virtual_test_sheriff.py@ 65802

Last change on this file since 65802 was 65802, checked in by vboxsync, 8 years ago
vsheriff: check out kvm_lock_spinning guest stacks.
Property svn:eol-style set to `LF` Property svn:executable set to ``* Property svn:keywords set to `Author Date Id Revision`
File size: 60.4 KB

Line
1	#!/usr/bin/env python
2	# -- coding: utf-8 --
3	# $Id: virtual_test_sheriff.py 65802 2017-02-17 12:59:39Z vboxsync $
4	# pylint: disable=C0301
5
6	"""
7	Virtual Test Sheriff.
8
9	Duties:
10	- Try to a assign failure reasons to recently failed tests.
11	- Reboot or disable bad test boxes.
12
13	"""
14
15	__copyright__ = \
16	"""
17	Copyright (C) 2012-2016 Oracle Corporation
18
19	This file is part of VirtualBox Open Source Edition (OSE), as
20	available from http://www.215389.xyz. This file is free software;
21	you can redistribute it and/or modify it under the terms of the GNU
22	General Public License (GPL) as published by the Free Software
23	Foundation, in version 2 as it comes in the "COPYING" file of the
24	VirtualBox OSE distribution. VirtualBox OSE is distributed in the
25	hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
26
27	The contents of this file may alternatively be used under the terms
28	of the Common Development and Distribution License Version 1.0
29	(CDDL) only, as it comes in the "COPYING.CDDL" file of the
30	VirtualBox OSE distribution, in which case the provisions of the
31	CDDL are applicable instead of those of the GPL.
32
33	You may elect to license modified versions of this file under the
34	terms and conditions of either the GPL or the CDDL or both.
35	"""
36	__version__ = "$Revision: 65802 $"
37
38
39	# Standard python imports
40	import sys;
41	import os;
42	import hashlib;
43	import StringIO;
44	from optparse import OptionParser;
45	from PIL import Image; # pylint: disable=import-error
46
47	# Add Test Manager's modules path
48	g_ksTestManagerDir = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))));
49	sys.path.append(g_ksTestManagerDir);
50
51	# Test Manager imports
52	from testmanager.core.db import TMDatabaseConnection;
53	from testmanager.core.build import BuildDataEx;
54	from testmanager.core.failurereason import FailureReasonLogic;
55	from testmanager.core.testbox import TestBoxLogic, TestBoxData;
56	from testmanager.core.testcase import TestCaseDataEx;
57	from testmanager.core.testgroup import TestGroupData;
58	from testmanager.core.testset import TestSetLogic, TestSetData;
59	from testmanager.core.testresults import TestResultLogic, TestResultFileData;
60	from testmanager.core.testresultfailures import TestResultFailureLogic, TestResultFailureData;
61	from testmanager.core.useraccount import UserAccountLogic;
62
63
64	class VirtualTestSheriffCaseFile(object):
65	"""
66	A failure investigation case file.
67
68	"""
69
70
71	## Max log file we'll read into memory. (256 MB)
72	kcbMaxLogRead = 0x10000000;
73
74	def __init__(self, oSheriff, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase):
75	self.oSheriff = oSheriff;
76	self.oTestSet = oTestSet; # TestSetData
77	self.oTree = oTree; # TestResultDataEx
78	self.oBuild = oBuild; # BuildDataEx
79	self.oTestBox = oTestBox; # TestBoxData
80	self.oTestGroup = oTestGroup; # TestGroupData
81	self.oTestCase = oTestCase; # TestCaseDataEx
82	self.sMainLog = ''; # The main log file. Empty string if not accessible.
83
84	# Generate a case file name.
85	self.sName = '#%u: %s' % (self.oTestSet.idTestSet, self.oTestCase.sName,)
86	self.sLongName = '#%u: "%s" on "%s" running %s %s (%s), "%s" by %s, using %s %s %s r%u' \
87	% ( self.oTestSet.idTestSet,
88	self.oTestCase.sName,
89	self.oTestBox.sName,
90	self.oTestBox.sOs,
91	self.oTestBox.sOsVersion,
92	self.oTestBox.sCpuArch,
93	self.oTestBox.sCpuName,
94	self.oTestBox.sCpuVendor,
95	self.oBuild.oCat.sProduct,
96	self.oBuild.oCat.sBranch,
97	self.oBuild.oCat.sType,
98	self.oBuild.iRevision, );
99
100	# Investigation notes.
101	self.tReason = None; # None or one of the ktReason_XXX constants.
102	self.dReasonForResultId = {}; # Reason assignments indexed by idTestResult.
103	self.dCommentForResultId = {}; # Comment assignments indexed by idTestResult.
104
105	#
106	# Reason.
107	#
108
109	def noteReason(self, tReason):
110	""" Notes down a possible reason. """
111	self.oSheriff.dprint(u'noteReason: %s -> %s' % (self.tReason, tReason,));
112	self.tReason = tReason;
113	return True;
114
115	def noteReasonForId(self, tReason, idTestResult, sComment = None):
116	""" Notes down a possible reason for a specific test result. """
117	self.oSheriff.dprint(u'noteReasonForId: %u: %s -> %s%s'
118	% (idTestResult, self.dReasonForResultId.get(idTestResult, None), tReason,
119	(u' (%s)' % (sComment,)) if sComment is not None else ''));
120	self.dReasonForResultId[idTestResult] = tReason;
121	if sComment is not None:
122	self.dCommentForResultId[idTestResult] = sComment;
123	return True;
124
125
126	#
127	# Test classification.
