virtual_test_sheriff.py@ 66431

Last change on this file since 66431 was 66431, checked in by vboxsync, 8 years ago
vsheriff: whitespace
Property svn:eol-style set to `LF` Property svn:executable set to ``* Property svn:keywords set to `Author Date Id Revision`
File size: 61.0 KB

Line
1	#!/usr/bin/env python
2	# -- coding: utf-8 --
3	# $Id: virtual_test_sheriff.py 66431 2017-04-05 12:04:01Z 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.virtualbox.org. 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: 66431 $"
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 self.sMainLog:
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:
295	self.oLogFile = open(self.oConfig.sLogFile, "a");
296	self.oLogFile.write('VirtualTestSheriff: $Revision: 66431 $ \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	# Frank: For now don't reboot boxes automatically
426	if True or oTestBoxLogic.hasTestBoxRecentlyBeenRebooted(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow):
427	self.vprint(u'Disabling testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
428	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
429	if self.oConfig.fRealRun is True:
430	try:
431	oTestBoxLogic.disableTestBox(idTestBox, self.uidSelf, fCommit = True,
432	sComment = 'Automatically disabled (iFirstOkay=%u cBad=%u cOkay=%u)'
433	% (iFirstOkay, cBad, cOkay),);
434	except Exception as oXcpt:
435	rcExit = self.eprint(u'Error disabling testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
436	else:
437	self.vprint(u'Rebooting testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
438	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
439	if self.oConfig.fRealRun is True:
440	try:
441	oTestBoxLogic.rebootTestBox(idTestBox, self.uidSelf, fCommit = True,
442	sComment = 'Automatically rebooted (iFirstOkay=%u cBad=%u cOkay=%u)'
443	% (iFirstOkay, cBad, cOkay),);
444	except Exception as oXcpt:
445	rcExit = self.eprint(u'Error rebooting testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
446	else:
447	self.dprint(u'badTestBoxManagement: #%u (%s) looks ok: iFirstOkay=%u cBad=%u cOkay=%u'
448	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
449	return rcExit;
450
451
452	## @name Failure reasons we know.
453	## @{
454	ktReason_BSOD_Recovery = ( 'BSOD', 'Recovery' );
455	ktReason_BSOD_Automatic_Repair = ( 'BSOD', 'Automatic Repair' );
456	ktReason_BSOD_C0000225 = ( 'BSOD', '0xC0000225 (boot)' );
457	ktReason_Guru_Generic = ( 'Guru Meditations', 'Generic Guru Meditation' );
458	ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_INSTR_NOT_IMPLEMENTED' );
459	ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' );
460	ktReason_Guru_VERR_TRPM_DONT_PANIC = ( 'Guru Meditations', 'VERR_TRPM_DONT_PANIC' );
461	ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED = ( 'Guru Meditations', 'VERR_PGM_PHYS_PAGE_RESERVED' );
462	ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE = ( 'Guru Meditations', 'VERR_VMX_INVALID_GUEST_STATE' );
463	ktReason_Guru_VINF_EM_TRIPLE_FAULT = ( 'Guru Meditations', 'VINF_EM_TRIPLE_FAULT' );
464	ktReason_Host_HostMemoryLow = ( 'Host', 'HostMemoryLow' );
465	ktReason_Host_DriverNotLoaded = ( 'Host', 'Driver not loaded' );
466	ktReason_Host_DriverNotUnloading = ( 'Host', 'Driver not unloading' );
467	ktReason_Host_NotSignedWithBuildCert = ( 'Host', 'Not signed with build cert' );
468	ktReason_Host_LeftoverService = ( 'Host', 'Leftover service' );
469	ktReason_Host_Reboot_OSX_Watchdog_Timeout = ( 'Host Reboot', 'OSX Watchdog Timeout' );
470	ktReason_Host_Modprobe_Failed = ( 'Host', 'Modprobe failed' );
471	ktReason_Networking_Nonexistent_host_nic = ( 'Networking', 'Nonexistent host networking interface' );
472	ktReason_OSInstall_GRUB_hang = ( 'O/S Install', 'GRUB hang' );
473	ktReason_Panic_BootManagerC000000F = ( 'Panic', 'Hardware Changed' );
474	ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
475	ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
476	ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
477	ktReason_Panic_TxUnitHang = ( 'Panic', 'Tx Unit Hang' );
478	ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
479	ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
480	ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
481	ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
482	ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
483	ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
484	ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
485	ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
486	ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
487	ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
488	ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
489	ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
490	## @}
491
492	## BSOD category.
493	ksBsodCategory = 'BSOD';
494	## Special reason indicating that the flesh and blood sheriff has work to do.
495	ksBsodAddNew = 'Add new BSOD';
496
497	## Unit test category.
498	ksUnitTestCategory = 'Unit';
499	## Special reason indicating that the flesh and blood sheriff has work to do.
500	ksUnitTestAddNew = 'Add new';
501
502	## Used for indica that we shouldn't report anything for this test result ID and
503	## consider promoting the previous error to test set level if it's the only one.
504	ktHarmless = ( 'Probably', 'Caused by previous error' );
505
506
507	def caseClosed(self, oCaseFile):
508	"""
509	Reports the findings in the case and closes it.
510	"""
511	#
512	# Log it and create a dReasonForReasultId we can use below.
513	#
514	dCommentForResultId = oCaseFile.dCommentForResultId;
515	if oCaseFile.dReasonForResultId:
516	# Must weed out ktHarmless.
517	dReasonForResultId = {};
518	for idKey, tReason in oCaseFile.dReasonForResultId.items():
519	if tReason is not self.ktHarmless:
520	dReasonForResultId[idKey] = tReason;
521	if not dReasonForResultId:
522	self.vprint(u'TODO: Closing %s without a real reason, only %s.'
523	% (oCaseFile.sName, oCaseFile.dReasonForResultId));
524	return False;
525
526	# Try promote to single reason.
527	atValues = dReasonForResultId.values();
528	fSingleReason = True;
529	if len(dReasonForResultId) == 1 and dReasonForResultId.keys()[0] != oCaseFile.oTestSet.idTestResult:
530	self.dprint(u'Promoting single reason to whole set: %s' % (atValues[0],));
531	elif len(dReasonForResultId) > 1 and len(atValues) == atValues.count(atValues[0]):
532	self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), atValues[0]));
533	else:
534	fSingleReason = False;
535	if fSingleReason:
536	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: atValues[0], };
537	if dCommentForResultId:
538	dCommentForResultId = { oCaseFile.oTestSet.idTestResult: dCommentForResultId.values()[0], };
539	elif oCaseFile.tReason is not None:
540	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
541	else:
542	self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
543	return False;
544
545	self.vprint(u'Closing %s with following reason%s: %s'
546	% ( oCaseFile.sName, 's' if dReasonForResultId > 0 else '', dReasonForResultId, ));
547
548	#
549	# Add the test failure reason record(s).
550	#
551	for idTestResult, tReason in dReasonForResultId.items():
552	oFailureReason = self.getFailureReason(tReason);
553	if oFailureReason is not None:
554	sComment = 'Set by $Revision: 66431 $' # Handy for reverting later.
555	if idTestResult in dCommentForResultId:
556	sComment += ': ' + dCommentForResultId[idTestResult];
557
558	oAdd = TestResultFailureData();
559	oAdd.initFromValues(idTestResult = idTestResult,
560	idFailureReason = oFailureReason.idFailureReason,
561	uidAuthor = self.uidSelf,
562	idTestSet = oCaseFile.oTestSet.idTestSet,
563	sComment = sComment,);
564	if self.oConfig.fRealRun:
565	try:
566	self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
567	except Exception as oXcpt:
568	self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
569	% (oXcpt, oAdd, oCaseFile.sLongName,));
570	else:
571	self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
572	return True;
573
574	#
575	# Tools for assiting log parsing.
576	#
577
578	@staticmethod
579	def matchFollowedByLines(sStr, off, asFollowingLines):
580	""" Worker for isThisFollowedByTheseLines. """
581
582	# Advance off to the end of the line.
583	off = sStr.find('\n', off);
584	if off < 0:
585	return False;
586	off += 1;
587
588	# Match each string with the subsequent lines.
589	for iLine, sLine in enumerate(asFollowingLines):
590	offEnd = sStr.find('\n', off);
591	if offEnd < 0:
592	return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
593	if sLine and sStr.find(sLine, off, offEnd) < 0:
594	return False;
595
596	# next line.
597	off = offEnd + 1;
598
599	return True;
600
601	@staticmethod
602	def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
603	"""
604	Looks for a line contining sFirst which is then followed by lines
605	with the strings in asFollowingLines. (No newline chars anywhere!)
606	Returns True / False.
607	"""
608	off = sStr.find(sFirst, 0);
609	while off >= 0:
610	if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
611	return True;
612	off = sStr.find(sFirst, off + 1);
613	return False;
614
615	@staticmethod
616	def findAndReturnResetOfLine(sHaystack, sNeedle):
617	"""
618	Looks for sNeedle in sHaystack.
619	Returns The text following the needle up to the end of the line.
620	Returns None if not found.
621	"""
622	if sHaystack is None:
623	return None;
624	off = sHaystack.find(sNeedle);
625	if off < 0:
626	return None;
627	off += len(sNeedle)
628	offEol = sHaystack.find('\n', off);
629	if offEol < 0:
630	offEol = len(sHaystack);
631	return sHaystack[off:offEol]
632
633	@staticmethod
634	def findInAnyAndReturnResetOfLine(asHaystacks, sNeedle):
635	"""
636	Looks for sNeedle in zeroe or more haystacks (asHaystack).
637	Returns The text following the first needed found up to the end of the line.
638	Returns None if not found.
639	"""
640	for sHaystack in asHaystacks:
641	sRet = VirtualTestSheriff.findAndReturnResetOfLine(sHaystack, sNeedle);
642	if sRet is not None:
643	return sRet;
644	return None;
645
646
647	#
648	# The investigative units.
649	#
650
651	katSimpleInstallUninstallMainLogReasons = [
652	# ( Whether to stop on hit, reason tuple, needle text. )
653	( False, ktReason_Host_LeftoverService,
654	'SERVICE_NAME: vbox' ),
655	( True, ktReason_Host_Modprobe_Failed,
656	'modprobe vboxdrv failed' )
657	];
658
659	kdatSimpleInstallUninstallMainLogReasonsPerOs = {
660	'darwin': [
661	# ( Whether to stop on hit, reason tuple, needle text. )
662	( True, ktReason_Host_DriverNotUnloading,
663	'Can\'t remove kext org.virtualbox.kext.VBoxDrv; services failed to terminate - 0xe00002c7' ),
664	],
665	};
666
667
668	def investigateInstallUninstallFailure(self, oCaseFile, oFailedResult, sResultLog, fInstall):
669	"""
670	Investigates an install or uninstall failure.
671
672	We lump the two together since the installation typically also performs
673	an uninstall first and will be seeing similar issues to the uninstall.
674	"""
675	_ = fInstall;
676
677	atSimple = self.katSimpleInstallUninstallMainLogReasons;
678	if oCaseFile.oTestBox.sOs in self.kdatSimpleInstallUninstallMainLogReasonsPerOs:
679	atSimple = self.kdatSimpleInstallUninstallMainLogReasonsPerOs[oCaseFile.oTestBox.sOs] + atSimple;
680
681	fFoundSomething = False;
682	for fStopOnHit, tReason, sNeedle in atSimple:
683	if sResultLog.find(sNeedle) > 0:
684	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
685	if fStopOnHit:
686	return True;
687	fFoundSomething = True;
688
689	return fFoundSomething if fFoundSomething else None;
690
691
692	def investigateBadTestBox(self, oCaseFile):
693	"""
694	Checks out bad-testbox statuses.
695	"""
696	_ = oCaseFile;
697	return False;
698
699
700	def investigateVBoxUnitTest(self, oCaseFile):
701	"""
702	Checks out a VBox unittest problem.
703	"""
704
705	#
706	# Process simple test case failures first, using their name as reason.
707	# We do the reason management just like for BSODs.
708	#
709	cRelevantOnes = 0;
710	sMainLog = oCaseFile.getMainLog();
711	aoFailedResults = oCaseFile.oTree.getListOfFailures();
712	for oFailedResult in aoFailedResults:
713	if oFailedResult is oCaseFile.oTree:
714	self.vprint('TODO: toplevel failure');
715	cRelevantOnes += 1
716
717	elif oFailedResult.sName == 'Installing VirtualBox':
718	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
719	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
720	cRelevantOnes += 1
721
722	elif oFailedResult.sName == 'Uninstalling VirtualBox':
723	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
724	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
725	cRelevantOnes += 1
726
727	elif oFailedResult.oParent is not None:
728	# Get the 2nd level node because that's where we'll find the unit test name.
729	while oFailedResult.oParent.oParent is not None:
730	oFailedResult = oFailedResult.oParent;
731
732	# Only report a failure once.
733	if oFailedResult.idTestResult not in oCaseFile.dReasonForResultId:
734	sKey = oFailedResult.sName;
735	if sKey.startswith('testcase/'):
736	sKey = sKey[9:];
737	if sKey in self.asUnitTestReasons:
738	tReason = ( self.ksUnitTestCategory, sKey );
739	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
740	else:
741	self.dprint(u'Unit test failure "%s" not found in %s;' % (sKey, self.asUnitTestReasons));
742	tReason = ( self.ksUnitTestCategory, self.ksUnitTestAddNew );
743	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sKey);
744	cRelevantOnes += 1
745	else:
746	self.vprint(u'Internal error: expected oParent to NOT be None for %s' % (oFailedResult,));
747
748	#
749	# If we've caught all the relevant ones by now, report the result.
750	#
751	if len(oCaseFile.dReasonForResultId) >= cRelevantOnes:
752	return self.caseClosed(oCaseFile);
753	return False;
754
755	def extractGuestCpuStack(self, sInfoText):
756	"""
757	Extracts the guest CPU stacks from the input file.
758
759	Returns a dictionary keyed by the CPU number, value being a list of
760	raw stack lines (no header).
761	Returns empty dictionary if no stacks where found.
762	"""
763	dRet = {};
764	off = 0;
765	while True:
766	# Find the stack.
767	offStart = sInfoText.find('=== start guest stack VCPU ', off);
768	if offStart < 0:
769	break;
770	offEnd = sInfoText.find('=== end guest stack', offStart + 20);
771	if offEnd >= 0:
772	offEnd += 3;
773	else:
774	offEnd = sInfoText.find('=== start guest stack VCPU', offStart + 20);
775	if offEnd < 0:
776	offEnd = len(sInfoText);
777
778	sStack = sInfoText[offStart : offEnd];
779	sStack = sStack.replace('\r',''); # paranoia
780	asLines = sStack.split('\n');
781
782	# Figure the CPU.
783	asWords = asLines[0].split();
784	if len(asWords) < 6 or not asWords[5].isdigit():
785	break;
786	iCpu = int(asWords[5]);
787
788	# Add it and advance.
789	dRet[iCpu] = [sLine.rstrip() for sLine in asLines[2:-1]]
790	off = offEnd;
791	return dRet;
792
793	def investigateInfoKvmLockSpinning(self, oCaseFile, sInfoText, dLogs):
794	""" Investigates kvm_lock_spinning deadlocks """
795	#
796	# Extract the stacks. We need more than one CPU to create a deadlock.
797	#
798	dStacks = self.extractGuestCpuStack(sInfoText);
799	self.dprint('kvm_lock_spinning: found %s stacks' % (len(dStacks),));
800	if len(dStacks) >= 2:
801	#
802	# Examin each of the stacks. Each must have kvm_lock_spinning in
803	# one of the first three entries.
804	#
805	cHits = 0;
806	for iCpu in dStacks:
807	asBacktrace = dStacks[iCpu];
808	for iFrame in xrange(min(3, len(asBacktrace))):
809	if asBacktrace[iFrame].find('kvm_lock_spinning') >= 0:
810	cHits += 1;
811	break;
812	self.dprint('kvm_lock_spinning: %s/%s hits' % (cHits, len(dStacks),));
813	if cHits == len(dStacks):
814	return (True, self.ktReason_VMM_kvm_lock_spinning);
815
816	_ = dLogs; _ = oCaseFile;
817	return (False, None);
818
819	## Things we search a main or VM log for to figure out why something went bust.
820	katSimpleMainAndVmLogReasons = [
821	# ( Whether to stop on hit, reason tuple, needle text. )
822	( False, ktReason_Guru_Generic, 'GuruMeditation' ),
823	( False, ktReason_Guru_Generic, 'Guru Meditation' ),
824	( True, ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED, 'VERR_IEM_INSTR_NOT_IMPLEMENTED' ),
825	( True, ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED, 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' ),
826	( True, ktReason_Guru_VERR_TRPM_DONT_PANIC, 'VERR_TRPM_DONT_PANIC' ),
827	( True, ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED, 'VERR_PGM_PHYS_PAGE_RESERVED' ),
828	( True, ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE, 'VERR_VMX_INVALID_GUEST_STATE' ),
829	( True, ktReason_Guru_VINF_EM_TRIPLE_FAULT, 'VINF_EM_TRIPLE_FAULT' ),
830	( True, ktReason_Networking_Nonexistent_host_nic,
831	'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
832	( True, ktReason_Host_Reboot_OSX_Watchdog_Timeout, ': "OSX Watchdog Timeout: ' ),
833	( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
834	'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
835	( True, ktReason_Host_HostMemoryLow, 'HostMemoryLow' ),
836	( True, ktReason_Host_HostMemoryLow, 'Failed to procure handy pages; rc=VERR_NO_MEMORY' ),
837	( True, ktReason_Unknown_File_Not_Found,
838	'Error: failed to start machine. Error message: File not found. (VERR_FILE_NOT_FOUND)' ),
839	( True, ktReason_Unknown_File_Not_Found, # lump it in with file-not-found for now.
840	'Error: failed to start machine. Error message: Not supported. (VERR_NOT_SUPPORTED)' ),
841	( False, ktReason_Unknown_VM_Crash, 'txsDoConnectViaTcp: Machine state: Aborted' ),
842	];
843
844	## Things we search a VBoxHardening.log file for to figure out why something went bust.
845	katSimpleVBoxHardeningLogReasons = [
846	# ( Whether to stop on hit, reason tuple, needle text. )
847	( True, ktReason_Host_DriverNotLoaded, 'Error opening VBoxDrvStub: STATUS_OBJECT_NAME_NOT_FOUND' ),
848	( True, ktReason_Host_NotSignedWithBuildCert, 'Not signed with the build certificate' ),
849	];
850
851	## Things we search a kernel.log file for to figure out why something went bust.
852	katSimpleKernelLogReasons = [
853	# ( Whether to stop on hit, reason tuple, needle text. )
854	( True, ktReason_Panic_HugeMemory, 'mm/huge_memory.c:1988' ),
855	( True, ktReason_Panic_IOAPICDoesntWork, 'IO-APIC + timer doesn''t work' ),
856	( True, ktReason_Panic_TxUnitHang, 'Detected Tx Unit Hang' ),
857	];
858
859	## Things we search the _RIGHT_ _STRIPPED_ vgatext for.
860	katSimpleVgaTextReasons = [
861	# ( Whether to stop on hit, reason tuple, needle text. )
862	( True, ktReason_Panic_MP_BIOS_IO_APIC,
863	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n\n" ),
864	( True, ktReason_Panic_MP_BIOS_IO_APIC,
865	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n"
866	"...trying to set up timer (IRQ0) through the 8259A ... failed.\n"
867	"...trying to set up timer as Virtual Wire IRQ... failed.\n"
868	"...trying to set up timer as ExtINT IRQ... failed :(.\n"
869	"Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug\n"
870	"and send a report. Then try booting with the 'noapic' option\n"
871	"\n" ),
872	( True, ktReason_OSInstall_GRUB_hang,
873	"-----\nGRUB Loading stage2..\n\n\n\n" ),
874	( True, ktReason_Panic_BootManagerC000000F,
875	"Windows failed to start. A recent hardware or software change might be the" ),
876	];
877
878	## Things we search for in the info.txt file. Require handlers for now.
879	katInfoTextHandlers = [
880	# ( Trigger text, handler method )
881	( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
882	];
883
884	## Mapping screenshot/failure SHA-256 hashes to failure reasons.
885	katSimpleScreenshotHashReasons = [
886	# ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
887	( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
888	( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
889	( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
890	( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
891	( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
892	( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
893	( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
894	( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
895	];
896
897	def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
898	"""
899	Investigates a failed VM run.
900	"""
901
902	def investigateLogSet():
903	"""
904	Investigates the current set of VM related logs.
905	"""
906	self.dprint('investigateLogSet: lengths: result log %u, VM log %u, kernel log %u, vga text %u, info text %u'
907	% ( len(sResultLog if sResultLog else ''),
908	len(sVMLog if sVMLog else ''),
909	len(sKrnlLog if sKrnlLog else ''),
910	len(sVgaText if sVgaText else ''),
911	len(sInfoText if sInfoText else ''), ));
912
913	#self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
914	#self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
915	#self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
916	#self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
917	#self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
918
919	# TODO: more
920
921	#
922	# Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
923	#
924	sDetails = self.findInAnyAndReturnResetOfLine([ sVMLog, sResultLog ],
925	'GIM: HyperV: Guest indicates a fatal condition! P0=');
926	if sDetails is not None:
927	# P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
928	sKey = sDetails.split(' ', 1)[0];
929	try: sKey = '0x%08X' % (int(sKey, 16),);
930	except: pass;
931	if sKey in self.asBsodReasons:
932	tReason = ( self.ksBsodCategory, sKey );
933	elif sKey.lower() in self.asBsodReasons: # just in case.
934	tReason = ( self.ksBsodCategory, sKey.lower() );
935	else:
936	self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
937	tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
938	return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
939
940	#
941	# Look for linux panic.
942	#
943	if sKrnlLog is not None:
944	for fStopOnHit, tReason, sNeedle in self.katSimpleKernelLogReasons:
945	if sKrnlLog.find(sNeedle) > 0:
946	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
947	if fStopOnHit:
948	return True;
949	fFoundSomething = True;
950
951	#
952	# Loop thru the simple stuff.
953	#
954	fFoundSomething = False;
955	for fStopOnHit, tReason, sNeedle in self.katSimpleMainAndVmLogReasons:
956	if sResultLog.find(sNeedle) > 0 or (sVMLog is not None and sVMLog.find(sNeedle) > 0):
957	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
958	if fStopOnHit:
959	return True;
960	fFoundSomething = True;
961
962	# Continue with vga text.
963	if sVgaText:
964	for fStopOnHit, tReason, sNeedle in self.katSimpleVgaTextReasons:
965	if sVgaText.find(sNeedle) > 0:
966	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
967	if fStopOnHit:
968	return True;
969	fFoundSomething = True;
970	_ = sInfoText;
971
972	# Continue with screen hashes.
973	if sScreenHash is not None:
974	for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
975	if sScreenHash == sHash:
976	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
977	if fStopOnHit:
978	return True;
979	fFoundSomething = True;
980
981	# Check VBoxHardening.log.
982	if sNtHardLog is not None:
983	for fStopOnHit, tReason, sNeedle in self.katSimpleVBoxHardeningLogReasons:
984	if sNtHardLog.find(sNeedle) > 0:
985	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
986	if fStopOnHit:
987	return True;
988	fFoundSomething = True;
989
990	#
991	# Complicated stuff.
992	#
993	dLogs = {
994	'sVMLog': sVMLog,
995	'sNtHardLog': sNtHardLog,
996	'sScreenHash': sScreenHash,
997	'sKrnlLog': sKrnlLog,
998	'sVgaText': sVgaText,
999	'sInfoText': sInfoText,
1000	};
1001
1002	# info.txt.
1003	if sInfoText:
1004	for sNeedle, fnHandler in self.katInfoTextHandlers:
1005	if sInfoText.find(sNeedle) > 0:
1006	(fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
1007	if tReason is not None:
1008	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1009	if fStop:
1010	return True;
1011	fFoundSomething = True;
1012
1013	#
1014	# Check for repeated reboots...
1015	#
1016	if sVMLog is not None:
1017	cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1018	if cResets > 10:
1019	return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1020	sComment = 'Counted %s reboots' % (cResets,));
1021
1022	return fFoundSomething;
1023
1024	#
1025	# Check if we got any VM or/and kernel logs. Treat them as sets in
1026	# case we run multiple VMs here (this is of course ASSUMING they
1027	# appear in the order that terminateVmBySession uploads them).
1028	#
1029	cTimes = 0;
1030	sVMLog = None;
1031	sNtHardLog = None;
1032	sScreenHash = None;
1033	sKrnlLog = None;
1034	sVgaText = None;
1035	sInfoText = None;
1036	for oFile in oFailedResult.aoFiles:
1037	if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1038	if 'VBoxHardening.log' not in oFile.sFile:
1039	if sVMLog is not None:
1040	if investigateLogSet() is True:
1041	return True;
1042	cTimes += 1;
1043	sInfoText = None;
1044	sVgaText = None;
1045	sKrnlLog = None;
1046	sScreenHash = None;
1047	sNtHardLog = None;
1048	sVMLog = oCaseFile.getLogFile(oFile);
1049	else:
1050	sNtHardLog = oCaseFile.getLogFile(oFile);
1051	elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1052	sKrnlLog = oCaseFile.getLogFile(oFile);
1053	elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1054	sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1055	elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1056	sInfoText = oCaseFile.getLogFile(oFile);
1057	elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1058	sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1059	if sScreenHash is not None:
1060	sScreenHash = sScreenHash.lower();
1061	self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1062
1063	if ( sVMLog is not None \
1064	or sNtHardLog is not None \
1065	or cTimes == 0) \
1066	and investigateLogSet() is True:
1067	return True;
1068
1069	return None;
1070
1071
1072	def isResultFromVMRun(self, oFailedResult, sResultLog):
1073	"""
1074	Checks if this result and corresponding log snippet looks like a VM run.
1075	"""
1076
1077	# Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1078	if sResultLog.find(' startVm') > 0:
1079	return True;
1080
1081	# Any other indicators? No?
1082	_ = oFailedResult;
1083	return False;
1084
1085	def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1086	"""
1087	Checks out a VBox VM test.
1088
1089	This is generic investigation of a test running one or more VMs, like
1090	for example a smoke test or a guest installation test.
1091
1092	The fSingleVM parameter is a hint, which probably won't come in useful.
1093	"""
1094	_ = fSingleVM;
1095
1096	#
1097	# Get a list of test result failures we should be looking into and the main log.
1098	#
1099	aoFailedResults = oCaseFile.oTree.getListOfFailures();
1100	sMainLog = oCaseFile.getMainLog();
1101
1102	#
1103	# There are a set of errors ending up on the top level result record.
1104	# Should deal with these first.
1105	#
1106	if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1107	# Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1108	# be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1109	# ignore other failures in the test if we're not a little bit careful here.
1110	if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1111	oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1112	return self.caseClosed(oCaseFile);
1113
1114	# Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1115	if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1116	(' Timeout: ', ' Attempting to abort child...',) ):
1117	if sMainLog.find('* glibc detected * /') > 0:
1118	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1119	else:
1120	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1121	return self.caseClosed(oCaseFile);
1122
1123	# Look for heap corruption without visible hang.
1124	if sMainLog.find('* glibc detected * /') > 0 \
1125	or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1126	oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1127	return self.caseClosed(oCaseFile);
1128
1129	# Out of memory w/ timeout.
1130	if sMainLog.find('sErrId=HostMemoryLow') > 0:
1131	oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1132	return self.caseClosed(oCaseFile);
1133
1134	# Stale files like vts_rm.exe (windows).
1135	offEnd = sMainLog.rfind('* The test driver exits successfully. *');
1136	if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1137	oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1138	return self.caseClosed(oCaseFile);
1139
1140	#
1141	# XPCOM screwup
1142	#
1143	if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1144	oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1145	return self.caseClosed(oCaseFile);
1146
1147	#
1148	# Go thru each failed result.
1149	#
1150	for oFailedResult in aoFailedResults:
1151	self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1152	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1153	if oFailedResult.sName == 'Installing VirtualBox':
1154	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1155
1156	elif oFailedResult.sName == 'Uninstalling VirtualBox':
1157	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1158
1159	elif self.isResultFromVMRun(oFailedResult, sResultLog):
1160	self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1161
1162	elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1163	oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1164
1165	elif sResultLog.find('The machine is not mutable (state is ') > 0:
1166	self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1167	oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1168
1169	elif sResultLog.find('** error: no action was specified') > 0 \
1170	or sResultLog.find('(len(self._asXml, asText))') > 0:
1171	oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1172
1173	else:
1174	self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1175	self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1176
1177	#
1178	# Report home and close the case if we got them all, otherwise log it.
1179	#
1180	if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1181	return self.caseClosed(oCaseFile);
1182
1183	if oCaseFile.dReasonForResultId:
1184	self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1185	% (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1186	else:
1187	self.vprint(u'XXX: Could not figure out anything at all! :-(');
1188	return False;
1189
1190
1191	def reasoningFailures(self):
1192	"""
1193	Guess the reason for failures.
1194	"""
1195	#
1196	# Get a list of failed test sets without any assigned failure reason.
1197	#
1198	cGot = 0;
1199	aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack, tsNow = self.tsNow);
1200	for oTestSet in aoTestSets:
1201	self.dprint(u'');
1202	self.dprint(u'reasoningFailures: Checking out test set #%u, status %s' % ( oTestSet.idTestSet, oTestSet.enmStatus,))
1203
1204	#
1205	# Open a case file and assign it to the right investigator.
1206	#
1207	(oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1208	oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1209	oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1210	oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1211	oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1212
1213	oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1214
1215	if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1216	self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1217	fRc = self.investigateBadTestBox(oCaseFile);
1218
1219	elif oCaseFile.isVBoxUnitTest():
1220	self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1221	fRc = self.investigateVBoxUnitTest(oCaseFile);
1222
1223	elif oCaseFile.isVBoxInstallTest():
1224	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1225	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1226
1227	elif oCaseFile.isVBoxUSBTest():
1228	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1229	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1230
1231	elif oCaseFile.isVBoxStorageTest():
1232	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1233	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1234
1235	elif oCaseFile.isVBoxGAsTest():
1236	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1237	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1238
1239	elif oCaseFile.isVBoxAPITest():
1240	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1241	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1242
1243	elif oCaseFile.isVBoxBenchmarkTest():
1244	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1245	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1246
1247	elif oCaseFile.isVBoxSmokeTest():
1248	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1249	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1250
1251	else:
1252	self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1253	fRc = False;
1254	cGot += fRc is True;
1255
1256	self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1257	return 0;
1258
1259
1260	def main(self):
1261	"""
1262	The 'main' function.
1263	Return exit code (0, 1, etc).
1264	"""
1265	# Database stuff.
1266	self.oDb = TMDatabaseConnection()
1267	self.oTestResultLogic = TestResultLogic(self.oDb);
1268	self.oTestSetLogic = TestSetLogic(self.oDb);
1269	self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1270	self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1271	self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1272	self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1273
1274	# Get a fix on our 'now' before we do anything..
1275	self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1276	self.tsNow = self.oDb.fetchOne();
1277
1278	# If we're suppost to commit anything we need to get our user ID.
1279	rcExit = 0;
1280	if self.oConfig.fRealRun:
1281	self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1282	if self.oLogin is None:
1283	rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1284	else:
1285	self.uidSelf = self.oLogin.uid;
1286
1287	#
1288	# Do the stuff.
1289	#
1290	if rcExit == 0:
1291	rcExit = self.selfCheck();
1292	if rcExit == 0:
1293	rcExit = self.badTestBoxManagement();
1294	rcExit2 = self.reasoningFailures();
1295	if rcExit == 0:
1296	rcExit = rcExit2;
1297	# Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1298	if rcExit == 0:
1299	rcExit = self.badTestBoxManagement();
1300
1301	# Cleanup.
1302	self.oFailureReasonLogic = None;
1303	self.oTestResultFailureLogic = None;
1304	self.oTestSetLogic = None;
1305	self.oTestResultLogic = None;
1306	self.oDb.close();
1307	self.oDb = None;
1308	if self.oLogFile is not None:
1309	self.oLogFile.close();
1310	self.oLogFile = None;
1311	return rcExit;
1312
1313	if __name__ == '__main__':
1314	sys.exit(VirtualTestSheriff().main());
1315

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

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 66431

Download in other formats: