virtual_test_sheriff.py@ 66817

Last change on this file since 66817 was 66817, checked in by vboxsync, 8 years ago
vsheriff: teach about BSOD / DRIVER_IRQL_NOT_LESS_OR_EQUAL
Property svn:eol-style set to `LF` Property svn:executable set to ``* Property svn:keywords set to `Author Date Id Revision`
File size: 61.6 KB

Line
1	#!/usr/bin/env python
2	# -- coding: utf-8 --
3	# $Id: virtual_test_sheriff.py 66817 2017-05-08 12:37:47Z 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: 66817 $"
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: 66817 $ \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_000000D1 = ( 'BSOD', '0x000000D1' );
457	ktReason_BSOD_C0000225 = ( 'BSOD', '0xC0000225 (boot)' );
458	ktReason_Guru_Generic = ( 'Guru Meditations', 'Generic Guru Meditation' );
459	ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_INSTR_NOT_IMPLEMENTED' );
460	ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' );
461	ktReason_Guru_VERR_TRPM_DONT_PANIC = ( 'Guru Meditations', 'VERR_TRPM_DONT_PANIC' );
462	ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED = ( 'Guru Meditations', 'VERR_PGM_PHYS_PAGE_RESERVED' );
463	ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE = ( 'Guru Meditations', 'VERR_VMX_INVALID_GUEST_STATE' );
464	ktReason_Guru_VINF_EM_TRIPLE_FAULT = ( 'Guru Meditations', 'VINF_EM_TRIPLE_FAULT' );
465	ktReason_Host_HostMemoryLow = ( 'Host', 'HostMemoryLow' );
466	ktReason_Host_DriverNotLoaded = ( 'Host', 'Driver not loaded' );
467	ktReason_Host_DriverNotUnloading = ( 'Host', 'Driver not unloading' );
468	ktReason_Host_NotSignedWithBuildCert = ( 'Host', 'Not signed with build cert' );
469	ktReason_Host_LeftoverService = ( 'Host', 'Leftover service' );
470	ktReason_Host_Reboot_OSX_Watchdog_Timeout = ( 'Host Reboot', 'OSX Watchdog Timeout' );
471	ktReason_Host_Modprobe_Failed = ( 'Host', 'Modprobe failed' );
472	ktReason_Networking_Nonexistent_host_nic = ( 'Networking', 'Nonexistent host networking interface' );
473	ktReason_OSInstall_GRUB_hang = ( 'O/S Install', 'GRUB hang' );
474	ktReason_Panic_BootManagerC000000F = ( 'Panic', 'Hardware Changed' );
475	ktReason_BootManager_Image_corrupt = ( 'Unknown', 'BOOTMGR Image corrupt' );
476	ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
477	ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
478	ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
479	ktReason_Panic_TxUnitHang = ( 'Panic', 'Tx Unit Hang' );
480	ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
481	ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
482	ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
483	ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
484	ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
485	ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
486	ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
487	ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
488	ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
489	ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
490	ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
491	ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
492	## @}
493
494	## BSOD category.
495	ksBsodCategory = 'BSOD';
496	## Special reason indicating that the flesh and blood sheriff has work to do.
497	ksBsodAddNew = 'Add new BSOD';
498
499	## Unit test category.
500	ksUnitTestCategory = 'Unit';
501	## Special reason indicating that the flesh and blood sheriff has work to do.
502	ksUnitTestAddNew = 'Add new';
503
504	## Used for indica that we shouldn't report anything for this test result ID and
505	## consider promoting the previous error to test set level if it's the only one.
506	ktHarmless = ( 'Probably', 'Caused by previous error' );
507
508
509	def caseClosed(self, oCaseFile):
510	"""
511	Reports the findings in the case and closes it.
512	"""
513	#
514	# Log it and create a dReasonForReasultId we can use below.
515	#
516	dCommentForResultId = oCaseFile.dCommentForResultId;
517	if oCaseFile.dReasonForResultId:
518	# Must weed out ktHarmless.
519	dReasonForResultId = {};
520	for idKey, tReason in oCaseFile.dReasonForResultId.items():
521	if tReason is not self.ktHarmless:
522	dReasonForResultId[idKey] = tReason;
523	if not dReasonForResultId:
524	self.vprint(u'TODO: Closing %s without a real reason, only %s.'
525	% (oCaseFile.sName, oCaseFile.dReasonForResultId));
526	return False;
527
528	# Try promote to single reason.
529	atValues = dReasonForResultId.values();
530	fSingleReason = True;
531	if len(dReasonForResultId) == 1 and dReasonForResultId.keys()[0] != oCaseFile.oTestSet.idTestResult:
532	self.dprint(u'Promoting single reason to whole set: %s' % (atValues[0],));
533	elif len(dReasonForResultId) > 1 and len(atValues) == atValues.count(atValues[0]):
534	self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), atValues[0]));
535	else:
536	fSingleReason = False;
537	if fSingleReason:
538	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: atValues[0], };
539	if dCommentForResultId:
540	dCommentForResultId = { oCaseFile.oTestSet.idTestResult: dCommentForResultId.values()[0], };
541	elif oCaseFile.tReason is not None:
542	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
543	else:
544	self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
545	return False;
546
547	self.vprint(u'Closing %s with following reason%s: %s'
548	% ( oCaseFile.sName, 's' if dReasonForResultId > 0 else '', dReasonForResultId, ));
549
550	#
551	# Add the test failure reason record(s).
552	#
553	for idTestResult, tReason in dReasonForResultId.items():
554	oFailureReason = self.getFailureReason(tReason);
555	if oFailureReason is not None:
556	sComment = 'Set by $Revision: 66817 $' # Handy for reverting later.
557	if idTestResult in dCommentForResultId:
558	sComment += ': ' + dCommentForResultId[idTestResult];
559
560	oAdd = TestResultFailureData();
561	oAdd.initFromValues(idTestResult = idTestResult,
562	idFailureReason = oFailureReason.idFailureReason,
563	uidAuthor = self.uidSelf,
564	idTestSet = oCaseFile.oTestSet.idTestSet,
565	sComment = sComment,);
566	if self.oConfig.fRealRun:
567	try:
568	self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
569	except Exception as oXcpt:
570	self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
571	% (oXcpt, oAdd, oCaseFile.sLongName,));
572	else:
573	self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
574	return True;
575
576	#
577	# Tools for assiting log parsing.
578	#
579
580	@staticmethod
581	def matchFollowedByLines(sStr, off, asFollowingLines):
582	""" Worker for isThisFollowedByTheseLines. """
583
584	# Advance off to the end of the line.
585	off = sStr.find('\n', off);
586	if off < 0:
587	return False;
588	off += 1;
589
590	# Match each string with the subsequent lines.
591	for iLine, sLine in enumerate(asFollowingLines):
592	offEnd = sStr.find('\n', off);
593	if offEnd < 0:
594	return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
595	if sLine and sStr.find(sLine, off, offEnd) < 0:
596	return False;
597
598	# next line.
599	off = offEnd + 1;
600
601	return True;
602
603	@staticmethod
604	def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
605	"""
606	Looks for a line contining sFirst which is then followed by lines
607	with the strings in asFollowingLines. (No newline chars anywhere!)
608	Returns True / False.
609	"""
610	off = sStr.find(sFirst, 0);
611	while off >= 0:
612	if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
613	return True;
614	off = sStr.find(sFirst, off + 1);
615	return False;
616
617	@staticmethod
618	def findAndReturnResetOfLine(sHaystack, sNeedle):
619	"""
620	Looks for sNeedle in sHaystack.
621	Returns The text following the needle up to the end of the line.
622	Returns None if not found.
623	"""
624	if sHaystack is None:
625	return None;
626	off = sHaystack.find(sNeedle);
627	if off < 0:
628	return None;
629	off += len(sNeedle)
630	offEol = sHaystack.find('\n', off);
631	if offEol < 0:
632	offEol = len(sHaystack);
633	return sHaystack[off:offEol]
634
635	@staticmethod
636	def findInAnyAndReturnResetOfLine(asHaystacks, sNeedle):
637	"""
638	Looks for sNeedle in zeroe or more haystacks (asHaystack).
639	Returns The text following the first needed found up to the end of the line.
640	Returns None if not found.
641	"""
642	for sHaystack in asHaystacks:
643	sRet = VirtualTestSheriff.findAndReturnResetOfLine(sHaystack, sNeedle);
644	if sRet is not None:
645	return sRet;
646	return None;
647
648
649	#
650	# The investigative units.
651	#
652
653	katSimpleInstallUninstallMainLogReasons = [
654	# ( Whether to stop on hit, reason tuple, needle text. )
655	( False, ktReason_Host_LeftoverService,
656	'SERVICE_NAME: vbox' ),
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	( True, ktReason_Host_Modprobe_Failed, 'Kernel driver not installed' )
843	];
844
845	## Things we search a VBoxHardening.log file for to figure out why something went bust.
846	katSimpleVBoxHardeningLogReasons = [
847	# ( Whether to stop on hit, reason tuple, needle text. )
848	( True, ktReason_Host_DriverNotLoaded, 'Error opening VBoxDrvStub: STATUS_OBJECT_NAME_NOT_FOUND' ),
849	( True, ktReason_Host_NotSignedWithBuildCert, 'Not signed with the build certificate' ),
850	];
851
852	## Things we search a kernel.log file for to figure out why something went bust.
853	katSimpleKernelLogReasons = [
854	# ( Whether to stop on hit, reason tuple, needle text. )
855	( True, ktReason_Panic_HugeMemory, 'mm/huge_memory.c:1988' ),
856	( True, ktReason_Panic_IOAPICDoesntWork, 'IO-APIC + timer doesn''t work' ),
857	( True, ktReason_Panic_TxUnitHang, 'Detected Tx Unit Hang' ),
858	];
859
860	## Things we search the _RIGHT_ _STRIPPED_ vgatext for.
861	katSimpleVgaTextReasons = [
862	# ( Whether to stop on hit, reason tuple, needle text. )
863	( True, ktReason_Panic_MP_BIOS_IO_APIC,
864	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n\n" ),
865	( True, ktReason_Panic_MP_BIOS_IO_APIC,
866	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n"
867	"...trying to set up timer (IRQ0) through the 8259A ... failed.\n"
868	"...trying to set up timer as Virtual Wire IRQ... failed.\n"
869	"...trying to set up timer as ExtINT IRQ... failed :(.\n"
870	"Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug\n"
871	"and send a report. Then try booting with the 'noapic' option\n"
872	"\n" ),
873	( True, ktReason_OSInstall_GRUB_hang,
874	"-----\nGRUB Loading stage2..\n\n\n\n" ),
875	( True, ktReason_Panic_BootManagerC000000F,
876	"Windows failed to start. A recent hardware or software change might be the" ),
877	( True, ktReason_BootManager_Image_corrupt,
878	"BOOTMGR image is corrupt. The system cannot boot." ),
879	];
880
881	## Things we search for in the info.txt file. Require handlers for now.
882	katInfoTextHandlers = [
883	# ( Trigger text, handler method )
884	( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
885	];
886
887	## Mapping screenshot/failure SHA-256 hashes to failure reasons.
888	katSimpleScreenshotHashReasons = [
889	# ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
890	( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
891	( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
892	( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
893	( True, ktReason_BSOD_000000D1, '134621281f00a3f8aeeb7660064bffbf6187ed56d5852142328d0bcb18ef0ede' ),
894	( True, ktReason_BSOD_000000D1, '279f11258150c9d2fef041eca65501f3141da8df39256d8f6377e897e3b45a93' ),
895	( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
896	( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
897	( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
898	( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
899	( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
900	];
901
902	def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
903	"""
904	Investigates a failed VM run.
905	"""
906
907	def investigateLogSet():
908	"""
909	Investigates the current set of VM related logs.
910	"""
911	self.dprint('investigateLogSet: lengths: result log %u, VM log %u, kernel log %u, vga text %u, info text %u'
912	% ( len(sResultLog if sResultLog else ''),
913	len(sVMLog if sVMLog else ''),
914	len(sKrnlLog if sKrnlLog else ''),
915	len(sVgaText if sVgaText else ''),
916	len(sInfoText if sInfoText else ''), ));
917
918	#self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
919	#self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
920	#self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
921	#self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
922	#self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
923
924	# TODO: more
925
926	#
927	# Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
928	#
929	sDetails = self.findInAnyAndReturnResetOfLine([ sVMLog, sResultLog ],
930	'GIM: HyperV: Guest indicates a fatal condition! P0=');
931	if sDetails is not None:
932	# P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
933	sKey = sDetails.split(' ', 1)[0];
934	try: sKey = '0x%08X' % (int(sKey, 16),);
935	except: pass;
936	if sKey in self.asBsodReasons:
937	tReason = ( self.ksBsodCategory, sKey );
938	elif sKey.lower() in self.asBsodReasons: # just in case.
939	tReason = ( self.ksBsodCategory, sKey.lower() );
940	else:
941	self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
942	tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
943	return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
944
945	#
946	# Look for linux panic.
947	#
948	if sKrnlLog is not None:
949	for fStopOnHit, tReason, sNeedle in self.katSimpleKernelLogReasons:
950	if sKrnlLog.find(sNeedle) > 0:
951	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
952	if fStopOnHit:
953	return True;
954	fFoundSomething = True;
955
956	#
957	# Loop thru the simple stuff.
958	#
959	fFoundSomething = False;
960	for fStopOnHit, tReason, sNeedle in self.katSimpleMainAndVmLogReasons:
961	if sResultLog.find(sNeedle) > 0 or (sVMLog is not None and sVMLog.find(sNeedle) > 0):
962	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
963	if fStopOnHit:
964	return True;
965	fFoundSomething = True;
966
967	# Continue with vga text.
968	if sVgaText:
969	for fStopOnHit, tReason, sNeedle in self.katSimpleVgaTextReasons:
970	if sVgaText.find(sNeedle) > 0:
971	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
972	if fStopOnHit:
973	return True;
974	fFoundSomething = True;
975	_ = sInfoText;
976
977	# Continue with screen hashes.
978	if sScreenHash is not None:
979	for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
980	if sScreenHash == sHash:
981	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
982	if fStopOnHit:
983	return True;
984	fFoundSomething = True;
985
986	# Check VBoxHardening.log.
987	if sNtHardLog is not None:
988	for fStopOnHit, tReason, sNeedle in self.katSimpleVBoxHardeningLogReasons:
989	if sNtHardLog.find(sNeedle) > 0:
990	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
991	if fStopOnHit:
992	return True;
993	fFoundSomething = True;
994
995	#
996	# Complicated stuff.
997	#
998	dLogs = {
999	'sVMLog': sVMLog,
1000	'sNtHardLog': sNtHardLog,
1001	'sScreenHash': sScreenHash,
1002	'sKrnlLog': sKrnlLog,
1003	'sVgaText': sVgaText,
1004	'sInfoText': sInfoText,
1005	};
1006
1007	# info.txt.
1008	if sInfoText:
1009	for sNeedle, fnHandler in self.katInfoTextHandlers:
1010	if sInfoText.find(sNeedle) > 0:
1011	(fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
1012	if tReason is not None:
1013	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1014	if fStop:
1015	return True;
1016	fFoundSomething = True;
1017
1018	#
1019	# Check for repeated reboots...
1020	#
1021	if sVMLog is not None:
1022	cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1023	if cResets > 10:
1024	return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1025	sComment = 'Counted %s reboots' % (cResets,));
1026
1027	return fFoundSomething;
1028
1029	#
1030	# Check if we got any VM or/and kernel logs. Treat them as sets in
1031	# case we run multiple VMs here (this is of course ASSUMING they
1032	# appear in the order that terminateVmBySession uploads them).
1033	#
1034	cTimes = 0;
1035	sVMLog = None;
1036	sNtHardLog = None;
1037	sScreenHash = None;
1038	sKrnlLog = None;
1039	sVgaText = None;
1040	sInfoText = None;
1041	for oFile in oFailedResult.aoFiles:
1042	if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1043	if 'VBoxHardening.log' not in oFile.sFile:
1044	if sVMLog is not None:
1045	if investigateLogSet() is True:
1046	return True;
1047	cTimes += 1;
1048	sInfoText = None;
1049	sVgaText = None;
1050	sKrnlLog = None;
1051	sScreenHash = None;
1052	sNtHardLog = None;
1053	sVMLog = oCaseFile.getLogFile(oFile);
1054	else:
1055	sNtHardLog = oCaseFile.getLogFile(oFile);
1056	elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1057	sKrnlLog = oCaseFile.getLogFile(oFile);
1058	elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1059	sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1060	elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1061	sInfoText = oCaseFile.getLogFile(oFile);
1062	elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1063	sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1064	if sScreenHash is not None:
1065	sScreenHash = sScreenHash.lower();
1066	self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1067
1068	if ( sVMLog is not None \
1069	or sNtHardLog is not None \
1070	or cTimes == 0) \
1071	and investigateLogSet() is True:
1072	return True;
1073
1074	return None;
1075
1076
1077	def isResultFromVMRun(self, oFailedResult, sResultLog):
1078	"""
1079	Checks if this result and corresponding log snippet looks like a VM run.
1080	"""
1081
1082	# Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1083	if sResultLog.find(' startVm') > 0:
1084	return True;
1085
1086	# Any other indicators? No?
1087	_ = oFailedResult;
1088	return False;
1089
1090	def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1091	"""
1092	Checks out a VBox VM test.
1093
1094	This is generic investigation of a test running one or more VMs, like
1095	for example a smoke test or a guest installation test.
1096
1097	The fSingleVM parameter is a hint, which probably won't come in useful.
1098	"""
1099	_ = fSingleVM;
1100
1101	#
1102	# Get a list of test result failures we should be looking into and the main log.
1103	#
1104	aoFailedResults = oCaseFile.oTree.getListOfFailures();
1105	sMainLog = oCaseFile.getMainLog();
1106
1107	#
1108	# There are a set of errors ending up on the top level result record.
1109	# Should deal with these first.
1110	#
1111	if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1112	# Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1113	# be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1114	# ignore other failures in the test if we're not a little bit careful here.
1115	if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1116	oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1117	return self.caseClosed(oCaseFile);
1118
1119	# Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1120	if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1121	(' Timeout: ', ' Attempting to abort child...',) ):
1122	if sMainLog.find('* glibc detected * /') > 0:
1123	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1124	else:
1125	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1126	return self.caseClosed(oCaseFile);
1127
1128	# Look for heap corruption without visible hang.
1129	if sMainLog.find('* glibc detected * /') > 0 \
1130	or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1131	oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1132	return self.caseClosed(oCaseFile);
1133
1134	# Out of memory w/ timeout.
1135	if sMainLog.find('sErrId=HostMemoryLow') > 0:
1136	oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1137	return self.caseClosed(oCaseFile);
1138
1139	# Stale files like vts_rm.exe (windows).
1140	offEnd = sMainLog.rfind('* The test driver exits successfully. *');
1141	if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1142	oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1143	return self.caseClosed(oCaseFile);
1144
1145	#
1146	# XPCOM screwup
1147	#
1148	if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1149	oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1150	return self.caseClosed(oCaseFile);
1151
1152	#
1153	# Go thru each failed result.
1154	#
1155	for oFailedResult in aoFailedResults:
1156	self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1157	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1158	if oFailedResult.sName == 'Installing VirtualBox':
1159	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1160
1161	elif oFailedResult.sName == 'Uninstalling VirtualBox':
1162	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1163
1164	elif self.isResultFromVMRun(oFailedResult, sResultLog):
1165	self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1166
1167	elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1168	oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1169
1170	elif sResultLog.find('The machine is not mutable (state is ') > 0:
1171	self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1172	oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1173
1174	elif sResultLog.find('** error: no action was specified') > 0 \
1175	or sResultLog.find('(len(self._asXml, asText))') > 0:
1176	oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1177
1178	else:
1179	self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1180	self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1181
1182	#
1183	# Report home and close the case if we got them all, otherwise log it.
1184	#
1185	if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1186	return self.caseClosed(oCaseFile);
1187
1188	if oCaseFile.dReasonForResultId:
1189	self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1190	% (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1191	else:
1192	self.vprint(u'XXX: Could not figure out anything at all! :-(');
1193	return False;
1194
1195
1196	def reasoningFailures(self):
1197	"""
1198	Guess the reason for failures.
1199	"""
1200	#
1201	# Get a list of failed test sets without any assigned failure reason.
1202	#
1203	cGot = 0;
1204	aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack, tsNow = self.tsNow);
1205	for oTestSet in aoTestSets:
1206	self.dprint(u'');
1207	self.dprint(u'reasoningFailures: Checking out test set #%u, status %s' % ( oTestSet.idTestSet, oTestSet.enmStatus,))
1208
1209	#
1210	# Open a case file and assign it to the right investigator.
1211	#
1212	(oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1213	oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1214	oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1215	oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1216	oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1217
1218	oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1219
1220	if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1221	self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1222	fRc = self.investigateBadTestBox(oCaseFile);
1223
1224	elif oCaseFile.isVBoxUnitTest():
1225	self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1226	fRc = self.investigateVBoxUnitTest(oCaseFile);
1227
1228	elif oCaseFile.isVBoxInstallTest():
1229	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1230	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1231
1232	elif oCaseFile.isVBoxUSBTest():
1233	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1234	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1235
1236	elif oCaseFile.isVBoxStorageTest():
1237	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1238	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1239
1240	elif oCaseFile.isVBoxGAsTest():
1241	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1242	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1243
1244	elif oCaseFile.isVBoxAPITest():
1245	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1246	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1247
1248	elif oCaseFile.isVBoxBenchmarkTest():
1249	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1250	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1251
1252	elif oCaseFile.isVBoxSmokeTest():
1253	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1254	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1255
1256	else:
1257	self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1258	fRc = False;
1259	cGot += fRc is True;
1260
1261	self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1262	return 0;
1263
1264
1265	def main(self):
1266	"""
1267	The 'main' function.
1268	Return exit code (0, 1, etc).
1269	"""
1270	# Database stuff.
1271	self.oDb = TMDatabaseConnection()
1272	self.oTestResultLogic = TestResultLogic(self.oDb);
1273	self.oTestSetLogic = TestSetLogic(self.oDb);
1274	self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1275	self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1276	self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1277	self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1278
1279	# Get a fix on our 'now' before we do anything..
1280	self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1281	self.tsNow = self.oDb.fetchOne();
1282
1283	# If we're suppost to commit anything we need to get our user ID.
1284	rcExit = 0;
1285	if self.oConfig.fRealRun:
1286	self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1287	if self.oLogin is None:
1288	rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1289	else:
1290	self.uidSelf = self.oLogin.uid;
1291
1292	#
1293	# Do the stuff.
1294	#
1295	if rcExit == 0:
1296	rcExit = self.selfCheck();
1297	if rcExit == 0:
1298	rcExit = self.badTestBoxManagement();
1299	rcExit2 = self.reasoningFailures();
1300	if rcExit == 0:
1301	rcExit = rcExit2;
1302	# Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1303	if rcExit == 0:
1304	rcExit = self.badTestBoxManagement();
1305
1306	# Cleanup.
1307	self.oFailureReasonLogic = None;
1308	self.oTestResultFailureLogic = None;
1309	self.oTestSetLogic = None;
1310	self.oTestResultLogic = None;
1311	self.oDb.close();
1312	self.oDb = None;
1313	if self.oLogFile is not None:
1314	self.oLogFile.close();
1315	self.oLogFile = None;
1316	return rcExit;
1317
1318	if __name__ == '__main__':
1319	sys.exit(VirtualTestSheriff().main());
1320

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

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 66817

Download in other formats: