-        @sh_in.puts(cmd)
-        @sh_in.flush
-        return
+    def callback(rd)
+      while s = rd.gets
+        @out.puts "#{@id}:#{rd.mode}:#{s.chomp}"
       end
-      term = "\necho '#{@terminator}':$? \necho '#{@terminator}' 1>&2"
-      @sh_in.puts(cmd+term)
-      @sh_in.flush
-      status = ""
-      io_set = [@sh_out,@sh_err]
-      loop do
-        io_sel, = IO.select(io_set,nil,nil)
-        for io in io_sel
-          s = io.gets.chomp
-          case io
-          when @sh_out
-            if s[0,TLEN] == @terminator
-              status = s[TLEN+1..-1]
-              io_set.delete(@sh_out)
-            else
-              yield "#{@id}:o:"+s
-            end
-          when @sh_err
-            if s[0,TLEN] == @terminator
-              io_set.delete(@sh_err)
-            else
-              yield "#{@id}:e:"+s
-            end
-          end
+      if rd.eof?
+        @selector.delete_reader(rd.io)
+        @rd_list.delete(rd)
+        if @rd_list.empty?  # process_end
+          @thread = @pid = nil
+          @log.info inspect_status
+          @out.puts "#{@id}:z:#{exit_status}"
+          @sh_in.close
+          @sh_out.close
+          @sh_err.close
+          start_process     # next process
         end
-        break if io_set.empty?
       end
-      yield "#{@id}:z:#{status}"
+    rescue => exc
+      @log.error(([exc.to_s]+exc.backtrace).join("\n"))
+      stop
     end
 
-    def close
-      execute(nil)  # threads end
+    def inspect_status
+      s = @status
+      case
+      when s.signaled?
+        if s.coredump?
+          "pid=#{s.pid} dumped core."
+        else
+          "pid=#{s.pid} was killed by signal #{s.termsig}"
+        end
+      when s.stopped?
+        "pid=#{s.pid} was stopped by signal #{s.stopsig}"
+      when s.exited?
+        "pid=#{s.pid} exited normally. status=#{s.exitstatus}"
+      else
+        "unknown status %#x" % s.to_i
+      end
     end
 
-    def join
-      LIST.delete(@id)
-      @exec_thread.join(15) if @exec_thread
+    def exit_status
+      s = @status
+      case
+      when s.signaled?
+        if s.coredump?
+          "core_dumped"
+        else
+          "killed:#{s.termsig}"
+        end
+      when s.stopped?
+        "stopped:#{s.stopsig}"
+      when s.exited?
+        "#{s.exitstatus}"
+      else
+        "unknown:%#x" % s.to_i
+      end
     end
 
     def kill(sig)
-      @queue.clear
+      stop
       if @pid
-        # kill process group
-        s = `ps ho pid --ppid=#{@pid}`
-        s.each_line do |x|
-          pid = x.to_i
-          Process.kill(sig,pid)
-          @log.warn "Executor(id=#{@id})#kill pid=#{pid} sig=#{sig}"
-        end
-        if s.empty?
-          @log.warn "Executor(id=#{@id})#kill nothing killed"
-        end
+        Process.kill(sig,-@pid)
+        @log.warn "Executor(id=#{@id})#kill pid=#{@pid} sig=#{sig}"
       end
-      @spawn_out.flush
-      @spawn_err.flush
     end
 
   end

data/lib/pwrake/worker/invoker.rb

@@ -1,24 +1,32 @@
 module Pwrake
 
   class Invoker
-    # using Michael Grosser's parallel
-    # https://github.com/grosser/parallel
-    include Parallel::ProcessorCount
+    begin
+      # use Michael Grosser's Parallel module
+      # https://github.com/grosser/parallel
+      include Parallel::ProcessorCount
+    rescue
+      def processor_count
+        # only for Linux
+        IO.read("/proc/cpuinfo").scan(/^processor/).size
+      end
+    end
 
     def initialize(dir_class, ncore, option)
       @dir_class = dir_class
       @option = option
+      @selector = Selector.new
+      @ex_list = {}
       @out = Writer.instance # firstly replace $stderr
       @log = LogExecutor.instance
       @log.init(@option)
       @log.open(@dir_class)
       @out.add_logger(@log)
-      ncore_max = processor_count()
       if ncore.kind_of?(Integer)
         if ncore > 0
           @ncore = ncore
         else
-          @ncore = ncore_max + ncore
+          @ncore = processor_count() + ncore
         end
         if @ncore <= 0
           m = "Out of range: ncore=#{ncore.inspect}"
@@ -26,7 +34,7 @@ module Pwrake
           raise ArgumentError,m
         end
       elsif ncore.nil?
-        @ncore = ncore_max
+        @ncore = processor_count()
       else
         m = "Invalid argument: ncore=#{ncore.inspect}"
         @out.puts "ncore:"+m
@@ -37,32 +45,31 @@ module Pwrake
       Signal.trap("PIPE", "SIG_IGN")
     end
 
-    def get_line
-      begin
-        line = $stdin.gets
-        exit if !line
+    def get_line(io)
+      line = io.gets
+      if line
         line.chomp!
         line.strip!
         @log.info ">#{line}"
-        return line
-      rescue
-        exit
       end
+      return line
     end
 
     def run
       setup_option
-      if setup_loop
-        start_heartbeat
-        command_loop
-      end
+      setup_loop
+      @rd = Reader.new($stdin)
+      @selector.add_reader($stdin){command_callback(@rd)}
+      @selector.loop
+    rescue => exc
+      @log.error(([exc.to_s]+exc.backtrace).join("\n"))
     ensure
       close_all
     end
 
     def setup_option
       @log.info @option.inspect
-      @heartbeat_interval = @option[:heartbeat]
+      @out.heartbeat = @option[:heartbeat]
       @shell_cmd = @option[:shell_command]
       @shell_rc = @option[:shell_rc] || []
       (@option[:pass_env]||{}).each do |k,v|
@@ -71,64 +78,54 @@ module Pwrake
     end
 
     def setup_loop
-      while line = get_line
+      while line = get_line($stdin)
         case line
         when /^(\d+):open$/o
           $1.split.each do |id|
-            Executor.new(@dir_class,id,@shell_cmd,@shell_rc)
+            @ex_list[id] = Executor.new(@selector,@dir_class,id)
           end
         when "setup_end"
-          return true
+          return
         else
-          return false if common_line(line)
-        end
-      end
-      false
-    end
-
-    def start_heartbeat
-      if @heartbeat_interval
-        @heartbeat_thread = Thread.new do
-          while true
-            @out.puts "heartbeat"
-            sleep @heartbeat_interval
+          if common_line(line)
+            raise RuntimeError,"exit during setup_loop"
           end
         end
       end
+      raise RuntimeError,"incomplete setup_loop"
     end
 
-    def command_loop
-      while line = get_line
+    def command_callback(rd)
+      while line = get_line(rd) # rd returns nil if line is incomplete
         case line
         when /^(\d+):(.*)$/o
           id,cmd = $1,$2
-          ex = Executor::LIST[id]
-          if ex.nil?
-            if cmd=="exit"
-              @out.puts "#{id}:end"
-              next
-            else
-              ex = Executor.new(@dir_class,id,@shell_cmd,@shell_rc)
-            end
-          end
-          ex.execute(cmd)
+          @ex_list[id].execute(cmd.chomp)
         else
           break if common_line(line)
         end
       end
+      if rd.eof?
+        # connection lost
+        raise RuntimeError,"lost connection to master"
+      end
     end
 
     def common_line(line)
       case line
       when /^exit$/o
+        @selector.delete_reader($stdin)
         return true
         #
       when /^kill:(.*)$/o
-        kill_all($1)
+        sig = $1
+        sig = sig.to_i if /^\d+$/o =~ sig
+        @log.warn "killing worker, signal=#{sig}"
+        @ex_list.each{|id,ex| ex.kill(sig)}
         return false
         #
       when /^p$/o
-        puts "Executor::LIST = #{Executor::LIST.inspect}"
+        $stderr.puts "@ex_list = #{@ex_list.inspect}"
         return false
         #
       else
@@ -138,26 +135,13 @@ module Pwrake
       end
     end
 
-    def kill_all(sig)
-      sig = sig.to_i if /^\d+$/o =~ sig
-      @log.warn "worker_killed:signal=#{sig}"
-      Executor::LIST.each{|id,exc| exc.kill(sig)}
-    end
-
     def close_all
       @log.info "close_all"
       @heartbeat_thread.kill if @heartbeat_thread
       Dir.chdir
-      id_list = Executor::LIST.keys
-      ex_list = Executor::LIST.values
-      ex_list.each{|ex| ex.close}
-      begin
-        ex_list.each{|ex| ex.join}
-      rescue => e
-        @log.error e
-        @log.error e.backtrace.join("\n")
-      end
-      @log.info "worker:end:#{id_list.inspect}"
+      @ex_list.each_value{|ex| ex.close}
+      @ex_list.each_value{|ex| ex.join}
+      @log.info "worker:end:#{@ex_list.keys.inspect}"
       begin
         Timeout.timeout(20){@log.close}
       rescue => e

data/lib/pwrake/worker/log_executor.rb

@@ -4,8 +4,17 @@ require "logger"
 
 module Pwrake
 
+  DELEGATE_METHODS = [
+    :debug, :info, :error, :fatal, :warn, :unknown,
+    :debug?, :info?, :error?, :fatal?, :warn?, :unknown?,
+    :level, :level=,
+    :formatter, :formatter=,
+    :datetime_format, :datetime_format=
+  ]
+
   class DummyLogger
-    def method_missing(id,*args)
+    DELEGATE_METHODS.each do |m|
+      define_method(m){|*a|}
     end
   end
 
@@ -13,15 +22,10 @@ module Pwrake
     include Singleton
     extend Forwardable
 
-    def_delegators :@logger, :debug, :info, :error, :fatal, :warn, :unknown
-    def_delegators :@logger, :debug?, :info?, :error?, :fatal?, :warn?, :unknown?
-    def_delegators :@logger, :level, :level=
-    def_delegators :@logger, :formatter, :formatter=
-    def_delegators :@logger, :datetime_format, :datetime_format=
+    def_delegators :@logger, *DELEGATE_METHODS
 
     def initialize
       @level = ::Logger::DEBUG
-      #@logger = @logger_stderr = ::Logger.new($stderr)
       @logger = @logger_stderr = DummyLogger.new
       @logger.level = @level
     end

data/lib/pwrake/worker/reader.rb

@@ -0,0 +1,73 @@
+module Pwrake
+
+  class Reader
+
+    def initialize(io,mode="")
+      @io = io
+      @buf = ''
+      @eof = false
+      @mode = mode
+    end
+
+    attr_reader :io, :mode
+
+    def eof?
+      @eof && @buf.empty?
+    end
+
+    def gets
+      read_until("\n")
+    end
+
+    def read_until(sep="\r\n", chunk_size=8192)
+      until i = @buf.index(sep)
+        if s = _read(chunk_size)
+          @buf += s
+        else
+          if !@buf.empty? && @eof
+            buf = @buf; @buf = ''
+            return buf
+          else
+            return nil
+          end
+        end
+      end
+      @buf.slice!(0, i+sep.bytesize)
+    end
+
+    def _read(sz)
+      @io.read_nonblock(sz)
+    rescue EOFError
+      @eof = true
+      nil
+    rescue IO::WaitReadable
+      nil
+    end
+
+  end
+
+
+  class Selector
+
+    def initialize
+      @readers = {}
+    end
+
+    def add_reader(io,&callback)
+      @readers[io] = callback
+    end
+
+    def delete_reader(io)
+      @readers.delete(io)
+    end
+
+    def loop
+      while !@readers.empty?
+        r, = IO.select(@readers.keys,nil,nil)
+        r.each{|io| @readers[io].call} if r
+      end
+    end
+
+  end
+
+end

data/lib/pwrake/worker/writer.rb

@@ -1,4 +1,5 @@
 require "singleton"
+require "timeout"
 
 module Pwrake
 
@@ -8,14 +9,26 @@ module Pwrake
     def initialize
       @out = $stdout
       @mutex = Mutex.new
-      pipe_in, pipe_out = IO.pipe
-      Thread.new(pipe_in,"log:") do |pin,pre|
-        while s = pin.gets
-          s.chomp!
-          @out.puts pre+s
+      @queue = Queue.new
+      @heartbeat = 0
+      @thread = Thread.new do
+        loop do
+          begin
+            Timeout.timeout(@heartbeat) do
+              if s = @queue.deq
+                _puts s
+              end
+            end
+          rescue Timeout::Error
+            _puts "heartbeat"
+          end
         end
       end
-      $stderr = pipe_out
+    end
+
+    def heartbeat=(heartbeat)
+      @heartbeat = heartbeat
+      @queue.enq(nil)
     end
 
     def add_logger(log)
@@ -23,20 +36,13 @@ module Pwrake
     end
 
     def puts(s)
-      begin
-        @mutex.synchronize do
-          @out.print s+"\n"
-        end
-        @out.flush
-      rescue Errno::EPIPE
-      end
-      @log.info "<#{s}" if @log
+      @queue.enq(s)
     end
 
-    def print(s)
+    def _puts(s)
       begin
         @mutex.synchronize do
-          @out.print s
+          @out.print s+"\n"
         end
         @out.flush
       rescue Errno::EPIPE

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: pwrake
 version: !ruby/object:Gem::Version
-  version: 2.1.2
+  version: 2.1.3
 platform: ruby
 authors:
 - Masahiro TANAKA
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2016-07-10 00:00:00.000000000 Z
+date: 2016-12-24 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: parallel
@@ -87,6 +87,7 @@ files:
 - lib/pwrake/worker/invoker.rb
 - lib/pwrake/worker/load.rb
 - lib/pwrake/worker/log_executor.rb
+- lib/pwrake/worker/reader.rb
 - lib/pwrake/worker/shared_directory.rb
 - lib/pwrake/worker/worker_main.rb
 - lib/pwrake/worker/writer.rb
@@ -129,7 +130,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubyforge_project: 
-rubygems_version: 2.5.1
+rubygems_version: 2.6.8
 signing_key: 
 specification_version: 4
 summary: Parallel Workflow engine based on Rake
@@ -151,4 +152,3 @@ test_files:
 - spec/helper.rb
 - spec/hosts
 - spec/pwrake_spec.rb
-has_rdoc: