virtual_test_sheriff.py@ 71088

Last change on this file since 71088 was 71088, checked in by vboxsync, 7 years ago
ValidationKit: Move the non-unique IP detection to the right place in vsheriff. This doesn't get close to a VM run.
Property svn:eol-style set to `LF` Property svn:executable set to ``* Property svn:keywords set to `Author Date Id Revision`
File size: 64.8 KB

Line
1	#!/usr/bin/env python
2	# -- coding: utf-8 --
3	# $Id: virtual_test_sheriff.py 71088 2018-02-21 16:37:40Z 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	from __future__ import print_function;
16
17	__copyright__ = \
18	"""
19	Copyright (C) 2012-2017 Oracle Corporation
20
21	This file is part of VirtualBox Open Source Edition (OSE), as
22	available from http://www.virtualbox.org. This file is free software;
23	you can redistribute it and/or modify it under the terms of the GNU
24	General Public License (GPL) as published by the Free Software
25	Foundation, in version 2 as it comes in the "COPYING" file of the
26	VirtualBox OSE distribution. VirtualBox OSE is distributed in the
27	hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
28
29	The contents of this file may alternatively be used under the terms
30	of the Common Development and Distribution License Version 1.0
31	(CDDL) only, as it comes in the "COPYING.CDDL" file of the
32	VirtualBox OSE distribution, in which case the provisions of the
33	CDDL are applicable instead of those of the GPL.
34
35	You may elect to license modified versions of this file under the
36	terms and conditions of either the GPL or the CDDL or both.
37	"""
38	__version__ = "$Revision: 71088 $"
39
40
41	# Standard python imports
42	import sys;
43	import os;
44	import hashlib;
45	if sys.version_info[0] >= 3:
46	from io import StringIO as StringIO; # pylint: disable=import-error,no-name-in-module
47	else:
48	from StringIO import StringIO as StringIO; # pylint: disable=import-error,no-name-in-module
49	from optparse import OptionParser; # pylint: disable=deprecated-module
50	from PIL import Image; # pylint: disable=import-error
51
52	# Add Test Manager's modules path
53	g_ksTestManagerDir = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))));
54	sys.path.append(g_ksTestManagerDir);
55
56	# Test Manager imports
57	from testmanager.core.db import TMDatabaseConnection;
58	from testmanager.core.build import BuildDataEx;
59	from testmanager.core.failurereason import FailureReasonLogic;
60	from testmanager.core.testbox import TestBoxLogic, TestBoxData;
61	from testmanager.core.testcase import TestCaseDataEx;
62	from testmanager.core.testgroup import TestGroupData;
63	from testmanager.core.testset import TestSetLogic, TestSetData;
64	from testmanager.core.testresults import TestResultLogic, TestResultFileData;
65	from testmanager.core.testresultfailures import TestResultFailureLogic, TestResultFailureData;
66	from testmanager.core.useraccount import UserAccountLogic;
67
68	# Python 3 hacks:
69	if sys.version_info[0] >= 3:
70	xrange = range; # pylint: disable=redefined-builtin,invalid-name
71
72
73	class VirtualTestSheriffCaseFile(object):
74	"""
75	A failure investigation case file.
76
77	"""
78
79
80	## Max log file we'll read into memory. (256 MB)
81	kcbMaxLogRead = 0x10000000;
82
83	def __init__(self, oSheriff, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase):
84	self.oSheriff = oSheriff;
85	self.oTestSet = oTestSet; # TestSetData
86	self.oTree = oTree; # TestResultDataEx
87	self.oBuild = oBuild; # BuildDataEx
88	self.oTestBox = oTestBox; # TestBoxData
89	self.oTestGroup = oTestGroup; # TestGroupData
90	self.oTestCase = oTestCase; # TestCaseDataEx
91	self.sMainLog = ''; # The main log file. Empty string if not accessible.
92
93	# Generate a case file name.
94	self.sName = '#%u: %s' % (self.oTestSet.idTestSet, self.oTestCase.sName,)
95	self.sLongName = '#%u: "%s" on "%s" running %s %s (%s), "%s" by %s, using %s %s %s r%u' \
96	% ( self.oTestSet.idTestSet,
97	self.oTestCase.sName,
98	self.oTestBox.sName,
99	self.oTestBox.sOs,
100	self.oTestBox.sOsVersion,
101	self.oTestBox.sCpuArch,
102	self.oTestBox.sCpuName,
103	self.oTestBox.sCpuVendor,
104	self.oBuild.oCat.sProduct,
105	self.oBuild.oCat.sBranch,
106	self.oBuild.oCat.sType,
107	self.oBuild.iRevision, );
108
109	# Investigation notes.
110	self.tReason = None; # None or one of the ktReason_XXX constants.
111	self.dReasonForResultId = {}; # Reason assignments indexed by idTestResult.
112	self.dCommentForResultId = {}; # Comment assignments indexed by idTestResult.
113
114	#
115	# Reason.
116	#
117
118	def noteReason(self, tReason):
119	""" Notes down a possible reason. """
120	self.oSheriff.dprint(u'noteReason: %s -> %s' % (self.tReason, tReason,));
121	self.tReason = tReason;
122	return True;
123
124	def noteReasonForId(self, tReason, idTestResult, sComment = None):
125	""" Notes down a possible reason for a specific test result. """
126	self.oSheriff.dprint(u'noteReasonForId: %u: %s -> %s%s'
127	% (idTestResult, self.dReasonForResultId.get(idTestResult, None), tReason,
128	(u' (%s)' % (sComment,)) if sComment is not None else ''));
129	self.dReasonForResultId[idTestResult] = tReason;
130	if sComment is not None:
131	self.dCommentForResultId[idTestResult] = sComment;
132	return True;
133
134
135	#
136	# Test classification.
137	#
138
139	def isVBoxTest(self):
140	""" Test classification: VirtualBox (using the build) """
141	return self.oBuild.oCat.sProduct.lower() in [ 'virtualbox', 'vbox' ];
142
143	def isVBoxUnitTest(self):
144	""" Test case classification: The unit test doing all our testcase/*.cpp stuff. """
145	return self.isVBoxTest() \
146	and (self.oTestCase.sName.lower() == 'unit tests' or self.oTestCase.sName.lower() == 'misc: unit tests');
147
148	def isVBoxInstallTest(self):
149	""" Test case classification: VirtualBox Guest installation test. """
150	return self.isVBoxTest() \
151	and self.oTestCase.sName.lower().startswith('install:');
152
153	def isVBoxUSBTest(self):
154	""" Test case classification: VirtualBox USB test. """
155	return self.isVBoxTest() \
156	and self.oTestCase.sName.lower().startswith('usb:');
157
158	def isVBoxStorageTest(self):
159	""" Test case classification: VirtualBox Storage test. """
160	return self.isVBoxTest() \
161	and self.oTestCase.sName.lower().startswith('storage:');
162
163	def isVBoxGAsTest(self):
164	""" Test case classification: VirtualBox Guest Additions test. """
165	return self.isVBoxTest() \
166	and self.oTestCase.sName.lower().startswith('ga\'s tests');
167
168	def isVBoxAPITest(self):
169	""" Test case classification: VirtualBox API test. """
170	return self.isVBoxTest() \
171	and self.oTestCase.sName.lower().startswith('api:');
172
173	def isVBoxBenchmarkTest(self):
174	""" Test case classification: VirtualBox Benchmark test. """
175	return self.isVBoxTest() \
176	and self.oTestCase.sName.lower().startswith('benchmark:');
177
178	def isVBoxSmokeTest(self):
179	""" Test case classification: Smoke test. """
180	return self.isVBoxTest() \
181	and self.oTestCase.sName.lower().startswith('smoketest');
182
183
184	#
185	# Utility methods.
186	#
187
188	def getMainLog(self):
189	"""
190	Tries to read the main log file since this will be the first source of information.
191	"""
192	if self.sMainLog:
193	return self.sMainLog;
194	(oFile, oSizeOrError, _) = self.oTestSet.openFile('main.log', 'rb');
195	if oFile is not None:
196	try:
197	self.sMainLog = oFile.read(min(self.kcbMaxLogRead, oSizeOrError)).decode('utf-8', 'replace');
198	except Exception as oXcpt:
199	self.oSheriff.vprint(u'Error reading main log file: %s' % (oXcpt,))
200	self.sMainLog = '';
201	else:
202	self.oSheriff.vprint(u'Error opening main log file: %s' % (oSizeOrError,));
203	return self.sMainLog;
204
205	def getLogFile(self, oFile):
206	"""
207	Tries to read the given file as a utf-8 log file.
208	oFile is a TestFileDataEx instance.
209	Returns empty string if problems opening or reading the file.
210	"""
211	sContent = '';
212	(oFile, oSizeOrError, _) = self.oTestSet.openFile(oFile.sFile, 'rb');
213	if oFile is not None:
214	try:
215	sContent = oFile.read(min(self.kcbMaxLogRead, oSizeOrError)).decode('utf-8', 'replace');
216	except Exception as oXcpt:
217	self.oSheriff.vprint(u'Error reading the "%s" log file: %s' % (oFile.sFile, oXcpt,))
218	else:
219	self.oSheriff.vprint(u'Error opening the "%s" log file: %s' % (oFile.sFile, oSizeOrError,));
220	return sContent;
221
222	def getScreenshotSha256(self, oFile):
223	"""
224	Tries to read the given screenshot file, uncompress it, and do SHA-2
225	on the raw pixels.
226	Returns SHA-2 digest string on success, None on failure.
227	"""
228	(oFile, _, _) = self.oTestSet.openFile(oFile.sFile, 'rb');
229	try:
230	abImageFile = oFile.read();
231	except Exception as oXcpt:
232	self.oSheriff.vprint(u'Error reading the "%s" image file: %s' % (oFile.sFile, oXcpt,))
233	else:
234	try:
235	oImage = Image.open(StringIO(abImageFile));
236	except Exception as oXcpt:
237	self.oSheriff.vprint(u'Error opening the "%s" image bytes using PIL.Image.open: %s' % (oFile.sFile, oXcpt,))
238	else:
239	try:
240	oHash = hashlib.sha256();
241	oHash.update(oImage.tostring());
242	except Exception as oXcpt:
243	self.oSheriff.vprint(u'Error hashing the uncompressed image bytes for "%s": %s' % (oFile.sFile, oXcpt,))
244	else:
245	return oHash.hexdigest();
246	return None;
247
248
249
250	def isSingleTestFailure(self):
251	"""
252	Figure out if this is a single test failing or if it's one of the
253	more complicated ones.
254	"""
255	if self.oTree.cErrors == 1:
256	return True;
257	if self.oTree.deepCountErrorContributers() <= 1:
258	return True;
259	return False;
260
261
262
263	class VirtualTestSheriff(object): # pylint: disable=R0903
264	"""
265	Add build info into Test Manager database.
266	"""
267
268	## The user account for the virtual sheriff.
269	ksLoginName = 'vsheriff';
270
271	def __init__(self):
272	"""
273	Parse command line.
274	"""
275	self.oDb = None;
276	self.tsNow = None;
277	self.oTestResultLogic = None;
278	self.oTestSetLogic = None;
279	self.oFailureReasonLogic = None; # FailureReasonLogic;
280	self.oTestResultFailureLogic = None; # TestResultFailureLogic
281	self.oLogin = None;
282	self.uidSelf = -1;
283	self.oLogFile = None;
284	self.asBsodReasons = [];
285	self.asUnitTestReasons = [];
286
287	oParser = OptionParser();
288	oParser.add_option('--start-hours-ago', dest = 'cStartHoursAgo', metavar = '<hours>', default = 0, type = 'int',
289	help = 'When to start specified as hours relative to current time. Defauls is right now.', );
290	oParser.add_option('--hours-period', dest = 'cHoursBack', metavar = '<period-in-hours>', default = 2, type = 'int',
291	help = 'Work period specified in hours. Defauls is 2 hours.');
292	oParser.add_option('--real-run-back', dest = 'fRealRun', action = 'store_true', default = False,
293	help = 'Whether to commit the findings to the database. Default is a dry run.');
294	oParser.add_option('-q', '--quiet', dest = 'fQuiet', action = 'store_true', default = False,
295	help = 'Quiet execution');
296	oParser.add_option('-l', '--log', dest = 'sLogFile', metavar = '<logfile>', default = None,
297	help = 'Where to log messages.');
298	oParser.add_option('--debug', dest = 'fDebug', action = 'store_true', default = False,
299	help = 'Enables debug mode.');
300
301	(self.oConfig, _) = oParser.parse_args();
302
303	if self.oConfig.sLogFile:
304	self.oLogFile = open(self.oConfig.sLogFile, "a");
305	self.oLogFile.write('VirtualTestSheriff: $Revision: 71088 $ \n');
306
307
308	def eprint(self, sText):
309	"""
310	Prints error messages.
311	Returns 1 (for exit code usage.)
312	"""
313	print('error: %s' % (sText,));
314	if self.oLogFile is not None:
315	self.oLogFile.write((u'error: %s\n' % (sText,)).encode('utf-8'));
316	return 1;
317
318	def dprint(self, sText):
319	"""
320	Prints debug info.
321	"""
322	if self.oConfig.fDebug:
323	if not self.oConfig.fQuiet:
324	print('debug: %s' % (sText, ));
325	if self.oLogFile is not None:
326	self.oLogFile.write((u'debug: %s\n' % (sText,)).encode('utf-8'));
327	return 0;
328
329	def vprint(self, sText):
330	"""
331	Prints verbose info.
332	"""
333	if not self.oConfig.fQuiet:
334	print('info: %s' % (sText,));
335	if self.oLogFile is not None:
336	self.oLogFile.write((u'info: %s\n' % (sText,)).encode('utf-8'));
337	return 0;
338
339	def getFailureReason(self, tReason):
340	""" Gets the failure reason object for tReason. """
341	return self.oFailureReasonLogic.cachedLookupByNameAndCategory(tReason[1], tReason[0]);
342
343	def selfCheck(self):
344	""" Does some self checks, looking up things we expect to be in the database and such. """
345	rcExit = 0;
346	for sAttr in dir(self.__class__):
347	if sAttr.startswith('ktReason_'):
348	tReason = getattr(self.__class__, sAttr);
349	oFailureReason = self.getFailureReason(tReason);
350	if oFailureReason is None:
351	rcExit = self.eprint(u'Failed to find failure reason "%s" in category "%s" in the database!'
352	% (tReason[1], tReason[0],));
353
354	# Check the user account as well.
355	if self.oLogin is None:
356	oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
357	if oLogin is None:
358	rcExit = self.eprint(u'Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
359	return rcExit;
360
361
362
363	def badTestBoxManagement(self):
364	"""
365	Looks for bad test boxes and first tries once to reboot them then disables them.
366	"""
367	rcExit = 0;
368
369	#
370	# We skip this entirely if we're running in the past and not in harmless debug mode.
371	#
372	if self.oConfig.cStartHoursAgo != 0 \
373	and (not self.oConfig.fDebug or self.oConfig.fRealRun):
374	return rcExit;
375	tsNow = self.tsNow if self.oConfig.fDebug else None;
376	cHoursBack = self.oConfig.cHoursBack if self.oConfig.fDebug else 2;
377	oTestBoxLogic = TestBoxLogic(self.oDb);
378
379	#
380	# Generate a list of failures reasons we consider bad-testbox behavior.
381	#
382	aidFailureReasons = [
383	self.getFailureReason(self.ktReason_Host_DriverNotUnloading).idFailureReason,
384	self.getFailureReason(self.ktReason_Host_DriverNotCompilable).idFailureReason,
385	self.getFailureReason(self.ktReason_Host_InstallationFailed).idFailureReason,
386	];
387
388	#
389	# Get list of bad test boxes for given period and check them out individually.
390	#
391	aidBadTestBoxes = self.oTestSetLogic.fetchBadTestBoxIds(cHoursBack = cHoursBack, tsNow = tsNow,
392	aidFailureReasons = aidFailureReasons);
393	for idTestBox in aidBadTestBoxes:
394	# Skip if the testbox is already disabled or has a pending reboot command.
395	try:
396	oTestBox = TestBoxData().initFromDbWithId(self.oDb, idTestBox);
397	except Exception as oXcpt:
398	rcExit = self.eprint('Failed to get data for test box #%u in badTestBoxManagement: %s' % (idTestBox, oXcpt,));
399	continue;
400	if not oTestBox.fEnabled:
401	self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has been disabled already.'
402	% ( idTestBox, oTestBox.sName, ));
403	continue;
404	if oTestBox.enmPendingCmd != TestBoxData.ksTestBoxCmd_None:
405	self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has a command pending: %s'
406	% ( idTestBox, oTestBox.sName, oTestBox.enmPendingCmd));
407	continue;
408
409	# Get the most recent testsets for this box (descending on tsDone) and see how bad it is.
410	aoSets = self.oTestSetLogic.fetchSetsForTestBox(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow);
411	cOkay = 0;
412	cBad = 0;
413	iFirstOkay = len(aoSets);
414	for iSet, oSet in enumerate(aoSets):
415	if oSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
416	cBad += 1;
417	else:
418	# Check for bad failure reasons.
419	oFailure = None;
420	if oSet.enmStatus in TestSetData.kasBadTestStatuses:
421	oFailure = self.oTestResultFailureLogic.getById(oSet.idTestResult);
422	if oFailure is not None and oFailure.idFailureReason in aidFailureReasons:
423	cBad += 1;
424	else:
425	# This is an okay test result then.
426	## @todo maybe check the elapsed time here, it could still be a bad run?
427	cOkay += 1;
428	if iFirstOkay > iSet:
429	iFirstOkay = iSet;
430	if iSet > 10:
431	break;
432
433	# We react if there are two or more bad-testbox statuses at the head of the
434	# history and at least three in the last 10 results.
435	if iFirstOkay >= 2 and cBad > 2:
436	# Frank: For now don't reboot boxes automatically
437	if True or oTestBoxLogic.hasTestBoxRecentlyBeenRebooted(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow):
438	self.vprint(u'Disabling testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
439	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
440	if self.oConfig.fRealRun is True:
441	try:
442	oTestBoxLogic.disableTestBox(idTestBox, self.uidSelf, fCommit = True,
443	sComment = 'Automatically disabled (iFirstOkay=%u cBad=%u cOkay=%u)'
444	% (iFirstOkay, cBad, cOkay),);
445	except Exception as oXcpt:
446	rcExit = self.eprint(u'Error disabling testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
447	else:
448	self.vprint(u'Rebooting testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
449	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
450	if self.oConfig.fRealRun is True:
451	try:
452	oTestBoxLogic.rebootTestBox(idTestBox, self.uidSelf, fCommit = True,
453	sComment = 'Automatically rebooted (iFirstOkay=%u cBad=%u cOkay=%u)'
454	% (iFirstOkay, cBad, cOkay),);
455	except Exception as oXcpt:
456	rcExit = self.eprint(u'Error rebooting testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
457	else:
458	self.dprint(u'badTestBoxManagement: #%u (%s) looks ok: iFirstOkay=%u cBad=%u cOkay=%u'
459	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
460	return rcExit;
461
462
463	## @name Failure reasons we know.
464	## @{
465	ktReason_BSOD_Recovery = ( 'BSOD', 'Recovery' );
466	ktReason_BSOD_Automatic_Repair = ( 'BSOD', 'Automatic Repair' );
467	ktReason_BSOD_0000007F = ( 'BSOD', '0x0000007F' );
468	ktReason_BSOD_000000D1 = ( 'BSOD', '0x000000D1' );
469	ktReason_BSOD_C0000225 = ( 'BSOD', '0xC0000225 (boot)' );
470	ktReason_Guru_Generic = ( 'Guru Meditations', 'Generic Guru Meditation' );
471	ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_INSTR_NOT_IMPLEMENTED' );
472	ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' );
473	ktReason_Guru_VERR_TRPM_DONT_PANIC = ( 'Guru Meditations', 'VERR_TRPM_DONT_PANIC' );
474	ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED = ( 'Guru Meditations', 'VERR_PGM_PHYS_PAGE_RESERVED' );
475	ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE = ( 'Guru Meditations', 'VERR_VMX_INVALID_GUEST_STATE' );
476	ktReason_Guru_VINF_EM_TRIPLE_FAULT = ( 'Guru Meditations', 'VINF_EM_TRIPLE_FAULT' );
477	ktReason_Host_HostMemoryLow = ( 'Host', 'HostMemoryLow' );
478	ktReason_Host_DriverNotLoaded = ( 'Host', 'Driver not loaded' );
479	ktReason_Host_DriverNotUnloading = ( 'Host', 'Driver not unloading' );
480	ktReason_Host_DriverNotCompilable = ( 'Host', 'Driver not compilable' );
481	ktReason_Host_InstallationFailed = ( 'Host', 'Installation failed' );
482	ktReason_Host_NotSignedWithBuildCert = ( 'Host', 'Not signed with build cert' );
483	ktReason_Host_DoubleFreeHeap = ( 'Host', 'Double free or corruption' );
484	ktReason_Host_LeftoverService = ( 'Host', 'Leftover service' );
485	ktReason_Host_Reboot_OSX_Watchdog_Timeout = ( 'Host Reboot', 'OSX Watchdog Timeout' );
486	ktReason_Host_Modprobe_Failed = ( 'Host', 'Modprobe failed' );
487	ktReason_Host_Install_Hang = ( 'Host', 'Install hang' );
488	ktReason_Host_NetworkMisconfiguration = ( 'Host', 'Network misconfiguration' );
489	ktReason_Networking_Nonexistent_host_nic = ( 'Networking', 'Nonexistent host networking interface' );
490	ktReason_OSInstall_GRUB_hang = ( 'O/S Install', 'GRUB hang' );
491	ktReason_OSInstall_Sata_no_BM = ( 'O/S Install', 'SATA busmaster bit not set' );
492	ktReason_Panic_BootManagerC000000F = ( 'Panic', 'Hardware Changed' );
493	ktReason_BootManager_Image_corrupt = ( 'Unknown', 'BOOTMGR Image corrupt' );
494	ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
495	ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
496	ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
497	ktReason_Panic_TxUnitHang = ( 'Panic', 'Tx Unit Hang' );
498	ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
499	ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
500	ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
501	ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
502	ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
503	ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
504	ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
505	ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
506	ktReason_Unknown_HalReturnToFirmware = ( 'Unknown', 'HalReturnToFirmware' );
507	ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
508	ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
509	ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
510	ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
511	## @}
512
513	## BSOD category.
514	ksBsodCategory = 'BSOD';
515	## Special reason indicating that the flesh and blood sheriff has work to do.
516	ksBsodAddNew = 'Add new BSOD';
517
518	## Unit test category.
519	ksUnitTestCategory = 'Unit';
520	## Special reason indicating that the flesh and blood sheriff has work to do.
521	ksUnitTestAddNew = 'Add new';
522
523	## Used for indica that we shouldn't report anything for this test result ID and
524	## consider promoting the previous error to test set level if it's the only one.
525	ktHarmless = ( 'Probably', 'Caused by previous error' );
526
527
528	def caseClosed(self, oCaseFile):
529	"""
530	Reports the findings in the case and closes it.
531	"""
532	#
533	# Log it and create a dReasonForReasultId we can use below.
534	#
535	dCommentForResultId = oCaseFile.dCommentForResultId;
536	if oCaseFile.dReasonForResultId:
537	# Must weed out ktHarmless.
538	dReasonForResultId = {};
539	for idKey, tReason in oCaseFile.dReasonForResultId.items():
540	if tReason is not self.ktHarmless:
541	dReasonForResultId[idKey] = tReason;
542	if not dReasonForResultId:
543	self.vprint(u'TODO: Closing %s without a real reason, only %s.'
544	% (oCaseFile.sName, oCaseFile.dReasonForResultId));
545	return False;
546
547	# Try promote to single reason.
548	atValues = dReasonForResultId.values();
549	fSingleReason = True;
550	if len(dReasonForResultId) == 1 and dReasonForResultId.keys()[0] != oCaseFile.oTestSet.idTestResult:
551	self.dprint(u'Promoting single reason to whole set: %s' % (atValues[0],));
552	elif len(dReasonForResultId) > 1 and len(atValues) == atValues.count(atValues[0]):
553	self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), atValues[0]));
554	else:
555	fSingleReason = False;
556	if fSingleReason:
557	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: atValues[0], };
558	if dCommentForResultId:
559	dCommentForResultId = { oCaseFile.oTestSet.idTestResult: dCommentForResultId.values()[0], };
560	elif oCaseFile.tReason is not None:
561	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
562	else:
563	self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
564	return False;
565
566	self.vprint(u'Closing %s with following reason%s: %s'
567	% ( oCaseFile.sName, 's' if dReasonForResultId > 0 else '', dReasonForResultId, ));
568
569	#
570	# Add the test failure reason record(s).
571	#
572	for idTestResult, tReason in dReasonForResultId.items():
573	oFailureReason = self.getFailureReason(tReason);
574	if oFailureReason is not None:
575	sComment = 'Set by $Revision: 71088 $' # Handy for reverting later.
576	if idTestResult in dCommentForResultId:
577	sComment += ': ' + dCommentForResultId[idTestResult];
578
579	oAdd = TestResultFailureData();
580	oAdd.initFromValues(idTestResult = idTestResult,
581	idFailureReason = oFailureReason.idFailureReason,
582	uidAuthor = self.uidSelf,
583	idTestSet = oCaseFile.oTestSet.idTestSet,
584	sComment = sComment,);
585	if self.oConfig.fRealRun:
586	try:
587	self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
588	except Exception as oXcpt:
589	self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
590	% (oXcpt, oAdd, oCaseFile.sLongName,));
591	else:
592	self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
593	return True;
594
595	#
596	# Tools for assiting log parsing.
597	#
598
599	@staticmethod
600	def matchFollowedByLines(sStr, off, asFollowingLines):
601	""" Worker for isThisFollowedByTheseLines. """
602
603	# Advance off to the end of the line.
604	off = sStr.find('\n', off);
605	if off < 0:
606	return False;
607	off += 1;
608
609	# Match each string with the subsequent lines.
610	for iLine, sLine in enumerate(asFollowingLines):
611	offEnd = sStr.find('\n', off);
612	if offEnd < 0:
613	return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
614	if sLine and sStr.find(sLine, off, offEnd) < 0:
615	return False;
616
617	# next line.
618	off = offEnd + 1;
619
620	return True;
621
622	@staticmethod
623	def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
624	"""
625	Looks for a line contining sFirst which is then followed by lines
626	with the strings in asFollowingLines. (No newline chars anywhere!)
627	Returns True / False.
628	"""
629	off = sStr.find(sFirst, 0);
630	while off >= 0:
631	if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
632	return True;
633	off = sStr.find(sFirst, off + 1);
634	return False;
635
636	@staticmethod
637	def findAndReturnRestOfLine(sHaystack, sNeedle):
638	"""
639	Looks for sNeedle in sHaystack.
640	Returns The text following the needle up to the end of the line.
641	Returns None if not found.
642	"""
643	if sHaystack is None:
644	return None;
645	off = sHaystack.find(sNeedle);
646	if off < 0:
647	return None;
648	off += len(sNeedle)
649	offEol = sHaystack.find('\n', off);
650	if offEol < 0:
651	offEol = len(sHaystack);
652	return sHaystack[off:offEol]
653
654	@staticmethod
655	def findInAnyAndReturnRestOfLine(asHaystacks, sNeedle):
656	"""
657	Looks for sNeedle in zeroe or more haystacks (asHaystack).
658	Returns The text following the first needed found up to the end of the line.
659	Returns None if not found.
660	"""
661	for sHaystack in asHaystacks:
662	sRet = VirtualTestSheriff.findAndReturnRestOfLine(sHaystack, sNeedle);
663	if sRet is not None:
664	return sRet;
665	return None;
666
667
668	#
669	# The investigative units.
670	#
671
672	katSimpleInstallUninstallMainLogReasons = [
673	# ( Whether to stop on hit, reason tuple, needle text. )
674	( False, ktReason_Host_LeftoverService,
675	'SERVICE_NAME: vbox' ),
676	];
677
678	kdatSimpleInstallUninstallMainLogReasonsPerOs = {
679	'darwin': [
680	# ( Whether to stop on hit, reason tuple, needle text. )
681	( True, ktReason_Host_DriverNotUnloading,
682	'Can\'t remove kext org.virtualbox.kext.VBoxDrv; services failed to terminate - 0xe00002c7' ),
683	],
684	'linux': [
685	# ( Whether to stop on hit, reason tuple, needle text. )
686	( True, ktReason_Host_DriverNotCompilable,
687	'This system is not currently set up to build kernel modules' ),
688	( True, ktReason_Host_DriverNotCompilable,
689	'This system is currently not set up to build kernel modules' ),
690	( True, ktReason_Host_InstallationFailed,
691	'vboxdrv.sh: failed: Look at /var/log/vbox-install.log to find out what went wrong.' ),
692	( True, ktReason_Host_DriverNotUnloading,
693	'Cannot unload module vboxdrv'),
694	],
695	};
696
697
698	def investigateInstallUninstallFailure(self, oCaseFile, oFailedResult, sResultLog, fInstall):
699	"""
700	Investigates an install or uninstall failure.
701
702	We lump the two together since the installation typically also performs
703	an uninstall first and will be seeing similar issues to the uninstall.
704	"""
705
706	if fInstall and oFailedResult.enmStatus == TestSetData.ksTestStatus_TimedOut:
707	oCaseFile.noteReasonForId(self.ktReason_Host_Install_Hang, oFailedResult.idTestResult)
708	return True;
709
710	atSimple = self.katSimpleInstallUninstallMainLogReasons;
711	if oCaseFile.oTestBox.sOs in self.kdatSimpleInstallUninstallMainLogReasonsPerOs:
712	atSimple = self.kdatSimpleInstallUninstallMainLogReasonsPerOs[oCaseFile.oTestBox.sOs] + atSimple;
713
714	fFoundSomething = False;
715	for fStopOnHit, tReason, sNeedle in atSimple:
716	if sResultLog.find(sNeedle) > 0:
717	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
718	if fStopOnHit:
719	return True;
720	fFoundSomething = True;
721
722	return fFoundSomething if fFoundSomething else None;
723
724
725	def investigateBadTestBox(self, oCaseFile):
726	"""
727	Checks out bad-testbox statuses.
728	"""
729	_ = oCaseFile;
730	return False;
731
732
733	def investigateVBoxUnitTest(self, oCaseFile):
734	"""
735	Checks out a VBox unittest problem.
736	"""
737
738	#
739	# Process simple test case failures first, using their name as reason.
740	# We do the reason management just like for BSODs.
741	#
742	cRelevantOnes = 0;
743	sMainLog = oCaseFile.getMainLog();
744	aoFailedResults = oCaseFile.oTree.getListOfFailures();
745	for oFailedResult in aoFailedResults:
746	if oFailedResult is oCaseFile.oTree:
747	self.vprint('TODO: toplevel failure');
748	cRelevantOnes += 1
749
750	elif oFailedResult.sName == 'Installing VirtualBox':
751	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
752	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
753	cRelevantOnes += 1
754
755	elif oFailedResult.sName == 'Uninstalling VirtualBox':
756	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
757	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
758	cRelevantOnes += 1
759
760	elif oFailedResult.oParent is not None:
761	# Get the 2nd level node because that's where we'll find the unit test name.
762	while oFailedResult.oParent.oParent is not None:
763	oFailedResult = oFailedResult.oParent;
764
765	# Only report a failure once.
766	if oFailedResult.idTestResult not in oCaseFile.dReasonForResultId:
767	sKey = oFailedResult.sName;
768	if sKey.startswith('testcase/'):
769	sKey = sKey[9:];
770	if sKey in self.asUnitTestReasons:
771	tReason = ( self.ksUnitTestCategory, sKey );
772	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
773	else:
774	self.dprint(u'Unit test failure "%s" not found in %s;' % (sKey, self.asUnitTestReasons));
775	tReason = ( self.ksUnitTestCategory, self.ksUnitTestAddNew );
776	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sKey);
777	cRelevantOnes += 1
778	else:
779	self.vprint(u'Internal error: expected oParent to NOT be None for %s' % (oFailedResult,));
780
781	#
782	# If we've caught all the relevant ones by now, report the result.
783	#
784	if len(oCaseFile.dReasonForResultId) >= cRelevantOnes:
785	return self.caseClosed(oCaseFile);
786	return False;
787
788	def extractGuestCpuStack(self, sInfoText):
789	"""
790	Extracts the guest CPU stacks from the input file.
791
792	Returns a dictionary keyed by the CPU number, value being a list of
793	raw stack lines (no header).
794	Returns empty dictionary if no stacks where found.
795	"""
796	dRet = {};
797	off = 0;
798	while True:
799	# Find the stack.
800	offStart = sInfoText.find('=== start guest stack VCPU ', off);
801	if offStart < 0:
802	break;
803	offEnd = sInfoText.find('=== end guest stack', offStart + 20);
804	if offEnd >= 0:
805	offEnd += 3;
806	else:
807	offEnd = sInfoText.find('=== start guest stack VCPU', offStart + 20);
808	if offEnd < 0:
809	offEnd = len(sInfoText);
810
811	sStack = sInfoText[offStart : offEnd];
812	sStack = sStack.replace('\r',''); # paranoia
813	asLines = sStack.split('\n');
814
815	# Figure the CPU.
816	asWords = asLines[0].split();
817	if len(asWords) < 6 or not asWords[5].isdigit():
818	break;
819	iCpu = int(asWords[5]);
820
821	# Add it and advance.
822	dRet[iCpu] = [sLine.rstrip() for sLine in asLines[2:-1]]
823	off = offEnd;
824	return dRet;
825
826	def investigateInfoKvmLockSpinning(self, oCaseFile, sInfoText, dLogs):
827	""" Investigates kvm_lock_spinning deadlocks """
828	#
829	# Extract the stacks. We need more than one CPU to create a deadlock.
830	#
831	dStacks = self.extractGuestCpuStack(sInfoText);
832	self.dprint('kvm_lock_spinning: found %s stacks' % (len(dStacks),));
833	if len(dStacks) >= 2:
834	#
835	# Examin each of the stacks. Each must have kvm_lock_spinning in
836	# one of the first three entries.
837	#
838	cHits = 0;
839	for iCpu in dStacks:
840	asBacktrace = dStacks[iCpu];
841	for iFrame in xrange(min(3, len(asBacktrace))):
842	if asBacktrace[iFrame].find('kvm_lock_spinning') >= 0:
843	cHits += 1;
844	break;
845	self.dprint('kvm_lock_spinning: %s/%s hits' % (cHits, len(dStacks),));
846	if cHits == len(dStacks):
847	return (True, self.ktReason_VMM_kvm_lock_spinning);
848
849	_ = dLogs; _ = oCaseFile;
850	return (False, None);
851
852	def investigateInfoHalReturnToFirmware(self, oCaseFile, sInfoText, dLogs):
853	""" Investigates HalReturnToFirmware hangs """
854	del oCaseFile
855	del sInfoText
856	del dLogs
857	# hope that's sufficient
858	return (True, self.ktReason_Unknown_HalReturnToFirmware);
859
860	## Things we search a main or VM log for to figure out why something went bust.
861	katSimpleMainAndVmLogReasons = [
862	# ( Whether to stop on hit, reason tuple, needle text. )
863	( False, ktReason_Guru_Generic, 'GuruMeditation' ),
864	( False, ktReason_Guru_Generic, 'Guru Meditation' ),
865	( True, ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED, 'VERR_IEM_INSTR_NOT_IMPLEMENTED' ),
866	( True, ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED, 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' ),
867	( True, ktReason_Guru_VERR_TRPM_DONT_PANIC, 'VERR_TRPM_DONT_PANIC' ),
868	( True, ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED, 'VERR_PGM_PHYS_PAGE_RESERVED' ),
869	( True, ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE, 'VERR_VMX_INVALID_GUEST_STATE' ),
870	( True, ktReason_Guru_VINF_EM_TRIPLE_FAULT, 'VINF_EM_TRIPLE_FAULT' ),
871	( True, ktReason_Networking_Nonexistent_host_nic,
872	'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
873	( True, ktReason_Host_Reboot_OSX_Watchdog_Timeout, ': "OSX Watchdog Timeout: ' ),
874	( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
875	'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
876	( True, ktReason_Host_HostMemoryLow, 'HostMemoryLow' ),
877	( True, ktReason_Host_HostMemoryLow, 'Failed to procure handy pages; rc=VERR_NO_MEMORY' ),
878	( True, ktReason_Unknown_File_Not_Found,
879	'Error: failed to start machine. Error message: File not found. (VERR_FILE_NOT_FOUND)' ),
880	( True, ktReason_Unknown_File_Not_Found, # lump it in with file-not-found for now.
881	'Error: failed to start machine. Error message: Not supported. (VERR_NOT_SUPPORTED)' ),
882	( False, ktReason_Unknown_VM_Crash, 'txsDoConnectViaTcp: Machine state: Aborted' ),
883	( True, ktReason_Host_Modprobe_Failed, 'Kernel driver not installed' ),
884	( True, ktReason_OSInstall_Sata_no_BM, 'PCHS=14128/14134/8224' ),
885	( True, ktReason_Host_DoubleFreeHeap, 'double free or corruption' ),
886	];
887
888	## Things we search a VBoxHardening.log file for to figure out why something went bust.
889	katSimpleVBoxHardeningLogReasons = [
890	# ( Whether to stop on hit, reason tuple, needle text. )
891	( True, ktReason_Host_DriverNotLoaded, 'Error opening VBoxDrvStub: STATUS_OBJECT_NAME_NOT_FOUND' ),
892	( True, ktReason_Host_NotSignedWithBuildCert, 'Not signed with the build certificate' ),
893	];
894
895	## Things we search a kernel.log file for to figure out why something went bust.
896	katSimpleKernelLogReasons = [
897	# ( Whether to stop on hit, reason tuple, needle text. )
898	( True, ktReason_Panic_HugeMemory, 'mm/huge_memory.c:1988' ),
899	( True, ktReason_Panic_IOAPICDoesntWork, 'IO-APIC + timer doesn''t work' ),
900	( True, ktReason_Panic_TxUnitHang, 'Detected Tx Unit Hang' ),
901	];
902
903	## Things we search the _RIGHT_ _STRIPPED_ vgatext for.
904	katSimpleVgaTextReasons = [
905	# ( Whether to stop on hit, reason tuple, needle text. )
906	( True, ktReason_Panic_MP_BIOS_IO_APIC,
907	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n\n" ),
908	( True, ktReason_Panic_MP_BIOS_IO_APIC,
909	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n"
910	"...trying to set up timer (IRQ0) through the 8259A ... failed.\n"
911	"...trying to set up timer as Virtual Wire IRQ... failed.\n"
912	"...trying to set up timer as ExtINT IRQ... failed :(.\n"
913	"Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug\n"
914	"and send a report. Then try booting with the 'noapic' option\n"
915	"\n" ),
916	( True, ktReason_OSInstall_GRUB_hang,
917	"-----\nGRUB Loading stage2..\n\n\n\n" ),
918	( True, ktReason_Panic_BootManagerC000000F,
919	"Windows failed to start. A recent hardware or software change might be the" ),
920	( True, ktReason_BootManager_Image_corrupt,
921	"BOOTMGR image is corrupt. The system cannot boot." ),
922	];
923
924	## Things we search for in the info.txt file. Require handlers for now.
925	katInfoTextHandlers = [
926	# ( Trigger text, handler method )
927	( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
928	( "HalReturnToFirmware", investigateInfoHalReturnToFirmware ),
929	];
930
931	## Mapping screenshot/failure SHA-256 hashes to failure reasons.
932	katSimpleScreenshotHashReasons = [
933	# ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
934	( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
935	( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
936	( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
937	( True, ktReason_BSOD_0000007F, '57e1880619e13042a87100e7a38c8974b85ce3866501be621bea0cc696bb2c63' ),
938	( True, ktReason_BSOD_000000D1, '134621281f00a3f8aeeb7660064bffbf6187ed56d5852142328d0bcb18ef0ede' ),
939	( True, ktReason_BSOD_000000D1, '279f11258150c9d2fef041eca65501f3141da8df39256d8f6377e897e3b45a93' ),
940	( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
941	( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
942	( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
943	( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
944	( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
945	];
946
947	def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
948	"""
949	Investigates a failed VM run.
950	"""
951
952	def investigateLogSet():
953	"""
954	Investigates the current set of VM related logs.
955	"""
956	self.dprint('investigateLogSet: lengths: result log %u, VM log %u, kernel log %u, vga text %u, info text %u'
957	% ( len(sResultLog if sResultLog else ''),
958	len(sVMLog if sVMLog else ''),
959	len(sKrnlLog if sKrnlLog else ''),
960	len(sVgaText if sVgaText else ''),
961	len(sInfoText if sInfoText else ''), ));
962
963	#self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
964	#self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
965	#self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
966	#self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
967	#self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
968
969	# TODO: more
970
971	#
972	# Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
973	#
974	sDetails = self.findInAnyAndReturnRestOfLine([ sVMLog, sResultLog ],
975	'GIM: HyperV: Guest indicates a fatal condition! P0=');
976	if sDetails is not None:
977	# P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
978	sKey = sDetails.split(' ', 1)[0];
979	try: sKey = '0x%08X' % (int(sKey, 16),);
980	except: pass;
981	if sKey in self.asBsodReasons:
982	tReason = ( self.ksBsodCategory, sKey );
983	elif sKey.lower() in self.asBsodReasons: # just in case.
984	tReason = ( self.ksBsodCategory, sKey.lower() );
985	else:
986	self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
987	tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
988	return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
989
990	#
991	# Look for linux panic.
992	#
993	if sKrnlLog is not None:
994	for fStopOnHit, tReason, sNeedle in self.katSimpleKernelLogReasons:
995	if sKrnlLog.find(sNeedle) > 0:
996	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
997	if fStopOnHit:
998	return True;
999	fFoundSomething = True;
1000
1001	#
1002	# Loop thru the simple stuff.
1003	#
1004	fFoundSomething = False;
1005	for fStopOnHit, tReason, sNeedle in self.katSimpleMainAndVmLogReasons:
1006	if sResultLog.find(sNeedle) > 0 or (sVMLog is not None and sVMLog.find(sNeedle) > 0):
1007	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1008	if fStopOnHit:
1009	return True;
1010	fFoundSomething = True;
1011
1012	# Continue with vga text.
1013	if sVgaText:
1014	for fStopOnHit, tReason, sNeedle in self.katSimpleVgaTextReasons:
1015	if sVgaText.find(sNeedle) > 0:
1016	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1017	if fStopOnHit:
1018	return True;
1019	fFoundSomething = True;
1020	_ = sInfoText;
1021
1022	# Continue with screen hashes.
1023	if sScreenHash is not None:
1024	for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
1025	if sScreenHash == sHash:
1026	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1027	if fStopOnHit:
1028	return True;
1029	fFoundSomething = True;
1030
1031	# Check VBoxHardening.log.
1032	if sNtHardLog is not None:
1033	for fStopOnHit, tReason, sNeedle in self.katSimpleVBoxHardeningLogReasons:
1034	if sNtHardLog.find(sNeedle) > 0:
1035	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1036	if fStopOnHit:
1037	return True;
1038	fFoundSomething = True;
1039
1040	#
1041	# Complicated stuff.
1042	#
1043	dLogs = {
1044	'sVMLog': sVMLog,
1045	'sNtHardLog': sNtHardLog,
1046	'sScreenHash': sScreenHash,
1047	'sKrnlLog': sKrnlLog,
1048	'sVgaText': sVgaText,
1049	'sInfoText': sInfoText,
1050	};
1051
1052	# info.txt.
1053	if sInfoText:
1054	for sNeedle, fnHandler in self.katInfoTextHandlers:
1055	if sInfoText.find(sNeedle) > 0:
1056	(fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
1057	if tReason is not None:
1058	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1059	if fStop:
1060	return True;
1061	fFoundSomething = True;
1062
1063	#
1064	# Check for repeated reboots...
1065	#
1066	if sVMLog is not None:
1067	cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1068	if cResets > 10:
1069	return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1070	sComment = 'Counted %s reboots' % (cResets,));
1071
1072	return fFoundSomething;
1073
1074	#
1075	# Check if we got any VM or/and kernel logs. Treat them as sets in
1076	# case we run multiple VMs here (this is of course ASSUMING they
1077	# appear in the order that terminateVmBySession uploads them).
1078	#
1079	cTimes = 0;
1080	sVMLog = None;
1081	sNtHardLog = None;
1082	sScreenHash = None;
1083	sKrnlLog = None;
1084	sVgaText = None;
1085	sInfoText = None;
1086	for oFile in oFailedResult.aoFiles:
1087	if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1088	if 'VBoxHardening.log' not in oFile.sFile:
1089	if sVMLog is not None:
1090	if investigateLogSet() is True:
1091	return True;
1092	cTimes += 1;
1093	sInfoText = None;
1094	sVgaText = None;
1095	sKrnlLog = None;
1096	sScreenHash = None;
1097	sNtHardLog = None;
1098	sVMLog = oCaseFile.getLogFile(oFile);
1099	else:
1100	sNtHardLog = oCaseFile.getLogFile(oFile);
1101	elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1102	sKrnlLog = oCaseFile.getLogFile(oFile);
1103	elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1104	sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1105	elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1106	sInfoText = oCaseFile.getLogFile(oFile);
1107	elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1108	sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1109	if sScreenHash is not None:
1110	sScreenHash = sScreenHash.lower();
1111	self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1112
1113	if ( sVMLog is not None \
1114	or sNtHardLog is not None \
1115	or cTimes == 0) \
1116	and investigateLogSet() is True:
1117	return True;
1118
1119	return None;
1120
1121
1122	def isResultFromVMRun(self, oFailedResult, sResultLog):
1123	"""
1124	Checks if this result and corresponding log snippet looks like a VM run.
1125	"""
1126
1127	# Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1128	if sResultLog.find(' startVm') > 0:
1129	return True;
1130
1131	# Any other indicators? No?
1132	_ = oFailedResult;
1133	return False;
1134
1135	def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1136	"""
1137	Checks out a VBox VM test.
1138
1139	This is generic investigation of a test running one or more VMs, like
1140	for example a smoke test or a guest installation test.
1141
1142	The fSingleVM parameter is a hint, which probably won't come in useful.
1143	"""
1144	_ = fSingleVM;
1145
1146	#
1147	# Get a list of test result failures we should be looking into and the main log.
1148	#
1149	aoFailedResults = oCaseFile.oTree.getListOfFailures();
1150	sMainLog = oCaseFile.getMainLog();
1151
1152	#
1153	# There are a set of errors ending up on the top level result record.
1154	# Should deal with these first.
1155	#
1156	if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1157	# Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1158	# be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1159	# ignore other failures in the test if we're not a little bit careful here.
1160	if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1161	oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1162	return self.caseClosed(oCaseFile);
1163
1164	# Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1165	if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1166	(' Timeout: ', ' Attempting to abort child...',) ):
1167	if sMainLog.find('* glibc detected * /') > 0:
1168	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1169	else:
1170	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1171	return self.caseClosed(oCaseFile);
1172
1173	# Look for heap corruption without visible hang.
1174	if sMainLog.find('* glibc detected * /') > 0 \
1175	or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1176	oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1177	return self.caseClosed(oCaseFile);
1178
1179	# Out of memory w/ timeout.
1180	if sMainLog.find('sErrId=HostMemoryLow') > 0:
1181	oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1182	return self.caseClosed(oCaseFile);
1183
1184	# Stale files like vts_rm.exe (windows).
1185	offEnd = sMainLog.rfind('* The test driver exits successfully. *');
1186	if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1187	oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1188	return self.caseClosed(oCaseFile);
1189
1190	#
1191	# XPCOM screwup
1192	#
1193	if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1194	oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1195	return self.caseClosed(oCaseFile);
1196
1197	#
1198	# Go thru each failed result.
1199	#
1200	for oFailedResult in aoFailedResults:
1201	self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1202	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1203	if oFailedResult.sName == 'Installing VirtualBox':
1204	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1205
1206	elif oFailedResult.sName == 'Uninstalling VirtualBox':
1207	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1208
1209	elif self.isResultFromVMRun(oFailedResult, sResultLog):
1210	self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1211
1212	elif sResultLog.find('most likely not unique') > 0:
1213	oCaseFile.noteReasonForId(self.ktReason_Host_NetworkMisconfiguration, oFailedResult.idTestResult)
1214	elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1215	oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1216
1217	elif sResultLog.find('The machine is not mutable (state is ') > 0:
1218	self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1219	oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1220
1221	elif sResultLog.find('** error: no action was specified') > 0 \
1222	or sResultLog.find('(len(self._asXml, asText))') > 0:
1223	oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1224
1225	else:
1226	self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1227	self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1228
1229	#
1230	# Report home and close the case if we got them all, otherwise log it.
1231	#
1232	if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1233	return self.caseClosed(oCaseFile);
1234
1235	if oCaseFile.dReasonForResultId:
1236	self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1237	% (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1238	else:
1239	self.vprint(u'XXX: Could not figure out anything at all! :-(');
1240	return False;
1241
1242
1243	def reasoningFailures(self):
1244	"""
1245	Guess the reason for failures.
1246	"""
1247	#
1248	# Get a list of failed test sets without any assigned failure reason.
1249	#
1250	cGot = 0;
1251	aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack, tsNow = self.tsNow);
1252	for oTestSet in aoTestSets:
1253	self.dprint(u'');
1254	self.dprint(u'reasoningFailures: Checking out test set #%u, status %s' % ( oTestSet.idTestSet, oTestSet.enmStatus,))
1255
1256	#
1257	# Open a case file and assign it to the right investigator.
1258	#
1259	(oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1260	oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1261	oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1262	oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1263	oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1264
1265	oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1266
1267	if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1268	self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1269	fRc = self.investigateBadTestBox(oCaseFile);
1270
1271	elif oCaseFile.isVBoxUnitTest():
1272	self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1273	fRc = self.investigateVBoxUnitTest(oCaseFile);
1274
1275	elif oCaseFile.isVBoxInstallTest():
1276	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1277	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1278
1279	elif oCaseFile.isVBoxUSBTest():
1280	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1281	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1282
1283	elif oCaseFile.isVBoxStorageTest():
1284	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1285	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1286
1287	elif oCaseFile.isVBoxGAsTest():
1288	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1289	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1290
1291	elif oCaseFile.isVBoxAPITest():
1292	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1293	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1294
1295	elif oCaseFile.isVBoxBenchmarkTest():
1296	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1297	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1298
1299	elif oCaseFile.isVBoxSmokeTest():
1300	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1301	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1302
1303	else:
1304	self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1305	fRc = False;
1306	cGot += fRc is True;
1307
1308	self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1309	return 0;
1310
1311
1312	def main(self):
1313	"""
1314	The 'main' function.
1315	Return exit code (0, 1, etc).
1316	"""
1317	# Database stuff.
1318	self.oDb = TMDatabaseConnection()
1319	self.oTestResultLogic = TestResultLogic(self.oDb);
1320	self.oTestSetLogic = TestSetLogic(self.oDb);
1321	self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1322	self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1323	self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1324	self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1325
1326	# Get a fix on our 'now' before we do anything..
1327	self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1328	self.tsNow = self.oDb.fetchOne();
1329
1330	# If we're suppost to commit anything we need to get our user ID.
1331	rcExit = 0;
1332	if self.oConfig.fRealRun:
1333	self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1334	if self.oLogin is None:
1335	rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1336	else:
1337	self.uidSelf = self.oLogin.uid;
1338
1339	#
1340	# Do the stuff.
1341	#
1342	if rcExit == 0:
1343	rcExit = self.selfCheck();
1344	if rcExit == 0:
1345	rcExit = self.badTestBoxManagement();
1346	rcExit2 = self.reasoningFailures();
1347	if rcExit == 0:
1348	rcExit = rcExit2;
1349	# Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1350	if rcExit == 0:
1351	rcExit = self.badTestBoxManagement();
1352
1353	# Cleanup.
1354	self.oFailureReasonLogic = None;
1355	self.oTestResultFailureLogic = None;
1356	self.oTestSetLogic = None;
1357	self.oTestResultLogic = None;
1358	self.oDb.close();
1359	self.oDb = None;
1360	if self.oLogFile is not None:
1361	self.oLogFile.close();
1362	self.oLogFile = None;
1363	return rcExit;
1364
1365	if __name__ == '__main__':
1366	sys.exit(VirtualTestSheriff().main());
1367

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

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 71088

Download in other formats: