Codestyle updates for locking code
[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-local-mirror-disk')
76 def TestInstanceAddWithLocalMirrorDisk(node):
77   """gnt-instance add -t local_raid1"""
78   return _DiskTest(node['primary'], 'local_raid1')
79
80
81 @qa_utils.DefineHook('instance-add-remote-raid-disk')
82 def TestInstanceAddWithRemoteRaidDisk(node, node2):
83   """gnt-instance add -t remote_raid1"""
84   return _DiskTest("%s:%s" % (node['primary'], node2['primary']),
85                    'remote_raid1')
86
87
88 @qa_utils.DefineHook('instance-add-drbd-disk')
89 def TestInstanceAddWithDrbdDisk(node, node2):
90   """gnt-instance add -t drbd"""
91   return _DiskTest("%s:%s" % (node['primary'], node2['primary']),
92                    'drbd')
93
94
95 @qa_utils.DefineHook('instance-remove')
96 def TestInstanceRemove(instance):
97   """gnt-instance remove"""
98   master = qa_config.GetMasterNode()
99
100   cmd = ['gnt-instance', 'remove', '-f', instance['name']]
101   AssertEqual(StartSSH(master['primary'],
102                        utils.ShellQuoteArgs(cmd)).wait(), 0)
103
104   qa_config.ReleaseInstance(instance)
105
106
107 @qa_utils.DefineHook('instance-startup')
108 def TestInstanceStartup(instance):
109   """gnt-instance startup"""
110   master = qa_config.GetMasterNode()
111
112   cmd = ['gnt-instance', 'startup', instance['name']]
113   AssertEqual(StartSSH(master['primary'],
114                        utils.ShellQuoteArgs(cmd)).wait(), 0)
115
116
117 @qa_utils.DefineHook('instance-shutdown')
118 def TestInstanceShutdown(instance):
119   """gnt-instance shutdown"""
120   master = qa_config.GetMasterNode()
121
122   cmd = ['gnt-instance', 'shutdown', instance['name']]
123   AssertEqual(StartSSH(master['primary'],
124                        utils.ShellQuoteArgs(cmd)).wait(), 0)
125
126
127 @qa_utils.DefineHook('instance-reinstall')
128 def TestInstanceReinstall(instance):
129   """gnt-instance reinstall"""
130   master = qa_config.GetMasterNode()
131
132   cmd = ['gnt-instance', 'reinstall', '-f', instance['name']]
133   AssertEqual(StartSSH(master['primary'],
134                        utils.ShellQuoteArgs(cmd)).wait(), 0)
135
136
137 @qa_utils.DefineHook('instance-failover')
138 def TestInstanceFailover(instance):
139   """gnt-instance failover"""
140   master = qa_config.GetMasterNode()
141
142   cmd = ['gnt-instance', 'failover', '--force', instance['name']]
143   AssertEqual(StartSSH(master['primary'],
144                        utils.ShellQuoteArgs(cmd)).wait(), 0)
145
146
147 @qa_utils.DefineHook('instance-info')
148 def TestInstanceInfo(instance):
149   """gnt-instance info"""
150   master = qa_config.GetMasterNode()
151
152   cmd = ['gnt-instance', 'info', instance['name']]
153   AssertEqual(StartSSH(master['primary'],
154                        utils.ShellQuoteArgs(cmd)).wait(), 0)
155
156
157 @qa_utils.DefineHook('instance-modify')
158 def TestInstanceModify(instance):
159   """gnt-instance modify"""
160   master = qa_config.GetMasterNode()
161
162   orig_memory = qa_config.get('mem')
163   orig_bridge = qa_config.get('bridge', 'xen-br0')
164   args = [
165     ["--memory", "128"],
166     ["--memory", str(orig_memory)],
167     ["--cpu", "2"],
168     ["--cpu", "1"],
169     ["--bridge", "xen-br1"],
170     ["--bridge", orig_bridge],
171     ["--kernel", "/dev/null"],
172     ["--kernel", "default"],
173     ["--initrd", "/dev/null"],
174     ["--initrd", "none"],
175     ["--initrd", "default"],
176     ["--hvm-boot-order", "acn"],
177     ["--hvm-boot-order", "default"],
178     ]
179   for alist in args:
180     cmd = ['gnt-instance', 'modify'] + alist + [instance['name']]
181     AssertEqual(StartSSH(master['primary'],
182                          utils.ShellQuoteArgs(cmd)).wait(), 0)
183
184   # check no-modify
185   cmd = ['gnt-instance', 'modify', instance['name']]
186   AssertNotEqual(StartSSH(master['primary'],
187                           utils.ShellQuoteArgs(cmd)).wait(), 0)
188
189
190 @qa_utils.DefineHook('instance-list')
191 def TestInstanceList():
192   """gnt-instance list"""
193   master = qa_config.GetMasterNode()
194
195   cmd = ['gnt-instance', 'list']
196   AssertEqual(StartSSH(master['primary'],
197                        utils.ShellQuoteArgs(cmd)).wait(), 0)
198
199
200 @qa_utils.DefineHook('backup-export')
201 def TestInstanceExport(instance, node):
202   """gnt-backup export"""
203   master = qa_config.GetMasterNode()
204
205   cmd = ['gnt-backup', 'export', '-n', node['primary'], instance['name']]
206   AssertEqual(StartSSH(master['primary'],
207                        utils.ShellQuoteArgs(cmd)).wait(), 0)
208
209   return qa_utils.ResolveInstanceName(instance)
210
211
212 @qa_utils.DefineHook('backup-import')
213 def TestInstanceImport(node, newinst, expnode, name):
214   """gnt-backup import"""
215   master = qa_config.GetMasterNode()
216
217   cmd = (['gnt-backup', 'import',
218           '--disk-template=plain',
219           '--no-ip-check',
220           '--src-node=%s' % expnode['primary'],
221           '--src-dir=%s/%s' % (constants.EXPORT_DIR, name),
222           '--node=%s' % node['primary']] +
223          _GetGenericAddParameters())
224   cmd.append(newinst['name'])
225   AssertEqual(StartSSH(master['primary'],
226                        utils.ShellQuoteArgs(cmd)).wait(), 0)
227
228
229 @qa_utils.DefineHook('backup-list')
230 def TestBackupList(expnode):
231   """gnt-backup list"""
232   master = qa_config.GetMasterNode()
233
234   cmd = ['gnt-backup', 'list', '--node=%s' % expnode['primary']]
235   AssertEqual(StartSSH(master['primary'],
236                        utils.ShellQuoteArgs(cmd)).wait(), 0)
237
238
239 def _TestInstanceDiskFailure(instance, node, node2, onmaster):
240   """Testing disk failure."""
241   master = qa_config.GetMasterNode()
242   sq = utils.ShellQuoteArgs
243
244   instance_full = qa_utils.ResolveInstanceName(instance)
245   node_full = qa_utils.ResolveNodeName(node)
246   node2_full = qa_utils.ResolveNodeName(node2)
247
248   print qa_utils.FormatInfo("Getting physical disk names")
249   cmd = ['gnt-node', 'volumes', '--separator=|', '--no-headers',
250          '--output=node,phys,instance',
251          node['primary'], node2['primary']]
252   output = qa_utils.GetCommandOutput(master['primary'], sq(cmd))
253
254   # Get physical disk names
255   re_disk = re.compile(r'^/dev/([a-z]+)\d+$')
256   node2disk = {}
257   for line in output.splitlines():
258     (node_name, phys, inst) = line.split('|')
259     if inst == instance_full:
260       if node_name not in node2disk:
261         node2disk[node_name] = []
262
263       m = re_disk.match(phys)
264       if not m:
265         raise qa_error.Error("Unknown disk name format: %s" % disk)
266
267       name = m.group(1)
268       if name not in node2disk[node_name]:
269         node2disk[node_name].append(name)
270
271   if [node2_full, node_full][int(onmaster)] not in node2disk:
272     raise qa_error.Error("Couldn't find physical disks used on"
273                          " %s node" % ["secondary", "master"][int(onmaster)])
274
275   print qa_utils.FormatInfo("Checking whether nodes have ability to stop"
276                             " disks")
277   for node_name, disks in node2disk.iteritems():
278     cmds = []
279     for disk in disks:
280       cmds.append(sq(["test", "-f", _GetDiskStatePath(disk)]))
281     AssertEqual(StartSSH(node_name, ' && '.join(cmds)).wait(), 0)
282
283   print qa_utils.FormatInfo("Getting device paths")
284   cmd = ['gnt-instance', 'activate-disks', instance['name']]
285   output = qa_utils.GetCommandOutput(master['primary'], sq(cmd))
286   devpath = []
287   for line in output.splitlines():
288     (_, _, tmpdevpath) = line.split(':')
289     devpath.append(tmpdevpath)
290   print devpath
291
292   print qa_utils.FormatInfo("Getting drbd device paths")
293   cmd = ['gnt-instance', 'info', instance['name']]
294   output = qa_utils.GetCommandOutput(master['primary'], sq(cmd))
295   pattern = (r'\s+-\s+sd[a-z]+,\s+type:\s+drbd8?,\s+.*$'
296              r'\s+primary:\s+(/dev/drbd\d+)\s+')
297   drbddevs = re.findall(pattern, output, re.M)
298   print drbddevs
299
300   halted_disks = []
301   try:
302     print qa_utils.FormatInfo("Deactivating disks")
303     cmds = []
304     for name in node2disk[[node2_full, node_full][int(onmaster)]]:
305       halted_disks.append(name)
306       cmds.append(sq(["echo", "offline"]) + " >%s" % _GetDiskStatePath(name))
307     AssertEqual(StartSSH([node2, node][int(onmaster)]['primary'],
308                          ' && '.join(cmds)).wait(), 0)
309
310     print qa_utils.FormatInfo("Write to disks and give some time to notice"
311                               " to notice the problem")
312     cmds = []
313     for disk in devpath:
314       cmds.append(sq(["dd", "count=1", "bs=512", "conv=notrunc",
315                       "if=%s" % disk, "of=%s" % disk]))
316     for _ in (0, 1, 2):
317       AssertEqual(StartSSH(node['primary'], ' && '.join(cmds)).wait(), 0)
318       time.sleep(3)
319
320     print qa_utils.FormatInfo("Debugging info")
321     for name in drbddevs:
322       cmd = ['drbdsetup', name, 'show']
323       AssertEqual(StartSSH(node['primary'], sq(cmd)).wait(), 0)
324
325     cmd = ['gnt-instance', 'info', instance['name']]
326     AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
327
328   finally:
329     print qa_utils.FormatInfo("Activating disks again")
330     cmds = []
331     for name in halted_disks:
332       cmds.append(sq(["echo", "running"]) + " >%s" % _GetDiskStatePath(name))
333     AssertEqual(StartSSH([node2, node][int(onmaster)]['primary'],
334                          '; '.join(cmds)).wait(), 0)
335
336   if onmaster:
337     for name in drbddevs:
338       cmd = ['drbdsetup', name, 'detach']
339       AssertEqual(StartSSH(node['primary'], sq(cmd)).wait(), 0)
340   else:
341     for name in drbddevs:
342       cmd = ['drbdsetup', name, 'disconnect']
343       AssertEqual(StartSSH(node2['primary'], sq(cmd)).wait(), 0)
344
345   # TODO
346   #cmd = ['vgs']
347   #AssertEqual(StartSSH([node2, node][int(onmaster)]['primary'],
348   #                     sq(cmd)).wait(), 0)
349
350   print qa_utils.FormatInfo("Making sure disks are up again")
351   cmd = ['gnt-instance', 'replace-disks', instance['name']]
352   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
353
354   print qa_utils.FormatInfo("Restarting instance")
355   cmd = ['gnt-instance', 'shutdown', instance['name']]
356   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
357
358   cmd = ['gnt-instance', 'startup', instance['name']]
359   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
360
361   cmd = ['gnt-cluster', 'verify']
362   AssertEqual(StartSSH(master['primary'], sq(cmd)).wait(), 0)
363
364
365 def TestInstanceMasterDiskFailure(instance, node, node2):
366   """Testing disk failure on master node."""
367   print qa_utils.FormatError("Disk failure on primary node cannot be"
368                              " tested due to potential crashes.")
369   # The following can cause crashes, thus it's disabled until fixed
370   #return _TestInstanceDiskFailure(instance, node, node2, True)
371
372
373 def TestInstanceSecondaryDiskFailure(instance, node, node2):
374   """Testing disk failure on secondary node."""
375   return _TestInstanceDiskFailure(instance, node, node2, False)