Class: Cnvrg::Helpers::Executer

Inherits:
Object
  • Object
show all
Defined in:
lib/cnvrg/helpers/executer.rb

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(owner: nil, machine_activity: nil, poll_every: 30, job_id: nil) ⇒ Executer

this class represent a machine_activity. it will poll the commands, communicate with the server (poll commands) and let the server know the status of this executer.



8
9
10
11
12
13
14
15
16
17
 
# File 'lib/cnvrg/helpers/executer.rb', line 8

def initialize(owner: nil, machine_activity: nil, poll_every: 30, job_id: nil)
  @owner = owner
  @job_id = job_id
  @poll_every = poll_every
  @machine_activity = machine_activity
  @commands_q = Queue.new
  @files_q = Queue.new
  @agent_id = nil
  @slave_id = nil
end

Instance Attribute Details

#agent_id ⇒ Object (readonly)

Returns the value of attribute agent_id.



3
4
5
 
# File 'lib/cnvrg/helpers/executer.rb', line 3

def agent_id
  @agent_id
end

#machine_activity ⇒ Object (readonly)

Returns the value of attribute machine_activity.



3
4
5
 
# File 'lib/cnvrg/helpers/executer.rb', line 3

def machine_activity
  @machine_activity
end

#slave_id ⇒ Object (readonly)

Returns the value of attribute slave_id.



3
4
5
 
# File 'lib/cnvrg/helpers/executer.rb', line 3

def slave_id
  @slave_id
end

Instance Method Details

#activity_url ⇒ Object 



33
34
35
 
# File 'lib/cnvrg/helpers/executer.rb', line 33

def activity_url
  ['users', @owner, 'machine_activities', @machine_activity].join("/")
end

#containers ⇒ Object 



74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
 
# File 'lib/cnvrg/helpers/executer.rb', line 74

def containers
  agent_id = nil
  slave_id = nil
  while agent_id.blank? or slave_id.blank?
    grep_by = @job_id
    grep_by = "$(hostname)" if ENV['KUBERNETES_PORT'].present?
    cntrs = `docker ps --format "table {{.ID}},{{.Names}}" | grep -i #{grep_by}`.split("\n").map{|x| x.strip}
    agent_id = cntrs.find{|container_name| container_name.include? "agent"}.split(",").first rescue nil
    slave_id = cntrs.find{|container_name| container_name.include? "slave"}.split(",").first rescue nil
    sleep(5)
  end
  if slave_id.blank?
    raise "Can't find slave id"
  end
  [agent_id, slave_id]
end

#create_file_cmd(path, content) ⇒ Object 



19
20
21
22
23
24
25
 
# File 'lib/cnvrg/helpers/executer.rb', line 19

def create_file_cmd(path, content)
  if path.include? "~"
    path = File.expand_path(path)
  end
  FileUtils.mkdir_p(File.dirname(path))
  File.open(path, "w+"){|f| f.write(content)}
end

#current_homedir ⇒ Object 



91
92
93
 
# File 'lib/cnvrg/helpers/executer.rb', line 91

def current_homedir
  `env | grep -w HOME`.strip.split("=").try(:last)
end

#execute_cmds ⇒ Object 



165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
 
# File 'lib/cnvrg/helpers/executer.rb', line 165

def execute_cmds
  pids = []
  while true
    if @commands_q.empty?
      sleep(5)
      next
    end
    cmd = @commands_q.pop.symbolize_keys
    command_json = Cnvrg::API.request([activity_url, "commands", cmd[:slug]].join('/'), "GET")

    cmd_status = command_json["status"] rescue ""

    if cmd_status == Cnvrg::Helpers::Agent::Status::ABORTED
      Cnvrg::Logger.log_info("stopping job because command #{cmd[:slug]} with status #{cmd_status}")
      next
    end
    pid = Process.fork do
      Cnvrg::Helpers::Agent.new(executer: self, **cmd).exec!
    end
    if cmd[:async].blank?
      Process.waitpid(pid)
    else
      Process.detach(pid)
    end
    pids << pid
    ######
  end
  pids
end

#executer_stats ⇒ Object 



37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
 
# File 'lib/cnvrg/helpers/executer.rb', line 37

def executer_stats
  return @stats if @stats.present?
  Cnvrg::Logger.log_info("getting containers")
  @agent_id, @slave_id = containers
  Cnvrg::Logger.log_info("got containers")
  pod_name, node_name = get_node_and_pod_names
  @stats = {
      pod_name: pod_name,
      node_name: node_name,
      agent: {
        container_id: @agent_id,
        workdir: `pwd`.strip,
        homedir: current_homedir,
        user: `whoami`.strip,
        user_id: `id -u`.strip,
        group_id: `id -g`.strip,
        cnvrg: Cnvrg::VERSION
      },
      slave: {
          container_id: @slave_id,
          workdir: run_in_slave('pwd'),
          homedir: slave_homedir,
          spark_path: spark_path,
          user: run_in_slave( 'whoami'),
          cnvrg: run_in_slave( 'which cnvrg'),
          has_bash: run_in_slave( 'which bash'),
          user_id: run_in_slave( 'id -u'),
          group_id: run_in_slave( 'id -g'),
          python_version: run_in_slave( 'python --version'),
          python3_version: run_in_slave( 'python3 --version'),
          pip_version: run_in_slave( 'pip --version'),
          pip3_version: run_in_slave( 'pip3 --version')
      },
  }
  @stats
end

#get_node_and_pod_names ⇒ Object 



200
201
202
203
204
205
206
207
208
209
 
# File 'lib/cnvrg/helpers/executer.rb', line 200

def get_node_and_pod_names
  pod_name = `hostname`.strip rescue nil
  node_name = nil
  if pod_name.present?
    pod_describe = `kubectl get pod #{pod_name} -o json` rescue nil
    pod_describe = JSON.parse(pod_describe) rescue {}
    node_name = pod_describe["spec"]["nodeName"] rescue nil
  end
  [pod_name, node_name]
end

#get_node_events(node_name) ⇒ Object 



223
224
225
226
 
# File 'lib/cnvrg/helpers/executer.rb', line 223

def get_node_events(node_name)
  return if node_name.blank?
  `kubectl get event --all-namespaces --field-selector involvedObject.name=#{node_name} -o json`
end

#get_pod_events(pod_name) ⇒ Object 



218
219
220
221
 
# File 'lib/cnvrg/helpers/executer.rb', line 218

def get_pod_events(pod_name)
  return if pod_name.blank?
  `kubectl get event --field-selector involvedObject.name=#{pod_name} -o json`
end

#handle_files(files) ⇒ Object 



27
28
29
30
31
 
# File 'lib/cnvrg/helpers/executer.rb', line 27

def handle_files(files)
  (files || {}).each do |path, content|
    create_file_cmd(path, content)
  end
end

#init ⇒ Object 



122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
 
# File 'lib/cnvrg/helpers/executer.rb', line 122

def init
  retries = 0
  success = false
  puts("Agent started, connecting to #{Cnvrg::API.get_api}")
  STDOUT.flush
  while !success and retries < 100
    begin
      resp = Cnvrg::API.request(activity_url, "PUT", {stats: executer_stats})
      machine_activity = resp["machine_activity"]
      success = true
      puts("Connected to server")
      STDOUT.flush
      Cnvrg::Logger.log_info("Got back machine activity #{machine_activity}")
      if machine_activity.present? and @machine_activity != machine_activity
        Cnvrg::Logger.log_info("Changing to machine activity #{machine_activity}")
        machine_activity_yml = {slug: machine_activity}
        File.open("/conf/.machine_activity.yml", "w+") {|f| f.write machine_activity_yml.to_yaml}
        @machine_activity = machine_activity
      end
    rescue => e
      Cnvrg::Logger.log_error(e)
      Cnvrg::Logger.info("Sleeping for #{5 * retries}")
      sleep(5 * retries)
      retries +=1
    end
  end
end

#main_thread ⇒ Object 



157
158
159
160
161
162
163
 
# File 'lib/cnvrg/helpers/executer.rb', line 157

def main_thread
  init
  Thread.new do
    polling_thread
  end
  execute_cmds
end

#merge_log_block(logs) ⇒ Object 



195
196
197
198
 
# File 'lib/cnvrg/helpers/executer.rb', line 195

def merge_log_block(logs)
  logs.group_by {|log| log[:timestamp].to_s}
      .map {|ts, logz| {timestamp: ts, logs: logz.map {|l| l[:log]}.join("\n")}}
end

#poll ⇒ Object 



112
113
114
115
116
117
118
119
120
 
# File 'lib/cnvrg/helpers/executer.rb', line 112

def poll
  resp = Cnvrg::API.request([activity_url, "commands"].join('/'), "POST")
  commands = resp["commands"]
  files = resp["files"]
  handle_files(files)
  commands.each{|cmd| @commands_q.push(cmd)}
rescue => e
  Cnvrg::Logger.log_error(e)
end

#polling_thread ⇒ Object 



150
151
152
153
154
155
 
# File 'lib/cnvrg/helpers/executer.rb', line 150

def polling_thread
  while true
    poll
    sleep(@poll_every)
  end
end

#pre_pod_stop ⇒ Object 



211
212
213
214
215
216
 
# File 'lib/cnvrg/helpers/executer.rb', line 211

def pre_pod_stop
  pod_name, node_name = get_node_and_pod_names
  pod_events = get_pod_events(pod_name)
  node_events = get_node_events(node_name)
  Cnvrg::API.request([activity_url, "job_events"].join('/'), "POST", {pod_events: pod_events, node_events: node_events})
end

#run_in_slave(command) ⇒ Object 



107
108
109
 
# File 'lib/cnvrg/helpers/executer.rb', line 107

def run_in_slave(command)
  `docker exec -i #{@slave_id} sh -c '#{command}'`.strip
end

#slave_env ⇒ Object 



103
104
105
 
# File 'lib/cnvrg/helpers/executer.rb', line 103

def slave_env
  run_in_slave("env").split("\n").map{|x| x.split("=")}
end

#slave_homedir ⇒ Object 



99
100
101
 
# File 'lib/cnvrg/helpers/executer.rb', line 99

def slave_homedir()
  run_in_slave("env | grep -w HOME").split("=").try(:last)
end

#spark_path ⇒ Object 



95
96
97
 
# File 'lib/cnvrg/helpers/executer.rb', line 95

def spark_path
  run_in_slave("env | grep SPARK_HOME").strip.split("=").try(:last)
end