128	#
129
130	def isVBoxTest(self):
131	""" Test classification: VirtualBox (using the build) """
132	return self.oBuild.oCat.sProduct.lower() in [ 'virtualbox', 'vbox' ];
133
134	def isVBoxUnitTest(self):
135	""" Test case classification: The unit test doing all our testcase/*.cpp stuff. """
136	return self.isVBoxTest() \
137	and self.oTestCase.sName.lower() == 'unit tests';
138
139	def isVBoxInstallTest(self):
140	""" Test case classification: VirtualBox Guest installation test. """
141	return self.isVBoxTest() \
142	and self.oTestCase.sName.lower().startswith('install:');
143
144	def isVBoxUSBTest(self):
145	""" Test case classification: VirtualBox USB test. """
146	return self.isVBoxTest() \
147	and self.oTestCase.sName.lower().startswith('usb:');
148
149	def isVBoxStorageTest(self):
150	""" Test case classification: VirtualBox Storage test. """
151	return self.isVBoxTest() \
152	and self.oTestCase.sName.lower().startswith('storage:');
153
154	def isVBoxGAsTest(self):
155	""" Test case classification: VirtualBox Guest Additions test. """
156	return self.isVBoxTest() \
157	and self.oTestCase.sName.lower().startswith('ga\'s tests');
158
159	def isVBoxAPITest(self):
160	""" Test case classification: VirtualBox API test. """
161	return self.isVBoxTest() \
162	and self.oTestCase.sName.lower().startswith('api:');
163
164	def isVBoxBenchmarkTest(self):
165	""" Test case classification: VirtualBox Benchmark test. """
166	return self.isVBoxTest() \
167	and self.oTestCase.sName.lower().startswith('benchmark:');
168
169	def isVBoxSmokeTest(self):
170	""" Test case classification: Smoke test. """
171	return self.isVBoxTest() \
172	and self.oTestCase.sName.lower().startswith('smoketest');
173
174
175	#
176	# Utility methods.
177	#
178
179	def getMainLog(self):
180	"""
181	Tries to reads the main log file since this will be the first source of information.
182	"""
183	if len(self.sMainLog) > 0:
184	return self.sMainLog;
185	(oFile, oSizeOrError, _) = self.oTestSet.openFile('main.log', 'rb');
186	if oFile is not None:
187	try:
188	self.sMainLog = oFile.read(min(self.kcbMaxLogRead, oSizeOrError)).decode('utf-8', 'replace');
189	except Exception as oXcpt:
190	self.oSheriff.vprint(u'Error reading main log file: %s' % (oXcpt,))
191	self.sMainLog = '';
192	else:
193	self.oSheriff.vprint(u'Error opening main log file: %s' % (oSizeOrError,));
194	return self.sMainLog;
195
196	def getLogFile(self, oFile):
197	"""
198	Tries to reads the given file as a utf-8 log file.
199	oFile is a TestFileDataEx instance.
200	Returns empty string if problems opening or reading the file.
201	"""
202	sContent = '';
203	(oFile, oSizeOrError, _) = self.oTestSet.openFile(oFile.sFile, 'rb');
204	if oFile is not None:
205	try:
206	sContent = oFile.read(min(self.kcbMaxLogRead, oSizeOrError)).decode('utf-8', 'replace');
207	except Exception as oXcpt:
208	self.oSheriff.vprint(u'Error reading the "%s" log file: %s' % (oFile.sFile, oXcpt,))
209	else:
210	self.oSheriff.vprint(u'Error opening the "%s" log file: %s' % (oFile.sFile, oSizeOrError,));
211	return sContent;
212
213	def getScreenshotSha256(self, oFile):
214	"""
215	Tries to read the given screenshot file, uncompress it, and do SHA-2
216	on the raw pixels.
217	Returns SHA-2 digest string on success, None on failure.
218	"""
219	(oFile, _, _) = self.oTestSet.openFile(oFile.sFile, 'rb');
220	try:
221	abImageFile = oFile.read();
222	except Exception as oXcpt:
223	self.oSheriff.vprint(u'Error reading the "%s" image file: %s' % (oFile.sFile, oXcpt,))
224	else:
225	try:
226	oImage = Image.open(StringIO.StringIO(abImageFile));
227	except Exception as oXcpt:
228	self.oSheriff.vprint(u'Error opening the "%s" image bytes using PIL.Image.open: %s' % (oFile.sFile, oXcpt,))
229	else:
230	try:
231	oHash = hashlib.sha256();
232	oHash.update(oImage.tostring());
233	except Exception as oXcpt:
234	self.oSheriff.vprint(u'Error hashing the uncompressed image bytes for "%s": %s' % (oFile.sFile, oXcpt,))
235	else:
236	return oHash.hexdigest();
237	return None;
238
239
240
241	def isSingleTestFailure(self):
242	"""
243	Figure out if this is a single test failing or if it's one of the
244	more complicated ones.
245	"""
246	if self.oTree.cErrors == 1:
247	return True;
248	if self.oTree.deepCountErrorContributers() <= 1:
249	return True;
250	return False;
251
252
253
254	class VirtualTestSheriff(object): # pylint: disable=R0903
255	"""
256	Add build info into Test Manager database.
257	"""
258
259	## The user account for the virtual sheriff.
260	ksLoginName = 'vsheriff';
261
262	def __init__(self):
263	"""
264	Parse command line.
265	"""
266	self.oDb = None;
267	self.tsNow = None;
268	self.oTestResultLogic = None;
269	self.oTestSetLogic = None;
270	self.oFailureReasonLogic = None; # FailureReasonLogic;
271	self.oTestResultFailureLogic = None; # TestResultFailureLogic
272	self.oLogin = None;
273	self.uidSelf = -1;
274	self.oLogFile = None;
275	self.asBsodReasons = [];
276	self.asUnitTestReasons = [];
277
278	oParser = OptionParser();
279	oParser.add_option('--start-hours-ago', dest = 'cStartHoursAgo', metavar = '<hours>', default = 0, type = 'int',
280	help = 'When to start specified as hours relative to current time. Defauls is right now.', );
281	oParser.add_option('--hours-period', dest = 'cHoursBack', metavar = '<period-in-hours>', default = 2, type = 'int',
282	help = 'Work period specified in hours. Defauls is 2 hours.');
283	oParser.add_option('--real-run-back', dest = 'fRealRun', action = 'store_true', default = False,
284	help = 'Whether to commit the findings to the database. Default is a dry run.');
285	oParser.add_option('-q', '--quiet', dest = 'fQuiet', action = 'store_true', default = False,
286	help = 'Quiet execution');
287	oParser.add_option('-l', '--log', dest = 'sLogFile', metavar = '<logfile>', default = None,
288	help = 'Where to log messages.');
289	oParser.add_option('--debug', dest = 'fDebug', action = 'store_true', default = False,
290	help = 'Enables debug mode.');
291
292	(self.oConfig, _) = oParser.parse_args();
293
294	if self.oConfig.sLogFile is not None and len(self.oConfig.sLogFile) > 0:
295	self.oLogFile = open(self.oConfig.sLogFile, "a");
296	self.oLogFile.write('VirtualTestSheriff: $Revision: 65802 $ \n');
297
298
299	def eprint(self, sText):
300	"""
301	Prints error messages.
302	Returns 1 (for exit code usage.)
303	"""
304	print 'error: %s' % (sText,);
305	if self.oLogFile is not None:
306	self.oLogFile.write((u'error: %s\n' % (sText,)).encode('utf-8'));
307	return 1;
308
309	def dprint(self, sText):
310	"""
311	Prints debug info.
312	"""
313	if self.oConfig.fDebug:
314	if not self.oConfig.fQuiet:
315	print 'debug: %s' % (sText, );
316	if self.oLogFile is not None:
317	self.oLogFile.write((u'debug: %s\n' % (sText,)).encode('utf-8'));
318	return 0;
319
320	def vprint(self, sText):
321	"""
322	Prints verbose info.
323	"""
324	if not self.oConfig.fQuiet:
325	print 'info: %s' % (sText,);
326	if self.oLogFile is not None:
327	self.oLogFile.write((u'info: %s\n' % (sText,)).encode('utf-8'));
328	return 0;
329
330	def getFailureReason(self, tReason):
331	""" Gets the failure reason object for tReason. """
332	return self.oFailureReasonLogic.cachedLookupByNameAndCategory(tReason[1], tReason[0]);
333
334	def selfCheck(self):
335	""" Does some self checks, looking up things we expect to be in the database and such. """
336	rcExit = 0;
337	for sAttr in dir(self.__class__):
338	if sAttr.startswith('ktReason_'):
339	tReason = getattr(self.__class__, sAttr);
340	oFailureReason = self.getFailureReason(tReason);
341	if oFailureReason is None:
342	rcExit = self.eprint(u'Failed to find failure reason "%s" in category "%s" in the database!'
343	% (tReason[1], tReason[0],));
344
345	# Check the user account as well.
346	if self.oLogin is None:
347	oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
348	if oLogin is None:
349	rcExit = self.eprint(u'Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
350	return rcExit;
351
352
353
354	def badTestBoxManagement(self):
355	"""
356	Looks for bad test boxes and first tries once to reboot them then disables them.
357	"""
358	rcExit = 0;
359
360	#
361	# We skip this entirely if we're running in the past and not in harmless debug mode.
362	#
363	if self.oConfig.cStartHoursAgo != 0 \
364	and (not self.oConfig.fDebug or self.oConfig.fRealRun):
365	return rcExit;
366	tsNow = self.tsNow if self.oConfig.fDebug else None;
367	cHoursBack = self.oConfig.cHoursBack if self.oConfig.fDebug else 2;
368	oTestBoxLogic = TestBoxLogic(self.oDb);
369
370	#
371	# Generate a list of failures reasons we consider bad-testbox behavior.
372	#
373	aidFailureReasons = [
374	self.getFailureReason(self.ktReason_Host_DriverNotUnloading).idFailureReason,
375	];
376
377	#
378	# Get list of bad test boxes for given period and check them out individually.
379	#
380	aidBadTestBoxes = self.oTestSetLogic.fetchBadTestBoxIds(cHoursBack = cHoursBack, tsNow = tsNow,
381	aidFailureReasons = aidFailureReasons);
382	for idTestBox in aidBadTestBoxes:
383	# Skip if the testbox is already disabled or has a pending reboot command.
384	try:
385	oTestBox = TestBoxData().initFromDbWithId(self.oDb, idTestBox);
386	except Exception as oXcpt:
387	rcExit = self.eprint('Failed to get data for test box #%u in badTestBoxManagement: %s' % (idTestBox, oXcpt,));
388	continue;
389	if not oTestBox.fEnabled:
390	self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has been disabled already.'
391	% ( idTestBox, oTestBox.sName, ));
392	continue;
393	if oTestBox.enmPendingCmd != TestBoxData.ksTestBoxCmd_None:
394	self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has a command pending: %s'
395	% ( idTestBox, oTestBox.sName, oTestBox.enmPendingCmd));
396	continue;
397
398	# Get the most recent testsets for this box (descending on tsDone) and see how bad it is.
399	aoSets = self.oTestSetLogic.fetchSetsForTestBox(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow);
400	cOkay = 0;
401	cBad = 0;
402	iFirstOkay = len(aoSets);
403	for iSet, oSet in enumerate(aoSets):
404	if oSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
405	cBad += 1;
406	else:
407	# Check for bad failure reasons.
408	oFailure = None;
409	if oSet.enmStatus in TestSetData.kasBadTestStatuses:
410	oFailure = self.oTestResultFailureLogic.getById(oSet.idTestResult);
411	if oFailure is not None and oFailure.idFailureReason in aidFailureReasons:
412	cBad += 1;
413	else:
414	# This is an okay test result then.
415	## @todo maybe check the elapsed time here, it could still be a bad run?
416	cOkay += 1;
417	if iFirstOkay > iSet:
418	iFirstOkay = iSet;
419	if iSet > 10:
420	break;
421
422	# We react if there are two or more bad-testbox statuses at the head of the
423	# history and at least three in the last 10 results.
424	if iFirstOkay >= 2 and cBad > 2:
425	if oTestBoxLogic.hasTestBoxRecentlyBeenRebooted(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow):
426	self.vprint(u'Disabling testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
427	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
428	if self.oConfig.fRealRun is True:
429	try:
430	oTestBoxLogic.disableTestBox(idTestBox, self.uidSelf, fCommit = True,
431	sComment = 'Automatically disabled (iFirstOkay=%u cBad=%u cOkay=%u)'
432	% (iFirstOkay, cBad, cOkay),);
433	except Exception as oXcpt:
434	rcExit = self.eprint(u'Error disabling testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
435	else:
436	self.vprint(u'Rebooting testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
437	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
438	if self.oConfig.fRealRun is True:
439	try:
440	oTestBoxLogic.rebootTestBox(idTestBox, self.uidSelf, fCommit = True,
441	sComment = 'Automatically rebooted (iFirstOkay=%u cBad=%u cOkay=%u)'
442	% (iFirstOkay, cBad, cOkay),);
443	except Exception as oXcpt:
444	rcExit = self.eprint(u'Error rebooting testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
445	else:
446	self.dprint(u'badTestBoxManagement: #%u (%s) looks ok: iFirstOkay=%u cBad=%u cOkay=%u'
447	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
448	return rcExit;
449
450
451	## @name Failure reasons we know.
452	## @{
453	ktReason_BSOD_Recovery = ( 'BSOD', 'Recovery' );
454	ktReason_BSOD_Automatic_Repair = ( 'BSOD', 'Automatic Repair' );
455	ktReason_BSOD_C0000225 = ( 'BSOD', '0xC0000225 (boot)' );
456	ktReason_Guru_Generic = ( 'Guru Meditations', 'Generic Guru Meditation' );
457	ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_INSTR_NOT_IMPLEMENTED' );
458	ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' );
459	ktReason_Guru_VERR_TRPM_DONT_PANIC = ( 'Guru Meditations', 'VERR_TRPM_DONT_PANIC' );
460	ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED = ( 'Guru Meditations', 'VERR_PGM_PHYS_PAGE_RESERVED' );
461	ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE = ( 'Guru Meditations', 'VERR_VMX_INVALID_GUEST_STATE' );
462	ktReason_Guru_VINF_EM_TRIPLE_FAULT = ( 'Guru Meditations', 'VINF_EM_TRIPLE_FAULT' );
463	ktReason_Host_HostMemoryLow = ( 'Host', 'HostMemoryLow' );
464	ktReason_Host_DriverNotLoaded = ( 'Host', 'Driver not loaded' );
465	ktReason_Host_DriverNotUnloading = ( 'Host', 'Driver not unloading' );
466	ktReason_Host_NotSignedWithBuildCert = ( 'Host', 'Not signed with build cert' );
467	ktReason_Host_Reboot_OSX_Watchdog_Timeout = ( 'Host Reboot', 'OSX Watchdog Timeout' );
468	ktReason_Networking_Nonexistent_host_nic = ( 'Networking', 'Nonexistent host networking interface' );
469	ktReason_OSInstall_GRUB_hang = ( 'O/S Install', 'GRUB hang' );
470	ktReason_Panic_BootManagerC000000F = ( 'Panic', 'Hardware Changed' );
471	ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
472	ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
473	ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
474	ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
475	ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
476	ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
477	ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
478	ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
479	ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
480	ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
481	ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
482	ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
483	ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
484	ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
485	ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
486	## @}
487
488	## BSOD category.
489	ksBsodCategory = 'BSOD';
490	## Special reason indicating that the flesh and blood sheriff has work to do.
491	ksBsodAddNew = 'Add new BSOD';
492
493	## Unit test category.
494	ksUnitTestCategory = 'Unit';
495	## Special reason indicating that the flesh and blood sheriff has work to do.
496	ksUnitTestAddNew = 'Add new';
497
498	## Used for indica that we shouldn't report anything for this test result ID and
499	## consider promoting the previous error to test set level if it's the only one.
500	ktHarmless = ( 'Probably', 'Caused by previous error' );
501
502
503	def caseClosed(self, oCaseFile):
504	"""
505	Reports the findings in the case and closes it.
506	"""
507	#
508	# Log it and create a dReasonForReasultId we can use below.
509	#
510	dCommentForResultId = oCaseFile.dCommentForResultId;
511	if len(oCaseFile.dReasonForResultId) > 0:
512	# Must weed out ktHarmless.
513	dReasonForResultId = {};
514	for idKey, tReason in oCaseFile.dReasonForResultId.items():
515	if tReason is not self.ktHarmless:
516	dReasonForResultId[idKey] = tReason;
517	if len(dReasonForResultId) == 0:
518	self.vprint(u'TODO: Closing %s without a real reason, only %s.'
519	% (oCaseFile.sName, oCaseFile.dReasonForResultId));
520	return False;
521
522	# Try promote to single reason.
523	atValues = dReasonForResultId.values();
524	fSingleReason = True;
525	if len(dReasonForResultId) == 1 and dReasonForResultId.keys()[0] != oCaseFile.oTestSet.idTestResult:
526	self.dprint(u'Promoting single reason to whole set: %s' % (atValues[0],));
527	elif len(dReasonForResultId) > 1 and len(atValues) == atValues.count(atValues[0]):
528	self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), atValues[0]));
529	else:
530	fSingleReason = False;
531	if fSingleReason:
532	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: atValues[0], };
533	if len(dCommentForResultId) > 0:
534	dCommentForResultId = { oCaseFile.oTestSet.idTestResult: dCommentForResultId.values()[0], };
535	elif oCaseFile.tReason is not None:
536	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
537	else:
538	self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
539	return False;
540
541	self.vprint(u'Closing %s with following reason%s: %s'
542	% ( oCaseFile.sName, 's' if dReasonForResultId > 0 else '', dReasonForResultId, ));
543
544	#
545	# Add the test failure reason record(s).
546	#
547	for idTestResult, tReason in dReasonForResultId.items():
548	oFailureReason = self.getFailureReason(tReason);
549	if oFailureReason is not None:
550	sComment = 'Set by $Revision: 65802 $' # Handy for reverting later.
551	if idTestResult in dCommentForResultId:
552	sComment += ': ' + dCommentForResultId[idTestResult];
553
554	oAdd = TestResultFailureData();
555	oAdd.initFromValues(idTestResult = idTestResult,
556	idFailureReason = oFailureReason.idFailureReason,
557	uidAuthor = self.uidSelf,
558	idTestSet = oCaseFile.oTestSet.idTestSet,
559	sComment = sComment,);
560	if self.oConfig.fRealRun:
561	try:
562	self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
563	except Exception as oXcpt:
564	self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
565	% (oXcpt, oAdd, oCaseFile.sLongName,));
566	else:
567	self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
568	return True;
569
570	#
571	# Tools for assiting log parsing.
572	#
573
574	@staticmethod
575	def matchFollowedByLines(sStr, off, asFollowingLines):
576	""" Worker for isThisFollowedByTheseLines. """
577
578	# Advance off to the end of the line.
579	off = sStr.find('\n', off);
580	if off < 0:
581	return False;
582	off += 1;
583
584	# Match each string with the subsequent lines.
585	for iLine, sLine in enumerate(asFollowingLines):
586	offEnd = sStr.find('\n', off);
587	if offEnd < 0:
588	return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
589	if len(sLine) > 0 and sStr.find(sLine, off, offEnd) < 0:
590	return False;
591
592	# next line.
593	off = offEnd + 1;
594
595	return True;
596
597	@staticmethod
598	def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
599	"""
600	Looks for a line contining sFirst which is then followed by lines
601	with the strings in asFollowingLines. (No newline chars anywhere!)
602	Returns True / False.
603	"""
604	off = sStr.find(sFirst, 0);
605	while off >= 0:
606	if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
607	return True;
608	off = sStr.find(sFirst, off + 1);
609	return False;
610
611	@staticmethod
612	def findAndReturnResetOfLine(sHaystack, sNeedle):
613	"""
614	Looks for sNeedle in sHaystack.
615	Returns The text following the needle up to the end of the line.
616	Returns None if not found.
617	"""
618	if sHaystack is None:
619	return None;
620	off = sHaystack.find(sNeedle);
621	if off < 0:
622	return None;
623	off += len(sNeedle)
624	offEol = sHaystack.find('\n', off);
625	if offEol < 0:
626	offEol = len(sHaystack);
627	return sHaystack[off:offEol]
628
629	@staticmethod
630	def findInAnyAndReturnResetOfLine(asHaystacks, sNeedle):
631	"""
632	Looks for sNeedle in zeroe or more haystacks (asHaystack).
633	Returns The text following the first needed found up to the end of the line.
634	Returns None if not found.
635	"""
636	for sHaystack in asHaystacks:
637	sRet = VirtualTestSheriff.findAndReturnResetOfLine(sHaystack, sNeedle);
638	if sRet is not None:
639	return sRet;
640	return None;
641
642
643	#
644	# The investigative units.
645	#
646
647	katSimpleInstallUninstallMainLogReasons = [
648	# ( Whether to stop on hit, reason tuple, needle text. )
649	];
650
651	kdatSimpleInstallUninstallMainLogReasonsPerOs = {
652	'darwin': [
653	# ( Whether to stop on hit, reason tuple, needle text. )
654	( True, ktReason_Host_DriverNotUnloading,
655	'Can\'t remove kext org.virtualbox.kext.VBoxDrv; services failed to terminate - 0xe00002c7' ),
656	],
657	};
658
659
660	def investigateInstallUninstallFailure(self, oCaseFile, oFailedResult, sResultLog, fInstall):
661	"""
662	Investigates an install or uninstall failure.
663
664	We lump the two together since the installation typically also performs
665	an uninstall first and will be seeing similar issues to the uninstall.
666	"""
667	_ = fInstall;
668
669	atSimple = self.katSimpleInstallUninstallMainLogReasons;
670	if oCaseFile.oTestBox.sOs in self.kdatSimpleInstallUninstallMainLogReasonsPerOs:
671	atSimple = self.kdatSimpleInstallUninstallMainLogReasonsPerOs[oCaseFile.oTestBox.sOs] + atSimple;
672
673	fFoundSomething = False;
674	for fStopOnHit, tReason, sNeedle in atSimple:
675	if sResultLog.find(sNeedle) > 0:
676	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
677	if fStopOnHit:
678	return True;
679	fFoundSomething = True;
680
681	return fFoundSomething if fFoundSomething else None;
682
683
684	def investigateBadTestBox(self, oCaseFile):
685	"""
686	Checks out bad-testbox statuses.
687	"""
688	_ = oCaseFile;
689	return False;
690
691
692	def investigateVBoxUnitTest(self, oCaseFile):
693	"""
694	Checks out a VBox unittest problem.
695	"""
696
697	#
698	# Process simple test case failures first, using their name as reason.
699	# We do the reason management just like for BSODs.
700	#
701	cRelevantOnes = 0;
702	sMainLog = oCaseFile.getMainLog();
703	aoFailedResults = oCaseFile.oTree.getListOfFailures();
704	for oFailedResult in aoFailedResults:
705	if oFailedResult is oCaseFile.oTree:
706	self.vprint('TODO: toplevel failure');
707	cRelevantOnes += 1
708
709	elif oFailedResult.sName == 'Installing VirtualBox':
710	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
711	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
712	cRelevantOnes += 1
713
714	elif oFailedResult.sName == 'Uninstalling VirtualBox':
715	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
716	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
717	cRelevantOnes += 1
718
719	elif oFailedResult.oParent is not None:
720	# Get the 2nd level node because that's where we'll find the unit test name.
721	while oFailedResult.oParent.oParent is not None:
722	oFailedResult = oFailedResult.oParent;
723
724	# Only report a failure once.
725	if oFailedResult.idTestResult not in oCaseFile.dReasonForResultId:
726	sKey = oFailedResult.sName;
727	if sKey.startswith('testcase/'):
728	sKey = sKey[9:];
729	if sKey in self.asUnitTestReasons:
730	tReason = ( self.ksUnitTestCategory, sKey );
731	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
732	else:
733	self.dprint(u'Unit test failure "%s" not found in %s;' % (sKey, self.asUnitTestReasons));
734	tReason = ( self.ksUnitTestCategory, self.ksUnitTestAddNew );
735	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sKey);
736	cRelevantOnes += 1
737	else:
738	self.vprint(u'Internal error: expected oParent to NOT be None for %s' % (oFailedResult,));
739
740	#
741	# If we've caught all the relevant ones by now, report the result.
742	#
743	if len(oCaseFile.dReasonForResultId) >= cRelevantOnes:
744	return self.caseClosed(oCaseFile);
745	return False;
746
747	@staticmethod
748	def extractGuestCpuStack(sInfoText):
749	"""
750	Extracts the guest CPU stacks from the input file.
751
752	Returns a dictionary keyed by the CPU number, value being a list of
753	raw stack lines (no header).
754	Returns empty dictionary if no stacks where found.
755	"""
756	dRet = {};
757	off = 0;
758	while True:
759	offStart = sInfoText.find('=== start guest stack VCPU ', off);
760	if offStart < 0:
761	break;
762	offEnd = sInfoText.find('=== end guest stack', offStart + 20);
763	if offEnd < 0:
764	offEnd = sInfoText.find('=== start guest stack VCPU', offStart + 20);
765	if offEnd >= 0:
766	offEnd += 3;
767	else:
768	offEnd = len(sInfoText);
769
770	sStack = sInfoText[offStart : offEnd];
771	sStack = sStack.replace('\r',''); # paranoia
772	asLines = sStack.split();
773
774	# figure the CPU.
775	asWords = asLines[0].split();
776	if asWords < 6 or not asWords[6].isdigit():
777	break;
778	iCpu = int(asWords[6]);
779
780	# add it.
781	dRet[iCpu] = [sLine.rstrip() for sLine in asLines[2:-1]]
782	return dRet;
783
784	def investigateInfoKvmLockSpinning(self, oCaseFile, sInfoText, dLogs):
785	""" Investigates kvm_lock_spinning deadlocks """
786	#
787	# Extract the stacks. We need more than one CPU to create a deadlock.
788	#
789	dStacks = self.extractGuestCpuStack(sInfoText);
790	if len(dStacks) >= 2:
791	#
792	# Examin each of the stacks. Each must have kvm_lock_spinning in
793	# one of the first three entries.
794	#
795	cHits = 0;
796	for iCpu in dStacks:
797	asBacktrace = dStacks[iCpu];
798	for iFrame in xrange(min(3, len(asBacktrace))):
799	if asBacktrace[iFrame].find('kvm_lock_spinning') >= 0:
800	cHits += 1;
801	break;
802	if cHits == len(dStacks):
803	return (True, self.ktReason_VMM_kvm_lock_spinning);
804
805	_ = dLogs; _ = oCaseFile;
806	return (False, None);
807
808	## Things we search a main or VM log for to figure out why something went bust.
809	katSimpleMainAndVmLogReasons = [
810	# ( Whether to stop on hit, reason tuple, needle text. )
811	( False, ktReason_Guru_Generic, 'GuruMeditation' ),
812	( False, ktReason_Guru_Generic, 'Guru Meditation' ),
813	( True, ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED, 'VERR_IEM_INSTR_NOT_IMPLEMENTED' ),
814	( True, ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED, 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' ),
815	( True, ktReason_Guru_VERR_TRPM_DONT_PANIC, 'VERR_TRPM_DONT_PANIC' ),
816	( True, ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED, 'VERR_PGM_PHYS_PAGE_RESERVED' ),
817	( True, ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE, 'VERR_VMX_INVALID_GUEST_STATE' ),
818	( True, ktReason_Guru_VINF_EM_TRIPLE_FAULT, 'VINF_EM_TRIPLE_FAULT' ),
819	( True, ktReason_Networking_Nonexistent_host_nic,
820	'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
821	( True, ktReason_Host_Reboot_OSX_Watchdog_Timeout, ': "OSX Watchdog Timeout: ' ),
822	( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
823	'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
824	( True, ktReason_Host_HostMemoryLow, 'HostMemoryLow' ),
825	( True, ktReason_Host_HostMemoryLow, 'Failed to procure handy pages; rc=VERR_NO_MEMORY' ),
826	( True, ktReason_Unknown_File_Not_Found,
827	'Error: failed to start machine. Error message: File not found. (VERR_FILE_NOT_FOUND)' ),
828	( True, ktReason_Unknown_File_Not_Found, # lump it in with file-not-found for now.
829	'Error: failed to start machine. Error message: Not supported. (VERR_NOT_SUPPORTED)' ),
830	( False, ktReason_Unknown_VM_Crash, 'txsDoConnectViaTcp: Machine state: Aborted' )
831	];
832
833	## Things we search a VBoxHardening.log file for to figure out why something went bust.
834	katSimpleVBoxHardeningLogReasons = [
835	# ( Whether to stop on hit, reason tuple, needle text. )
836	( True, ktReason_Host_DriverNotLoaded, 'Error opening VBoxDrvStub: STATUS_OBJECT_NAME_NOT_FOUND' ),
837	( True, ktReason_Host_NotSignedWithBuildCert, 'Not signed with the build certificate' ),
838	];
839
840	## Things we search a kernel.log file for to figure out why something went bust.
841	katSimpleKernelLogReasons = [
842	# ( Whether to stop on hit, reason tuple, needle text. )
843	( True, ktReason_Panic_HugeMemory, 'mm/huge_memory.c:1988' ),
844	( True, ktReason_Panic_IOAPICDoesntWork, 'IO-APIC + timer doesn''t work' ),
845	];
846
847	## Things we search the _RIGHT_ _STRIPPED_ vgatext for.
848	katSimpleVgaTextReasons = [
849	# ( Whether to stop on hit, reason tuple, needle text. )
850	( True, ktReason_Panic_MP_BIOS_IO_APIC,
851	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n\n" ),
852	( True, ktReason_Panic_MP_BIOS_IO_APIC,
853	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n"
854	"...trying to set up timer (IRQ0) through the 8259A ... failed.\n"
855	"...trying to set up timer as Virtual Wire IRQ... failed.\n"
856	"...trying to set up timer as ExtINT IRQ... failed :(.\n"
857	"Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug\n"
858	"and send a report. Then try booting with the 'noapic' option\n"
859	"\n" ),
860	( True, ktReason_OSInstall_GRUB_hang,
861	"-----\nGRUB Loading stage2..\n\n\n\n" ),
862	( True, ktReason_Panic_BootManagerC000000F,
863	"Windows failed to start. A recent hardware or software change might be the" ),
864	];
865
866	## Things we search for in the info.txt file. Require handlers for now.
867	katInfoTextHandlers = [
868	# ( Trigger text, handler method )
869	( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
870	];
871
872	## Mapping screenshot/failure SHA-256 hashes to failure reasons.
873	katSimpleScreenshotHashReasons = [
874	# ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
875	( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
876	( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
877	( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
878	( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
879	( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
880	( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
881	( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
882	( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
883	];
884
885	def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
886	"""
887	Investigates a failed VM run.
888	"""
889
890	def investigateLogSet():
891	"""
892	Investigates the current set of VM related logs.
893	"""
894	self.dprint('investigateLogSet: lengths: result log %u, VM log %u, kernel log %u, vga text %u, info text %u'
895	% ( len(sResultLog) if sResultLog is not None else 0,
896	len(sVMLog) if sVMLog is not None else 0,
897	len(sKrnlLog) if sKrnlLog is not None else 0,
898	len(sVgaText) if sVgaText is not None else 0,
899	len(sInfoText) if sInfoText is not None else 0, ));
900
901	#self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
902	#self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
903	#self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
904	#self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
905	#self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
906
907	# TODO: more
908
909	#
910	# Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
911	#
912	sDetails = self.findInAnyAndReturnResetOfLine([ sVMLog, sResultLog ],
913	'GIM: HyperV: Guest indicates a fatal condition! P0=');
914	if sDetails is not None:
915	# P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
916	sKey = sDetails.split(' ', 1)[0];
917	try: sKey = '0x%08X' % (int(sKey, 16),);
918	except: pass;
919	if sKey in self.asBsodReasons:
920	tReason = ( self.ksBsodCategory, sKey );
921	elif sKey.lower() in self.asBsodReasons: # just in case.
922	tReason = ( self.ksBsodCategory, sKey.lower() );
923	else:
924	self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
925	tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
926	return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
927
928	#
929	# Look for linux panic.
930	#
931	if sKrnlLog is not None:
932	for fStopOnHit, tReason, sNeedle in self.katSimpleKernelLogReasons:
933	if sKrnlLog.find(sNeedle) > 0:
934	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
935	if fStopOnHit:
936	return True;
937	fFoundSomething = True;
938
939	#
940	# Loop thru the simple stuff.
941	#
942	fFoundSomething = False;
943	for fStopOnHit, tReason, sNeedle in self.katSimpleMainAndVmLogReasons:
944	if sResultLog.find(sNeedle) > 0 or (sVMLog is not None and sVMLog.find(sNeedle) > 0):
945	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
946	if fStopOnHit:
947	return True;
948	fFoundSomething = True;
949
950	# Continue with vga text.
951	if sVgaText is not None and len(sVgaText) > 0:
952	for fStopOnHit, tReason, sNeedle in self.katSimpleVgaTextReasons:
953	if sVgaText.find(sNeedle) > 0:
954	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
955	if fStopOnHit:
956	return True;
957	fFoundSomething = True;
958	_ = sInfoText;
959
960	# Continue with screen hashes.
961	if sScreenHash is not None:
962	for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
963	if sScreenHash == sHash:
964	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
965	if fStopOnHit:
966	return True;
967	fFoundSomething = True;
968
969	# Check VBoxHardening.log.
970	if sNtHardLog is not None:
971	for fStopOnHit, tReason, sNeedle in self.katSimpleVBoxHardeningLogReasons:
972	if sNtHardLog.find(sNeedle) > 0:
973	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
974	if fStopOnHit:
975	return True;
976	fFoundSomething = True;
977
978	#
979	# Complicated stuff.
980	#
981	dLogs = {
982	'sVMLog': sVMLog,
983	'sNtHardLog': sNtHardLog,
984	'sScreenHash': sScreenHash,
985	'sKrnlLog': sKrnlLog,
986	'sVgaText': sVgaText,
987	'sInfoText': sInfoText,
988	};
989
990	# info.txt.
991	if sInfoText is not None and len(sInfoText) > 0:
992	for sNeedle, fnHandler in self.katInfoTextHandlers:
993	if sInfoText.find(sNeedle) > 0:
994	(fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
995	if tReason is not None:
996	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
997	if fStop:
998	return True;
999	fFoundSomething = True;
1000
1001	#
1002	# Check for repeated reboots...
1003	#
1004	if sVMLog is not None:
1005	cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1006	if cResets > 10:
1007	return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1008	sComment = 'Counted %s reboots' % (cResets,));
1009
1010	return fFoundSomething;
1011
1012	#
1013	# Check if we got any VM or/and kernel logs. Treat them as sets in
1014	# case we run multiple VMs here (this is of course ASSUMING they
1015	# appear in the order that terminateVmBySession uploads them).
1016	#
1017	cTimes = 0;
1018	sVMLog = None;
1019	sNtHardLog = None;
1020	sScreenHash = None;
1021	sKrnlLog = None;
1022	sVgaText = None;
1023	sInfoText = None;
1024	for oFile in oFailedResult.aoFiles:
1025	if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1026	if 'VBoxHardening.log' not in oFile.sFile:
1027	if sVMLog is not None:
1028	if investigateLogSet() is True:
1029	return True;
1030	cTimes += 1;
1031	sInfoText = None;
1032	sVgaText = None;
1033	sKrnlLog = None;
1034	sScreenHash = None;
1035	sNtHardLog = None;
1036	sVMLog = oCaseFile.getLogFile(oFile);
1037	else:
1038	sNtHardLog = oCaseFile.getLogFile(oFile);
1039	elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1040	sKrnlLog = oCaseFile.getLogFile(oFile);
1041	elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1042	sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1043	elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1044	sInfoText = oCaseFile.getLogFile(oFile);
1045	elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1046	sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1047	if sScreenHash is not None:
1048	sScreenHash = sScreenHash.lower();
1049	self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1050
1051	if ( sVMLog is not None \
1052	or sNtHardLog is not None \
1053	or cTimes == 0) \
1054	and investigateLogSet() is True:
1055	return True;
1056
1057	return None;
1058
1059
1060	def isResultFromVMRun(self, oFailedResult, sResultLog):
1061	"""
1062	Checks if this result and corresponding log snippet looks like a VM run.
1063	"""
1064
1065	# Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1066	if sResultLog.find(' startVm') > 0:
1067	return True;
1068
1069	# Any other indicators? No?
1070	_ = oFailedResult;
1071	return False;
1072
1073	def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1074	"""
1075	Checks out a VBox VM test.
1076
1077	This is generic investigation of a test running one or more VMs, like
1078	for example a smoke test or a guest installation test.
1079
1080	The fSingleVM parameter is a hint, which probably won't come in useful.
1081	"""
1082	_ = fSingleVM;
1083
1084	#
1085	# Get a list of test result failures we should be looking into and the main log.
1086	#
1087	aoFailedResults = oCaseFile.oTree.getListOfFailures();
1088	sMainLog = oCaseFile.getMainLog();
1089
1090	#
1091	# There are a set of errors ending up on the top level result record.
1092	# Should deal with these first.
1093	#
1094	if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1095	# Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1096	# be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1097	# ignore other failures in the test if we're not a little bit careful here.
1098	if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1099	oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1100	return self.caseClosed(oCaseFile);
1101
1102	# Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1103	if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1104	(' Timeout: ', ' Attempting to abort child...',) ):
1105	if sMainLog.find('* glibc detected * /') > 0:
1106	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1107	else:
1108	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1109	return self.caseClosed(oCaseFile);
1110
1111	# Look for heap corruption without visible hang.
1112	if sMainLog.find('* glibc detected * /') > 0 \
1113	or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1114	oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1115	return self.caseClosed(oCaseFile);
1116
1117	# Out of memory w/ timeout.
1118	if sMainLog.find('sErrId=HostMemoryLow') > 0:
1119	oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1120	return self.caseClosed(oCaseFile);
1121
1122	# Stale files like vts_rm.exe (windows).
1123	offEnd = sMainLog.rfind('* The test driver exits successfully. *');
1124	if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1125	oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1126	return self.caseClosed(oCaseFile);
1127
1128	#
1129	# XPCOM screwup
1130	#
1131	if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1132	oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1133	return self.caseClosed(oCaseFile);
1134
1135	#
1136	# Go thru each failed result.
1137	#
1138	for oFailedResult in aoFailedResults:
1139	self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1140	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1141	if oFailedResult.sName == 'Installing VirtualBox':
1142	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1143
1144	elif oFailedResult.sName == 'Uninstalling VirtualBox':
1145	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1146
1147	elif self.isResultFromVMRun(oFailedResult, sResultLog):
1148	self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1149
1150	elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1151	oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1152
1153	elif sResultLog.find('The machine is not mutable (state is ') > 0:
1154	self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1155	oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1156
1157	elif sResultLog.find('** error: no action was specified') > 0 \
1158	or sResultLog.find('(len(self._asXml, asText))') > 0:
1159	oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1160
1161	else:
1162	self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1163	self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1164
1165	#
1166	# Report home and close the case if we got them all, otherwise log it.
1167	#
1168	if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1169	return self.caseClosed(oCaseFile);
1170
1171	if len(oCaseFile.dReasonForResultId) > 0:
1172	self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1173	% (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1174	else:
1175	self.vprint(u'XXX: Could not figure out anything at all! :-(');
1176	return False;
1177
1178
1179	def reasoningFailures(self):
1180	"""
1181	Guess the reason for failures.
1182	"""
1183	#
1184	# Get a list of failed test sets without any assigned failure reason.
1185	#
1186	cGot = 0;
1187	aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack, tsNow = self.tsNow);
1188	for oTestSet in aoTestSets:
1189	self.dprint(u'');
1190	self.dprint(u'reasoningFailures: Checking out test set #%u, status %s' % ( oTestSet.idTestSet, oTestSet.enmStatus,))
1191
1192	#
1193	# Open a case file and assign it to the right investigator.
1194	#
1195	(oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1196	oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1197	oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1198	oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1199	oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1200
1201	oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1202
1203	if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1204	self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1205	fRc = self.investigateBadTestBox(oCaseFile);
1206
1207	elif oCaseFile.isVBoxUnitTest():
1208	self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1209	fRc = self.investigateVBoxUnitTest(oCaseFile);
1210
1211	elif oCaseFile.isVBoxInstallTest():
1212	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1213	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1214
1215	elif oCaseFile.isVBoxUSBTest():
1216	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1217	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1218
1219	elif oCaseFile.isVBoxStorageTest():
1220	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1221	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1222
1223	elif oCaseFile.isVBoxGAsTest():
1224	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1225	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1226
1227	elif oCaseFile.isVBoxAPITest():
1228	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1229	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1230
1231	elif oCaseFile.isVBoxBenchmarkTest():
1232	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1233	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1234
1235	elif oCaseFile.isVBoxSmokeTest():
1236	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1237	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1238
1239	else:
1240	self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1241	fRc = False;
1242	cGot += fRc is True;
1243
1244	self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1245	return 0;
1246
1247
1248	def main(self):
1249	"""
1250	The 'main' function.
1251	Return exit code (0, 1, etc).
1252	"""
1253	# Database stuff.
1254	self.oDb = TMDatabaseConnection()
1255	self.oTestResultLogic = TestResultLogic(self.oDb);
1256	self.oTestSetLogic = TestSetLogic(self.oDb);
1257	self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1258	self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1259	self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1260	self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1261
1262	# Get a fix on our 'now' before we do anything..
1263	self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1264	self.tsNow = self.oDb.fetchOne();
1265
1266	# If we're suppost to commit anything we need to get our user ID.
1267	rcExit = 0;
1268	if self.oConfig.fRealRun:
1269	self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1270	if self.oLogin is None:
1271	rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1272	else:
1273	self.uidSelf = self.oLogin.uid;
1274
1275	#
1276	# Do the stuff.
1277	#
1278	if rcExit == 0:
1279	rcExit = self.selfCheck();
1280	if rcExit == 0:
1281	rcExit = self.badTestBoxManagement();
1282	rcExit2 = self.reasoningFailures();
1283	if rcExit == 0:
1284	rcExit = rcExit2;
1285	# Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1286	if rcExit == 0:
1287	rcExit = self.badTestBoxManagement();
1288
1289	# Cleanup.
1290	self.oFailureReasonLogic = None;
1291	self.oTestResultFailureLogic = None;
1292	self.oTestSetLogic = None;
1293	self.oTestResultLogic = None;
1294	self.oDb.close();
1295	self.oDb = None;
1296	if self.oLogFile is not None:
1297	self.oLogFile.close();
1298	self.oLogFile = None;
1299	return rcExit;
1300
1301	if __name__ == '__main__':
1302	sys.exit(VirtualTestSheriff().main());
1303

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 65802

Download in other formats: