kcluster 1.1.0

data/bin/kcluster

@@ -0,0 +1,255 @@
+#!/usr/bin/env ruby
+
+$:.push(File.dirname($0))
+require 'optparse'
+require 'socket'
+require 'net/http'
+require 'json'
+
+# FIXME this really only works inside twitter. it would be nice to support
+# a general serverset interface (or opensource colony).
+$colony_host = ENV["COLONY_HOST"] || "colony.smf1.twitter.com"
+$colony_port = ENV["COLONY_PORT"] || 9080
+
+$options = {
+  :config_filename => ENV['HOME'] + "/.kestrel_cluster",
+  :server_list => [],
+  :port => 22133,
+  :verbose => false,
+}
+
+def verbose(s)
+  puts s if $options[:verbose]
+end
+
+def fetch_stats(host, port, data)
+  verbose "--- Fetching stats from #{host}:#{port}"
+  sock = TCPSocket.open(host, port)
+  sock.puts("stats")
+  done = false
+  while !done && line = sock.gets.chomp
+    if (line == 'END') then
+      done = true
+    elsif line =~ /STAT queue_([\w\+\-]+) (\d+)/
+      key = $1
+      value = $2.to_i
+      (stat, queue_name) = case key
+        when /([\w\+\-]+)_total_items/   then [:total_items, $1]
+        when /([\w\+\-]+)_expired_items/ then [:expired_items, $1]
+        when /([\w\+\-]+)_mem_items/     then [:mem_items, $1]
+        when /([\w\+\-]+)_items/         then [:items, $1]
+        when /([\w\+\-]+)_mem_bytes/     then [:mem_bytes, $1]
+        when /([\w\+\-]+)_bytes/         then [:bytes, $1]
+        when /([\w\+\-]+)_age/           then [:age, $1]
+      end
+
+      if (queue_name)
+        queue_name = queue_name.split('+', 2).first if $options[:rollup_fanouts]
+
+        if (stat == :age)
+          data[:min_age][queue_name] = value if value < data[:min_age][queue_name]
+          data[:max_age][queue_name] = value if value > data[:max_age][queue_name]
+        else
+          data[stat][queue_name] += value
+        end
+      end
+    end
+  end
+  sock.close
+end
+
+def fetch_all
+  data = Hash.new do |h, data_type|
+    h[data_type] = Hash.new do |h, k|
+      h[k] = (data_type == :min_age) ? 2**31 : 0
+    end
+  end
+  $options[:server_list].each do |server|
+    begin
+      fetch_stats(server, $options[:port], data)
+    rescue => e
+      puts "Could not connect to host #{server}: #{e}"
+    end
+  end
+  data
+end
+
+def sort_data(data)
+  rv = {}
+  data.each { |key, v| rv[key] = v.sort_by { |h| [ h[1], h[0] ] } }
+  rv
+end
+
+def report(data, key)
+  puts ""
+  format = "%14s %s\n"
+  printf(format, key, "queue")
+  printf(format, "============", "====================")
+  stats = data[key] || {}
+  stats.each { |queue, value| printf("%14d %s\n", value, queue) }
+end
+
+def report_all(data, keys)
+  keys.each { |key| report(data, key) }
+  puts ""
+end
+
+def delete_all(queue_name)
+  $options[:server_list].each do |server|
+    print "--- Deleting queue #{queue_name} from #{server}:#{$options[:port]} ... "
+    STDOUT.flush
+    sock = TCPSocket.open(server, $options[:port])
+    sock.puts("delete " + queue_name)
+    puts sock.readline.chomp
+    sock.close
+  end
+  puts "Done."
+end
+
+def flush_all(queue_name)
+  $options[:server_list].each do |server|
+    print "--- Flushing queue #{queue_name} from #{server}:#{$options[:port]} ... "
+    STDOUT.flush
+    sock = TCPSocket.open(server, $options[:port])
+    sock.puts("flush " + queue_name)
+    puts sock.readline.chomp
+    sock.close
+  end
+  puts "Done."
+end
+
+def keep_unchanged_data(last, current)
+  rv = {}
+  current.keys.each { |key| rv[key] = keep_unchanged_hash(last[key], current[key]) }
+  rv
+end
+
+def keep_unchanged_hash(last, current)
+  rv = {}
+  last.each do |key, value|
+    rv[key] = value if current[key] == value
+  end
+  rv
+end
+
+# dump queues that appear to have neither :items nor :total_items changing.
+def find_stale(rounds)
+  puts "Check 1..."
+  last = fetch_all
+  (1...rounds).each do |i|
+    sleep 5
+    puts "Check #{i + 1}..."
+    current = fetch_all
+    last = keep_unchanged_data(last, current)
+  end
+
+  stale = (last[:total_items].keys & last[:items].keys).sort!
+  sorted = sort_data(last)
+
+  puts ""
+  puts "stale queue prediction"
+  puts "======================"
+  printf("%11s %11s %s\n", "total_items", "items", "queue")
+  printf("%11s %11s %s\n", "-----------", "-----------", "--------------------")
+  stale.each do |queue_name|
+    items = last[:items][queue_name]
+    total_items = last[:total_items][queue_name]
+    printf("%11d %11d %s\n", total_items, items, queue_name)
+  end
+end
+
+
+# ----------
+
+parser = OptionParser.new do |opts|
+  opts.banner = "Usage: #{$0} [options] <command>"
+  opts.separator "Example: #{$0} -f kestrel max_age"
+
+  opts.on("-f", "--file=FILENAME", "load kestrel server list from file (use '-' for stdin) (default: #{$options[:config_filename]})") do |filename|
+    $options[:config_filename] = filename
+  end
+  opts.on("-p", "--port=N", "use port (default: #{$options[:port]})") do |port|
+    $options[:port] = port.to_i
+  end
+  opts.on("-r", "--role=ROLE", "load kestrel server list from colony audubon role (requires colony)") do |role|
+    $options[:role] = role
+  end
+  opts.on("-v", "--verbose", "print verbose debugging info as we work") do
+    $options[:verbose] = true 
+  end
+  opts.on("-F", "--rollup-fanouts", "roll up stats for fanout queues into a single count") do
+    $options[:rollup_fanouts] = true
+  end
+
+  opts.separator ""
+  opts.separator "Commands:"
+  opts.separator "    all             show all of the stats summaries below (items, etc)"
+  opts.separator "    items           show item counts per queue"
+  opts.separator "    bytes           show byte counts per queue"
+  opts.separator "    mem             show in-memory byte counts per queue"
+  opts.separator "    min_age         show minimum age (msec) per queue"
+  opts.separator "    max_age         show maximum age (msec) per queue"
+  opts.separator ""
+  opts.separator "    delete <name>   remove a queue across the cluster"
+  opts.separator "    flush <name>    drain a queue across the cluster"
+  opts.separator "    stale [rounds]  try to detect stale queues (default: 5 rounds)"
+  opts.separator ""
+end
+
+parser.parse!(ARGV)
+
+if ARGV.size < 1
+  puts
+  puts parser
+  exit 1
+end
+
+if $options[:role]
+  data = Net::HTTP.get URI.parse("http://#{$colony_host}:#{$colony_port}/query?q=mo+audubon.role.#{$options[:role]}")
+  $options[:server_list] = JSON.parse(data).values.flatten
+else
+  server_file = $options[:config_filename] == "-" ? STDIN : File.open($options[:config_filename], "r")
+  $options[:server_list] = server_file.readlines.map { |line| line.chomp }
+  server_file.close
+end
+
+command = ARGV[0].downcase
+
+if command == "all"
+  $show = [ :items, :bytes, :mem_bytes, :min_age, :max_age ]
+elsif command == "items"
+  $show = [ :items ]
+elsif command == "bytes"
+  $show = [ :bytes ]
+elsif command == "mem"
+  $show = [ :mem_bytes ]
+elsif command == "min_age"
+  $show = [ :min_age ]
+elsif command == "max_age"
+  $show = [ :max_age ]
+elsif command == "delete" && ARGV.size == 2
+  queue_name = ARGV[1]
+  print "Really delete #{queue_name}? (yes/no) "
+  STDOUT.flush
+  delete_all(queue_name) if STDIN.gets.chomp == 'yes'
+  exit 0
+elsif command == "flush" && ARGV.size == 2
+  queue_name = ARGV[1]
+  print "Really flush #{queue_name}? (yes/no) "
+  STDOUT.flush
+  flush_all(queue_name) if STDIN.gets.chomp == 'yes'
+  exit 0
+elsif command == "stale"
+  rounds = ARGV[1] ? ARGV[1].to_i : 5
+  find_stale(rounds)
+  exit 0
+end
+
+if $show
+  report_all(sort_data(fetch_all), $show)
+  exit 0
+else
+  puts
+  puts parser
+  exit 1
+end

metadata

@@ -0,0 +1,83 @@
+--- !ruby/object:Gem::Specification 
+name: kcluster
+version: !ruby/object:Gem::Version 
+  hash: 19
+  prerelease: 
+  segments: 
+  - 1
+  - 1
+  - 0
+  version: 1.1.0
+platform: ruby
+authors: 
+- Robey Pointer
+autorequire: 
+bindir: bin
+cert_chain: []
+
+date: 2012-03-02 00:00:00 Z
+dependencies: 
+- !ruby/object:Gem::Dependency 
+  name: json
+  prerelease: false
+  requirement: &id001 !ruby/object:Gem::Requirement 
+    none: false
+    requirements: 
+    - - ">="
+      - !ruby/object:Gem::Version 
+        hash: 3
+        segments: 
+        - 0
+        version: "0"
+  type: :runtime
+  version_requirements: *id001
+description: |
+  A handy little CLI tool for browsing quick stats across your kestrel cluster.
+  Kcluster collects stats from each machine in a cluster, then summarizes them
+  into one table.
+
+email: robeypointer@gmail.com
+executables: 
+- kcluster
+extensions: []
+
+extra_rdoc_files: []
+
+files: 
+- lib/nothing.rb
+- bin/kcluster
+homepage: http://robey.github.com/kestrel
+licenses: []
+
+post_install_message: 
+rdoc_options: []
+
+require_paths: 
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement 
+  none: false
+  requirements: 
+  - - ">="
+    - !ruby/object:Gem::Version 
+      hash: 3
+      segments: 
+      - 0
+      version: "0"
+required_rubygems_version: !ruby/object:Gem::Requirement 
+  none: false
+  requirements: 
+  - - ">="
+    - !ruby/object:Gem::Version 
+      hash: 3
+      segments: 
+      - 0
+      version: "0"
+requirements: []
+
+rubyforge_project: 
+rubygems_version: 1.8.15
+signing_key: 
+specification_version: 3
+summary: Kestrel cluster management tool
+test_files: []
+