xpflow 0.1b
Sign up to get free protection for your applications and to get access to all the features.
- data/bin/xpflow +96 -0
- data/lib/colorado.rb +198 -0
- data/lib/json/add/core.rb +243 -0
- data/lib/json/add/rails.rb +8 -0
- data/lib/json/common.rb +423 -0
- data/lib/json/editor.rb +1369 -0
- data/lib/json/ext.rb +28 -0
- data/lib/json/pure/generator.rb +442 -0
- data/lib/json/pure/parser.rb +320 -0
- data/lib/json/pure.rb +15 -0
- data/lib/json/version.rb +8 -0
- data/lib/json.rb +62 -0
- data/lib/mime/types.rb +881 -0
- data/lib/mime-types.rb +3 -0
- data/lib/restclient/abstract_response.rb +106 -0
- data/lib/restclient/exceptions.rb +193 -0
- data/lib/restclient/net_http_ext.rb +55 -0
- data/lib/restclient/payload.rb +235 -0
- data/lib/restclient/raw_response.rb +34 -0
- data/lib/restclient/request.rb +316 -0
- data/lib/restclient/resource.rb +169 -0
- data/lib/restclient/response.rb +24 -0
- data/lib/restclient.rb +174 -0
- data/lib/xpflow/bash.rb +341 -0
- data/lib/xpflow/bundle.rb +113 -0
- data/lib/xpflow/cmdline.rb +249 -0
- data/lib/xpflow/collection.rb +122 -0
- data/lib/xpflow/concurrency.rb +79 -0
- data/lib/xpflow/data.rb +393 -0
- data/lib/xpflow/dsl.rb +816 -0
- data/lib/xpflow/engine.rb +574 -0
- data/lib/xpflow/ensemble.rb +135 -0
- data/lib/xpflow/events.rb +56 -0
- data/lib/xpflow/experiment.rb +65 -0
- data/lib/xpflow/exts/facter.rb +30 -0
- data/lib/xpflow/exts/g5k.rb +931 -0
- data/lib/xpflow/exts/g5k_use.rb +50 -0
- data/lib/xpflow/exts/gui.rb +140 -0
- data/lib/xpflow/exts/model.rb +155 -0
- data/lib/xpflow/graph.rb +1603 -0
- data/lib/xpflow/graph_xpflow.rb +251 -0
- data/lib/xpflow/import.rb +196 -0
- data/lib/xpflow/library.rb +349 -0
- data/lib/xpflow/logging.rb +153 -0
- data/lib/xpflow/manager.rb +147 -0
- data/lib/xpflow/nodes.rb +1250 -0
- data/lib/xpflow/runs.rb +773 -0
- data/lib/xpflow/runtime.rb +125 -0
- data/lib/xpflow/scope.rb +168 -0
- data/lib/xpflow/ssh.rb +186 -0
- data/lib/xpflow/stat.rb +50 -0
- data/lib/xpflow/stdlib.rb +381 -0
- data/lib/xpflow/structs.rb +369 -0
- data/lib/xpflow/taktuk.rb +193 -0
- data/lib/xpflow/templates/ssh-config.basic +14 -0
- data/lib/xpflow/templates/ssh-config.inria +18 -0
- data/lib/xpflow/templates/ssh-config.proxy +13 -0
- data/lib/xpflow/templates/taktuk +6590 -0
- data/lib/xpflow/templates/utils/batch +4 -0
- data/lib/xpflow/templates/utils/bootstrap +12 -0
- data/lib/xpflow/templates/utils/hostname +3 -0
- data/lib/xpflow/templates/utils/ping +3 -0
- data/lib/xpflow/templates/utils/rsync +12 -0
- data/lib/xpflow/templates/utils/scp +17 -0
- data/lib/xpflow/templates/utils/scp_many +8 -0
- data/lib/xpflow/templates/utils/ssh +3 -0
- data/lib/xpflow/templates/utils/ssh-interactive +4 -0
- data/lib/xpflow/templates/utils/taktuk +19 -0
- data/lib/xpflow/threads.rb +187 -0
- data/lib/xpflow/utils.rb +569 -0
- data/lib/xpflow/visual.rb +230 -0
- data/lib/xpflow/with_g5k.rb +7 -0
- data/lib/xpflow.rb +349 -0
- metadata +135 -0
data/lib/xpflow/nodes.rb
ADDED
@@ -0,0 +1,1250 @@
|
|
1
|
+
|
2
|
+
require 'erb'
|
3
|
+
require 'ostruct'
|
4
|
+
require 'xpflow/exts/g5k'
|
5
|
+
require 'yaml'
|
6
|
+
require 'thread'
|
7
|
+
require 'shellwords'
|
8
|
+
|
9
|
+
def get_g5k_username
|
10
|
+
raise "No G5K username!" if $g5k_user.nil?
|
11
|
+
return $g5k_user
|
12
|
+
end
|
13
|
+
|
14
|
+
module XPFlow
|
15
|
+
|
16
|
+
# manages all nodes
|
17
|
+
|
18
|
+
class NodesManager
|
19
|
+
|
20
|
+
def initialize(directory)
|
21
|
+
@directory = directory
|
22
|
+
@mutex = Mutex.new
|
23
|
+
@node_counter = 0
|
24
|
+
end
|
25
|
+
|
26
|
+
def synchronize(&block)
|
27
|
+
return @mutex.synchronize(&block)
|
28
|
+
end
|
29
|
+
|
30
|
+
def subdir(name)
|
31
|
+
return @directory.subdir(name)
|
32
|
+
end
|
33
|
+
|
34
|
+
def get_node(user, host, factory, opts = {})
|
35
|
+
synchronize do
|
36
|
+
@node_counter += 1
|
37
|
+
node_directory = subdir("#{host}--#{user}--#{factory.name}--#{@node_counter}")
|
38
|
+
opts[:factory] = factory
|
39
|
+
factory.build(user, host, node_directory, opts)
|
40
|
+
end
|
41
|
+
end
|
42
|
+
|
43
|
+
end
|
44
|
+
|
45
|
+
class SimpleNodeFactory
|
46
|
+
# build a directly-reachable host
|
47
|
+
|
48
|
+
def name
|
49
|
+
return "normal"
|
50
|
+
end
|
51
|
+
|
52
|
+
def build(*args)
|
53
|
+
return SimpleNode.new(*args)
|
54
|
+
end
|
55
|
+
|
56
|
+
end
|
57
|
+
|
58
|
+
class G5KNodeFactory
|
59
|
+
|
60
|
+
def name
|
61
|
+
return "grid5000"
|
62
|
+
end
|
63
|
+
|
64
|
+
def build(user, host, node_directory, opts)
|
65
|
+
opts = { :group => _get_group(host) }.merge(opts)
|
66
|
+
return G5KNode.new(user, host, node_directory, opts)
|
67
|
+
end
|
68
|
+
|
69
|
+
def _get_group(host)
|
70
|
+
m = /^(\w+)-(\d+).+$/.match(host) # <cluster>-<nodeid> ...
|
71
|
+
if m
|
72
|
+
return m.captures.first
|
73
|
+
else
|
74
|
+
return nil
|
75
|
+
end
|
76
|
+
end
|
77
|
+
|
78
|
+
end
|
79
|
+
|
80
|
+
class ProxiedFactory
|
81
|
+
|
82
|
+
def initialize(node)
|
83
|
+
@node = node
|
84
|
+
end
|
85
|
+
|
86
|
+
def name
|
87
|
+
return "proxy"
|
88
|
+
end
|
89
|
+
|
90
|
+
def build(user, host, directory, opts)
|
91
|
+
opts[:proxy] = @node
|
92
|
+
return ProxiedNode.new(user, host, directory, opts)
|
93
|
+
end
|
94
|
+
|
95
|
+
end
|
96
|
+
|
97
|
+
class AbstractNode
|
98
|
+
|
99
|
+
# a node that is installed using a set of templates
|
100
|
+
|
101
|
+
attr_reader :user
|
102
|
+
attr_reader :host
|
103
|
+
attr_reader :directory
|
104
|
+
|
105
|
+
def initialize(user, host, directory, opts = {})
|
106
|
+
@user = user
|
107
|
+
@host = host
|
108
|
+
@directory = directory
|
109
|
+
@mutex = Mutex.new
|
110
|
+
@opts = opts
|
111
|
+
|
112
|
+
__setup__()
|
113
|
+
end
|
114
|
+
|
115
|
+
def options
|
116
|
+
return @opts
|
117
|
+
end
|
118
|
+
|
119
|
+
def domain
|
120
|
+
# gets accessibility domain of a node (nodes from within one domain
|
121
|
+
# are pairwise accessible and reachable)
|
122
|
+
return @opts[:factory].name
|
123
|
+
end
|
124
|
+
|
125
|
+
def synchronize(&block)
|
126
|
+
return @mutex.synchronize(&block)
|
127
|
+
end
|
128
|
+
|
129
|
+
def group
|
130
|
+
return @opts[:group]
|
131
|
+
end
|
132
|
+
|
133
|
+
def userhost
|
134
|
+
return "#{@user}@#{@host}"
|
135
|
+
end
|
136
|
+
|
137
|
+
def tmpfile
|
138
|
+
return execute("mktemp").stdout.strip
|
139
|
+
end
|
140
|
+
|
141
|
+
def to_s
|
142
|
+
return "#{self.class}('#{userhost}')"
|
143
|
+
end
|
144
|
+
|
145
|
+
def path
|
146
|
+
return @directory.path
|
147
|
+
end
|
148
|
+
|
149
|
+
def md5sum(files)
|
150
|
+
h = {}
|
151
|
+
output = execute("md5sum #{files.join(' ')}").stdout.strip
|
152
|
+
output.lines.each do |line|
|
153
|
+
hash, filename = line.split
|
154
|
+
h[filename] = hash
|
155
|
+
end
|
156
|
+
return h
|
157
|
+
end
|
158
|
+
|
159
|
+
def run(*args)
|
160
|
+
return @directory.run(*args)
|
161
|
+
end
|
162
|
+
|
163
|
+
def execute(cmd, env = {}, opts = {})
|
164
|
+
# opts will become an env of execution
|
165
|
+
wd = opts[:wd]
|
166
|
+
env = env.each_pair.select { |k, v| v.is_a?(String) }.map { |k, v| [ k.to_s, v ] }
|
167
|
+
env = Hash[env]
|
168
|
+
return @directory.run_ssh(cmd, :env => env, :wd => wd)
|
169
|
+
end
|
170
|
+
|
171
|
+
def execute_with_files(cmd, out, err)
|
172
|
+
return @directory.run_ssh(cmd, :out => out, :err => err)
|
173
|
+
end
|
174
|
+
|
175
|
+
def ping()
|
176
|
+
@directory.run("ping")
|
177
|
+
end
|
178
|
+
|
179
|
+
def hostname()
|
180
|
+
@directory.run("hostname")
|
181
|
+
end
|
182
|
+
|
183
|
+
def scp(from, to)
|
184
|
+
@directory.run("scp #{from} #{to}")
|
185
|
+
end
|
186
|
+
|
187
|
+
def scp_many(files, to_dir)
|
188
|
+
args = files.join(" ") # TODO: escaping?
|
189
|
+
return @directory.run("scp_many #{to_dir} #{args}")
|
190
|
+
end
|
191
|
+
|
192
|
+
def file(path, &block)
|
193
|
+
# creates a file here and then does scp to the node
|
194
|
+
filename = @directory.mktemp(&block)
|
195
|
+
scp(filename, path)
|
196
|
+
end
|
197
|
+
|
198
|
+
def proxy_factory
|
199
|
+
# generates a node factory for nodes proxied through this one
|
200
|
+
return ProxiedFactory.new(self)
|
201
|
+
end
|
202
|
+
|
203
|
+
def context
|
204
|
+
return {
|
205
|
+
:user => self.user,
|
206
|
+
:host => self.host,
|
207
|
+
:path => self.path
|
208
|
+
}
|
209
|
+
end
|
210
|
+
|
211
|
+
end
|
212
|
+
|
213
|
+
class TemplateNode < AbstractNode
|
214
|
+
|
215
|
+
def __setup__
|
216
|
+
raise "No template!" if template_name().nil?
|
217
|
+
NodeUtils.install_templates(path(), template_name(), context)
|
218
|
+
end
|
219
|
+
|
220
|
+
end
|
221
|
+
|
222
|
+
class SimpleNode < TemplateNode
|
223
|
+
|
224
|
+
# a simple, directly reachable node
|
225
|
+
def template_name
|
226
|
+
"basic"
|
227
|
+
end
|
228
|
+
|
229
|
+
end
|
230
|
+
|
231
|
+
class ProxiedNode < TemplateNode
|
232
|
+
|
233
|
+
def template_name
|
234
|
+
"proxy"
|
235
|
+
end
|
236
|
+
|
237
|
+
def context
|
238
|
+
return super.merge({ :proxy => "#{@opts[:proxy].path}/ssh" })
|
239
|
+
end
|
240
|
+
|
241
|
+
end
|
242
|
+
|
243
|
+
class G5KNode < AbstractNode
|
244
|
+
|
245
|
+
def labels
|
246
|
+
# TODO: I don't know if it is a good way to do it
|
247
|
+
sites = %w{bordeaux grenoble lille luxembourg lyon nancy reims rennes sophia toulouse}
|
248
|
+
site = sites.select { |x| @host.include?(x) }.map(&:to_sym)
|
249
|
+
return [ :g5k ] + site[0..0]
|
250
|
+
end
|
251
|
+
|
252
|
+
def self.inside_g5k
|
253
|
+
return $hostname.end_with?('grid5000.fr')
|
254
|
+
end
|
255
|
+
|
256
|
+
def __setup__
|
257
|
+
# here we make some magic; there are 3 cases:
|
258
|
+
# 1. Inside grid5000, we make efficient bootstrap using basic-templates
|
259
|
+
# 2. Inside inria, we have efficient access to G5K
|
260
|
+
# 3. Really outside (``chez moi'' for example)
|
261
|
+
|
262
|
+
is_g5k = G5KNode.inside_g5k()
|
263
|
+
return install_inside_g5k() if is_g5k
|
264
|
+
if File.exist?("/etc/resolv.conf")
|
265
|
+
resolv = IO.read("/etc/resolv.conf").lines.
|
266
|
+
select { |line| line.start_with?("domain ") or line.start_with?("search ") }
|
267
|
+
domain = resolv.first
|
268
|
+
if domain.nil? == false
|
269
|
+
domain = domain.split[1]
|
270
|
+
gw = inria_gateway(domain)
|
271
|
+
return install_inside_inria(gw) if (gw.nil? == false)
|
272
|
+
end
|
273
|
+
end
|
274
|
+
return install_generic()
|
275
|
+
end
|
276
|
+
|
277
|
+
def inria_gateway(domain)
|
278
|
+
# more stuff has to be implemented here
|
279
|
+
return "grid5000.loria.fr" if domain == "loria.fr"
|
280
|
+
return nil
|
281
|
+
end
|
282
|
+
|
283
|
+
def install_inside_g5k
|
284
|
+
NodeUtils.install_templates(path(), "basic",
|
285
|
+
context.merge({ :g5k_user => get_g5k_username() })
|
286
|
+
)
|
287
|
+
end
|
288
|
+
|
289
|
+
def install_inside_inria(gw)
|
290
|
+
NodeUtils.install_templates(path(), "inria",
|
291
|
+
context.merge({ :g5k_user => get_g5k_username(), :gw => gw })
|
292
|
+
)
|
293
|
+
end
|
294
|
+
|
295
|
+
def install_generic
|
296
|
+
NodeUtils.install_templates(path(), "inria",
|
297
|
+
context.merge({ :g5k_user => get_g5k_username(), :gw => "access.grid5000.fr" })
|
298
|
+
)
|
299
|
+
end
|
300
|
+
|
301
|
+
def self.kavlan(job, manager)
|
302
|
+
link = job["links"].select { |x| x["rel"] == "parent" }.first["href"]
|
303
|
+
site = link.split("/").last
|
304
|
+
front = manager.get_node(get_g5k_username(), "#{site}.grid5000.fr", G5KNodeFactory.new)
|
305
|
+
ns = job["assigned_nodes"]
|
306
|
+
begin
|
307
|
+
ns = front.execute("kavlan -l -j #{job["uid"]}").lines.map(&:strip)
|
308
|
+
rescue
|
309
|
+
nil
|
310
|
+
end
|
311
|
+
return ns.map { |x| manager.get_node("root", x, G5KNodeFactory.new) }
|
312
|
+
end
|
313
|
+
|
314
|
+
def self.get_ssh_config
|
315
|
+
return %x(echo ~/.ssh/config).strip
|
316
|
+
end
|
317
|
+
|
318
|
+
def self.obtain_ssh_pubkey_path
|
319
|
+
return %x(echo ~/.ssh/id_rsa.pub).strip
|
320
|
+
end
|
321
|
+
|
322
|
+
end
|
323
|
+
|
324
|
+
### TYPES OF RESULTS ###
|
325
|
+
|
326
|
+
class BasicRemoteResult
|
327
|
+
|
328
|
+
attr_reader :opts
|
329
|
+
attr_reader :node
|
330
|
+
|
331
|
+
def initialize(node, cmd, stdout, stderr, opts = {})
|
332
|
+
@node = node
|
333
|
+
@cmd = cmd
|
334
|
+
@stdout = stdout
|
335
|
+
@stderr = stderr
|
336
|
+
@opts = opts
|
337
|
+
end
|
338
|
+
|
339
|
+
def stdout
|
340
|
+
return @stdout
|
341
|
+
end
|
342
|
+
|
343
|
+
def stderr
|
344
|
+
return @stderr
|
345
|
+
end
|
346
|
+
|
347
|
+
def command
|
348
|
+
return @cmd
|
349
|
+
end
|
350
|
+
|
351
|
+
def save_stdout(filename)
|
352
|
+
IO.write(filename, @stdout)
|
353
|
+
end
|
354
|
+
|
355
|
+
def save_stderr(filename)
|
356
|
+
IO.write(filename, @stderr)
|
357
|
+
end
|
358
|
+
|
359
|
+
def to_s
|
360
|
+
return "BasicRemoteResult('#{@cmd}' on #{@node})"
|
361
|
+
end
|
362
|
+
|
363
|
+
end
|
364
|
+
|
365
|
+
class FileRemoteResult < BasicRemoteResult
|
366
|
+
|
367
|
+
def initialize(node, cmd, stdout, stderr, opts = {})
|
368
|
+
# however! stdout & stderr are *paths*!
|
369
|
+
super
|
370
|
+
end
|
371
|
+
|
372
|
+
def stdout
|
373
|
+
return @node.execute("cat #{@stdout}").stdout
|
374
|
+
end
|
375
|
+
|
376
|
+
def stderr
|
377
|
+
return @node.execute("cat #{@stderr}").stdout
|
378
|
+
end
|
379
|
+
|
380
|
+
def stdout_file
|
381
|
+
return @stdout
|
382
|
+
end
|
383
|
+
|
384
|
+
def stderr_file
|
385
|
+
return @stderr
|
386
|
+
end
|
387
|
+
|
388
|
+
def save_stdout(filename)
|
389
|
+
return @node.execute_with_files("cat #{@stdout}", filename, "/dev/null")
|
390
|
+
end
|
391
|
+
|
392
|
+
def save_stderr(filename)
|
393
|
+
return @node.execute_with_files("cat #{@stderr}", filename, "/dev/null")
|
394
|
+
end
|
395
|
+
|
396
|
+
def to_s
|
397
|
+
return "FileRemoteResult('#{@cmd}' on #{@node}, out => #{@stdout}, err => #{@stderr})"
|
398
|
+
end
|
399
|
+
|
400
|
+
end
|
401
|
+
|
402
|
+
class FileRemote
|
403
|
+
# a remote file (but usually already local)
|
404
|
+
# result is probably: LocalExecutionResult
|
405
|
+
|
406
|
+
attr_reader :path
|
407
|
+
attr_reader :result
|
408
|
+
|
409
|
+
def initialize(path, result)
|
410
|
+
@path = path
|
411
|
+
@result = result
|
412
|
+
end
|
413
|
+
|
414
|
+
end
|
415
|
+
|
416
|
+
class ManyExecutionResult
|
417
|
+
|
418
|
+
def initialize(list, cmd)
|
419
|
+
@list = list
|
420
|
+
@command = cmd
|
421
|
+
end
|
422
|
+
|
423
|
+
def to_list
|
424
|
+
return @list
|
425
|
+
end
|
426
|
+
|
427
|
+
def length
|
428
|
+
return @list.length
|
429
|
+
end
|
430
|
+
|
431
|
+
def to_s
|
432
|
+
return "ManyResult('#{@command}' on #{@list.length} nodes)"
|
433
|
+
end
|
434
|
+
|
435
|
+
def each(&block)
|
436
|
+
return @list.each(&block)
|
437
|
+
end
|
438
|
+
|
439
|
+
end
|
440
|
+
|
441
|
+
|
442
|
+
|
443
|
+
# we obtain a global grid5000 user
|
444
|
+
$hostname = %x(hostname).strip
|
445
|
+
$ssh_key = G5KNode.obtain_ssh_pubkey_path
|
446
|
+
|
447
|
+
## XPFLOW LIBRARY
|
448
|
+
|
449
|
+
class NodesLibrary < ActivityLibrary
|
450
|
+
|
451
|
+
activities :node_list, :execute, :copy, :run_script, :check_node, :file,
|
452
|
+
:g5k_get_avail, :proxy_node, :broadcast, :g5k_site, :g5k_job,
|
453
|
+
:g5k_nodes, :monitor_node,
|
454
|
+
:g5k_kavlan_id, :g5k_kavlan_nodes_file, :g5k_frontend_from_job,
|
455
|
+
:g5k_kavlan_nodes, :nodes_file, :execute_funny,
|
456
|
+
:g5k_node, :execute_many, :execute_many_local, :all_prefixes, :execute_one, :distribute_one,
|
457
|
+
:execute_many_ignore_errors, :g5k_kadeploy, :execute_many_here,
|
458
|
+
:bootstrap_taktuk, :simple_node, :node_range, :taktuk_raw, :test_connectivity,
|
459
|
+
:nodes_from_file, :nodes_from_result, :distribute, :chain_copy, :ssh_key,
|
460
|
+
:nodes_from_machinefile, :g5k_deploy_keys, :localhost, :file_consistency,
|
461
|
+
:ping_localhost, :ping_node, :g5k_reserve_nodes
|
462
|
+
|
463
|
+
def setup
|
464
|
+
nil
|
465
|
+
end
|
466
|
+
|
467
|
+
def ping_node(node, target)
|
468
|
+
result = execute_one(node,"ping #{target} -c 1")
|
469
|
+
result.stdout[/time=(\d+.*) /,1].to_f
|
470
|
+
end
|
471
|
+
|
472
|
+
def ping_localhost(node = nil)
|
473
|
+
node = self.localhost() if node.nil?
|
474
|
+
ping_node(node, "localhost")
|
475
|
+
end
|
476
|
+
|
477
|
+
def get_g5k_tmpfile(prefix = "tmp")
|
478
|
+
hash = 16.times.map { |x| (rand * 16).to_i.to_s(16) }.join
|
479
|
+
return "/tmp/.#{prefix}-#{get_g5k_username()}-#{hash}"
|
480
|
+
end
|
481
|
+
|
482
|
+
def all_prefixes(nodes, inc = 1)
|
483
|
+
arr = []
|
484
|
+
i = inc - 1
|
485
|
+
while i < nodes.length
|
486
|
+
arr.push(nodes[0..i])
|
487
|
+
i += inc
|
488
|
+
end
|
489
|
+
return arr
|
490
|
+
end
|
491
|
+
|
492
|
+
def nodes
|
493
|
+
return Scope.current[:__nodes__]
|
494
|
+
end
|
495
|
+
|
496
|
+
def _transform_nodes(x)
|
497
|
+
if x.is_a?(String)
|
498
|
+
return x.strip.split
|
499
|
+
elsif x.is_a?(Hash)
|
500
|
+
h = x.map { |k, v| [ k.strip, _transform_nodes(v) ] }
|
501
|
+
return Hash[h]
|
502
|
+
elsif x.is_a?(Array)
|
503
|
+
return x.map { |x| _transform_nodes(x) }
|
504
|
+
else
|
505
|
+
raise "Error!"
|
506
|
+
end
|
507
|
+
end
|
508
|
+
|
509
|
+
def _get_node_via_proxy(name, parent = nil)
|
510
|
+
name = name.strip
|
511
|
+
name = "nancy.g5k" if name == "g5k"
|
512
|
+
if /^(.+)\.g5k$/.match(name)
|
513
|
+
raise "G5K proxies must be topmost" if !parent.nil?
|
514
|
+
# special syntax for G5K
|
515
|
+
site = name.split(".").first
|
516
|
+
proxy = g5k_site(site)
|
517
|
+
return proxy
|
518
|
+
else
|
519
|
+
user, host = name.split("@")
|
520
|
+
if parent.nil?
|
521
|
+
proxy = simple_node(name)
|
522
|
+
else
|
523
|
+
proxy = proxy_node(parent, user, host)
|
524
|
+
end
|
525
|
+
return proxy
|
526
|
+
end
|
527
|
+
end
|
528
|
+
|
529
|
+
def __transform_with_proxy(structure, proxy, nodes)
|
530
|
+
# proxy_node(via, user, host)
|
531
|
+
if structure.is_a?(String)
|
532
|
+
node = _get_node_via_proxy(structure, proxy)
|
533
|
+
nodes[:nodes].push(node)
|
534
|
+
elsif structure.is_a?(Array)
|
535
|
+
structure.each { |x| __transform_with_proxy(x, proxy, nodes) }
|
536
|
+
elsif structure.is_a?(Hash)
|
537
|
+
structure.each_pair do |p, sub|
|
538
|
+
new_proxy = _get_node_via_proxy(p, proxy)
|
539
|
+
nodes[:proxies].push(new_proxy)
|
540
|
+
__transform_with_proxy(sub, new_proxy, nodes)
|
541
|
+
end
|
542
|
+
else
|
543
|
+
raise "Error!"
|
544
|
+
end
|
545
|
+
end
|
546
|
+
|
547
|
+
def _transform_with_proxy(tree)
|
548
|
+
nodes = { :nodes => [], :proxies => [] }
|
549
|
+
__transform_with_proxy(tree, nil, nodes)
|
550
|
+
return nodes
|
551
|
+
end
|
552
|
+
|
553
|
+
def nodes_from_machinefile(filename, opts = {})
|
554
|
+
nodes = IO.read(filename).strip.split
|
555
|
+
nodes = nodes.map { |x| "#{opts[:user]}@#{x}" }
|
556
|
+
return nodes.map { |x| simple_node(x) }
|
557
|
+
end
|
558
|
+
|
559
|
+
def nodes_from_file(filename, opts = {})
|
560
|
+
contents = IO.read(filename)
|
561
|
+
yaml = YAML.load(contents)
|
562
|
+
tree = _transform_nodes(yaml)
|
563
|
+
nodes = _transform_with_proxy(tree)
|
564
|
+
return nodes[:nodes]
|
565
|
+
end
|
566
|
+
|
567
|
+
def _parse_opts(array)
|
568
|
+
h = {}
|
569
|
+
array.each do |o|
|
570
|
+
k, v = o.split("=")
|
571
|
+
h[k.to_sym] = v
|
572
|
+
end
|
573
|
+
return h
|
574
|
+
end
|
575
|
+
|
576
|
+
def nodes_from_result(result, proxy = nil)
|
577
|
+
lines = result.stdout.strip.lines.map(&:strip)
|
578
|
+
r = lines.map do |line|
|
579
|
+
userhost = line.split.first
|
580
|
+
opts = _parse_opts(line.split[1..-1])
|
581
|
+
if proxy.nil?
|
582
|
+
simple_node(userhost, opts)
|
583
|
+
else
|
584
|
+
u, h = userhost.split("@")
|
585
|
+
proxy_node(proxy, u, h, opts)
|
586
|
+
end
|
587
|
+
end
|
588
|
+
return r
|
589
|
+
end
|
590
|
+
|
591
|
+
# activities
|
592
|
+
|
593
|
+
def node_list()
|
594
|
+
return nodes()
|
595
|
+
end
|
596
|
+
|
597
|
+
def get_node_list(args)
|
598
|
+
# extracts nodes and a command from arguments
|
599
|
+
if args.length == 1
|
600
|
+
return [ node_list(), args.first ]
|
601
|
+
end
|
602
|
+
if args.length == 2
|
603
|
+
cmd = args.last
|
604
|
+
return [ arrayize(args.first), cmd ]
|
605
|
+
end
|
606
|
+
raise "Wrong number of arguments"
|
607
|
+
end
|
608
|
+
|
609
|
+
def arrayize(nodes)
|
610
|
+
# turns the argument into a list of nodes
|
611
|
+
if !nodes.is_a?(Array)
|
612
|
+
nodes = [ nodes ]
|
613
|
+
end
|
614
|
+
return nodes
|
615
|
+
end
|
616
|
+
|
617
|
+
def execute(nodes, cmd, env = {})
|
618
|
+
wd = env.delete(:wd)
|
619
|
+
nodes = arrayize(nodes)
|
620
|
+
arr = []
|
621
|
+
nodes.each do |node|
|
622
|
+
res = node.execute(cmd, env, :wd => wd)
|
623
|
+
arr.push(res)
|
624
|
+
end
|
625
|
+
return arr
|
626
|
+
end
|
627
|
+
|
628
|
+
def execute_one(node, cmd, env = {})
|
629
|
+
return execute(node, cmd, env).first
|
630
|
+
end
|
631
|
+
|
632
|
+
def _execute_many_parse_args(args)
|
633
|
+
opts = {}
|
634
|
+
opts = args.pop if args.last.is_a?(Hash)
|
635
|
+
nodes, cmd = get_node_list(args)
|
636
|
+
return [ nodes, cmd, opts ]
|
637
|
+
end
|
638
|
+
|
639
|
+
def _get_taktuk(nodes, options = {})
|
640
|
+
# domains = nodes.map(&:domain).uniq
|
641
|
+
# raise "TakTuk: nodes span different domains" if domains.length != 1
|
642
|
+
|
643
|
+
master = nodes.first
|
644
|
+
nodes = nodes.tail if options[:exclude_master]
|
645
|
+
taktuk = File.join(master.directory.path, "ssh taktuk")
|
646
|
+
directory = proxy.engine.main_directory
|
647
|
+
opts = {
|
648
|
+
:stdout => directory.mktemp(),
|
649
|
+
:stderr => directory.mktemp(),
|
650
|
+
:filename => directory.mktemp()
|
651
|
+
}.merge(options)
|
652
|
+
return TakTukRun.new(taktuk, nodes, opts)
|
653
|
+
end
|
654
|
+
|
655
|
+
def bootstrap_taktuk(nodes)
|
656
|
+
|
657
|
+
if nodes.is_a?(AbstractNode)
|
658
|
+
nodes = [ nodes ]
|
659
|
+
end
|
660
|
+
|
661
|
+
if nodes.length == 0
|
662
|
+
return
|
663
|
+
end
|
664
|
+
|
665
|
+
cmd = "(dpkg -l | grep taktuk) || apt-get install -y --force-yes taktuk"
|
666
|
+
escaped_cmd = Shellwords.escape(cmd)
|
667
|
+
master = nodes.first
|
668
|
+
master.execute(escaped_cmd)
|
669
|
+
|
670
|
+
proxy.log("#{master} has TakTuk now")
|
671
|
+
|
672
|
+
return execute_many(nodes, cmd, :propagate => true)
|
673
|
+
end
|
674
|
+
|
675
|
+
def execute_many_here(*args)
|
676
|
+
nodes, cmd, opts = _execute_many_parse_args(args)
|
677
|
+
|
678
|
+
return ManyExecutionResult.new([], cmd) if nodes.length == 0
|
679
|
+
|
680
|
+
taktuk = _get_taktuk(nodes, opts)
|
681
|
+
|
682
|
+
# command has to be escape 2 times: local shell and remote shell
|
683
|
+
succ, fail = taktuk.execute(cmd, :escape => 2)
|
684
|
+
|
685
|
+
if succ.length != nodes.length
|
686
|
+
raise "TakTuk: Some nodes failed (success: #{succ.length}/#{nodes.length}). See #{taktuk.stdout}"
|
687
|
+
end
|
688
|
+
|
689
|
+
results = succ.map do |x|
|
690
|
+
BasicRemoteResult.new(x[:node], cmd, x[:stdout], x[:stderr])
|
691
|
+
end
|
692
|
+
return ManyExecutionResult.new(results, cmd)
|
693
|
+
end
|
694
|
+
|
695
|
+
def execute_many(*args)
|
696
|
+
nodes, cmd, opts = _execute_many_parse_args(args)
|
697
|
+
|
698
|
+
return ManyExecutionResult.new([], cmd) if nodes.length == 0
|
699
|
+
|
700
|
+
taktuk = _get_taktuk(nodes, opts)
|
701
|
+
|
702
|
+
succ, fail = taktuk.execute_remote(cmd, :escape => 2)
|
703
|
+
|
704
|
+
if succ.length != nodes.length
|
705
|
+
msg = "TakTuk: Some nodes failed (success: #{succ.length}/#{nodes.length})."
|
706
|
+
if fail.length > 0
|
707
|
+
msg += " See #{fail.first[:stdout_file]}"
|
708
|
+
end
|
709
|
+
msg += " See #{taktuk.stdout}"
|
710
|
+
raise msg
|
711
|
+
end
|
712
|
+
|
713
|
+
results = succ.map do |x|
|
714
|
+
FileRemoteResult.new(x[:node], cmd, x[:stdout_file], x[:stderr_file])
|
715
|
+
end
|
716
|
+
return ManyExecutionResult.new(results, cmd)
|
717
|
+
end
|
718
|
+
|
719
|
+
def distribute_one(f, nodes, dest, opts = {})
|
720
|
+
|
721
|
+
nodes = [ nodes ] unless nodes.is_a?(Array)
|
722
|
+
|
723
|
+
if f.is_a?(String)
|
724
|
+
# nothing
|
725
|
+
elsif f.is_a?(LocalExecutionResult)
|
726
|
+
f = f.stdout_file
|
727
|
+
elsif f.is_a?(FileRemote)
|
728
|
+
f = f.result.stdout_file
|
729
|
+
else
|
730
|
+
raise "I don't know how to distribute #{f.class}"
|
731
|
+
end
|
732
|
+
|
733
|
+
if dest.end_with?("/") # a directory
|
734
|
+
dest = File.join(dest, File.basename(f))
|
735
|
+
else # a file
|
736
|
+
# it's fine
|
737
|
+
end
|
738
|
+
|
739
|
+
proxy.log("Saving to: #{dest}")
|
740
|
+
|
741
|
+
master = nodes.first
|
742
|
+
|
743
|
+
return nil if nodes.length == 0
|
744
|
+
|
745
|
+
master = nodes.first
|
746
|
+
master.scp_many([ f ], dest)
|
747
|
+
|
748
|
+
return nil if nodes.length == 1
|
749
|
+
|
750
|
+
taktuk = _get_taktuk(nodes, :exclude_master => true)
|
751
|
+
results = taktuk.put(dest, dest, :escape => 2)
|
752
|
+
|
753
|
+
if results.length + 1 != nodes.length
|
754
|
+
raise "Some nodes did not respond."
|
755
|
+
end
|
756
|
+
|
757
|
+
thelist = results.map { |x| x[:hash] }.uniq
|
758
|
+
raise "Some hashes were different." if thelist.length != 1
|
759
|
+
|
760
|
+
orig_hash = md5sum(nodes, [ dest ]).map { |x| x[:hash] }.uniq
|
761
|
+
|
762
|
+
if orig_hash.length != 1
|
763
|
+
raise "Hashes could not be verified."
|
764
|
+
end
|
765
|
+
|
766
|
+
return nil
|
767
|
+
end
|
768
|
+
|
769
|
+
def file_consistency(nodes, fs)
|
770
|
+
if !fs.is_a?(Array)
|
771
|
+
fs = [ fs ]
|
772
|
+
end
|
773
|
+
sums = md5sum(nodes, fs)
|
774
|
+
hashes = Hash.new { |h, k| h[k] = [] }
|
775
|
+
sums.each do |h|
|
776
|
+
hashes[h[:filename]].push(h[:hash])
|
777
|
+
end
|
778
|
+
hashes.each_pair do |f, h|
|
779
|
+
raise "File #{f} is not consistent" if h.uniq.length != 1
|
780
|
+
end
|
781
|
+
end
|
782
|
+
|
783
|
+
def md5sum(nodes, files)
|
784
|
+
h = []
|
785
|
+
results = execute_many_here nodes, "md5sum #{files.join(' ')}"
|
786
|
+
results.to_list.each do |r|
|
787
|
+
r.stdout.strip.lines do |line|
|
788
|
+
hash, filename = line.split
|
789
|
+
h.push({ :filename => filename, :hash => hash })
|
790
|
+
end
|
791
|
+
end
|
792
|
+
return h
|
793
|
+
end
|
794
|
+
|
795
|
+
def distribute(glob, nodes, dest, opts = {})
|
796
|
+
|
797
|
+
nodes = [ nodes ] unless nodes.is_a?(Array)
|
798
|
+
glob = glob.first if glob.is_a?(Array)
|
799
|
+
|
800
|
+
if glob.is_a?(String)
|
801
|
+
files = Dir[glob] # get a list of files
|
802
|
+
elsif glob.is_a?(LocalExecutionResult)
|
803
|
+
files = [ glob.stdout_file ]
|
804
|
+
elsif glob.is_a?(FileRemote)
|
805
|
+
files = [ glob.result.stdout_file ]
|
806
|
+
else
|
807
|
+
raise "I don't know how to distribute #{glob.class}"
|
808
|
+
end
|
809
|
+
remote_files = files.map { |f| File.join(dest, File.basename(f)) }
|
810
|
+
|
811
|
+
proxy.log("Found #{files.length} files to distribute.")
|
812
|
+
|
813
|
+
return nil if nodes.length == 0
|
814
|
+
|
815
|
+
master = nodes.first
|
816
|
+
master.scp_many(files, dest)
|
817
|
+
|
818
|
+
proxy.log("Files copied to #{master}:#{dest}")
|
819
|
+
|
820
|
+
if nodes.length == 1
|
821
|
+
return nil
|
822
|
+
end
|
823
|
+
|
824
|
+
md5sum = master.execute("md5sum #{remote_files.join(' ')}").stdout
|
825
|
+
orig_hashes = {}
|
826
|
+
md5sum.strip.lines do |line|
|
827
|
+
hash, filename = line.strip.split
|
828
|
+
orig_hashes[filename] = hash
|
829
|
+
end
|
830
|
+
|
831
|
+
hashes = {}
|
832
|
+
remote_files.each do |filepath|
|
833
|
+
proxy.log("Distributing #{filepath}")
|
834
|
+
taktuk = _get_taktuk(nodes, :exclude_master => true)
|
835
|
+
results = taktuk.put(filepath, filepath, :escape => 2)
|
836
|
+
|
837
|
+
if results.length + 1 != nodes.length
|
838
|
+
raise "Some nodes did not respond."
|
839
|
+
end
|
840
|
+
|
841
|
+
thelist = results.map { |x| x[:hash] }.uniq
|
842
|
+
raise "Some hashes were different." if thelist.length != 1
|
843
|
+
|
844
|
+
hashes[filepath] = thelist.first
|
845
|
+
end
|
846
|
+
|
847
|
+
raise "Weird?" if hashes.keys.sort != orig_hashes.keys.sort
|
848
|
+
|
849
|
+
orig_hashes.keys.each do |filepath|
|
850
|
+
raise "Hash for #{filepath} does not match." \
|
851
|
+
if orig_hashes[filepath] != hashes[filepath]
|
852
|
+
end
|
853
|
+
|
854
|
+
proxy.log("MD5s match.")
|
855
|
+
|
856
|
+
end
|
857
|
+
|
858
|
+
def ssh_key
|
859
|
+
return $ssh_key
|
860
|
+
end
|
861
|
+
|
862
|
+
def test_connectivity(nodes)
|
863
|
+
execute_many nodes, "true"
|
864
|
+
end
|
865
|
+
|
866
|
+
def chain_copy(nodes, src, dest, opts = {})
|
867
|
+
|
868
|
+
nodes = [ nodes ] unless nodes.is_a?(Array)
|
869
|
+
|
870
|
+
raise "'#{src}' does not exist!" unless File.exist?(src)
|
871
|
+
|
872
|
+
dest = File.join(dest, File.basename(src))
|
873
|
+
|
874
|
+
return nil if nodes.length == 0
|
875
|
+
|
876
|
+
master = nodes.first
|
877
|
+
laster = nodes.last
|
878
|
+
|
879
|
+
master.scp(src, dest)
|
880
|
+
|
881
|
+
proxy.log("#{src} copied to #{master}:#{dest}")
|
882
|
+
|
883
|
+
if nodes.length == 1
|
884
|
+
return nil
|
885
|
+
end
|
886
|
+
|
887
|
+
orig_hash = master.execute("md5sum #{dest}").stdout.strip.split.first
|
888
|
+
|
889
|
+
execute_many(nodes, "apt-get install -y mbuffer")
|
890
|
+
|
891
|
+
# mbuffer creates a buffer of size 2% * memory (by default)
|
892
|
+
# we do chainsend thing
|
893
|
+
|
894
|
+
second = nodes[1].host
|
895
|
+
ssh_opts = "-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"
|
896
|
+
master.file("/tmp/.chainsend") do |f|
|
897
|
+
f.puts "set -eu"
|
898
|
+
f.puts "cat $1 | ssh #{ssh_opts} #{second} \"bash /tmp/.chainsend $1\""
|
899
|
+
end
|
900
|
+
laster.file("/tmp/.chainsend") do |f|
|
901
|
+
f.puts "set -eu"
|
902
|
+
f.puts "cat > $1"
|
903
|
+
end
|
904
|
+
nodes.each_with_index do |node, i|
|
905
|
+
next if i == 0 or i == nodes.length - 1
|
906
|
+
next_host = nodes[i+1].host
|
907
|
+
node.file("/tmp/.chainsend") do |f|
|
908
|
+
f.puts "set -eu"
|
909
|
+
f.puts "tee $1 | mbuffer -q | ssh #{ssh_opts} #{next_host} \"bash /tmp/.chainsend $1\""
|
910
|
+
end
|
911
|
+
end
|
912
|
+
|
913
|
+
proxy.log("Chain prepared.")
|
914
|
+
|
915
|
+
master.execute("bash /tmp/.chainsend #{dest}")
|
916
|
+
results = execute_many_here(nodes, "md5sum #{dest}")
|
917
|
+
|
918
|
+
hashes = results.to_list.map { |x| x.stdout.split.first }.uniq
|
919
|
+
|
920
|
+
raise "Hashes differ." if hashes.length != 1
|
921
|
+
|
922
|
+
return nil
|
923
|
+
end
|
924
|
+
|
925
|
+
def file(node, path)
|
926
|
+
r = execute_one(node, "cat #{path}")
|
927
|
+
return FileRemote.new(path, r)
|
928
|
+
end
|
929
|
+
|
930
|
+
def copy(name, nodes, where)
|
931
|
+
# make a copy of a file named "name" to "nodes"
|
932
|
+
# at path "where"; name can be:
|
933
|
+
# * a path (at local fs)
|
934
|
+
# * a LocalExecutionResult (as a result of remote execution)
|
935
|
+
# * RemoteFile - pointer to a file on a remote node
|
936
|
+
|
937
|
+
name = name.first if name.is_a?(Array)
|
938
|
+
|
939
|
+
if name.is_a?(String)
|
940
|
+
# ok
|
941
|
+
elsif name.is_a?(LocalExecutionResult)
|
942
|
+
name = name.stdout_file
|
943
|
+
elsif name.is_a?(FileRemote)
|
944
|
+
name = name.result.stdout_file
|
945
|
+
else
|
946
|
+
raise "Unknown file source: #{name.class}"
|
947
|
+
end
|
948
|
+
|
949
|
+
nodes = arrayize(nodes)
|
950
|
+
result = nodes.map do |node|
|
951
|
+
node.scp(name, where)
|
952
|
+
end
|
953
|
+
return result
|
954
|
+
end
|
955
|
+
|
956
|
+
def run_script(name, label = :all)
|
957
|
+
path = $files[name]
|
958
|
+
node_list(label).each do |node|
|
959
|
+
f = node.tmpfile
|
960
|
+
node.scp(path, f)
|
961
|
+
node.chmod(f, "700")
|
962
|
+
out = node.execute(f)
|
963
|
+
node.rm(f)
|
964
|
+
end
|
965
|
+
end
|
966
|
+
|
967
|
+
def check_node(node)
|
968
|
+
debug "Checking #{node}..."
|
969
|
+
failed = []
|
970
|
+
begin
|
971
|
+
node.ping()
|
972
|
+
rescue ExecutionError => e
|
973
|
+
return false
|
974
|
+
end
|
975
|
+
return true
|
976
|
+
end
|
977
|
+
|
978
|
+
def g5k_reserve_nodes(*args)
|
979
|
+
p = proxy()
|
980
|
+
lib = G5K::Library.new
|
981
|
+
lib.logging = proc { |x| p.engine.log(x) }
|
982
|
+
lib.proxy = proxy
|
983
|
+
return lib.reserve_nodes(*args)
|
984
|
+
end
|
985
|
+
|
986
|
+
def g5k_get_avail(opts = {})
|
987
|
+
p = proxy()
|
988
|
+
lib = G5K::Library.new
|
989
|
+
lib.logging = proc { |x| p.engine.log(x) }
|
990
|
+
lib.proxy = proxy
|
991
|
+
job = lib.pick_reservation(opts)
|
992
|
+
return job
|
993
|
+
end
|
994
|
+
|
995
|
+
def nodes_file(user, filepath)
|
996
|
+
nodes = IO.read(filepath).chomp.lines.map(&:chomp)
|
997
|
+
nodes = nodes.map do |host|
|
998
|
+
simple_node("#{user}@#{host}")
|
999
|
+
end
|
1000
|
+
return nodes
|
1001
|
+
end
|
1002
|
+
|
1003
|
+
def g5k_job(opts = {})
|
1004
|
+
lib = G5K::Library.new
|
1005
|
+
lib.logging = proc { |x| puts x }
|
1006
|
+
job = lib.job(opts[:site], opts[:id])
|
1007
|
+
return job
|
1008
|
+
end
|
1009
|
+
|
1010
|
+
def g5k_nodes(job)
|
1011
|
+
hosts = job["assigned_nodes"]
|
1012
|
+
username = get_g5k_username()
|
1013
|
+
return hosts.map { |h| g5k_node(username, h) }
|
1014
|
+
end
|
1015
|
+
|
1016
|
+
def g5k_site_from_job(job)
|
1017
|
+
nodes = job["assigned_nodes"]
|
1018
|
+
link = job["links"].select { |x| x["rel"] == "parent" }.first
|
1019
|
+
site = link["href"].split("/").last
|
1020
|
+
return site
|
1021
|
+
end
|
1022
|
+
|
1023
|
+
def _filter_vlan(ok_nodes, vlan_nodes)
|
1024
|
+
h = {}
|
1025
|
+
hosts = ok_nodes.each { |x| h[x.split(".").first] = true }
|
1026
|
+
# puts hosts.inspect
|
1027
|
+
good_ones = vlan_nodes.select { |x| h.key?(x.split("-kavlan").first) }
|
1028
|
+
# puts good_ones.inspect
|
1029
|
+
return good_ones
|
1030
|
+
end
|
1031
|
+
|
1032
|
+
def g5k_deploy_keys(nodes, site)
|
1033
|
+
# assumes you are on the frontend
|
1034
|
+
key = get_g5k_tmpfile("ssh_key")
|
1035
|
+
frontend = g5k_site(site)
|
1036
|
+
frontend.execute("rm -f #{key} #{key}.pub; ssh-keygen -f #{key} -q -N \'\'")
|
1037
|
+
nodes.each do |n|
|
1038
|
+
proxy.log "Sending key to: #{n.userhost}"
|
1039
|
+
frontend.execute("scp -o 'BatchMode=yes' -o 'UserKnownHostsFile=/dev/null' #{key} #{n.userhost}:.ssh/id_rsa")
|
1040
|
+
frontend.execute("ssh-copy-id -i #{key}.pub #{n.userhost}")
|
1041
|
+
end
|
1042
|
+
frontend.execute("rm -f #{key} #{key}.pub")
|
1043
|
+
end
|
1044
|
+
|
1045
|
+
def g5k_kadeploy(job, env, custom = "", opts = {})
|
1046
|
+
|
1047
|
+
site = g5k_site_from_job(job)
|
1048
|
+
frontend = g5k_site(site)
|
1049
|
+
nodes = job["assigned_nodes"]
|
1050
|
+
|
1051
|
+
nodes = opts.key?(:count) ? nodes[0...opts[:count]] : nodes
|
1052
|
+
final_nodes = opts[:real_nodes]
|
1053
|
+
|
1054
|
+
proxy.log("Using #{nodes.length} machines.")
|
1055
|
+
|
1056
|
+
machinefile = get_g5k_tmpfile("machines")
|
1057
|
+
nodes_ok = get_g5k_tmpfile("good_nodes")
|
1058
|
+
|
1059
|
+
IO.write(machinefile, nodes.join("\n"))
|
1060
|
+
frontend.scp(machinefile, machinefile)
|
1061
|
+
kadeploy = "kadeploy3 -f #{machinefile} -e #{env} -k #{custom} -o #{nodes_ok}"
|
1062
|
+
proxy.log("Running deployment: #{kadeploy}")
|
1063
|
+
frontend.execute(kadeploy)
|
1064
|
+
frontend.execute("rm -f #{machinefile}")
|
1065
|
+
|
1066
|
+
key = get_g5k_tmpfile("ssh_key")
|
1067
|
+
frontend.execute("rm -f #{key} #{key}.pub; ssh-keygen -f #{key} -q -N \'\'")
|
1068
|
+
|
1069
|
+
ok_nodes = frontend.execute("sort -V #{nodes_ok}").stdout.split
|
1070
|
+
frontend.execute("rm -f #{nodes_ok}")
|
1071
|
+
|
1072
|
+
proxy.log("Nodes that survived: #{ok_nodes.length}/#{nodes.length}")
|
1073
|
+
proxy.log("Final nodes: #{final_nodes}")
|
1074
|
+
|
1075
|
+
if final_nodes.nil?
|
1076
|
+
final_nodes = ok_nodes
|
1077
|
+
else
|
1078
|
+
final_nodes = _filter_vlan(ok_nodes, final_nodes)
|
1079
|
+
end
|
1080
|
+
|
1081
|
+
# we have to install SSH keys
|
1082
|
+
final_nodes.each do |n|
|
1083
|
+
host = "root@#{n}"
|
1084
|
+
frontend.execute("scp -o 'BatchMode=yes' -o 'UserKnownHostsFile=/dev/null' #{key} #{host}:.ssh/id_rsa")
|
1085
|
+
frontend.execute("ssh-copy-id -i #{key}.pub #{host}")
|
1086
|
+
end
|
1087
|
+
frontend.execute("rm -f #{key} #{key}.pub")
|
1088
|
+
|
1089
|
+
all_nodes = final_nodes.map { |x| g5k_node("root", x) }
|
1090
|
+
return all_nodes
|
1091
|
+
end
|
1092
|
+
|
1093
|
+
def g5k_frontend_from_job(job)
|
1094
|
+
site = g5k_site_from_job(job)
|
1095
|
+
frontend = g5k_site(site)
|
1096
|
+
return frontend
|
1097
|
+
end
|
1098
|
+
|
1099
|
+
def g5k_kavlan_id(job)
|
1100
|
+
uid = job['uid']
|
1101
|
+
frontend = g5k_frontend_from_job(job)
|
1102
|
+
out = frontend.execute("kavlan -V -j #{uid}")
|
1103
|
+
return out.stdout.strip.to_i
|
1104
|
+
end
|
1105
|
+
|
1106
|
+
def g5k_kavlan_nodes_file(job)
|
1107
|
+
# TODO
|
1108
|
+
uid = job['uid']
|
1109
|
+
kavlan_nodes = get_g5k_tmpfile("kavlan_nodes")
|
1110
|
+
frontend = g5k_frontend_from_job(job)
|
1111
|
+
out = frontend.execute("kavlan -l -j #{uid}")
|
1112
|
+
IO.write(kavlan_nodes, out.stdout)
|
1113
|
+
frontend.scp(kavlan_nodes, kavlan_nodes)
|
1114
|
+
return kavlan_nodes
|
1115
|
+
end
|
1116
|
+
|
1117
|
+
def g5k_kavlan_nodes(job)
|
1118
|
+
uid = job['uid']
|
1119
|
+
frontend = g5k_frontend_from_job(job)
|
1120
|
+
out = frontend.execute("kavlan -l -j #{uid}")
|
1121
|
+
hosts = out.stdout.strip.lines.map(&:strip)
|
1122
|
+
nodes = hosts.map { |x| g5k_node("root", x) }
|
1123
|
+
return nodes
|
1124
|
+
end
|
1125
|
+
|
1126
|
+
def g5k_site(site)
|
1127
|
+
manager = proxy.engine.nodes_manager
|
1128
|
+
return manager.get_node(get_g5k_username(), "#{site}.grid5000.fr", G5KNodeFactory.new)
|
1129
|
+
end
|
1130
|
+
|
1131
|
+
def g5k_node(user, host)
|
1132
|
+
manager = proxy.engine.nodes_manager
|
1133
|
+
return manager.get_node(user, host, G5KNodeFactory.new)
|
1134
|
+
end
|
1135
|
+
|
1136
|
+
def proxy_node(via, user, host, opts = {})
|
1137
|
+
manager = proxy.engine.nodes_manager
|
1138
|
+
return manager.get_node(user, host, ProxiedFactory.new(via), opts)
|
1139
|
+
end
|
1140
|
+
|
1141
|
+
def split(pattern)
|
1142
|
+
pattern = pattern.strip
|
1143
|
+
if pattern == ""
|
1144
|
+
raise "Empty host specification"
|
1145
|
+
end
|
1146
|
+
parts = pattern.split("@")
|
1147
|
+
if parts.length == 2
|
1148
|
+
return parts
|
1149
|
+
elsif parts.length > 2
|
1150
|
+
raise "Invalid host specification: #{pattern}"
|
1151
|
+
else
|
1152
|
+
proxy.log "User not specified. This is not reproducible."
|
1153
|
+
return [ Etc.getlogin, pattern ]
|
1154
|
+
end
|
1155
|
+
end
|
1156
|
+
|
1157
|
+
def broadcast(pattern)
|
1158
|
+
manager = proxy.engine.nodes_manager
|
1159
|
+
user, address = split(pattern)
|
1160
|
+
nodes = NodeUtils.broadcast_ping(address)
|
1161
|
+
proxy.log("Found #{nodes.length} hosts via ICMP broadcast")
|
1162
|
+
return nodes.map { |n| manager.get_node(user, n) }
|
1163
|
+
end
|
1164
|
+
|
1165
|
+
def node_range(ip_start, count)
|
1166
|
+
user, host = split(ip_start)
|
1167
|
+
parts = host.split(".").map(&:to_i)
|
1168
|
+
ips = []
|
1169
|
+
count.times do |i|
|
1170
|
+
j = 3
|
1171
|
+
while j > 0 and parts[j] == 256
|
1172
|
+
parts[j] = 0
|
1173
|
+
parts[j - 1] += 1
|
1174
|
+
j -= 1
|
1175
|
+
end
|
1176
|
+
ip = parts.map(&:to_s).join(".")
|
1177
|
+
ips.push(ip)
|
1178
|
+
parts[3] += 1
|
1179
|
+
end
|
1180
|
+
return ips.map { |x| simple_node("#{user}@#{x}") }
|
1181
|
+
end
|
1182
|
+
|
1183
|
+
def simple_node(pattern, opts = {})
|
1184
|
+
user, host = split(pattern)
|
1185
|
+
manager = proxy.engine.nodes_manager
|
1186
|
+
return manager.get_node(user, host, SimpleNodeFactory.new, opts)
|
1187
|
+
end
|
1188
|
+
|
1189
|
+
def localhost()
|
1190
|
+
user = ENV["USER"]
|
1191
|
+
return simple_node("#{user}@127.0.0.1")
|
1192
|
+
end
|
1193
|
+
|
1194
|
+
end
|
1195
|
+
|
1196
|
+
module NodeUtils
|
1197
|
+
|
1198
|
+
def self.get_templates_dir(name)
|
1199
|
+
here = File.dirname(__FILE__)
|
1200
|
+
return realpath(File.join(here, name))
|
1201
|
+
end
|
1202
|
+
|
1203
|
+
def self.render_file(t, path, ctx)
|
1204
|
+
out = Erb.render(IO.read(t), ctx)
|
1205
|
+
File.open(path, "wb") do |f|
|
1206
|
+
f.write(out)
|
1207
|
+
f.chmod(0700)
|
1208
|
+
end
|
1209
|
+
end
|
1210
|
+
|
1211
|
+
def self.install_templates(path, name, ctx)
|
1212
|
+
templates = get_templates_dir("templates")
|
1213
|
+
ctx = ctx.merge({ :templates => templates })
|
1214
|
+
tdir = get_templates_dir("templates/utils")
|
1215
|
+
Dir.entries(tdir).each do |f|
|
1216
|
+
template = File.join(tdir, f)
|
1217
|
+
output = File.join(path, f)
|
1218
|
+
render_file(template, output, ctx) if File.file?(template)
|
1219
|
+
end
|
1220
|
+
config = get_templates_dir("templates/ssh-config.#{name}")
|
1221
|
+
ssh_config = File.join(path, "ssh-config")
|
1222
|
+
render_file(config, ssh_config, ctx)
|
1223
|
+
end
|
1224
|
+
|
1225
|
+
# uses a broadcast ping to get a list of nodes that respond
|
1226
|
+
def self.broadcast_ping(address, timeout = 2)
|
1227
|
+
output = %x(ping -c 3 -n -b #{address} -w #{timeout} 2> /dev/null)
|
1228
|
+
# raise "Ping returned error #{$?.exitstatus}." if $?.exitstatus != 0
|
1229
|
+
ms = output.scan(/from ([\.0-9]+): icmp_seq=/)
|
1230
|
+
addresses = ms.flatten.uniq.sort
|
1231
|
+
raise "No nodes found" if addresses.length == 0
|
1232
|
+
return addresses
|
1233
|
+
end
|
1234
|
+
|
1235
|
+
class Erb < OpenStruct
|
1236
|
+
|
1237
|
+
def render(hash)
|
1238
|
+
ERB.new(hash).result(binding)
|
1239
|
+
end
|
1240
|
+
|
1241
|
+
def self.render(template, hash)
|
1242
|
+
x = Erb.new(hash)
|
1243
|
+
return x.render(template)
|
1244
|
+
end
|
1245
|
+
|
1246
|
+
end
|
1247
|
+
|
1248
|
+
end
|
1249
|
+
|
1250
|
+
end
|