virtual_test_sheriff.py@ 66788

Last change on this file since 66788 was 66788, checked in by vboxsync, 8 years ago
vsheriff: try to detect "BOOTMGR image is corrupt"
Property svn:eol-style set to `LF` Property svn:executable set to ``* Property svn:keywords set to `Author Date Id Revision`
File size: 61.3 KB

Line
1	#!/usr/bin/env python
2	# -- coding: utf-8 --
3	# $Id: virtual_test_sheriff.py 66788 2017-05-04 11:58:57Z 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: 66788 $"
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: 66788 $ \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_BootManager_Image_corrupt = ( 'Unknown', 'BOOTMGR Image corrupt' );
475	ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
476	ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
477	ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
478	ktReason_Panic_TxUnitHang = ( 'Panic', 'Tx Unit Hang' );
479	ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
480	ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
481	ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
482	ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
483	ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
484	ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
485	ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
486	ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
487	ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
488	ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
489	ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
490	ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
491	## @}
492
493	## BSOD category.
494	ksBsodCategory = 'BSOD';
495	## Special reason indicating that the flesh and blood sheriff has work to do.
496	ksBsodAddNew = 'Add new BSOD';
497
498	## Unit test category.
499	ksUnitTestCategory = 'Unit';
500	## Special reason indicating that the flesh and blood sheriff has work to do.
501	ksUnitTestAddNew = 'Add new';
502
503	## Used for indica that we shouldn't report anything for this test result ID and
504	## consider promoting the previous error to test set level if it's the only one.
505	ktHarmless = ( 'Probably', 'Caused by previous error' );
506
507
508	def caseClosed(self, oCaseFile):
509	"""
510	Reports the findings in the case and closes it.
511	"""
512	#
513	# Log it and create a dReasonForReasultId we can use below.
514	#
515	dCommentForResultId = oCaseFile.dCommentForResultId;
516	if oCaseFile.dReasonForResultId:
517	# Must weed out ktHarmless.
518	dReasonForResultId = {};
519	for idKey, tReason in oCaseFile.dReasonForResultId.items():
520	if tReason is not self.ktHarmless:
521	dReasonForResultId[idKey] = tReason;
522	if not dReasonForResultId:
523	self.vprint(u'TODO: Closing %s without a real reason, only %s.'
524	% (oCaseFile.sName, oCaseFile.dReasonForResultId));
525	return False;
526
527	# Try promote to single reason.
528	atValues = dReasonForResultId.values();
529	fSingleReason = True;
530	if len(dReasonForResultId) == 1 and dReasonForResultId.keys()[0] != oCaseFile.oTestSet.idTestResult:
531	self.dprint(u'Promoting single reason to whole set: %s' % (atValues[0],));
532	elif len(dReasonForResultId) > 1 and len(atValues) == atValues.count(atValues[0]):
533	self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), atValues[0]));
534	else:
535	fSingleReason = False;
536	if fSingleReason:
537	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: atValues[0], };
538	if dCommentForResultId:
539	dCommentForResultId = { oCaseFile.oTestSet.idTestResult: dCommentForResultId.values()[0], };
540	elif oCaseFile.tReason is not None:
541	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
542	else:
543	self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
544	return False;
545
546	self.vprint(u'Closing %s with following reason%s: %s'
547	% ( oCaseFile.sName, 's' if dReasonForResultId > 0 else '', dReasonForResultId, ));
548
549	#
550	# Add the test failure reason record(s).
551	#
552	for idTestResult, tReason in dReasonForResultId.items():
553	oFailureReason = self.getFailureReason(tReason);
554	if oFailureReason is not None:
555	sComment = 'Set by $Revision: 66788 $' # Handy for reverting later.
556	if idTestResult in dCommentForResultId:
557	sComment += ': ' + dCommentForResultId[idTestResult];
558
559	oAdd = TestResultFailureData();
560	oAdd.initFromValues(idTestResult = idTestResult,
561	idFailureReason = oFailureReason.idFailureReason,
562	uidAuthor = self.uidSelf,
563	idTestSet = oCaseFile.oTestSet.idTestSet,
564	sComment = sComment,);
565	if self.oConfig.fRealRun:
566	try:
567	self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
568	except Exception as oXcpt:
569	self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
570	% (oXcpt, oAdd, oCaseFile.sLongName,));
571	else:
572	self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
573	return True;
574
575	#
576	# Tools for assiting log parsing.
577	#
578
579	@staticmethod
580	def matchFollowedByLines(sStr, off, asFollowingLines):
581	""" Worker for isThisFollowedByTheseLines. """
582
583	# Advance off to the end of the line.
584	off = sStr.find('\n', off);
585	if off < 0:
586	return False;
587	off += 1;
588
589	# Match each string with the subsequent lines.
590	for iLine, sLine in enumerate(asFollowingLines):
591	offEnd = sStr.find('\n', off);
592	if offEnd < 0:
593	return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
594	if sLine and sStr.find(sLine, off, offEnd) < 0:
595	return False;
596
597	# next line.
598	off = offEnd + 1;
599
600	return True;
601
602	@staticmethod
603	def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
604	"""
605	Looks for a line contining sFirst which is then followed by lines
606	with the strings in asFollowingLines. (No newline chars anywhere!)
607	Returns True / False.
608	"""
609	off = sStr.find(sFirst, 0);
610	while off >= 0:
611	if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
612	return True;
613	off = sStr.find(sFirst, off + 1);
614	return False;
615
616	@staticmethod
617	def findAndReturnResetOfLine(sHaystack, sNeedle):
618	"""
619	Looks for sNeedle in sHaystack.
620	Returns The text following the needle up to the end of the line.
621	Returns None if not found.
622	"""
623	if sHaystack is None:
624	return None;
625	off = sHaystack.find(sNeedle);
626	if off < 0:
627	return None;
628	off += len(sNeedle)
629	offEol = sHaystack.find('\n', off);
630	if offEol < 0:
631	offEol = len(sHaystack);
632	return sHaystack[off:offEol]
633
634	@staticmethod
635	def findInAnyAndReturnResetOfLine(asHaystacks, sNeedle):
636	"""
637	Looks for sNeedle in zeroe or more haystacks (asHaystack).
638	Returns The text following the first needed found up to the end of the line.
639	Returns None if not found.
640	"""
641	for sHaystack in asHaystacks:
642	sRet = VirtualTestSheriff.findAndReturnResetOfLine(sHaystack, sNeedle);
643	if sRet is not None:
644	return sRet;
645	return None;
646
647
648	#
649	# The investigative units.
650	#
651
652	katSimpleInstallUninstallMainLogReasons = [
653	# ( Whether to stop on hit, reason tuple, needle text. )
654	( False, ktReason_Host_LeftoverService,
655	'SERVICE_NAME: vbox' ),
656	];
657
658	kdatSimpleInstallUninstallMainLogReasonsPerOs = {
659	'darwin': [
660	# ( Whether to stop on hit, reason tuple, needle text. )
661	( True, ktReason_Host_DriverNotUnloading,
662	'Can\'t remove kext org.virtualbox.kext.VBoxDrv; services failed to terminate - 0xe00002c7' ),
663	],
664	};
665
666
667	def investigateInstallUninstallFailure(self, oCaseFile, oFailedResult, sResultLog, fInstall):
668	"""
669	Investigates an install or uninstall failure.
670
671	We lump the two together since the installation typically also performs
672	an uninstall first and will be seeing similar issues to the uninstall.
673	"""
674	_ = fInstall;
675
676	atSimple = self.katSimpleInstallUninstallMainLogReasons;
677	if oCaseFile.oTestBox.sOs in self.kdatSimpleInstallUninstallMainLogReasonsPerOs:
678	atSimple = self.kdatSimpleInstallUninstallMainLogReasonsPerOs[oCaseFile.oTestBox.sOs] + atSimple;
679
680	fFoundSomething = False;
681	for fStopOnHit, tReason, sNeedle in atSimple:
682	if sResultLog.find(sNeedle) > 0:
683	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
684	if fStopOnHit:
685	return True;
686	fFoundSomething = True;
687
688	return fFoundSomething if fFoundSomething else None;
689
690
691	def investigateBadTestBox(self, oCaseFile):
692	"""
693	Checks out bad-testbox statuses.
694	"""
695	_ = oCaseFile;
696	return False;
697
698
699	def investigateVBoxUnitTest(self, oCaseFile):
700	"""
701	Checks out a VBox unittest problem.
702	"""
703
704	#
705	# Process simple test case failures first, using their name as reason.
706	# We do the reason management just like for BSODs.
707	#
708	cRelevantOnes = 0;
709	sMainLog = oCaseFile.getMainLog();
710	aoFailedResults = oCaseFile.oTree.getListOfFailures();
711	for oFailedResult in aoFailedResults:
712	if oFailedResult is oCaseFile.oTree:
713	self.vprint('TODO: toplevel failure');
714	cRelevantOnes += 1
715
716	elif oFailedResult.sName == 'Installing VirtualBox':
717	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
718	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
719	cRelevantOnes += 1
720
721	elif oFailedResult.sName == 'Uninstalling VirtualBox':
722	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
723	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
724	cRelevantOnes += 1
725
726	elif oFailedResult.oParent is not None:
727	# Get the 2nd level node because that's where we'll find the unit test name.
728	while oFailedResult.oParent.oParent is not None:
729	oFailedResult = oFailedResult.oParent;
730
731	# Only report a failure once.
732	if oFailedResult.idTestResult not in oCaseFile.dReasonForResultId:
733	sKey = oFailedResult.sName;
734	if sKey.startswith('testcase/'):
735	sKey = sKey[9:];
736	if sKey in self.asUnitTestReasons:
737	tReason = ( self.ksUnitTestCategory, sKey );
738	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
739	else:
740	self.dprint(u'Unit test failure "%s" not found in %s;' % (sKey, self.asUnitTestReasons));
741	tReason = ( self.ksUnitTestCategory, self.ksUnitTestAddNew );
742	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sKey);
743	cRelevantOnes += 1
744	else:
745	self.vprint(u'Internal error: expected oParent to NOT be None for %s' % (oFailedResult,));
746
747	#
748	# If we've caught all the relevant ones by now, report the result.
749	#
750	if len(oCaseFile.dReasonForResultId) >= cRelevantOnes:
751	return self.caseClosed(oCaseFile);
752	return False;
753
754	def extractGuestCpuStack(self, sInfoText):
755	"""
756	Extracts the guest CPU stacks from the input file.
757
758	Returns a dictionary keyed by the CPU number, value being a list of
759	raw stack lines (no header).
760	Returns empty dictionary if no stacks where found.
761	"""
762	dRet = {};
763	off = 0;
764	while True:
765	# Find the stack.
766	offStart = sInfoText.find('=== start guest stack VCPU ', off);
767	if offStart < 0:
768	break;
769	offEnd = sInfoText.find('=== end guest stack', offStart + 20);
770	if offEnd >= 0:
771	offEnd += 3;
772	else:
773	offEnd = sInfoText.find('=== start guest stack VCPU', offStart + 20);
774	if offEnd < 0:
775	offEnd = len(sInfoText);
776
777	sStack = sInfoText[offStart : offEnd];
778	sStack = sStack.replace('\r',''); # paranoia
779	asLines = sStack.split('\n');
780
781	# Figure the CPU.
782	asWords = asLines[0].split();
783	if len(asWords) < 6 or not asWords[5].isdigit():
784	break;
785	iCpu = int(asWords[5]);
786
787	# Add it and advance.
788	dRet[iCpu] = [sLine.rstrip() for sLine in asLines[2:-1]]
789	off = offEnd;
790	return dRet;
791
792	def investigateInfoKvmLockSpinning(self, oCaseFile, sInfoText, dLogs):
793	""" Investigates kvm_lock_spinning deadlocks """
794	#
795	# Extract the stacks. We need more than one CPU to create a deadlock.
796	#
797	dStacks = self.extractGuestCpuStack(sInfoText);
798	self.dprint('kvm_lock_spinning: found %s stacks' % (len(dStacks),));
799	if len(dStacks) >= 2:
800	#
801	# Examin each of the stacks. Each must have kvm_lock_spinning in
802	# one of the first three entries.
803	#
804	cHits = 0;
805	for iCpu in dStacks:
806	asBacktrace = dStacks[iCpu];
807	for iFrame in xrange(min(3, len(asBacktrace))):
808	if asBacktrace[iFrame].find('kvm_lock_spinning') >= 0:
809	cHits += 1;
810	break;
811	self.dprint('kvm_lock_spinning: %s/%s hits' % (cHits, len(dStacks),));
812	if cHits == len(dStacks):
813	return (True, self.ktReason_VMM_kvm_lock_spinning);
814
815	_ = dLogs; _ = oCaseFile;
816	return (False, None);
817
818	## Things we search a main or VM log for to figure out why something went bust.
819	katSimpleMainAndVmLogReasons = [
820	# ( Whether to stop on hit, reason tuple, needle text. )
821	( False, ktReason_Guru_Generic, 'GuruMeditation' ),
822	( False, ktReason_Guru_Generic, 'Guru Meditation' ),
823	( True, ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED, 'VERR_IEM_INSTR_NOT_IMPLEMENTED' ),
824	( True, ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED, 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' ),
825	( True, ktReason_Guru_VERR_TRPM_DONT_PANIC, 'VERR_TRPM_DONT_PANIC' ),
826	( True, ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED, 'VERR_PGM_PHYS_PAGE_RESERVED' ),
827	( True, ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE, 'VERR_VMX_INVALID_GUEST_STATE' ),
828	( True, ktReason_Guru_VINF_EM_TRIPLE_FAULT, 'VINF_EM_TRIPLE_FAULT' ),
829	( True, ktReason_Networking_Nonexistent_host_nic,
830	'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
831	( True, ktReason_Host_Reboot_OSX_Watchdog_Timeout, ': "OSX Watchdog Timeout: ' ),
832	( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
833	'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
834	( True, ktReason_Host_HostMemoryLow, 'HostMemoryLow' ),
835	( True, ktReason_Host_HostMemoryLow, 'Failed to procure handy pages; rc=VERR_NO_MEMORY' ),
836	( True, ktReason_Unknown_File_Not_Found,
837	'Error: failed to start machine. Error message: File not found. (VERR_FILE_NOT_FOUND)' ),
838	( True, ktReason_Unknown_File_Not_Found, # lump it in with file-not-found for now.
839	'Error: failed to start machine. Error message: Not supported. (VERR_NOT_SUPPORTED)' ),
840	( False, ktReason_Unknown_VM_Crash, 'txsDoConnectViaTcp: Machine state: Aborted' ),
841	( True, ktReason_Host_Modprobe_Failed, 'Kernel driver not installed' )
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	( True, ktReason_BootManager_Image_corrupt,
877	"BOOTMGR image is corrupt. The system cannot boot." ),
878	];
879
880	## Things we search for in the info.txt file. Require handlers for now.
881	katInfoTextHandlers = [
882	# ( Trigger text, handler method )
883	( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
884	];
885
886	## Mapping screenshot/failure SHA-256 hashes to failure reasons.
887	katSimpleScreenshotHashReasons = [
888	# ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
889	( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
890	( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
891	( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
892	( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
893	( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
894	( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
895	( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
896	( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
897	];
898
899	def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
900	"""
901	Investigates a failed VM run.
902	"""
903
904	def investigateLogSet():
905	"""
906	Investigates the current set of VM related logs.
907	"""
908	self.dprint('investigateLogSet: lengths: result log %u, VM log %u, kernel log %u, vga text %u, info text %u'
909	% ( len(sResultLog if sResultLog else ''),
910	len(sVMLog if sVMLog else ''),
911	len(sKrnlLog if sKrnlLog else ''),
912	len(sVgaText if sVgaText else ''),
913	len(sInfoText if sInfoText else ''), ));
914
915	#self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
916	#self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
917	#self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
918	#self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
919	#self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
920
921	# TODO: more
922
923	#
924	# Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
925	#
926	sDetails = self.findInAnyAndReturnResetOfLine([ sVMLog, sResultLog ],
927	'GIM: HyperV: Guest indicates a fatal condition! P0=');
928	if sDetails is not None:
929	# P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
930	sKey = sDetails.split(' ', 1)[0];
931	try: sKey = '0x%08X' % (int(sKey, 16),);
932	except: pass;
933	if sKey in self.asBsodReasons:
934	tReason = ( self.ksBsodCategory, sKey );
935	elif sKey.lower() in self.asBsodReasons: # just in case.
936	tReason = ( self.ksBsodCategory, sKey.lower() );
937	else:
938	self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
939	tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
940	return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
941
942	#
943	# Look for linux panic.
944	#
945	if sKrnlLog is not None:
946	for fStopOnHit, tReason, sNeedle in self.katSimpleKernelLogReasons:
947	if sKrnlLog.find(sNeedle) > 0:
948	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
949	if fStopOnHit:
950	return True;
951	fFoundSomething = True;
952
953	#
954	# Loop thru the simple stuff.
955	#
956	fFoundSomething = False;
957	for fStopOnHit, tReason, sNeedle in self.katSimpleMainAndVmLogReasons:
958	if sResultLog.find(sNeedle) > 0 or (sVMLog is not None and sVMLog.find(sNeedle) > 0):
959	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
960	if fStopOnHit:
961	return True;
962	fFoundSomething = True;
963
964	# Continue with vga text.
965	if sVgaText:
966	for fStopOnHit, tReason, sNeedle in self.katSimpleVgaTextReasons:
967	if sVgaText.find(sNeedle) > 0:
968	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
969	if fStopOnHit:
970	return True;
971	fFoundSomething = True;
972	_ = sInfoText;
973
974	# Continue with screen hashes.
975	if sScreenHash is not None:
976	for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
977	if sScreenHash == sHash:
978	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
979	if fStopOnHit:
980	return True;
981	fFoundSomething = True;
982
983	# Check VBoxHardening.log.
984	if sNtHardLog is not None:
985	for fStopOnHit, tReason, sNeedle in self.katSimpleVBoxHardeningLogReasons:
986	if sNtHardLog.find(sNeedle) > 0:
987	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
988	if fStopOnHit:
989	return True;
990	fFoundSomething = True;
991
992	#
993	# Complicated stuff.
994	#
995	dLogs = {
996	'sVMLog': sVMLog,
997	'sNtHardLog': sNtHardLog,
998	'sScreenHash': sScreenHash,
999	'sKrnlLog': sKrnlLog,
1000	'sVgaText': sVgaText,
1001	'sInfoText': sInfoText,
1002	};
1003
1004	# info.txt.
1005	if sInfoText:
1006	for sNeedle, fnHandler in self.katInfoTextHandlers:
1007	if sInfoText.find(sNeedle) > 0:
1008	(fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
1009	if tReason is not None:
1010	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1011	if fStop:
1012	return True;
1013	fFoundSomething = True;
1014
1015	#
1016	# Check for repeated reboots...
1017	#
1018	if sVMLog is not None:
1019	cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1020	if cResets > 10:
1021	return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1022	sComment = 'Counted %s reboots' % (cResets,));
1023
1024	return fFoundSomething;
1025
1026	#
1027	# Check if we got any VM or/and kernel logs. Treat them as sets in
1028	# case we run multiple VMs here (this is of course ASSUMING they
1029	# appear in the order that terminateVmBySession uploads them).
1030	#
1031	cTimes = 0;
1032	sVMLog = None;
1033	sNtHardLog = None;
1034	sScreenHash = None;
1035	sKrnlLog = None;
1036	sVgaText = None;
1037	sInfoText = None;
1038	for oFile in oFailedResult.aoFiles:
1039	if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1040	if 'VBoxHardening.log' not in oFile.sFile:
1041	if sVMLog is not None:
1042	if investigateLogSet() is True:
1043	return True;
1044	cTimes += 1;
1045	sInfoText = None;
1046	sVgaText = None;
1047	sKrnlLog = None;
1048	sScreenHash = None;
1049	sNtHardLog = None;
1050	sVMLog = oCaseFile.getLogFile(oFile);
1051	else:
1052	sNtHardLog = oCaseFile.getLogFile(oFile);
1053	elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1054	sKrnlLog = oCaseFile.getLogFile(oFile);
1055	elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1056	sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1057	elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1058	sInfoText = oCaseFile.getLogFile(oFile);
1059	elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1060	sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1061	if sScreenHash is not None:
1062	sScreenHash = sScreenHash.lower();
1063	self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1064
1065	if ( sVMLog is not None \
1066	or sNtHardLog is not None \
1067	or cTimes == 0) \
1068	and investigateLogSet() is True:
1069	return True;
1070
1071	return None;
1072
1073
1074	def isResultFromVMRun(self, oFailedResult, sResultLog):
1075	"""
1076	Checks if this result and corresponding log snippet looks like a VM run.
1077	"""
1078
1079	# Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1080	if sResultLog.find(' startVm') > 0:
1081	return True;
1082
1083	# Any other indicators? No?
1084	_ = oFailedResult;
1085	return False;
1086
1087	def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1088	"""
1089	Checks out a VBox VM test.
1090
1091	This is generic investigation of a test running one or more VMs, like
1092	for example a smoke test or a guest installation test.
1093
1094	The fSingleVM parameter is a hint, which probably won't come in useful.
1095	"""
1096	_ = fSingleVM;
1097
1098	#
1099	# Get a list of test result failures we should be looking into and the main log.
1100	#
1101	aoFailedResults = oCaseFile.oTree.getListOfFailures();
1102	sMainLog = oCaseFile.getMainLog();
1103
1104	#
1105	# There are a set of errors ending up on the top level result record.
1106	# Should deal with these first.
1107	#
1108	if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1109	# Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1110	# be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1111	# ignore other failures in the test if we're not a little bit careful here.
1112	if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1113	oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1114	return self.caseClosed(oCaseFile);
1115
1116	# Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1117	if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1118	(' Timeout: ', ' Attempting to abort child...',) ):
1119	if sMainLog.find('* glibc detected * /') > 0:
1120	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1121	else:
1122	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1123	return self.caseClosed(oCaseFile);
1124
1125	# Look for heap corruption without visible hang.
1126	if sMainLog.find('* glibc detected * /') > 0 \
1127	or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1128	oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1129	return self.caseClosed(oCaseFile);
1130
1131	# Out of memory w/ timeout.
1132	if sMainLog.find('sErrId=HostMemoryLow') > 0:
1133	oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1134	return self.caseClosed(oCaseFile);
1135
1136	# Stale files like vts_rm.exe (windows).
1137	offEnd = sMainLog.rfind('* The test driver exits successfully. *');
1138	if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1139	oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1140	return self.caseClosed(oCaseFile);
1141
1142	#
1143	# XPCOM screwup
1144	#
1145	if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1146	oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1147	return self.caseClosed(oCaseFile);
1148
1149	#
1150	# Go thru each failed result.
1151	#
1152	for oFailedResult in aoFailedResults:
1153	self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1154	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1155	if oFailedResult.sName == 'Installing VirtualBox':
1156	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1157
1158	elif oFailedResult.sName == 'Uninstalling VirtualBox':
1159	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1160
1161	elif self.isResultFromVMRun(oFailedResult, sResultLog):
1162	self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1163
1164	elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1165	oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1166
1167	elif sResultLog.find('The machine is not mutable (state is ') > 0:
1168	self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1169	oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1170
1171	elif sResultLog.find('** error: no action was specified') > 0 \
1172	or sResultLog.find('(len(self._asXml, asText))') > 0:
1173	oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1174
1175	else:
1176	self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1177	self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1178
1179	#
1180	# Report home and close the case if we got them all, otherwise log it.
1181	#
1182	if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1183	return self.caseClosed(oCaseFile);
1184
1185	if oCaseFile.dReasonForResultId:
1186	self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1187	% (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1188	else:
1189	self.vprint(u'XXX: Could not figure out anything at all! :-(');
1190	return False;
1191
1192
1193	def reasoningFailures(self):
1194	"""
1195	Guess the reason for failures.
1196	"""
1197	#
1198	# Get a list of failed test sets without any assigned failure reason.
1199	#
1200	cGot = 0;
1201	aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack, tsNow = self.tsNow);
1202	for oTestSet in aoTestSets:
1203	self.dprint(u'');
1204	self.dprint(u'reasoningFailures: Checking out test set #%u, status %s' % ( oTestSet.idTestSet, oTestSet.enmStatus,))
1205
1206	#
1207	# Open a case file and assign it to the right investigator.
1208	#
1209	(oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1210	oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1211	oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1212	oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1213	oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1214
1215	oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1216
1217	if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1218	self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1219	fRc = self.investigateBadTestBox(oCaseFile);
1220
1221	elif oCaseFile.isVBoxUnitTest():
1222	self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1223	fRc = self.investigateVBoxUnitTest(oCaseFile);
1224
1225	elif oCaseFile.isVBoxInstallTest():
1226	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1227	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1228
1229	elif oCaseFile.isVBoxUSBTest():
1230	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1231	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1232
1233	elif oCaseFile.isVBoxStorageTest():
1234	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1235	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1236
1237	elif oCaseFile.isVBoxGAsTest():
1238	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1239	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1240
1241	elif oCaseFile.isVBoxAPITest():
1242	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1243	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1244
1245	elif oCaseFile.isVBoxBenchmarkTest():
1246	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1247	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1248
1249	elif oCaseFile.isVBoxSmokeTest():
1250	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1251	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1252
1253	else:
1254	self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1255	fRc = False;
1256	cGot += fRc is True;
1257
1258	self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1259	return 0;
1260
1261
1262	def main(self):
1263	"""
1264	The 'main' function.
1265	Return exit code (0, 1, etc).
1266	"""
1267	# Database stuff.
1268	self.oDb = TMDatabaseConnection()
1269	self.oTestResultLogic = TestResultLogic(self.oDb);
1270	self.oTestSetLogic = TestSetLogic(self.oDb);
1271	self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1272	self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1273	self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1274	self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1275
1276	# Get a fix on our 'now' before we do anything..
1277	self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1278	self.tsNow = self.oDb.fetchOne();
1279
1280	# If we're suppost to commit anything we need to get our user ID.
1281	rcExit = 0;
1282	if self.oConfig.fRealRun:
1283	self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1284	if self.oLogin is None:
1285	rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1286	else:
1287	self.uidSelf = self.oLogin.uid;
1288
1289	#
1290	# Do the stuff.
1291	#
1292	if rcExit == 0:
1293	rcExit = self.selfCheck();
1294	if rcExit == 0:
1295	rcExit = self.badTestBoxManagement();
1296	rcExit2 = self.reasoningFailures();
1297	if rcExit == 0:
1298	rcExit = rcExit2;
1299	# Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1300	if rcExit == 0:
1301	rcExit = self.badTestBoxManagement();
1302
1303	# Cleanup.
1304	self.oFailureReasonLogic = None;
1305	self.oTestResultFailureLogic = None;
1306	self.oTestSetLogic = None;
1307	self.oTestResultLogic = None;
1308	self.oDb.close();
1309	self.oDb = None;
1310	if self.oLogFile is not None:
1311	self.oLogFile.close();
1312	self.oLogFile = None;
1313	return rcExit;
1314
1315	if __name__ == '__main__':
1316	sys.exit(VirtualTestSheriff().main());
1317

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

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 66788

Download in other formats: