virtual_test_sheriff.py@ 67715

Last change on this file since 67715 was 67715, checked in by vboxsync, 7 years ago
typo
Property svn:eol-style set to `LF` Property svn:executable set to ``* Property svn:keywords set to `Author Date Id Revision`
File size: 62.7 KB

Line
1	#!/usr/bin/env python
2	# -- coding: utf-8 --
3	# $Id: virtual_test_sheriff.py 67715 2017-06-30 07:48:40Z 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	__copyright__ = \
16	"""
17	Copyright (C) 2012-2016 Oracle Corporation
18
19	This file is part of VirtualBox Open Source Edition (OSE), as
20	available from http://www.virtualbox.org. This file is free software;
21	you can redistribute it and/or modify it under the terms of the GNU
22	General Public License (GPL) as published by the Free Software
23	Foundation, in version 2 as it comes in the "COPYING" file of the
24	VirtualBox OSE distribution. VirtualBox OSE is distributed in the
25	hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
26
27	The contents of this file may alternatively be used under the terms
28	of the Common Development and Distribution License Version 1.0
29	(CDDL) only, as it comes in the "COPYING.CDDL" file of the
30	VirtualBox OSE distribution, in which case the provisions of the
31	CDDL are applicable instead of those of the GPL.
32
33	You may elect to license modified versions of this file under the
34	terms and conditions of either the GPL or the CDDL or both.
35	"""
36	__version__ = "$Revision: 67715 $"
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: 67715 $ \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	self.getFailureReason(self.ktReason_Host_DriverNotCompilable).idFailureReason,
376	];
377
378	#
379	# Get list of bad test boxes for given period and check them out individually.
380	#
381	aidBadTestBoxes = self.oTestSetLogic.fetchBadTestBoxIds(cHoursBack = cHoursBack, tsNow = tsNow,
382	aidFailureReasons = aidFailureReasons);
383	for idTestBox in aidBadTestBoxes:
384	# Skip if the testbox is already disabled or has a pending reboot command.
385	try:
386	oTestBox = TestBoxData().initFromDbWithId(self.oDb, idTestBox);
387	except Exception as oXcpt:
388	rcExit = self.eprint('Failed to get data for test box #%u in badTestBoxManagement: %s' % (idTestBox, oXcpt,));
389	continue;
390	if not oTestBox.fEnabled:
391	self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has been disabled already.'
392	% ( idTestBox, oTestBox.sName, ));
393	continue;
394	if oTestBox.enmPendingCmd != TestBoxData.ksTestBoxCmd_None:
395	self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has a command pending: %s'
396	% ( idTestBox, oTestBox.sName, oTestBox.enmPendingCmd));
397	continue;
398
399	# Get the most recent testsets for this box (descending on tsDone) and see how bad it is.
400	aoSets = self.oTestSetLogic.fetchSetsForTestBox(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow);
401	cOkay = 0;
402	cBad = 0;
403	iFirstOkay = len(aoSets);
404	for iSet, oSet in enumerate(aoSets):
405	if oSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
406	cBad += 1;
407	else:
408	# Check for bad failure reasons.
409	oFailure = None;
410	if oSet.enmStatus in TestSetData.kasBadTestStatuses:
411	oFailure = self.oTestResultFailureLogic.getById(oSet.idTestResult);
412	if oFailure is not None and oFailure.idFailureReason in aidFailureReasons:
413	cBad += 1;
414	else:
415	# This is an okay test result then.
416	## @todo maybe check the elapsed time here, it could still be a bad run?
417	cOkay += 1;
418	if iFirstOkay > iSet:
419	iFirstOkay = iSet;
420	if iSet > 10:
421	break;
422
423	# We react if there are two or more bad-testbox statuses at the head of the
424	# history and at least three in the last 10 results.
425	if iFirstOkay >= 2 and cBad > 2:
426	# Frank: For now don't reboot boxes automatically
427	if True or oTestBoxLogic.hasTestBoxRecentlyBeenRebooted(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow):
428	self.vprint(u'Disabling testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u'
429	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
430	if self.oConfig.fRealRun is True:
431	try:
432	oTestBoxLogic.disableTestBox(idTestBox, self.uidSelf, fCommit = True,
433	sComment = 'Automatically disabled (iFirstOkay=%u cBad=%u cOkay=%u)'
434	% (iFirstOkay, cBad, cOkay),);
435	except Exception as oXcpt:
436	rcExit = self.eprint(u'Error disabling testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
437	else:
438	self.vprint(u'Rebooting 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.rebootTestBox(idTestBox, self.uidSelf, fCommit = True,
443	sComment = 'Automatically rebooted (iFirstOkay=%u cBad=%u cOkay=%u)'
444	% (iFirstOkay, cBad, cOkay),);
445	except Exception as oXcpt:
446	rcExit = self.eprint(u'Error rebooting testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
447	else:
448	self.dprint(u'badTestBoxManagement: #%u (%s) looks ok: iFirstOkay=%u cBad=%u cOkay=%u'
449	% ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
450	return rcExit;
451
452
453	## @name Failure reasons we know.
454	## @{
455	ktReason_BSOD_Recovery = ( 'BSOD', 'Recovery' );
456	ktReason_BSOD_Automatic_Repair = ( 'BSOD', 'Automatic Repair' );
457	ktReason_BSOD_000000D1 = ( 'BSOD', '0x000000D1' );
458	ktReason_BSOD_C0000225 = ( 'BSOD', '0xC0000225 (boot)' );
459	ktReason_Guru_Generic = ( 'Guru Meditations', 'Generic Guru Meditation' );
460	ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_INSTR_NOT_IMPLEMENTED' );
461	ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' );
462	ktReason_Guru_VERR_TRPM_DONT_PANIC = ( 'Guru Meditations', 'VERR_TRPM_DONT_PANIC' );
463	ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED = ( 'Guru Meditations', 'VERR_PGM_PHYS_PAGE_RESERVED' );
464	ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE = ( 'Guru Meditations', 'VERR_VMX_INVALID_GUEST_STATE' );
465	ktReason_Guru_VINF_EM_TRIPLE_FAULT = ( 'Guru Meditations', 'VINF_EM_TRIPLE_FAULT' );
466	ktReason_Host_HostMemoryLow = ( 'Host', 'HostMemoryLow' );
467	ktReason_Host_DriverNotLoaded = ( 'Host', 'Driver not loaded' );
468	ktReason_Host_DriverNotUnloading = ( 'Host', 'Driver not unloading' );
469	ktReason_Host_DriverNotCompilable = ( 'Host', 'Driver not compilable' );
470	ktReason_Host_NotSignedWithBuildCert = ( 'Host', 'Not signed with build cert' );
471	ktReason_Host_LeftoverService = ( 'Host', 'Leftover service' );
472	ktReason_Host_Reboot_OSX_Watchdog_Timeout = ( 'Host Reboot', 'OSX Watchdog Timeout' );
473	ktReason_Host_Modprobe_Failed = ( 'Host', 'Modprobe failed' );
474	ktReason_Host_Install_Hang = ( 'Host', 'Install hang' );
475	ktReason_Networking_Nonexistent_host_nic = ( 'Networking', 'Nonexistent host networking interface' );
476	ktReason_OSInstall_GRUB_hang = ( 'O/S Install', 'GRUB hang' );
477	ktReason_OSInstall_Sata_no_BM = (' O/S Install', 'SATA busmaster bit not set' );
478	ktReason_Panic_BootManagerC000000F = ( 'Panic', 'Hardware Changed' );
479	ktReason_BootManager_Image_corrupt = ( 'Unknown', 'BOOTMGR Image corrupt' );
480	ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
481	ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
482	ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
483	ktReason_Panic_TxUnitHang = ( 'Panic', 'Tx Unit Hang' );
484	ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
485	ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
486	ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
487	ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
488	ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
489	ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
490	ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
491	ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
492	ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
493	ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
494	ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
495	ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
496	## @}
497
498	## BSOD category.
499	ksBsodCategory = 'BSOD';
500	## Special reason indicating that the flesh and blood sheriff has work to do.
501	ksBsodAddNew = 'Add new BSOD';
502
503	## Unit test category.
504	ksUnitTestCategory = 'Unit';
505	## Special reason indicating that the flesh and blood sheriff has work to do.
506	ksUnitTestAddNew = 'Add new';
507
508	## Used for indica that we shouldn't report anything for this test result ID and
509	## consider promoting the previous error to test set level if it's the only one.
510	ktHarmless = ( 'Probably', 'Caused by previous error' );
511
512
513	def caseClosed(self, oCaseFile):
514	"""
515	Reports the findings in the case and closes it.
516	"""
517	#
518	# Log it and create a dReasonForReasultId we can use below.
519	#
520	dCommentForResultId = oCaseFile.dCommentForResultId;
521	if oCaseFile.dReasonForResultId:
522	# Must weed out ktHarmless.
523	dReasonForResultId = {};
524	for idKey, tReason in oCaseFile.dReasonForResultId.items():
525	if tReason is not self.ktHarmless:
526	dReasonForResultId[idKey] = tReason;
527	if not dReasonForResultId:
528	self.vprint(u'TODO: Closing %s without a real reason, only %s.'
529	% (oCaseFile.sName, oCaseFile.dReasonForResultId));
530	return False;
531
532	# Try promote to single reason.
533	atValues = dReasonForResultId.values();
534	fSingleReason = True;
535	if len(dReasonForResultId) == 1 and dReasonForResultId.keys()[0] != oCaseFile.oTestSet.idTestResult:
536	self.dprint(u'Promoting single reason to whole set: %s' % (atValues[0],));
537	elif len(dReasonForResultId) > 1 and len(atValues) == atValues.count(atValues[0]):
538	self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), atValues[0]));
539	else:
540	fSingleReason = False;
541	if fSingleReason:
542	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: atValues[0], };
543	if dCommentForResultId:
544	dCommentForResultId = { oCaseFile.oTestSet.idTestResult: dCommentForResultId.values()[0], };
545	elif oCaseFile.tReason is not None:
546	dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
547	else:
548	self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
549	return False;
550
551	self.vprint(u'Closing %s with following reason%s: %s'
552	% ( oCaseFile.sName, 's' if dReasonForResultId > 0 else '', dReasonForResultId, ));
553
554	#
555	# Add the test failure reason record(s).
556	#
557	for idTestResult, tReason in dReasonForResultId.items():
558	oFailureReason = self.getFailureReason(tReason);
559	if oFailureReason is not None:
560	sComment = 'Set by $Revision: 67715 $' # Handy for reverting later.
561	if idTestResult in dCommentForResultId:
562	sComment += ': ' + dCommentForResultId[idTestResult];
563
564	oAdd = TestResultFailureData();
565	oAdd.initFromValues(idTestResult = idTestResult,
566	idFailureReason = oFailureReason.idFailureReason,
567	uidAuthor = self.uidSelf,
568	idTestSet = oCaseFile.oTestSet.idTestSet,
569	sComment = sComment,);
570	if self.oConfig.fRealRun:
571	try:
572	self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
573	except Exception as oXcpt:
574	self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
575	% (oXcpt, oAdd, oCaseFile.sLongName,));
576	else:
577	self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
578	return True;
579
580	#
581	# Tools for assiting log parsing.
582	#
583
584	@staticmethod
585	def matchFollowedByLines(sStr, off, asFollowingLines):
586	""" Worker for isThisFollowedByTheseLines. """
587
588	# Advance off to the end of the line.
589	off = sStr.find('\n', off);
590	if off < 0:
591	return False;
592	off += 1;
593
594	# Match each string with the subsequent lines.
595	for iLine, sLine in enumerate(asFollowingLines):
596	offEnd = sStr.find('\n', off);
597	if offEnd < 0:
598	return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
599	if sLine and sStr.find(sLine, off, offEnd) < 0:
600	return False;
601
602	# next line.
603	off = offEnd + 1;
604
605	return True;
606
607	@staticmethod
608	def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
609	"""
610	Looks for a line contining sFirst which is then followed by lines
611	with the strings in asFollowingLines. (No newline chars anywhere!)
612	Returns True / False.
613	"""
614	off = sStr.find(sFirst, 0);
615	while off >= 0:
616	if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
617	return True;
618	off = sStr.find(sFirst, off + 1);
619	return False;
620
621	@staticmethod
622	def findAndReturnRestOfLine(sHaystack, sNeedle):
623	"""
624	Looks for sNeedle in sHaystack.
625	Returns The text following the needle up to the end of the line.
626	Returns None if not found.
627	"""
628	if sHaystack is None:
629	return None;
630	off = sHaystack.find(sNeedle);
631	if off < 0:
632	return None;
633	off += len(sNeedle)
634	offEol = sHaystack.find('\n', off);
635	if offEol < 0:
636	offEol = len(sHaystack);
637	return sHaystack[off:offEol]
638
639	@staticmethod
640	def findInAnyAndReturnRestOfLine(asHaystacks, sNeedle):
641	"""
642	Looks for sNeedle in zeroe or more haystacks (asHaystack).
643	Returns The text following the first needed found up to the end of the line.
644	Returns None if not found.
645	"""
646	for sHaystack in asHaystacks:
647	sRet = VirtualTestSheriff.findAndReturnRestOfLine(sHaystack, sNeedle);
648	if sRet is not None:
649	return sRet;
650	return None;
651
652
653	#
654	# The investigative units.
655	#
656
657	katSimpleInstallUninstallMainLogReasons = [
658	# ( Whether to stop on hit, reason tuple, needle text. )
659	( False, ktReason_Host_LeftoverService,
660	'SERVICE_NAME: vbox' ),
661	];
662
663	kdatSimpleInstallUninstallMainLogReasonsPerOs = {
664	'darwin': [
665	# ( Whether to stop on hit, reason tuple, needle text. )
666	( True, ktReason_Host_DriverNotUnloading,
667	'Can\'t remove kext org.virtualbox.kext.VBoxDrv; services failed to terminate - 0xe00002c7' ),
668	],
669	'linux': [
670	# ( Whether to stop on hit, reason tuple, needle text. )
671	( True, ktReason_Host_DriverNotCompilable,
672	'This system is not currently set up to build kernel modules' ),
673	( True, ktReason_Host_DriverNotCompilable,
674	'This system is currently not set up to build kernel modules' ),
675	],
676	};
677
678
679	def investigateInstallUninstallFailure(self, oCaseFile, oFailedResult, sResultLog, fInstall):
680	"""
681	Investigates an install or uninstall failure.
682
683	We lump the two together since the installation typically also performs
684	an uninstall first and will be seeing similar issues to the uninstall.
685	"""
686
687	if fInstall and oFailedResult.enmStatus == TestSetData.ksTestStatus_TimedOut:
688	oCaseFile.noteReasonForId(self.ktReason_Host_Install_Hang, oFailedResult.idTestResult)
689	return True;
690
691	atSimple = self.katSimpleInstallUninstallMainLogReasons;
692	if oCaseFile.oTestBox.sOs in self.kdatSimpleInstallUninstallMainLogReasonsPerOs:
693	atSimple = self.kdatSimpleInstallUninstallMainLogReasonsPerOs[oCaseFile.oTestBox.sOs] + atSimple;
694
695	fFoundSomething = False;
696	for fStopOnHit, tReason, sNeedle in atSimple:
697	if sResultLog.find(sNeedle) > 0:
698	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
699	if fStopOnHit:
700	return True;
701	fFoundSomething = True;
702
703	return fFoundSomething if fFoundSomething else None;
704
705
706	def investigateBadTestBox(self, oCaseFile):
707	"""
708	Checks out bad-testbox statuses.
709	"""
710	_ = oCaseFile;
711	return False;
712
713
714	def investigateVBoxUnitTest(self, oCaseFile):
715	"""
716	Checks out a VBox unittest problem.
717	"""
718
719	#
720	# Process simple test case failures first, using their name as reason.
721	# We do the reason management just like for BSODs.
722	#
723	cRelevantOnes = 0;
724	sMainLog = oCaseFile.getMainLog();
725	aoFailedResults = oCaseFile.oTree.getListOfFailures();
726	for oFailedResult in aoFailedResults:
727	if oFailedResult is oCaseFile.oTree:
728	self.vprint('TODO: toplevel failure');
729	cRelevantOnes += 1
730
731	elif oFailedResult.sName == 'Installing VirtualBox':
732	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
733	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
734	cRelevantOnes += 1
735
736	elif oFailedResult.sName == 'Uninstalling VirtualBox':
737	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
738	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
739	cRelevantOnes += 1
740
741	elif oFailedResult.oParent is not None:
742	# Get the 2nd level node because that's where we'll find the unit test name.
743	while oFailedResult.oParent.oParent is not None:
744	oFailedResult = oFailedResult.oParent;
745
746	# Only report a failure once.
747	if oFailedResult.idTestResult not in oCaseFile.dReasonForResultId:
748	sKey = oFailedResult.sName;
749	if sKey.startswith('testcase/'):
750	sKey = sKey[9:];
751	if sKey in self.asUnitTestReasons:
752	tReason = ( self.ksUnitTestCategory, sKey );
753	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
754	else:
755	self.dprint(u'Unit test failure "%s" not found in %s;' % (sKey, self.asUnitTestReasons));
756	tReason = ( self.ksUnitTestCategory, self.ksUnitTestAddNew );
757	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sKey);
758	cRelevantOnes += 1
759	else:
760	self.vprint(u'Internal error: expected oParent to NOT be None for %s' % (oFailedResult,));
761
762	#
763	# If we've caught all the relevant ones by now, report the result.
764	#
765	if len(oCaseFile.dReasonForResultId) >= cRelevantOnes:
766	return self.caseClosed(oCaseFile);
767	return False;
768
769	def extractGuestCpuStack(self, sInfoText):
770	"""
771	Extracts the guest CPU stacks from the input file.
772
773	Returns a dictionary keyed by the CPU number, value being a list of
774	raw stack lines (no header).
775	Returns empty dictionary if no stacks where found.
776	"""
777	dRet = {};
778	off = 0;
779	while True:
780	# Find the stack.
781	offStart = sInfoText.find('=== start guest stack VCPU ', off);
782	if offStart < 0:
783	break;
784	offEnd = sInfoText.find('=== end guest stack', offStart + 20);
785	if offEnd >= 0:
786	offEnd += 3;
787	else:
788	offEnd = sInfoText.find('=== start guest stack VCPU', offStart + 20);
789	if offEnd < 0:
790	offEnd = len(sInfoText);
791
792	sStack = sInfoText[offStart : offEnd];
793	sStack = sStack.replace('\r',''); # paranoia
794	asLines = sStack.split('\n');
795
796	# Figure the CPU.
797	asWords = asLines[0].split();
798	if len(asWords) < 6 or not asWords[5].isdigit():
799	break;
800	iCpu = int(asWords[5]);
801
802	# Add it and advance.
803	dRet[iCpu] = [sLine.rstrip() for sLine in asLines[2:-1]]
804	off = offEnd;
805	return dRet;
806
807	def investigateInfoKvmLockSpinning(self, oCaseFile, sInfoText, dLogs):
808	""" Investigates kvm_lock_spinning deadlocks """
809	#
810	# Extract the stacks. We need more than one CPU to create a deadlock.
811	#
812	dStacks = self.extractGuestCpuStack(sInfoText);
813	self.dprint('kvm_lock_spinning: found %s stacks' % (len(dStacks),));
814	if len(dStacks) >= 2:
815	#
816	# Examin each of the stacks. Each must have kvm_lock_spinning in
817	# one of the first three entries.
818	#
819	cHits = 0;
820	for iCpu in dStacks:
821	asBacktrace = dStacks[iCpu];
822	for iFrame in xrange(min(3, len(asBacktrace))):
823	if asBacktrace[iFrame].find('kvm_lock_spinning') >= 0:
824	cHits += 1;
825	break;
826	self.dprint('kvm_lock_spinning: %s/%s hits' % (cHits, len(dStacks),));
827	if cHits == len(dStacks):
828	return (True, self.ktReason_VMM_kvm_lock_spinning);
829
830	_ = dLogs; _ = oCaseFile;
831	return (False, None);
832
833	## Things we search a main or VM log for to figure out why something went bust.
834	katSimpleMainAndVmLogReasons = [
835	# ( Whether to stop on hit, reason tuple, needle text. )
836	( False, ktReason_Guru_Generic, 'GuruMeditation' ),
837	( False, ktReason_Guru_Generic, 'Guru Meditation' ),
838	( True, ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED, 'VERR_IEM_INSTR_NOT_IMPLEMENTED' ),
839	( True, ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED, 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' ),
840	( True, ktReason_Guru_VERR_TRPM_DONT_PANIC, 'VERR_TRPM_DONT_PANIC' ),
841	( True, ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED, 'VERR_PGM_PHYS_PAGE_RESERVED' ),
842	( True, ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE, 'VERR_VMX_INVALID_GUEST_STATE' ),
843	( True, ktReason_Guru_VINF_EM_TRIPLE_FAULT, 'VINF_EM_TRIPLE_FAULT' ),
844	( True, ktReason_Networking_Nonexistent_host_nic,
845	'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
846	( True, ktReason_Host_Reboot_OSX_Watchdog_Timeout, ': "OSX Watchdog Timeout: ' ),
847	( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
848	'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
849	( True, ktReason_Host_HostMemoryLow, 'HostMemoryLow' ),
850	( True, ktReason_Host_HostMemoryLow, 'Failed to procure handy pages; rc=VERR_NO_MEMORY' ),
851	( True, ktReason_Unknown_File_Not_Found,
852	'Error: failed to start machine. Error message: File not found. (VERR_FILE_NOT_FOUND)' ),
853	( True, ktReason_Unknown_File_Not_Found, # lump it in with file-not-found for now.
854	'Error: failed to start machine. Error message: Not supported. (VERR_NOT_SUPPORTED)' ),
855	( False, ktReason_Unknown_VM_Crash, 'txsDoConnectViaTcp: Machine state: Aborted' ),
856	( True, ktReason_Host_Modprobe_Failed, 'Kernel driver not installed' ),
857	( True, ktReason_OSInstall_Sata_no_BM, 'PCHS=14128/14134/8224' )
858	];
859
860	## Things we search a VBoxHardening.log file for to figure out why something went bust.
861	katSimpleVBoxHardeningLogReasons = [
862	# ( Whether to stop on hit, reason tuple, needle text. )
863	( True, ktReason_Host_DriverNotLoaded, 'Error opening VBoxDrvStub: STATUS_OBJECT_NAME_NOT_FOUND' ),
864	( True, ktReason_Host_NotSignedWithBuildCert, 'Not signed with the build certificate' ),
865	];
866
867	## Things we search a kernel.log file for to figure out why something went bust.
868	katSimpleKernelLogReasons = [
869	# ( Whether to stop on hit, reason tuple, needle text. )
870	( True, ktReason_Panic_HugeMemory, 'mm/huge_memory.c:1988' ),
871	( True, ktReason_Panic_IOAPICDoesntWork, 'IO-APIC + timer doesn''t work' ),
872	( True, ktReason_Panic_TxUnitHang, 'Detected Tx Unit Hang' ),
873	];
874
875	## Things we search the _RIGHT_ _STRIPPED_ vgatext for.
876	katSimpleVgaTextReasons = [
877	# ( Whether to stop on hit, reason tuple, needle text. )
878	( True, ktReason_Panic_MP_BIOS_IO_APIC,
879	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n\n" ),
880	( True, ktReason_Panic_MP_BIOS_IO_APIC,
881	"..MP-BIOS bug: 8254 timer not connected to IO-APIC\n"
882	"...trying to set up timer (IRQ0) through the 8259A ... failed.\n"
883	"...trying to set up timer as Virtual Wire IRQ... failed.\n"
884	"...trying to set up timer as ExtINT IRQ... failed :(.\n"
885	"Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug\n"
886	"and send a report. Then try booting with the 'noapic' option\n"
887	"\n" ),
888	( True, ktReason_OSInstall_GRUB_hang,
889	"-----\nGRUB Loading stage2..\n\n\n\n" ),
890	( True, ktReason_Panic_BootManagerC000000F,
891	"Windows failed to start. A recent hardware or software change might be the" ),
892	( True, ktReason_BootManager_Image_corrupt,
893	"BOOTMGR image is corrupt. The system cannot boot." ),
894	];
895
896	## Things we search for in the info.txt file. Require handlers for now.
897	katInfoTextHandlers = [
898	# ( Trigger text, handler method )
899	( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
900	];
901
902	## Mapping screenshot/failure SHA-256 hashes to failure reasons.
903	katSimpleScreenshotHashReasons = [
904	# ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
905	( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
906	( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
907	( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
908	( True, ktReason_BSOD_000000D1, '134621281f00a3f8aeeb7660064bffbf6187ed56d5852142328d0bcb18ef0ede' ),
909	( True, ktReason_BSOD_000000D1, '279f11258150c9d2fef041eca65501f3141da8df39256d8f6377e897e3b45a93' ),
910	( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
911	( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
912	( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
913	( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
914	( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
915	];
916
917	def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
918	"""
919	Investigates a failed VM run.
920	"""
921
922	def investigateLogSet():
923	"""
924	Investigates the current set of VM related logs.
925	"""
926	self.dprint('investigateLogSet: lengths: result log %u, VM log %u, kernel log %u, vga text %u, info text %u'
927	% ( len(sResultLog if sResultLog else ''),
928	len(sVMLog if sVMLog else ''),
929	len(sKrnlLog if sKrnlLog else ''),
930	len(sVgaText if sVgaText else ''),
931	len(sInfoText if sInfoText else ''), ));
932
933	#self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
934	#self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
935	#self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
936	#self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
937	#self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
938
939	# TODO: more
940
941	#
942	# Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
943	#
944	sDetails = self.findInAnyAndReturnRestOfLine([ sVMLog, sResultLog ],
945	'GIM: HyperV: Guest indicates a fatal condition! P0=');
946	if sDetails is not None:
947	# P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
948	sKey = sDetails.split(' ', 1)[0];
949	try: sKey = '0x%08X' % (int(sKey, 16),);
950	except: pass;
951	if sKey in self.asBsodReasons:
952	tReason = ( self.ksBsodCategory, sKey );
953	elif sKey.lower() in self.asBsodReasons: # just in case.
954	tReason = ( self.ksBsodCategory, sKey.lower() );
955	else:
956	self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
957	tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
958	return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
959
960	#
961	# Look for linux panic.
962	#
963	if sKrnlLog is not None:
964	for fStopOnHit, tReason, sNeedle in self.katSimpleKernelLogReasons:
965	if sKrnlLog.find(sNeedle) > 0:
966	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
967	if fStopOnHit:
968	return True;
969	fFoundSomething = True;
970
971	#
972	# Loop thru the simple stuff.
973	#
974	fFoundSomething = False;
975	for fStopOnHit, tReason, sNeedle in self.katSimpleMainAndVmLogReasons:
976	if sResultLog.find(sNeedle) > 0 or (sVMLog is not None and sVMLog.find(sNeedle) > 0):
977	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
978	if fStopOnHit:
979	return True;
980	fFoundSomething = True;
981
982	# Continue with vga text.
983	if sVgaText:
984	for fStopOnHit, tReason, sNeedle in self.katSimpleVgaTextReasons:
985	if sVgaText.find(sNeedle) > 0:
986	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
987	if fStopOnHit:
988	return True;
989	fFoundSomething = True;
990	_ = sInfoText;
991
992	# Continue with screen hashes.
993	if sScreenHash is not None:
994	for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
995	if sScreenHash == sHash:
996	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
997	if fStopOnHit:
998	return True;
999	fFoundSomething = True;
1000
1001	# Check VBoxHardening.log.
1002	if sNtHardLog is not None:
1003	for fStopOnHit, tReason, sNeedle in self.katSimpleVBoxHardeningLogReasons:
1004	if sNtHardLog.find(sNeedle) > 0:
1005	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1006	if fStopOnHit:
1007	return True;
1008	fFoundSomething = True;
1009
1010	#
1011	# Complicated stuff.
1012	#
1013	dLogs = {
1014	'sVMLog': sVMLog,
1015	'sNtHardLog': sNtHardLog,
1016	'sScreenHash': sScreenHash,
1017	'sKrnlLog': sKrnlLog,
1018	'sVgaText': sVgaText,
1019	'sInfoText': sInfoText,
1020	};
1021
1022	# info.txt.
1023	if sInfoText:
1024	for sNeedle, fnHandler in self.katInfoTextHandlers:
1025	if sInfoText.find(sNeedle) > 0:
1026	(fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
1027	if tReason is not None:
1028	oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1029	if fStop:
1030	return True;
1031	fFoundSomething = True;
1032
1033	#
1034	# Check for repeated reboots...
1035	#
1036	if sVMLog is not None:
1037	cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1038	if cResets > 10:
1039	return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1040	sComment = 'Counted %s reboots' % (cResets,));
1041
1042	return fFoundSomething;
1043
1044	#
1045	# Check if we got any VM or/and kernel logs. Treat them as sets in
1046	# case we run multiple VMs here (this is of course ASSUMING they
1047	# appear in the order that terminateVmBySession uploads them).
1048	#
1049	cTimes = 0;
1050	sVMLog = None;
1051	sNtHardLog = None;
1052	sScreenHash = None;
1053	sKrnlLog = None;
1054	sVgaText = None;
1055	sInfoText = None;
1056	for oFile in oFailedResult.aoFiles:
1057	if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1058	if 'VBoxHardening.log' not in oFile.sFile:
1059	if sVMLog is not None:
1060	if investigateLogSet() is True:
1061	return True;
1062	cTimes += 1;
1063	sInfoText = None;
1064	sVgaText = None;
1065	sKrnlLog = None;
1066	sScreenHash = None;
1067	sNtHardLog = None;
1068	sVMLog = oCaseFile.getLogFile(oFile);
1069	else:
1070	sNtHardLog = oCaseFile.getLogFile(oFile);
1071	elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1072	sKrnlLog = oCaseFile.getLogFile(oFile);
1073	elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1074	sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1075	elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1076	sInfoText = oCaseFile.getLogFile(oFile);
1077	elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1078	sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1079	if sScreenHash is not None:
1080	sScreenHash = sScreenHash.lower();
1081	self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1082
1083	if ( sVMLog is not None \
1084	or sNtHardLog is not None \
1085	or cTimes == 0) \
1086	and investigateLogSet() is True:
1087	return True;
1088
1089	return None;
1090
1091
1092	def isResultFromVMRun(self, oFailedResult, sResultLog):
1093	"""
1094	Checks if this result and corresponding log snippet looks like a VM run.
1095	"""
1096
1097	# Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1098	if sResultLog.find(' startVm') > 0:
1099	return True;
1100
1101	# Any other indicators? No?
1102	_ = oFailedResult;
1103	return False;
1104
1105	def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1106	"""
1107	Checks out a VBox VM test.
1108
1109	This is generic investigation of a test running one or more VMs, like
1110	for example a smoke test or a guest installation test.
1111
1112	The fSingleVM parameter is a hint, which probably won't come in useful.
1113	"""
1114	_ = fSingleVM;
1115
1116	#
1117	# Get a list of test result failures we should be looking into and the main log.
1118	#
1119	aoFailedResults = oCaseFile.oTree.getListOfFailures();
1120	sMainLog = oCaseFile.getMainLog();
1121
1122	#
1123	# There are a set of errors ending up on the top level result record.
1124	# Should deal with these first.
1125	#
1126	if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1127	# Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1128	# be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1129	# ignore other failures in the test if we're not a little bit careful here.
1130	if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1131	oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1132	return self.caseClosed(oCaseFile);
1133
1134	# Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1135	if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1136	(' Timeout: ', ' Attempting to abort child...',) ):
1137	if sMainLog.find('* glibc detected * /') > 0:
1138	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1139	else:
1140	oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1141	return self.caseClosed(oCaseFile);
1142
1143	# Look for heap corruption without visible hang.
1144	if sMainLog.find('* glibc detected * /') > 0 \
1145	or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1146	oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1147	return self.caseClosed(oCaseFile);
1148
1149	# Out of memory w/ timeout.
1150	if sMainLog.find('sErrId=HostMemoryLow') > 0:
1151	oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1152	return self.caseClosed(oCaseFile);
1153
1154	# Stale files like vts_rm.exe (windows).
1155	offEnd = sMainLog.rfind('* The test driver exits successfully. *');
1156	if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1157	oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1158	return self.caseClosed(oCaseFile);
1159
1160	#
1161	# XPCOM screwup
1162	#
1163	if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1164	oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1165	return self.caseClosed(oCaseFile);
1166
1167	#
1168	# Go thru each failed result.
1169	#
1170	for oFailedResult in aoFailedResults:
1171	self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1172	sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1173	if oFailedResult.sName == 'Installing VirtualBox':
1174	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1175
1176	elif oFailedResult.sName == 'Uninstalling VirtualBox':
1177	self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1178
1179	elif self.isResultFromVMRun(oFailedResult, sResultLog):
1180	self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1181
1182	elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1183	oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1184
1185	elif sResultLog.find('The machine is not mutable (state is ') > 0:
1186	self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1187	oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1188
1189	elif sResultLog.find('** error: no action was specified') > 0 \
1190	or sResultLog.find('(len(self._asXml, asText))') > 0:
1191	oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1192
1193	else:
1194	self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1195	self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1196
1197	#
1198	# Report home and close the case if we got them all, otherwise log it.
1199	#
1200	if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1201	return self.caseClosed(oCaseFile);
1202
1203	if oCaseFile.dReasonForResultId:
1204	self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1205	% (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1206	else:
1207	self.vprint(u'XXX: Could not figure out anything at all! :-(');
1208	return False;
1209
1210
1211	def reasoningFailures(self):
1212	"""
1213	Guess the reason for failures.
1214	"""
1215	#
1216	# Get a list of failed test sets without any assigned failure reason.
1217	#
1218	cGot = 0;
1219	aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack, tsNow = self.tsNow);
1220	for oTestSet in aoTestSets:
1221	self.dprint(u'');
1222	self.dprint(u'reasoningFailures: Checking out test set #%u, status %s' % ( oTestSet.idTestSet, oTestSet.enmStatus,))
1223
1224	#
1225	# Open a case file and assign it to the right investigator.
1226	#
1227	(oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1228	oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1229	oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1230	oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1231	oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1232
1233	oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1234
1235	if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1236	self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1237	fRc = self.investigateBadTestBox(oCaseFile);
1238
1239	elif oCaseFile.isVBoxUnitTest():
1240	self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1241	fRc = self.investigateVBoxUnitTest(oCaseFile);
1242
1243	elif oCaseFile.isVBoxInstallTest():
1244	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1245	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1246
1247	elif oCaseFile.isVBoxUSBTest():
1248	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1249	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1250
1251	elif oCaseFile.isVBoxStorageTest():
1252	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1253	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1254
1255	elif oCaseFile.isVBoxGAsTest():
1256	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1257	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1258
1259	elif oCaseFile.isVBoxAPITest():
1260	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1261	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1262
1263	elif oCaseFile.isVBoxBenchmarkTest():
1264	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1265	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1266
1267	elif oCaseFile.isVBoxSmokeTest():
1268	self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1269	fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1270
1271	else:
1272	self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1273	fRc = False;
1274	cGot += fRc is True;
1275
1276	self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1277	return 0;
1278
1279
1280	def main(self):
1281	"""
1282	The 'main' function.
1283	Return exit code (0, 1, etc).
1284	"""
1285	# Database stuff.
1286	self.oDb = TMDatabaseConnection()
1287	self.oTestResultLogic = TestResultLogic(self.oDb);
1288	self.oTestSetLogic = TestSetLogic(self.oDb);
1289	self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1290	self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1291	self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1292	self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1293
1294	# Get a fix on our 'now' before we do anything..
1295	self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1296	self.tsNow = self.oDb.fetchOne();
1297
1298	# If we're suppost to commit anything we need to get our user ID.
1299	rcExit = 0;
1300	if self.oConfig.fRealRun:
1301	self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1302	if self.oLogin is None:
1303	rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1304	else:
1305	self.uidSelf = self.oLogin.uid;
1306
1307	#
1308	# Do the stuff.
1309	#
1310	if rcExit == 0:
1311	rcExit = self.selfCheck();
1312	if rcExit == 0:
1313	rcExit = self.badTestBoxManagement();
1314	rcExit2 = self.reasoningFailures();
1315	if rcExit == 0:
1316	rcExit = rcExit2;
1317	# Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1318	if rcExit == 0:
1319	rcExit = self.badTestBoxManagement();
1320
1321	# Cleanup.
1322	self.oFailureReasonLogic = None;
1323	self.oTestResultFailureLogic = None;
1324	self.oTestSetLogic = None;
1325	self.oTestResultLogic = None;
1326	self.oDb.close();
1327	self.oDb = None;
1328	if self.oLogFile is not None:
1329	self.oLogFile.close();
1330	self.oLogFile = None;
1331	return rcExit;
1332
1333	if __name__ == '__main__':
1334	sys.exit(VirtualTestSheriff().main());
1335

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

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 67715

Download in other formats: