Modify Disk.GetNodes() to support LD_FILE
[ganeti-local] / qa / qa_instance.py
1 #
2 #
3
4 # Copyright (C) 2007 Google Inc.
5 #
6 # This program is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2 of the License, or
9 # (at your option) any later version.
10 #
11 # This program is distributed in the hope that it will be useful, but
12 # WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 # General Public License for more details.
15 #
16 # You should have received a copy of the GNU General Public License
17 # along with this program; if not, write to the Free Software
18 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 # 02110-1301, USA.
20
21
22 """Instance related QA tests.
23
24 """
25
26 import re
27 import time
28
29 from ganeti import utils
30 from ganeti import constants
31
32 import qa_config
33 import qa_utils
34 import qa_error
35
36 from qa_utils import AssertEqual, AssertNotEqual, StartSSH
37
38
39 def _GetDiskStatePath(disk):
40   return "/sys/block/%s/device/state" % disk
41
42
43 def _GetGenericAddParameters():
44   return ['--os-size=%s' % qa_config.get('os-size'),
45           '--swap-size=%s' % qa_config.get('swap-size'),
46           '--memory=%s' % qa_config.get('mem')]
47
48
49 def _DiskTest(node, disk_template):
50   master = qa_config.GetMasterNode()
51
52   instance = qa_config.AcquireInstance()
53   try:
54     cmd = (['gnt-instance', 'add',
55             '--os-type=%s' % qa_config.get('os'),
56             '--disk-template=%s' % disk_template,
57             '--node=%s' % node] +
58            _GetGenericAddParameters())
59     cmd.append(instance['name'])
60
61     AssertEqual(StartSSH(master['primary'],
62                          utils.ShellQuoteArgs(cmd)).wait(), 0)
63     return instance
64   except:
65     qa_config.ReleaseInstance(instance)
66     raise
67
68
69 @qa_utils.DefineHook('instance-add-plain-disk')
70 def TestInstanceAddWithPlainDisk(node):
71   """gnt-instance add -t plain"""
72   return _DiskTest(node['primary'], 'plain')
73
74
75 @qa_utils.DefineHook('instance-add-drbd-disk')
76 def TestInstanceAddWithDrbdDisk(node, node2):
77   """gnt-instance add -t drbd"""
78   return _DiskTest("%s:%s" % (node['primary'], node2['primary']),
79                    'drbd')
80
81
82 @qa_utils.DefineHook('instance-remove')
83 def TestInstanceRemove(instance):
84   """gnt-instance remove"""
85   master = qa_config.GetMasterNode()
86
87   cmd = ['gnt-instance', 'remove', '-f', instance['name']]
88   AssertEqual(StartSSH(master['primary'],
89                        utils.ShellQuoteArgs(cmd)).wait(), 0)
90
91   qa_config.ReleaseInstance(instance)
92
93
94 @qa_utils.DefineHook('instance-startup')
95 def TestInstanceStartup(instance):
96   """gnt-instance startup"""
97   master = qa_config.GetMasterNode()
98
99   cmd = ['gnt-instance', 'startup', instance['name']]
100   AssertEqual(StartSSH(master['primary'],
101                        utils.ShellQuoteArgs(cmd)).wait(), 0)
102
103
104 @qa_utils.DefineHook('instance-shutdown')
105 def TestInstanceShutdown(instance):
106   """gnt-instance shutdown"""
107   master = qa_config.GetMasterNode()
108
109   cmd = ['gnt-instance', 'shutdown', instance['name']]
110   AssertEqual(StartSSH(master['primary'],
111                        utils.ShellQuoteArgs(cmd)).wait(), 0)
112
113
114 @qa_utils.DefineHook('instance-reinstall')
115 def TestInstanceReinstall(instance):
116   """gnt-instance reinstall"""
117   master = qa_config.GetMasterNode()
118
119   cmd = ['gnt-instance', 'reinstall', '-f', instance['name']]
120   AssertEqual(StartSSH(master['primary'],
121                        utils.ShellQuoteArgs(cmd)).wait(), 0)
122
123
124 @qa_utils.DefineHook('instance-failover')
125 def TestInstanceFailover(instance):
126   """gnt-instance failover"""
127   master = qa_config.GetMasterNode()
128
129   cmd = ['gnt-instance', 'failover', '--force', instance['name']]
130   AssertEqual(StartSSH(master['primary'],
131                        utils.ShellQuoteArgs(cmd)).wait(), 0)
132
133
134 @qa_utils.DefineHook('instance-info')
135 def TestInstanceInfo(instance):
136   """gnt-instance info"""
137   master = qa_config.GetMasterNode()
138
139   cmd = ['gnt-instance', 'info', instance['name']]
140   AssertEqual(StartSSH(master['primary'],
141                        utils.ShellQuoteArgs(cmd)).wait(), 0)
142
143
144 @qa_utils.DefineHook('instance-modify')
145 def TestInstanceModify(instance):
146   """gnt-instance modify"""
147   master = qa_config.GetMasterNode()
148
149   orig_memory = qa_config.get('mem')
150   orig_bridge = qa_config.get('bridge', 'xen-br0')
151   args = [
152     ["--memory", "128"],
153     ["--memory", str(orig_memory)],
154     ["--cpu", "2"],
155     ["--cpu", "1"],
156     ["--bridge", "xen-br1"],
157     ["--bridge", orig_bridge],
158     ["--kernel", "/dev/null"],
159     ["--kernel", "default"],
160     ["--initrd", "/dev/null"],
161     ["--initrd", "none"],
162     ["--initrd", "default"],
163     ["--hvm-boot-order", "acn"],
164     ["--hvm-boot-order", "default"],
165     ]
166   for alist in args:
167     cmd = ['gnt-instance', 'modify'] + alist + [instance['name']]
168     AssertEqual(StartSSH(master['primary'],
169                          utils.ShellQuoteArgs(cmd)).wait(), 0)
170
171   # check no-modify
172   cmd = ['gnt-instance', 'modify', instance['name']]
173   AssertNotEqual(StartSSH(master['primary'],
174                           utils.ShellQuoteArgs(cmd)).wait(), 0)
175
176
177 @qa_utils.DefineHook('instance-list')
178 def TestInstanceList():
179   """gnt-instance list"""
180   master = qa_config.GetMasterNode()
181
182   cmd = ['gnt-instance', 'list']
183   AssertEqual(StartSSH(master['primary'],
184                        utils.ShellQuoteArgs(cmd)).wait(), 0)
185
186
187 @qa_utils.DefineHook('backup-export')
188 def TestInstanceExport(instance, node):
189   """gnt-backup export"""
190   master = qa_config.GetMasterNode()
191
192   cmd = ['gnt-backup', 'export', '-n', node['primary'], instance['name']]
193   AssertEqual(StartSSH(master['primary'],
194                        utils.ShellQuoteArgs(cmd)).wait(), 0)
195
196   return qa_utils.ResolveInstanceName(instance)
197
198
199 @qa_utils.DefineHook('backup-import')
200 def TestInstanceImport(node, newinst, expnode, name):
201   """gnt-backup import"""
202   master = qa_config.GetMasterNode()
203
204   cmd = (['gnt-backup', 'import',
205           '--disk-template=plain',
206           '--no-ip-check',
207           '--src-node=%s' % expnode['primary'],
208           '--src-dir=%s/%s' % (constants.EXPORT_DIR, name),
209           '--node=%s' % node['primary']] +
210          _GetGenericAddParameters())
211   cmd.append(newinst['name'])
212   AssertEqual(StartSSH(master['primary'],
213                        utils.ShellQuoteArgs(cmd)).wait(), 0)
214
215
216 @qa_utils.DefineHook('backup-list')
217 def TestBackupList(expnode):
218   """gnt-backup list"""
219   master = qa_config.GetMasterNode()
220
221   cmd = ['gnt-backup', 'list', '--node=%s' % expnode['primary']]
222   AssertEqual(StartSSH(master['primary'],
223                        utils.ShellQuoteArgs(cmd)).wait(), 0)
224
225
226 def _TestInstanceDiskFailure(instance, node, node2, onmaster):
227   """Testing disk failure."""
228   master = qa_config.GetMasterNode()
229   sq = utils.ShellQuoteArgs
230
231   instance_full = qa_utils.ResolveInstanceName(instance)
232   node_full = qa_utils.ResolveNodeName(node)
233   node2_full = qa_utils.ResolveNodeName(node2)
234
235   print qa_utils.FormatInfo("Getting physical disk names")
236   cmd = ['gnt-node', 'volumes', '--separator=|', '--no-headers',
237          '--output=node,phys,instance',
238          node['primary'], node2['primary']]
239   output = qa_utils.GetCommandOutput(master['primary'], sq(cmd))
240
241   # Get physical disk names
242   re_disk = re.compile(r'^/dev/([a-z]+)\d+$')
243   node2disk = {}
244   for line in output.splitlines():
245     (node_name, phys, inst) = line.split('|')
246     if inst == instance_full:
247       if node_name not in node2disk:
248         node2disk[node_name] = []
249
250       m = re_disk.match(phys)
251       if not m:
252         raise qa_error.Error("Unknown disk name format: %s" % disk)
253
254       name = m.group(1)
255       if name not in node2disk[node_name]:
256         node2disk[node_name].append(name)
257
258   if [node2_full, node_full][int(onmaster)] not in node2disk:
259     raise qa_error.Error("Couldn't find physical disks used on"
260                          " %s node" % ["secondary", "master"][int(onmaster)])
261
262   print qa_utils.FormatInfo("Checking whether nodes have ability to stop"
263                             " disks")
264   for node_name, disks in node2disk.iteritems():
265     cmds = []
266     for disk in disks:
267       cmds.append(sq(["test", "-f", _GetDiskStatePath(disk)]))
268     AssertEqual(StartSSH(node_name, ' && '.join(cmds)).wait(), 0)
269
270   print qa_utils.FormatInfo("Getting device paths")
271   cmd = ['gnt-instance', 'activate-disks', instance['name']]
272   output = qa_utils.GetCommandOutput(master['primary'], sq(cmd))
273   devpath = []
274   for line in output.splitlines():
275     (_, _, tmpdevpath) = line.split(':')
276     devpath.append(tmpdevpath)
277   print devpath
278
279   print qa_utils.FormatInfo("Getting drbd device paths")
280   cmd = ['gnt-instance', 'info', instance['name']]
281   output = qa_utils.GetCommandOutput(master['primary'], sq(cmd))
282   pattern = (r'\s+-\s+sd[a-z]+,\s+type:\s+drbd8?,\s+.*$'
283              r'\s+primary:\s+(/dev/drbd\d+)\s+')
284   drbddevs = re.findall(pattern, output, re.M)
285   print drbddevs
286
287   halted_disks = []
288   try:
289     print qa_utils.FormatInfo("Deactivating disks")
290     cmds = []
291     for name in node2disk[[node2_full, node_full][int(onmaster)]]:
292       halted_disks.append(name)
293       cmds.append(sq(["echo", "offline"]) + " >%s" % _GetDiskStatePath(name))
294     AssertEqual(StartSSH([node2, node][int(onmaster)]['primary'],
295                          ' && '.join(cmds)).wait(), 0)
296
297     print qa_utils.FormatInfo("Write to disks and give some time to notice"
298                               " to notice the problem")
299     cmds = []
300     for disk in devpath:
301       cmds.append(sq(["dd", "count=1", "bs=512", "conv=notrunc",
302                       "if=%s" % disk, "of=%s" % disk]))
303     for _ in (0, 1, 2):
304       AssertEqual(StartSSH(node['primary'], ' && '.join(cmds)).wait(), 0)
305       time.sleep(3)
306
307     print qa_utils.FormatInfo("Debugging info")
308     for name in drbddevs:
309       cmd = ['drbdsetup', name, 'show']
310       AssertEqual(StartSSH(node['primary'], sq(cmd)).wait(), 0)
311
312     cmd = ['gnt-instance', 'info', instance['name']]
313     AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
314
315   finally:
316     print qa_utils.FormatInfo("Activating disks again")
317     cmds = []
318     for name in halted_disks:
319       cmds.append(sq(["echo", "running"]) + " >%s" % _GetDiskStatePath(name))
320     AssertEqual(StartSSH([node2, node][int(onmaster)]['primary'],
321                          '; '.join(cmds)).wait(), 0)
322
323   if onmaster:
324     for name in drbddevs:
325       cmd = ['drbdsetup', name, 'detach']
326       AssertEqual(StartSSH(node['primary'], sq(cmd)).wait(), 0)
327   else:
328     for name in drbddevs:
329       cmd = ['drbdsetup', name, 'disconnect']
330       AssertEqual(StartSSH(node2['primary'], sq(cmd)).wait(), 0)
331
332   # TODO
333   #cmd = ['vgs']
334   #AssertEqual(StartSSH([node2, node][int(onmaster)]['primary'],
335   #                     sq(cmd)).wait(), 0)
336
337   print qa_utils.FormatInfo("Making sure disks are up again")
338   cmd = ['gnt-instance', 'replace-disks', instance['name']]
339   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
340
341   print qa_utils.FormatInfo("Restarting instance")
342   cmd = ['gnt-instance', 'shutdown', instance['name']]
343   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
344
345   cmd = ['gnt-instance', 'startup', instance['name']]
346   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
347
348   cmd = ['gnt-cluster', 'verify']
349   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
350
351
352 def TestInstanceMasterDiskFailure(instance, node, node2):
353   """Testing disk failure on master node."""
354   print qa_utils.FormatError("Disk failure on primary node cannot be"
355                              " tested due to potential crashes.")
356   # The following can cause crashes, thus it's disabled until fixed
357   #return _TestInstanceDiskFailure(instance, node, node2, True)
358
359
360 def TestInstanceSecondaryDiskFailure(instance, node, node2):
361   """Testing disk failure on secondary node."""
362   return _TestInstanceDiskFailure(instance, node, node2, False)