nitos_testbed_rc 2.0.3 → 2.0.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +8 -8
- data/README.md +2 -2
- data/bin/cm_proxy +13 -6
- data/bin/omf6 +372 -339
- data/lib/nitos_testbed_rc/cm.rb +493 -0
- data/lib/nitos_testbed_rc/cm_factory.rb +439 -457
- data/lib/nitos_testbed_rc/frisbee.rb +89 -24
- data/lib/nitos_testbed_rc/frisbeed.rb +1 -1
- data/lib/version.rb +1 -1
- metadata +3 -2
@@ -0,0 +1,493 @@
|
|
1
|
+
#this resource is used to control chassis managers.
|
2
|
+
require 'rubygems'
|
3
|
+
require 'yaml'
|
4
|
+
require 'open-uri'
|
5
|
+
require 'nokogiri'
|
6
|
+
require 'net/ssh'
|
7
|
+
|
8
|
+
REBOOT_CMD = "reboot"
|
9
|
+
SHUTDOWN_CMD = "shutdown -P now"
|
10
|
+
|
11
|
+
module OmfRc::ResourceProxy::CM
|
12
|
+
include OmfRc::ResourceProxyDSL
|
13
|
+
|
14
|
+
@config = YAML.load_file('/etc/nitos_testbed_rc/cm_proxy_conf.yaml')
|
15
|
+
# @config = YAML.load_file(File.join(File.dirname(File.expand_path(__FILE__)), '../etc/cm_proxy_conf.yaml'))
|
16
|
+
@@timeout = @config[:timeout]
|
17
|
+
|
18
|
+
register_proxy :cm, :create_by => :cm_factory
|
19
|
+
|
20
|
+
property :node, default: "node000"
|
21
|
+
|
22
|
+
hook :after_initial_configured do |res|
|
23
|
+
puts "*******************************************"
|
24
|
+
puts "Node: #{res.property.node}"
|
25
|
+
puts "*******************************************"
|
26
|
+
end
|
27
|
+
|
28
|
+
configure :state do |res, value|
|
29
|
+
debug "Received message '#{value.inspect}'"
|
30
|
+
if error_msg = value.error_msg
|
31
|
+
res.inform(:error,{
|
32
|
+
event_type: "AUTH",
|
33
|
+
exit_code: "-1",
|
34
|
+
node_name: value[:node],
|
35
|
+
msg: error_msg
|
36
|
+
}, :ALL)
|
37
|
+
next
|
38
|
+
end
|
39
|
+
nod = value.node
|
40
|
+
# nod = {node_name: "node120", node_ip: "10.0.1.120", node_mac: "00-03-1d-0d-4b-96", node_cm_ip: "10.1.0.120"} if value.node == 'node120'
|
41
|
+
# nod = {node_name: "node121", node_ip: "10.0.1.121", node_mac: "00-03-1d-0d-40-98", node_cm_ip: "10.1.0.121"} if value.node == 'node121'
|
42
|
+
|
43
|
+
case value[:status].to_sym
|
44
|
+
when :on then res.start_node(nod, value[:wait])
|
45
|
+
when :off then res.stop_node(nod, value[:wait])
|
46
|
+
when :reset then res.reset_node(nod, value[:wait])
|
47
|
+
when :start_on_pxe then res.start_node_pxe(nod)
|
48
|
+
when :start_without_pxe then res.start_node_pxe_off(nod, value[:last_action])
|
49
|
+
when :get_status then res.status(nod)
|
50
|
+
else
|
51
|
+
res.log_inform_warn "Cannot switch node to unknown state '#{value[:status].to_s}'!"
|
52
|
+
end
|
53
|
+
end
|
54
|
+
|
55
|
+
#this is used by the get status call
|
56
|
+
work("status") do |res, node|
|
57
|
+
debug "Status url: http://#{node[:node_cm_ip].to_s}/state"
|
58
|
+
begin
|
59
|
+
resp = open("http://#{node[:node_cm_ip].to_s}/state")
|
60
|
+
rescue
|
61
|
+
res.inform(:error, {
|
62
|
+
event_type: "HTTP",
|
63
|
+
exit_code: "-1",
|
64
|
+
node_name: "#{node[:node_name].to_s}",
|
65
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
66
|
+
}, :ALL)
|
67
|
+
next
|
68
|
+
end
|
69
|
+
ans = res.parse_responce(resp, "//Response//line//value")
|
70
|
+
|
71
|
+
res.inform(:status, {
|
72
|
+
current: "#{ans}",
|
73
|
+
node_name: "#{node[:node_name].to_s}"
|
74
|
+
}, :ALL)
|
75
|
+
sleep 1 #this solves the getting stuck problem.
|
76
|
+
end
|
77
|
+
|
78
|
+
work("start_node") do |res, node, wait|
|
79
|
+
node[:node_mac] = node[:node_mac].downcase.gsub(/:/, '-')
|
80
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
81
|
+
if File.exists?(symlink_name)
|
82
|
+
File.delete(symlink_name)
|
83
|
+
end
|
84
|
+
debug "Start_node url: http://#{node[:node_cm_ip].to_s}/on"
|
85
|
+
begin
|
86
|
+
resp = open("http://#{node[:node_cm_ip].to_s}/on")
|
87
|
+
rescue
|
88
|
+
res.inform(:error, {
|
89
|
+
event_type: "HTTP",
|
90
|
+
exit_code: "-1",
|
91
|
+
node_name: "#{node[:node_name].to_s}",
|
92
|
+
msg: "#{node[:name]} failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
93
|
+
}, :ALL)
|
94
|
+
next
|
95
|
+
end
|
96
|
+
|
97
|
+
ans = res.parse_responce(resp, "//Response")
|
98
|
+
|
99
|
+
if ans == 'ok'
|
100
|
+
res.inform(:status, {
|
101
|
+
node_name: "#{node[:node_name].to_s}",
|
102
|
+
current: :booting,
|
103
|
+
desired: :running
|
104
|
+
}, :ALL)
|
105
|
+
elsif ans == 'already on'
|
106
|
+
res.inform(:status, {
|
107
|
+
node_name: "#{node[:node_name].to_s}",
|
108
|
+
current: :running,
|
109
|
+
desired: :running
|
110
|
+
}, :ALL)
|
111
|
+
end
|
112
|
+
|
113
|
+
if wait
|
114
|
+
if res.wait_until_ping(node[:node_ip])
|
115
|
+
res.inform(:status, {
|
116
|
+
node_name: "#{node[:node_name].to_s}",
|
117
|
+
current: :running,
|
118
|
+
desired: :running
|
119
|
+
}, :ALL)
|
120
|
+
else
|
121
|
+
res.inform(:error, {
|
122
|
+
event_type: "TIME_OUT",
|
123
|
+
exit_code: "-1",
|
124
|
+
node_name: "#{node[:node_name].to_s}",
|
125
|
+
msg: "Node '#{node[:node_name].to_s}' timed out while booting."
|
126
|
+
}, :ALL)
|
127
|
+
end
|
128
|
+
end
|
129
|
+
sleep 1
|
130
|
+
end
|
131
|
+
|
132
|
+
work("stop_node") do |res, node, wait|
|
133
|
+
node[:node_mac] = node[:node_mac].downcase.gsub(/:/, '-')
|
134
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
135
|
+
if File.exists?(symlink_name)
|
136
|
+
File.delete(symlink_name)
|
137
|
+
end
|
138
|
+
begin
|
139
|
+
debug "Shutting down node '#{node[:node_name]}' through ssh."
|
140
|
+
ssh = Net::SSH.start(node[:node_ip], 'root')#, :password => @password)
|
141
|
+
resp = ssh.exec!(SHUTDOWN_CMD)
|
142
|
+
ssh.close
|
143
|
+
debug "shutting down completed with ssh."
|
144
|
+
res.inform(:status, {
|
145
|
+
node_name: "#{node[:node_name].to_s}",
|
146
|
+
current: :running,
|
147
|
+
desired: :stopped
|
148
|
+
}, :ALL)
|
149
|
+
rescue
|
150
|
+
begin
|
151
|
+
debug "ssh failed, using CM card instead."
|
152
|
+
debug "Stop_node url: http://#{node[:node_cm_ip].to_s}/off"
|
153
|
+
|
154
|
+
begin
|
155
|
+
resp = open("http://#{node[:node_cm_ip].to_s}/off")
|
156
|
+
rescue
|
157
|
+
res.inform(:error, {
|
158
|
+
event_type: "HTTP",
|
159
|
+
exit_code: "-1",
|
160
|
+
node_name: "#{node[:node_name].to_s}",
|
161
|
+
msg: "#{node[:name]} failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
162
|
+
}, :ALL)
|
163
|
+
next
|
164
|
+
end
|
165
|
+
|
166
|
+
ans = res.parse_responce(resp, "//Response")
|
167
|
+
|
168
|
+
if ans == 'ok'
|
169
|
+
res.inform(:status, {
|
170
|
+
node_name: "#{node[:node_name].to_s}",
|
171
|
+
current: :running,
|
172
|
+
desired: :stopped
|
173
|
+
}, :ALL)
|
174
|
+
elsif ans == 'already off'
|
175
|
+
res.inform(:status, {
|
176
|
+
node_name: "#{node[:node_name].to_s}",
|
177
|
+
current: :stopped,
|
178
|
+
desired: :stopped
|
179
|
+
}, :ALL)
|
180
|
+
end
|
181
|
+
rescue
|
182
|
+
res.inform(:error, {
|
183
|
+
event_type: "HTTP",
|
184
|
+
exit_code: "-1",
|
185
|
+
node_name: "#{node[:node_name].to_s}",
|
186
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
187
|
+
}, :ALL)
|
188
|
+
next
|
189
|
+
end
|
190
|
+
end
|
191
|
+
|
192
|
+
if wait
|
193
|
+
if res.wait_until_no_ping(node[:node_ip])
|
194
|
+
res.inform(:status, {
|
195
|
+
node_name: "#{node[:node_name].to_s}",
|
196
|
+
current: :stopped,
|
197
|
+
desired: :stopped
|
198
|
+
}, :ALL)
|
199
|
+
else
|
200
|
+
res.inform(:error, {
|
201
|
+
event_type: "TIME_OUT",
|
202
|
+
exit_code: "-1",
|
203
|
+
node_name: "#{node[:node_name].to_s}",
|
204
|
+
msg: "Node '#{node[:node_name].to_s}' timed out while shutting down."
|
205
|
+
}, :ALL)
|
206
|
+
end
|
207
|
+
end
|
208
|
+
sleep 1
|
209
|
+
end
|
210
|
+
|
211
|
+
work("reset_node") do |res, node, wait|
|
212
|
+
node[:node_mac] = node[:node_mac].downcase.gsub(/:/, '-')
|
213
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
214
|
+
if File.exists?(symlink_name)
|
215
|
+
File.delete(symlink_name)
|
216
|
+
end
|
217
|
+
begin
|
218
|
+
debug "Rebooting node '#{node[:node_name]}' through ssh."
|
219
|
+
ssh = Net::SSH.start(node[:node_ip], 'root')#, :password => @password)
|
220
|
+
resp = ssh.exec!(REBOOT_CMD)
|
221
|
+
ssh.close
|
222
|
+
debug "Rebooting completed with ssh."
|
223
|
+
res.inform(:status, {
|
224
|
+
node_name: "#{node[:node_name].to_s}",
|
225
|
+
current: :running,
|
226
|
+
desired: :resetted
|
227
|
+
}, :ALL)
|
228
|
+
rescue
|
229
|
+
begin
|
230
|
+
debug "ssh failed, using CM card instead."
|
231
|
+
debug "Reset_node url: http://#{node[:node_cm_ip].to_s}/reset"
|
232
|
+
begin
|
233
|
+
resp = open("http://#{node[:node_cm_ip].to_s}/reset")
|
234
|
+
rescue
|
235
|
+
res.inform(:error, {
|
236
|
+
event_type: "HTTP",
|
237
|
+
exit_code: "-1",
|
238
|
+
node_name: "#{node[:node_name].to_s}",
|
239
|
+
msg: "#{node[:name]} failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
240
|
+
}, :ALL)
|
241
|
+
next
|
242
|
+
end
|
243
|
+
|
244
|
+
ans = res.parse_responce(resp, "//Response")
|
245
|
+
if ans == 'ok'
|
246
|
+
res.inform(:status, {
|
247
|
+
node_name: "#{node[:node_name].to_s}",
|
248
|
+
current: :resetted,
|
249
|
+
desired: :resetted
|
250
|
+
}, :ALL)
|
251
|
+
end
|
252
|
+
rescue
|
253
|
+
res.inform(:error, {
|
254
|
+
event_type: "HTTP",
|
255
|
+
exit_code: "-1",
|
256
|
+
node_name: "#{node[:node_name].to_s}",
|
257
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
258
|
+
}, :ALL)
|
259
|
+
next
|
260
|
+
end
|
261
|
+
end
|
262
|
+
|
263
|
+
if wait
|
264
|
+
if res.wait_until_ping(node[:node_ip])
|
265
|
+
res.inform(:status, {
|
266
|
+
node_name: "#{node[:node_name].to_s}",
|
267
|
+
current: :resetted,
|
268
|
+
desired: :resetted
|
269
|
+
}, :ALL)
|
270
|
+
else
|
271
|
+
res.inform(:error, {
|
272
|
+
event_type: "TIME_OUT",
|
273
|
+
exit_code: "-1",
|
274
|
+
node_name: "#{node[:node_name].to_s}",
|
275
|
+
msg: "Node '#{node[:node_name].to_s}' timed out while reseting."
|
276
|
+
}, :ALL)
|
277
|
+
end
|
278
|
+
end
|
279
|
+
sleep 1
|
280
|
+
end
|
281
|
+
|
282
|
+
work("start_node_pxe") do |res, node|
|
283
|
+
resp = res.get_status(node)
|
284
|
+
node[:node_mac] = node[:node_mac].downcase.gsub(/:/, '-')
|
285
|
+
if resp == :on
|
286
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
287
|
+
if !File.exists?("#{symlink_name}")
|
288
|
+
File.symlink("/tftpboot/pxelinux.cfg/#{@config[:pxeSymLinkConfFile]}", "#{symlink_name}")
|
289
|
+
end
|
290
|
+
debug "Start_node_pxe RESET: http://#{node[:node_cm_ip].to_s}/reset"
|
291
|
+
begin
|
292
|
+
open("http://#{node[:node_cm_ip].to_s}/reset")
|
293
|
+
rescue
|
294
|
+
res.inform(:error, {
|
295
|
+
event_type: "HTTP",
|
296
|
+
exit_code: "-1",
|
297
|
+
node_name: "#{node[:node_name].to_s}",
|
298
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
299
|
+
}, :ALL)
|
300
|
+
next
|
301
|
+
end
|
302
|
+
elsif resp == :off
|
303
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
304
|
+
if !File.exists?("#{symlink_name}")
|
305
|
+
File.symlink("/tftpboot/pxelinux.cfg/#{@config[:pxeSymLinkConfFile]}", "#{symlink_name}")
|
306
|
+
end
|
307
|
+
debug "Start_node_pxe ON: http://#{node[:node_cm_ip].to_s}/on"
|
308
|
+
begin
|
309
|
+
open("http://#{node[:node_cm_ip].to_s}/on")
|
310
|
+
rescue
|
311
|
+
res.inform(:error, {
|
312
|
+
event_type: "HTTP",
|
313
|
+
exit_code: "-1",
|
314
|
+
node_name: "#{node[:node_name].to_s}",
|
315
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
316
|
+
}, :ALL)
|
317
|
+
next
|
318
|
+
end
|
319
|
+
elsif resp == :started_on_pxe
|
320
|
+
debug "Start_node_pxe STARTED: http://#{node[:node_cm_ip].to_s}/reset"
|
321
|
+
begin
|
322
|
+
open("http://#{node[:node_cm_ip].to_s}/reset")
|
323
|
+
rescue
|
324
|
+
res.inform(:error, {
|
325
|
+
event_type: "HTTP",
|
326
|
+
exit_code: "-1",
|
327
|
+
node_name: "#{node[:node_name].to_s}",
|
328
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
329
|
+
}, :ALL)
|
330
|
+
next
|
331
|
+
end
|
332
|
+
end
|
333
|
+
|
334
|
+
if res.wait_until_ping(res, node[:node_ip])
|
335
|
+
res.inform(:status, {
|
336
|
+
node_name: "#{node[:node_name].to_s}",
|
337
|
+
current: :pxe_on,
|
338
|
+
desired: :pxe_on
|
339
|
+
}, :ALL)
|
340
|
+
else
|
341
|
+
node[:node_mac] = node[:node_mac].downcase.gsub(/:/, '-')
|
342
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
343
|
+
if File.exists?(symlink_name)
|
344
|
+
File.delete(symlink_name)
|
345
|
+
end
|
346
|
+
res.inform(:error, {
|
347
|
+
event_type: "TIME_OUT",
|
348
|
+
exit_code: "-1",
|
349
|
+
node_name: "#{node[:node_name].to_s}",
|
350
|
+
msg: "Node '#{node[:node_name].to_s}' timed out while trying to boot on PXE."
|
351
|
+
}, :ALL)
|
352
|
+
end
|
353
|
+
sleep 1
|
354
|
+
end
|
355
|
+
|
356
|
+
work("start_node_pxe_off") do |res, node, action|
|
357
|
+
node[:node_mac] = node[:node_mac].downcase.gsub(/:/, '-')
|
358
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
359
|
+
if File.exists?(symlink_name)
|
360
|
+
File.delete(symlink_name)
|
361
|
+
end
|
362
|
+
if action == "reset"
|
363
|
+
debug "Start_node_pxe_off RESET: http://#{node[:node_cm_ip].to_s}/reset"
|
364
|
+
begin
|
365
|
+
open("http://#{node[:node_cm_ip].to_s}/reset")
|
366
|
+
rescue
|
367
|
+
res.inform(:error, {
|
368
|
+
event_type: "HTTP",
|
369
|
+
exit_code: "-1",
|
370
|
+
node_name: "#{node[:node_name].to_s}",
|
371
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
372
|
+
}, :ALL)
|
373
|
+
next
|
374
|
+
end
|
375
|
+
|
376
|
+
t = 0
|
377
|
+
if res.wait_until_ping(node[:node_ip])
|
378
|
+
res.inform(:status, {
|
379
|
+
node_name: "#{node[:node_name].to_s}",
|
380
|
+
current: :pxe_off,
|
381
|
+
desired: :pxe_off
|
382
|
+
}, :ALL)
|
383
|
+
else
|
384
|
+
res.inform(:error, {
|
385
|
+
event_type: "TIME_OUT",
|
386
|
+
exit_code: "-1",
|
387
|
+
node_name: "#{node[:node_name].to_s}",
|
388
|
+
msg: "Node '#{node[:node_name].to_s}' timed out while booting."
|
389
|
+
}, :ALL)
|
390
|
+
end
|
391
|
+
elsif action == "shutdown"
|
392
|
+
debug "Start_node_pxe_off SHUTDOWN: http://#{node[:node_cm_ip].to_s}/off"
|
393
|
+
begin
|
394
|
+
open("http://#{node[:node_cm_ip].to_s}/off")
|
395
|
+
rescue
|
396
|
+
res.inform(:error, {
|
397
|
+
event_type: "HTTP",
|
398
|
+
exit_code: "-1",
|
399
|
+
node_name: "#{node[:node_name].to_s}",
|
400
|
+
msg: "failed to reach cm, ip: #{node[:node_cm_ip].to_s}."
|
401
|
+
}, :ALL)
|
402
|
+
next
|
403
|
+
end
|
404
|
+
|
405
|
+
if res.wait_until_no_ping(node[:node_ip])
|
406
|
+
res.inform(:status, {
|
407
|
+
node_name: "#{node[:node_name].to_s}",
|
408
|
+
current: :pxe_off,
|
409
|
+
desired: :pxe_off
|
410
|
+
}, :ALL)
|
411
|
+
else
|
412
|
+
res.inform(:error, {
|
413
|
+
event_type: "TIME_OUT",
|
414
|
+
exit_code: "-1",
|
415
|
+
node_name: "#{node[:node_name].to_s}",
|
416
|
+
msg: "Node '#{node[:node_name].to_s}' timed out while shutting down."
|
417
|
+
}, :ALL)
|
418
|
+
end
|
419
|
+
end
|
420
|
+
sleep 1
|
421
|
+
end
|
422
|
+
|
423
|
+
#this is used by other methods in this scope
|
424
|
+
def wait_until_ping(res, ip)
|
425
|
+
t = 0
|
426
|
+
resp = false
|
427
|
+
loop do
|
428
|
+
sleep 2
|
429
|
+
status = system("ping #{ip} -c 2 -w 2")
|
430
|
+
if t < @@timeout
|
431
|
+
if status == true
|
432
|
+
resp = true
|
433
|
+
break
|
434
|
+
end
|
435
|
+
else
|
436
|
+
resp = false
|
437
|
+
break
|
438
|
+
end
|
439
|
+
t += 2
|
440
|
+
end
|
441
|
+
resp
|
442
|
+
end
|
443
|
+
|
444
|
+
#this is used by other methods in this scope
|
445
|
+
work("wait_until_no_ping") do |res, ip|
|
446
|
+
t = 0
|
447
|
+
resp = false
|
448
|
+
loop do
|
449
|
+
sleep 2
|
450
|
+
status = system("ping #{ip} -c 2 -w 2")
|
451
|
+
if t < @@timeout
|
452
|
+
if status == false
|
453
|
+
resp = true
|
454
|
+
break
|
455
|
+
end
|
456
|
+
else
|
457
|
+
resp = false
|
458
|
+
break
|
459
|
+
end
|
460
|
+
t += 2
|
461
|
+
end
|
462
|
+
resp
|
463
|
+
end
|
464
|
+
|
465
|
+
#this is used by other methods in this scope
|
466
|
+
work("get_status") do |res, node|
|
467
|
+
debug "http://#{node[:node_cm_ip].to_s}/state"
|
468
|
+
resp = open("http://#{node[:node_cm_ip].to_s}/state")
|
469
|
+
resp = res.parse_responce(resp, "//Response//line//value")
|
470
|
+
debug "state response: #{resp}"
|
471
|
+
|
472
|
+
if resp == 'on'
|
473
|
+
symlink_name = "/tftpboot/pxelinux.cfg/01-#{node[:node_mac]}"
|
474
|
+
if File.exists?("#{symlink_name}")
|
475
|
+
:on_pxe
|
476
|
+
else
|
477
|
+
:on
|
478
|
+
end
|
479
|
+
elsif resp == 'off'
|
480
|
+
:off
|
481
|
+
end
|
482
|
+
end
|
483
|
+
|
484
|
+
work("parse_responce") do |res, input, path|
|
485
|
+
input = input.string if input.kind_of? StringIO
|
486
|
+
if input[0] == "<"
|
487
|
+
output = Nokogiri::XML(input).xpath(path).text.strip
|
488
|
+
else
|
489
|
+
output = input.strip
|
490
|
+
end
|
491
|
+
output
|
492
|
+
end
|
493
|
+
end
|