VirtualBox

source: vbox/trunk/src/VBox/ValidationKit/testmanager/batch/virtual_test_sheriff.py@ 98103

Last change on this file since 98103 was 98103, checked in by vboxsync, 2 years ago

Copyright year updates by scm.

  • Property svn:eol-style set to LF
  • Property svn:executable set to *
  • Property svn:keywords set to Author Date Id Revision
File size: 90.3 KB
Line 
1#!/usr/bin/env python
2# -*- coding: utf-8 -*-
3# $Id: virtual_test_sheriff.py 98103 2023-01-17 14:15:46Z vboxsync $
4# pylint: disable=line-too-long
5
6"""
7Virtual Test Sheriff.
8
9Duties:
10 - Try to a assign failure reasons to recently failed tests.
11 - Reboot or disable bad test boxes.
12
13"""
14
15from __future__ import print_function;
16
17__copyright__ = \
18"""
19Copyright (C) 2012-2023 Oracle and/or its affiliates.
20
21This file is part of VirtualBox base platform packages, as
22available from https://www.virtualbox.org.
23
24This program is free software; you can redistribute it and/or
25modify it under the terms of the GNU General Public License
26as published by the Free Software Foundation, in version 3 of the
27License.
28
29This program is distributed in the hope that it will be useful, but
30WITHOUT ANY WARRANTY; without even the implied warranty of
31MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
32General Public License for more details.
33
34You should have received a copy of the GNU General Public License
35along with this program; if not, see <https://www.gnu.org/licenses>.
36
37The contents of this file may alternatively be used under the terms
38of the Common Development and Distribution License Version 1.0
39(CDDL), a copy of it is provided in the "COPYING.CDDL" file included
40in the VirtualBox distribution, in which case the provisions of the
41CDDL are applicable instead of those of the GPL.
42
43You may elect to license modified versions of this file under the
44terms and conditions of either the GPL or the CDDL or both.
45
46SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
47"""
48__version__ = "$Revision: 98103 $"
49
50
51# Standard python imports
52import hashlib;
53import os;
54import re;
55import smtplib;
56#import subprocess;
57import sys;
58from email.mime.multipart import MIMEMultipart;
59from email.mime.text import MIMEText;
60from email.utils import COMMASPACE;
61
62if sys.version_info[0] >= 3:
63 from io import BytesIO as BytesIO; # pylint: disable=import-error,no-name-in-module,useless-import-alias
64else:
65 from StringIO import StringIO as BytesIO; # pylint: disable=import-error,no-name-in-module,useless-import-alias
66from optparse import OptionParser; # pylint: disable=deprecated-module
67from PIL import Image; # pylint: disable=import-error
68
69# Add Test Manager's modules path
70g_ksTestManagerDir = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))));
71sys.path.append(g_ksTestManagerDir);
72
73# Test Manager imports
74from common import utils;
75from testmanager.core.db import TMDatabaseConnection;
76from testmanager.core.build import BuildDataEx;
77from testmanager.core.failurereason import FailureReasonLogic;
78from testmanager.core.testbox import TestBoxLogic, TestBoxData;
79from testmanager.core.testcase import TestCaseDataEx;
80from testmanager.core.testgroup import TestGroupData;
81from testmanager.core.testset import TestSetLogic, TestSetData;
82from testmanager.core.testresults import TestResultLogic, TestResultFileData;
83from testmanager.core.testresultfailures import TestResultFailureLogic, TestResultFailureData;
84from testmanager.core.useraccount import UserAccountLogic;
85from testmanager.config import g_ksSmtpHost, g_kcSmtpPort, g_ksAlertFrom, \
86 g_ksAlertSubject, g_asAlertList #, g_ksLomPassword;
87
88# Python 3 hacks:
89if sys.version_info[0] >= 3:
90 xrange = range; # pylint: disable=redefined-builtin,invalid-name
91
92
93class VirtualTestSheriffCaseFile(object):
94 """
95 A failure investigation case file.
96
97 """
98
99
100 ## Max log file we'll read into memory. (256 MB)
101 kcbMaxLogRead = 0x10000000;
102
103 def __init__(self, oSheriff, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase):
104 self.oSheriff = oSheriff;
105 self.oTestSet = oTestSet; # TestSetData
106 self.oTree = oTree; # TestResultDataEx
107 self.oBuild = oBuild; # BuildDataEx
108 self.oTestBox = oTestBox; # TestBoxData
109 self.oTestGroup = oTestGroup; # TestGroupData
110 self.oTestCase = oTestCase; # TestCaseDataEx
111 self.sMainLog = ''; # The main log file. Empty string if not accessible.
112 self.sSvcLog = ''; # The VBoxSVC log file. Empty string if not accessible.
113
114 # Generate a case file name.
115 self.sName = '#%u: %s' % (self.oTestSet.idTestSet, self.oTestCase.sName,)
116 self.sLongName = '#%u: "%s" on "%s" running %s %s (%s), "%s" by %s, using %s %s %s r%u' \
117 % ( self.oTestSet.idTestSet,
118 self.oTestCase.sName,
119 self.oTestBox.sName,
120 self.oTestBox.sOs,
121 self.oTestBox.sOsVersion,
122 self.oTestBox.sCpuArch,
123 self.oTestBox.sCpuName,
124 self.oTestBox.sCpuVendor,
125 self.oBuild.oCat.sProduct,
126 self.oBuild.oCat.sBranch,
127 self.oBuild.oCat.sType,
128 self.oBuild.iRevision, );
129
130 # Investigation notes.
131 self.tReason = None; # None or one of the ktReason_XXX constants.
132 self.dReasonForResultId = {}; # Reason assignments indexed by idTestResult.
133 self.dCommentForResultId = {}; # Comment assignments indexed by idTestResult.
134
135 #
136 # Reason.
137 #
138
139 def noteReason(self, tReason):
140 """ Notes down a possible reason. """
141 self.oSheriff.dprint(u'noteReason: %s -> %s' % (self.tReason, tReason,));
142 self.tReason = tReason;
143 return True;
144
145 def noteReasonForId(self, tReason, idTestResult, sComment = None):
146 """ Notes down a possible reason for a specific test result. """
147 self.oSheriff.dprint(u'noteReasonForId: %u: %s -> %s%s'
148 % (idTestResult, self.dReasonForResultId.get(idTestResult, None), tReason,
149 (u' (%s)' % (sComment,)) if sComment is not None else ''));
150 self.dReasonForResultId[idTestResult] = tReason;
151 if sComment is not None:
152 self.dCommentForResultId[idTestResult] = sComment;
153 return True;
154
155
156 #
157 # Test classification.
158 #
159
160 def isVBoxTest(self):
161 """ Test classification: VirtualBox (using the build) """
162 return self.oBuild.oCat.sProduct.lower() in [ 'virtualbox', 'vbox' ];
163
164 def isVBoxUnitTest(self):
165 """ Test case classification: The unit test doing all our testcase/*.cpp stuff. """
166 return self.isVBoxTest() \
167 and (self.oTestCase.sName.lower() == 'unit tests' or self.oTestCase.sName.lower().startswith('misc: unit tests'));
168
169 def isVBoxInstallTest(self):
170 """ Test case classification: VirtualBox Guest installation test. """
171 return self.isVBoxTest() \
172 and self.oTestCase.sName.lower().startswith('install:');
173
174 def isVBoxUnattendedInstallTest(self):
175 """ Test case classification: VirtualBox Guest installation test. """
176 return self.isVBoxTest() \
177 and self.oTestCase.sName.lower().startswith('uinstall:');
178
179 def isVBoxUSBTest(self):
180 """ Test case classification: VirtualBox USB test. """
181 return self.isVBoxTest() \
182 and self.oTestCase.sName.lower().startswith('usb:');
183
184 def isVBoxStorageTest(self):
185 """ Test case classification: VirtualBox Storage test. """
186 return self.isVBoxTest() \
187 and self.oTestCase.sName.lower().startswith('storage:');
188
189 def isVBoxGAsTest(self):
190 """ Test case classification: VirtualBox Guest Additions test. """
191 return self.isVBoxTest() \
192 and ( self.oTestCase.sName.lower().startswith('guest additions')
193 or self.oTestCase.sName.lower().startswith('ga\'s tests'));
194
195 def isVBoxAPITest(self):
196 """ Test case classification: VirtualBox API test. """
197 return self.isVBoxTest() \
198 and self.oTestCase.sName.lower().startswith('api:');
199
200 def isVBoxBenchmarkTest(self):
201 """ Test case classification: VirtualBox Benchmark test. """
202 return self.isVBoxTest() \
203 and self.oTestCase.sName.lower().startswith('benchmark:');
204
205 def isVBoxSmokeTest(self):
206 """ Test case classification: Smoke test. """
207 return self.isVBoxTest() \
208 and self.oTestCase.sName.lower().startswith('smoketest');
209
210 def isVBoxSerialTest(self):
211 """ Test case classification: Smoke test. """
212 return self.isVBoxTest() \
213 and self.oTestCase.sName.lower().startswith('serial:');
214
215
216 #
217 # Utility methods.
218 #
219
220 def getMainLog(self):
221 """
222 Tries to read the main log file since this will be the first source of information.
223 """
224 if self.sMainLog:
225 return self.sMainLog;
226 (oFile, oSizeOrError, _) = self.oTestSet.openFile('main.log', 'rb');
227 if oFile is not None:
228 try:
229 self.sMainLog = oFile.read(min(self.kcbMaxLogRead, oSizeOrError)).decode('utf-8', 'replace');
230 except Exception as oXcpt:
231 self.oSheriff.vprint(u'Error reading main log file: %s' % (oXcpt,))
232 self.sMainLog = '';
233 else:
234 self.oSheriff.vprint(u'Error opening main log file: %s' % (oSizeOrError,));
235 return self.sMainLog;
236
237 def getLogFile(self, oFile):
238 """
239 Tries to read the given file as a utf-8 log file.
240 oFile is a TestFileDataEx instance.
241 Returns empty string if problems opening or reading the file.
242 """
243 sContent = '';
244 (oFile, oSizeOrError, _) = self.oTestSet.openFile(oFile.sFile, 'rb');
245 if oFile is not None:
246 try:
247 sContent = oFile.read(min(self.kcbMaxLogRead, oSizeOrError)).decode('utf-8', 'replace');
248 except Exception as oXcpt:
249 self.oSheriff.vprint(u'Error reading the "%s" log file: %s' % (oFile.sFile, oXcpt,))
250 else:
251 self.oSheriff.vprint(u'Error opening the "%s" log file: %s' % (oFile.sFile, oSizeOrError,));
252 return sContent;
253
254 def getSvcLog(self):
255 """
256 Tries to read the VBoxSVC log file as it typically not associated with a failing test result.
257 Note! Returns the first VBoxSVC log file we find.
258 """
259 if not self.sSvcLog:
260 aoSvcLogFiles = self.oTree.getListOfLogFilesByKind(TestResultFileData.ksKind_LogReleaseSvc);
261 if aoSvcLogFiles:
262 self.sSvcLog = self.getLogFile(aoSvcLogFiles[0]);
263 return self.sSvcLog;
264
265 def getScreenshotSha256(self, oFile):
266 """
267 Tries to read the given screenshot file, uncompress it, and do SHA-2
268 on the raw pixels.
269 Returns SHA-2 digest string on success, None on failure.
270 """
271 (oImgFile, _, _) = self.oTestSet.openFile(oFile.sFile, 'rb');
272 try:
273 abImageFile = oImgFile.read();
274 except Exception as oXcpt:
275 self.oSheriff.vprint(u'Error reading the "%s" image file: %s' % (oFile.sFile, oXcpt,))
276 else:
277 try:
278 oImage = Image.open(BytesIO(abImageFile));
279 except Exception as oXcpt:
280 self.oSheriff.vprint(u'Error opening the "%s" image bytes using PIL.Image.open: %s' % (oFile.sFile, oXcpt,))
281 else:
282 try:
283 oHash = hashlib.sha256();
284 if hasattr(oImage, 'tobytes'):
285 oHash.update(oImage.tobytes());
286 else:
287 oHash.update(oImage.tostring()); # pylint: disable=no-member
288 except Exception as oXcpt:
289 self.oSheriff.vprint(u'Error hashing the uncompressed image bytes for "%s": %s' % (oFile.sFile, oXcpt,))
290 else:
291 return oHash.hexdigest();
292 return None;
293
294
295
296 def isSingleTestFailure(self):
297 """
298 Figure out if this is a single test failing or if it's one of the
299 more complicated ones.
300 """
301 if self.oTree.cErrors == 1:
302 return True;
303 if self.oTree.deepCountErrorContributers() <= 1:
304 return True;
305 return False;
306
307
308
309class VirtualTestSheriff(object): # pylint: disable=too-few-public-methods
310 """
311 Add build info into Test Manager database.
312 """
313
314 ## The user account for the virtual sheriff.
315 ksLoginName = 'vsheriff';
316
317 def __init__(self):
318 """
319 Parse command line.
320 """
321 self.oDb = None;
322 self.tsNow = None;
323 self.oTestResultLogic = None;
324 self.oTestSetLogic = None;
325 self.oFailureReasonLogic = None; # FailureReasonLogic;
326 self.oTestResultFailureLogic = None; # TestResultFailureLogic
327 self.oLogin = None;
328 self.uidSelf = -1;
329 self.oLogFile = None;
330 self.asBsodReasons = [];
331 self.asUnitTestReasons = [];
332
333 oParser = OptionParser();
334 oParser.add_option('--start-hours-ago', dest = 'cStartHoursAgo', metavar = '<hours>', default = 0, type = 'int',
335 help = 'When to start specified as hours relative to current time. Defauls is right now.', );
336 oParser.add_option('--hours-period', dest = 'cHoursBack', metavar = '<period-in-hours>', default = 2, type = 'int',
337 help = 'Work period specified in hours. Defauls is 2 hours.');
338 oParser.add_option('--real-run-back', dest = 'fRealRun', action = 'store_true', default = False,
339 help = 'Whether to commit the findings to the database. Default is a dry run.');
340 oParser.add_option('--testset', dest = 'aidTestSets', metavar = '<id>', default = [], type = 'int', action = 'append',
341 help = 'Only investigate this one. Accumulates IDs when repeated.');
342 oParser.add_option('-q', '--quiet', dest = 'fQuiet', action = 'store_true', default = False,
343 help = 'Quiet execution');
344 oParser.add_option('-l', '--log', dest = 'sLogFile', metavar = '<logfile>', default = None,
345 help = 'Where to log messages.');
346 oParser.add_option('--debug', dest = 'fDebug', action = 'store_true', default = False,
347 help = 'Enables debug mode.');
348
349 (self.oConfig, _) = oParser.parse_args();
350
351 if self.oConfig.sLogFile:
352 self.oLogFile = open(self.oConfig.sLogFile, "a"); # pylint: disable=consider-using-with,unspecified-encoding
353 self.oLogFile.write('VirtualTestSheriff: $Revision: 98103 $ \n');
354
355
356 def eprint(self, sText):
357 """
358 Prints error messages.
359 Returns 1 (for exit code usage.)
360 """
361 print('error: %s' % (sText,));
362 if self.oLogFile is not None:
363 if sys.version_info[0] >= 3:
364 self.oLogFile.write(u'error: %s\n' % (sText,));
365 else:
366 self.oLogFile.write((u'error: %s\n' % (sText,)).encode('utf-8'));
367 return 1;
368
369 def dprint(self, sText):
370 """
371 Prints debug info.
372 """
373 if self.oConfig.fDebug:
374 if not self.oConfig.fQuiet:
375 print('debug: %s' % (sText, ));
376 if self.oLogFile is not None:
377 if sys.version_info[0] >= 3:
378 self.oLogFile.write(u'debug: %s\n' % (sText,));
379 else:
380 self.oLogFile.write((u'debug: %s\n' % (sText,)).encode('utf-8'));
381 return 0;
382
383 def vprint(self, sText):
384 """
385 Prints verbose info.
386 """
387 if not self.oConfig.fQuiet:
388 print('info: %s' % (sText,));
389 if self.oLogFile is not None:
390 if sys.version_info[0] >= 3:
391 self.oLogFile.write(u'info: %s\n' % (sText,));
392 else:
393 self.oLogFile.write((u'info: %s\n' % (sText,)).encode('utf-8'));
394 return 0;
395
396 def getFailureReason(self, tReason):
397 """ Gets the failure reason object for tReason. """
398 return self.oFailureReasonLogic.cachedLookupByNameAndCategory(tReason[1], tReason[0]);
399
400 def selfCheck(self):
401 """ Does some self checks, looking up things we expect to be in the database and such. """
402 rcExit = 0;
403 for sAttr in dir(self.__class__):
404 if sAttr.startswith('ktReason_'):
405 tReason = getattr(self.__class__, sAttr);
406 oFailureReason = self.getFailureReason(tReason);
407 if oFailureReason is None:
408 rcExit = self.eprint(u'Failed to find failure reason "%s" in category "%s" in the database!'
409 % (tReason[1], tReason[0],));
410
411 # Check the user account as well.
412 if self.oLogin is None:
413 oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
414 if oLogin is None:
415 rcExit = self.eprint(u'Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
416 return rcExit;
417
418 def sendEmailAlert(self, uidAuthor, sBodyText):
419 """
420 Sends email alert.
421 """
422
423 # Get author email
424 self.oDb.execute('SELECT sEmail FROM Users WHERE uid=%s', (uidAuthor,));
425 sFrom = self.oDb.fetchOne();
426 if sFrom is not None:
427 sFrom = sFrom[0];
428 else:
429 sFrom = g_ksAlertFrom;
430
431 # Gather recipient list.
432 asEmailList = [];
433 for sUser in g_asAlertList:
434 self.oDb.execute('SELECT sEmail FROM Users WHERE sUsername=%s', (sUser,));
435 sEmail = self.oDb.fetchOne();
436 if sEmail:
437 asEmailList.append(sEmail[0]);
438 if not asEmailList:
439 return self.eprint('No email addresses to send alter to!');
440
441 # Compose the message.
442 oMsg = MIMEMultipart();
443 oMsg['From'] = sFrom;
444 oMsg['To'] = COMMASPACE.join(asEmailList);
445 oMsg['Subject'] = g_ksAlertSubject;
446 oMsg.attach(MIMEText(sBodyText, 'plain'))
447
448 # Try send it.
449 try:
450 oSMTP = smtplib.SMTP(g_ksSmtpHost, g_kcSmtpPort);
451 oSMTP.sendmail(sFrom, asEmailList, oMsg.as_string())
452 oSMTP.quit()
453 except smtplib.SMTPException as oXcpt:
454 return self.eprint('Failed to send mail: %s' % (oXcpt,));
455
456 return 0;
457
458 def badTestBoxManagement(self):
459 """
460 Looks for bad test boxes and first tries once to reboot them then disables them.
461 """
462 rcExit = 0;
463
464 #
465 # We skip this entirely if we're running in the past and not in harmless debug mode.
466 #
467 if self.oConfig.cStartHoursAgo != 0 \
468 and (not self.oConfig.fDebug or self.oConfig.fRealRun):
469 return rcExit;
470 tsNow = self.tsNow if self.oConfig.fDebug else None;
471 cHoursBack = self.oConfig.cHoursBack if self.oConfig.fDebug else 2;
472 oTestBoxLogic = TestBoxLogic(self.oDb);
473
474 #
475 # Generate a list of failures reasons we consider bad-testbox behavior.
476 #
477 aidFailureReasons = [
478 self.getFailureReason(self.ktReason_Host_DriverNotLoaded).idFailureReason,
479 self.getFailureReason(self.ktReason_Host_DriverNotUnloading).idFailureReason,
480 self.getFailureReason(self.ktReason_Host_DriverNotCompilable).idFailureReason,
481 self.getFailureReason(self.ktReason_Host_InstallationFailed).idFailureReason,
482 ];
483
484 #
485 # Get list of bad test boxes for given period and check them out individually.
486 #
487 aidBadTestBoxes = self.oTestSetLogic.fetchBadTestBoxIds(cHoursBack = cHoursBack, tsNow = tsNow,
488 aidFailureReasons = aidFailureReasons);
489 for idTestBox in aidBadTestBoxes:
490 # Skip if the testbox is already disabled or has a pending reboot command.
491 try:
492 oTestBox = TestBoxData().initFromDbWithId(self.oDb, idTestBox);
493 except Exception as oXcpt:
494 rcExit = self.eprint('Failed to get data for test box #%u in badTestBoxManagement: %s' % (idTestBox, oXcpt,));
495 continue;
496 if not oTestBox.fEnabled:
497 self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has been disabled already.'
498 % ( idTestBox, oTestBox.sName, ));
499 continue;
500 if oTestBox.enmPendingCmd != TestBoxData.ksTestBoxCmd_None:
501 self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has a command pending: %s'
502 % ( idTestBox, oTestBox.sName, oTestBox.enmPendingCmd));
503 continue;
504
505 # Get the most recent testsets for this box (descending on tsDone) and see how bad it is.
506 aoSets = self.oTestSetLogic.fetchSetsForTestBox(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow);
507 cOkay = 0;
508 cBad = 0;
509 iFirstOkay = len(aoSets);
510 for iSet, oSet in enumerate(aoSets):
511 if oSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
512 cBad += 1;
513 else:
514 # Check for bad failure reasons.
515 oFailure = None;
516 if oSet.enmStatus in TestSetData.kasBadTestStatuses:
517 (oTree, _ ) = self.oTestResultLogic.fetchResultTree(oSet.idTestSet)
518 aoFailedResults = oTree.getListOfFailures();
519 for oFailedResult in aoFailedResults:
520 oFailure = self.oTestResultFailureLogic.getById(oFailedResult.idTestResult);
521 if oFailure is not None and oFailure.idFailureReason in aidFailureReasons:
522 break;
523 oFailure = None;
524 if oFailure is not None:
525 cBad += 1;
526 else:
527 # This is an okay test result then.
528 ## @todo maybe check the elapsed time here, it could still be a bad run?
529 cOkay += 1;
530 iFirstOkay = min(iFirstOkay, iSet);
531 if iSet > 10:
532 break;
533
534 # We react if there are two or more bad-testbox statuses at the head of the
535 # history and at least three in the last 10 results.
536 if iFirstOkay >= 2 and cBad > 2:
537 if oTestBoxLogic.hasTestBoxRecentlyBeenRebooted(idTestBox, cHoursBack = cHoursBack, tsNow = tsNow):
538 sComment = u'Disabling testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u' \
539 % (idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay);
540 self.vprint(sComment);
541 self.sendEmailAlert(self.uidSelf, sComment);
542 if self.oConfig.fRealRun is True:
543 try:
544 oTestBoxLogic.disableTestBox(idTestBox, self.uidSelf, fCommit = True,
545 sComment = 'Automatically disabled (iFirstOkay=%u cBad=%u cOkay=%u)'
546 % (iFirstOkay, cBad, cOkay),);
547 except Exception as oXcpt:
548 rcExit = self.eprint(u'Error disabling testbox #%u (%u): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
549 else:
550 sComment = u'Rebooting testbox #%u (%s) - iFirstOkay=%u cBad=%u cOkay=%u' \
551 % (idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay);
552 self.vprint(sComment);
553 self.sendEmailAlert(self.uidSelf, sComment);
554 if self.oConfig.fRealRun is True:
555 try:
556 oTestBoxLogic.rebootTestBox(idTestBox, self.uidSelf, fCommit = True,
557 sComment = 'Automatically rebooted (iFirstOkay=%u cBad=%u cOkay=%u)'
558 % (iFirstOkay, cBad, cOkay),);
559 except Exception as oXcpt:
560 rcExit = self.eprint(u'Error rebooting testbox #%u (%s): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
561 else:
562 self.dprint(u'badTestBoxManagement: #%u (%s) looks ok: iFirstOkay=%u cBad=%u cOkay=%u'
563 % ( idTestBox, oTestBox.sName, iFirstOkay, cBad, cOkay));
564
565 ## @todo r=bird: review + rewrite;
566 ## - no selecting here, that belongs in the core/*.py files.
567 ## - preserve existing comments.
568 ## - doing way too much in the try/except block.
569 ## - No password quoting in the sshpass command that always fails (127).
570 ## - Timeout is way to low. testboxmem1 need more than 10 min to take a dump, ages to
571 ## get thru POST and another 5 just to time out in grub. Should be an hour or so.
572 ## Besides, it need to be constant elsewhere in the file, not a variable here.
573 ##
574 ##
575 ## Reset hanged testboxes
576 ##
577 #cStatusTimeoutMins = 10;
578 #
579 #self.oDb.execute('SELECT TestBoxStatuses.idTestBox\n'
580 # ' FROM TestBoxStatuses, TestBoxes\n'
581 # ' WHERE TestBoxStatuses.tsUpdated >= (CURRENT_TIMESTAMP - interval \'%s hours\')\n'
582 # ' AND TestBoxStatuses.tsUpdated < (CURRENT_TIMESTAMP - interval \'%s minutes\')\n'
583 # ' AND TestBoxStatuses.idTestBox = TestBoxes.idTestBox\n'
584 # ' AND Testboxes.tsExpire = \'infinity\'::timestamp', (cHoursBack,cStatusTimeoutMins));
585 #for idTestBox in self.oDb.fetchAll():
586 # idTestBox = idTestBox[0];
587 # try:
588 # oTestBox = TestBoxData().initFromDbWithId(self.oDb, idTestBox);
589 # except Exception as oXcpt:
590 # rcExit = self.eprint('Failed to get data for test box #%u in badTestBoxManagement: %s' % (idTestBox, oXcpt,));
591 # continue;
592 # # Skip if the testbox is already disabled, already reset or there's no iLOM
593 # if not oTestBox.fEnabled or oTestBox.ipLom is None or oTestBox.sComment is not None and oTestBox.sComment.find('Automatically reset') >= 0:
594 # self.dprint(u'badTestBoxManagement: Skipping test box #%u (%s) as it has been disabled already.'
595 # % ( idTestBox, oTestBox.sName, ));
596 # continue;
597 # ## @todo get iLOM credentials from a table?
598 # sCmd = 'sshpass -p%s ssh -oStrictHostKeyChecking=no root@%s show /SP && reset /SYS' % (g_ksLomPassword, oTestBox.ipLom,);
599 # try:
600 # oPs = subprocess.Popen(sCmd, stdout=subprocess.PIPE, shell=True);
601 # sStdout = oPs.communicate()[0];
602 # iRC = oPs.wait();
603 #
604 # oTestBox.sComment = 'Automatically reset (iRC=%u sStdout=%s)' % (iRC, sStdout,);
605 # oTestBoxLogic.editEntry(oTestBox, self.uidSelf, fCommit = True);
606 #
607 # sComment = u'Reset testbox #%u (%s) - iRC=%u sStduot=%s' % ( idTestBox, oTestBox.sName, iRC, sStdout);
608 # self.vprint(sComment);
609 # self.sendEmailAlert(self.uidSelf, sComment);
610 #
611 # except Exception as oXcpt:
612 # rcExit = self.eprint(u'Error resetting testbox #%u (%s): %s\n' % (idTestBox, oTestBox.sName, oXcpt,));
613 #
614 return rcExit;
615
616
617 ## @name Failure reasons we know.
618 ## @{
619
620 ktReason_Add_Installer_Win_Failed = ( 'Additions', 'Win GA install' );
621 ktReason_Add_ShFl_Automount = ( 'Additions', 'Automounting' );
622 ktReason_Add_ShFl_FsPerf = ( 'Additions', 'FsPerf' );
623 ktReason_Add_ShFl_FsPerf_Abend = ( 'Additions', 'FsPerf abend' );
624 ktReason_Add_GstCtl_Preparations = ( 'Additions', 'GstCtl preparations' );
625 ktReason_Add_GstCtl_SessionBasics = ( 'Additions', 'Session basics' );
626 ktReason_Add_GstCtl_SessionProcRefs = ( 'Additions', 'Session process' );
627 ktReason_Add_GstCtl_Session_Reboot = ( 'Additions', 'Session reboot' );
628 ktReason_Add_GstCtl_CopyFromGuest_Timeout = ( 'Additions', 'CopyFromGuest timeout' );
629 ktReason_Add_GstCtl_CopyToGuest_Timeout = ( 'Additions', 'CopyToGuest timeout' );
630 ktReason_Add_GstCtl_CopyToGuest_DstEmpty = ( 'Additions', 'CopyToGuest dst empty' );
631 ktReason_Add_GstCtl_CopyToGuest_DstExists = ( 'Additions', 'CopyToGuest dst exists' );
632 ktReason_Add_FlushViewOfFile = ( 'Additions', 'FlushViewOfFile' );
633 ktReason_Add_Mmap_Coherency = ( 'Additions', 'mmap coherency' );
634 ktReason_BSOD_Recovery = ( 'BSOD', 'Recovery' );
635 ktReason_BSOD_Automatic_Repair = ( 'BSOD', 'Automatic Repair' );
636 ktReason_BSOD_0000007F = ( 'BSOD', '0x0000007F' );
637 ktReason_BSOD_000000D1 = ( 'BSOD', '0x000000D1' );
638 ktReason_BSOD_C0000225 = ( 'BSOD', '0xC0000225 (boot)' );
639 ktReason_Guru_Generic = ( 'Guru Meditations', 'Generic Guru Meditation' );
640 ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_INSTR_NOT_IMPLEMENTED' );
641 ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED = ( 'Guru Meditations', 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' );
642 ktReason_Guru_VERR_TRPM_DONT_PANIC = ( 'Guru Meditations', 'VERR_TRPM_DONT_PANIC' );
643 ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED = ( 'Guru Meditations', 'VERR_PGM_PHYS_PAGE_RESERVED' );
644 ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE = ( 'Guru Meditations', 'VERR_VMX_INVALID_GUEST_STATE' );
645 ktReason_Guru_VINF_EM_TRIPLE_FAULT = ( 'Guru Meditations', 'VINF_EM_TRIPLE_FAULT' );
646 ktReason_Host_HostMemoryLow = ( 'Host', 'HostMemoryLow' );
647 ktReason_Host_DriverNotLoaded = ( 'Host', 'Driver not loaded' );
648 ktReason_Host_DriverNotUnloading = ( 'Host', 'Driver not unloading' );
649 ktReason_Host_DriverNotCompilable = ( 'Host', 'Driver not compilable' );
650 ktReason_Host_InstallationFailed = ( 'Host', 'Installation failed' );
651 ktReason_Host_InstallationWantReboot = ( 'Host', 'Installation want reboot' );
652 ktReason_Host_InvalidPackage = ( 'Host', 'ERROR_INSTALL_PACKAGE_INVALID' );
653 ktReason_Host_InstallSourceAbsent = ( 'Host', 'ERROR_INSTALL_SOURCE_ABSENT' );
654 ktReason_Host_NotSignedWithBuildCert = ( 'Host', 'Not signed with build cert' );
655 ktReason_Host_DiskFull = ( 'Host', 'Host disk full' );
656 ktReason_Host_DoubleFreeHeap = ( 'Host', 'Double free or corruption' );
657 ktReason_Host_LeftoverService = ( 'Host', 'Leftover service' );
658 ktReason_Host_win32com_gen_py = ( 'Host', 'win32com.gen_py' );
659 ktReason_Host_Reboot_OSX_Watchdog_Timeout = ( 'Host Reboot', 'OSX Watchdog Timeout' );
660 ktReason_Host_Modprobe_Failed = ( 'Host', 'Modprobe failed' );
661 ktReason_Host_Install_Hang = ( 'Host', 'Install hang' );
662 ktReason_Host_NetworkMisconfiguration = ( 'Host', 'Network misconfiguration' );
663 ktReason_Host_TSTInfo_Accuracy_OOR = ( 'Host', 'TSTInfo accuracy out of range' );
664 ktReason_Networking_Nonexistent_host_nic = ( 'Networking', 'Nonexistent host networking interface' );
665 ktReason_Networking_VERR_INTNET_FLT_IF_NOT_FOUND = ( 'Networking', 'VERR_INTNET_FLT_IF_NOT_FOUND' );
666 ktReason_OSInstall_GRUB_hang = ( 'O/S Install', 'GRUB hang' );
667 ktReason_OSInstall_Udev_hang = ( 'O/S Install', 'udev hang' );
668 ktReason_OSInstall_Sata_no_BM = ( 'O/S Install', 'SATA busmaster bit not set' );
669 ktReason_Panic_BootManagerC000000F = ( 'Panic', 'Hardware Changed' );
670 ktReason_Panic_MP_BIOS_IO_APIC = ( 'Panic', 'MP-BIOS/IO-APIC' );
671 ktReason_Panic_HugeMemory = ( 'Panic', 'Huge memory assertion' );
672 ktReason_Panic_IOAPICDoesntWork = ( 'Panic', 'IO-APIC and timer does not work' );
673 ktReason_Panic_TxUnitHang = ( 'Panic', 'Tx Unit Hang' );
674 ktReason_API_std_bad_alloc = ( 'API / (XP)COM', 'std::bad_alloc' );
675 ktReason_API_Digest_Mismatch = ( 'API / (XP)COM', 'Digest mismatch' );
676 ktReason_API_MoveVM_SharingViolation = ( 'API / (XP)COM', 'MoveVM sharing violation' );
677 ktReason_API_MoveVM_InvalidParameter = ( 'API / (XP)COM', 'MoveVM invalid parameter' );
678 ktReason_API_Open_Session_Failed = ( 'API / (XP)COM', 'Open session failed' );
679 ktReason_XPCOM_Exit_Minus_11 = ( 'API / (XP)COM', 'exit -11' );
680 ktReason_XPCOM_VBoxSVC_Hang = ( 'API / (XP)COM', 'VBoxSVC hang' );
681 ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption = ( 'API / (XP)COM', 'VBoxSVC hang + heap corruption' );
682 ktReason_XPCOM_NS_ERROR_CALL_FAILED = ( 'API / (XP)COM', 'NS_ERROR_CALL_FAILED' );
683 ktReason_BootManager_Image_corrupt = ( 'Unknown', 'BOOTMGR Image corrupt' );
684 ktReason_Unknown_Heap_Corruption = ( 'Unknown', 'Heap corruption' );
685 ktReason_Unknown_Reboot_Loop = ( 'Unknown', 'Reboot loop' );
686 ktReason_Unknown_File_Not_Found = ( 'Unknown', 'File not found' );
687 ktReason_Unknown_HalReturnToFirmware = ( 'Unknown', 'HalReturnToFirmware' );
688 ktReason_Unknown_VM_Crash = ( 'Unknown', 'VM crash' );
689 ktReason_Unknown_VM_Terminated = ( 'Unknown', 'VM terminated' );
690 ktReason_Unknown_VM_Start_Error = ( 'Unknown', 'VM Start Error' );
691 ktReason_Unknown_VM_Runtime_Error = ( 'Unknown', 'VM Runtime Error' );
692 ktReason_VMM_kvm_lock_spinning = ( 'VMM', 'kvm_lock_spinning' );
693 ktReason_Ignore_Buggy_Test_Driver = ( 'Ignore', 'Buggy test driver' );
694 ktReason_Ignore_Stale_Files = ( 'Ignore', 'Stale files' );
695 ktReason_Buggy_Build_Broken_Build = ( 'Broken Build', 'Buggy build' );
696 ktReason_GuestBug_CompizVBoxQt = ( 'Guest Bug', 'Compiz + VirtualBox Qt GUI crash' );
697 ## @}
698
699 ## BSOD category.
700 ksBsodCategory = 'BSOD';
701 ## Special reason indicating that the flesh and blood sheriff has work to do.
702 ksBsodAddNew = 'Add new BSOD';
703
704 ## Unit test category.
705 ksUnitTestCategory = 'Unit';
706 ## Special reason indicating that the flesh and blood sheriff has work to do.
707 ksUnitTestAddNew = 'Add new';
708
709 ## Used for indica that we shouldn't report anything for this test result ID and
710 ## consider promoting the previous error to test set level if it's the only one.
711 ktHarmless = ( 'Probably', 'Caused by previous error' );
712
713
714 def caseClosed(self, oCaseFile):
715 """
716 Reports the findings in the case and closes it.
717 """
718 #
719 # Log it and create a dReasonForReasultId we can use below.
720 #
721 dCommentForResultId = oCaseFile.dCommentForResultId;
722 if oCaseFile.dReasonForResultId:
723 # Must weed out ktHarmless.
724 dReasonForResultId = {};
725 for idKey, tReason in oCaseFile.dReasonForResultId.items():
726 if tReason is not self.ktHarmless:
727 dReasonForResultId[idKey] = tReason;
728 if not dReasonForResultId:
729 self.vprint(u'TODO: Closing %s without a real reason, only %s.'
730 % (oCaseFile.sName, oCaseFile.dReasonForResultId));
731 return False;
732
733 # Try promote to single reason.
734 atValues = dReasonForResultId.values();
735 fSingleReason = True;
736 if len(dReasonForResultId) == 1 and next(iter(dReasonForResultId.keys())) != oCaseFile.oTestSet.idTestResult:
737 self.dprint(u'Promoting single reason to whole set: %s' % (next(iter(atValues)),));
738 elif len(dReasonForResultId) > 1 and len(atValues) == list(atValues).count(next(iter(atValues))):
739 self.dprint(u'Merged %d reasons to a single one: %s' % (len(atValues), next(iter(atValues))));
740 else:
741 fSingleReason = False;
742 if fSingleReason:
743 dReasonForResultId = { oCaseFile.oTestSet.idTestResult: next(iter(atValues)), };
744 if dCommentForResultId:
745 dCommentForResultId = { oCaseFile.oTestSet.idTestResult: next(iter(dCommentForResultId.values())), };
746 elif oCaseFile.tReason is not None:
747 dReasonForResultId = { oCaseFile.oTestSet.idTestResult: oCaseFile.tReason, };
748 else:
749 self.vprint(u'Closing %s without a reason - this should not happen!' % (oCaseFile.sName,));
750 return False;
751
752 self.vprint(u'Closing %s with following reason%s: %s'
753 % ( oCaseFile.sName, 's' if len(dReasonForResultId) > 1 else '', dReasonForResultId, ));
754
755 #
756 # Add the test failure reason record(s).
757 #
758 for idTestResult, tReason in dReasonForResultId.items():
759 oFailureReason = self.getFailureReason(tReason);
760 if oFailureReason is not None:
761 sComment = 'Set by $Revision: 98103 $' # Handy for reverting later.
762 if idTestResult in dCommentForResultId:
763 sComment += ': ' + dCommentForResultId[idTestResult];
764
765 oAdd = TestResultFailureData();
766 oAdd.initFromValues(idTestResult = idTestResult,
767 idFailureReason = oFailureReason.idFailureReason,
768 uidAuthor = self.uidSelf,
769 idTestSet = oCaseFile.oTestSet.idTestSet,
770 sComment = sComment,);
771 if self.oConfig.fRealRun:
772 try:
773 self.oTestResultFailureLogic.addEntry(oAdd, self.uidSelf, fCommit = True);
774 except Exception as oXcpt:
775 self.eprint(u'caseClosed: Exception "%s" while adding reason %s for %s'
776 % (oXcpt, oAdd, oCaseFile.sLongName,));
777 else:
778 self.eprint(u'caseClosed: Cannot locate failure reason: %s / %s' % ( tReason[0], tReason[1],));
779 return True;
780
781 #
782 # Tools for assiting log parsing.
783 #
784
785 @staticmethod
786 def matchFollowedByLines(sStr, off, asFollowingLines):
787 """ Worker for isThisFollowedByTheseLines. """
788
789 # Advance off to the end of the line.
790 off = sStr.find('\n', off);
791 if off < 0:
792 return False;
793 off += 1;
794
795 # Match each string with the subsequent lines.
796 for iLine, sLine in enumerate(asFollowingLines):
797 offEnd = sStr.find('\n', off);
798 if offEnd < 0:
799 return iLine + 1 == len(asFollowingLines) and sStr.find(sLine, off) < 0;
800 if sLine and sStr.find(sLine, off, offEnd) < 0:
801 return False;
802
803 # next line.
804 off = offEnd + 1;
805
806 return True;
807
808 @staticmethod
809 def isThisFollowedByTheseLines(sStr, sFirst, asFollowingLines):
810 """
811 Looks for a line contining sFirst which is then followed by lines
812 with the strings in asFollowingLines. (No newline chars anywhere!)
813 Returns True / False.
814 """
815 off = sStr.find(sFirst, 0);
816 while off >= 0:
817 if VirtualTestSheriff.matchFollowedByLines(sStr, off, asFollowingLines):
818 return True;
819 off = sStr.find(sFirst, off + 1);
820 return False;
821
822 @staticmethod
823 def findAndReturnRestOfLine(sHaystack, sNeedle):
824 """
825 Looks for sNeedle in sHaystack.
826 Returns The text following the needle up to the end of the line.
827 Returns None if not found.
828 """
829 if sHaystack is None:
830 return None;
831 off = sHaystack.find(sNeedle);
832 if off < 0:
833 return None;
834 off += len(sNeedle)
835 offEol = sHaystack.find('\n', off);
836 if offEol < 0:
837 offEol = len(sHaystack);
838 return sHaystack[off:offEol]
839
840 @staticmethod
841 def findInAnyAndReturnRestOfLine(asHaystacks, sNeedle):
842 """
843 Looks for sNeedle in zeroe or more haystacks (asHaystack).
844 Returns The text following the first needed found up to the end of the line.
845 Returns None if not found.
846 """
847 for sHaystack in asHaystacks:
848 sRet = VirtualTestSheriff.findAndReturnRestOfLine(sHaystack, sNeedle);
849 if sRet is not None:
850 return sRet;
851 return None;
852
853
854 #
855 # The investigative units.
856 #
857
858 katSimpleInstallUninstallMainLogReasons = [
859 # ( Whether to stop on hit, reason tuple, needle text. )
860 ( False, ktReason_Host_LeftoverService,
861 'SERVICE_NAME: vbox' ),
862 ( False, ktReason_Host_LeftoverService,
863 'Seems installation was skipped. Old version lurking behind? Not the fault of this build/test run!'),
864 ];
865
866 kdatSimpleInstallUninstallMainLogReasonsPerOs = {
867 'darwin': [
868 # ( Whether to stop on hit, reason tuple, needle text. )
869 ( True, ktReason_Host_DriverNotUnloading,
870 'Can\'t remove kext org.virtualbox.kext.VBoxDrv; services failed to terminate - 0xe00002c7' ),
871 ],
872 'linux': [
873 # ( Whether to stop on hit, reason tuple, needle text. )
874 ( True, ktReason_Host_DriverNotCompilable,
875 'This system is not currently set up to build kernel modules' ),
876 ( True, ktReason_Host_DriverNotCompilable,
877 'This system is currently not set up to build kernel modules' ),
878 ( True, ktReason_Host_InstallationFailed,
879 'vboxdrv.sh: failed: Look at /var/log/vbox-install.log to find out what went wrong.' ),
880 ( True, ktReason_Host_DriverNotUnloading,
881 'Cannot unload module vboxdrv'),
882 ],
883 'solaris': [
884 # ( Whether to stop on hit, reason tuple, needle text. )
885 ( True, ktReason_Host_DriverNotUnloading, 'can\'t unload the module: Device busy' ),
886 ( True, ktReason_Host_DriverNotUnloading, 'Unloading: Host module ...FAILED!' ),
887 ( True, ktReason_Host_DriverNotUnloading, 'Unloading: NetFilter (Crossbow) module ...FAILED!' ),
888 ( True, ktReason_Host_InstallationFailed, 'svcadm: Couldn\'t bind to svc.configd.' ),
889 ( True, ktReason_Host_InstallationFailed, 'pkgadd: ERROR: postinstall script did not complete successfully' ),
890 ],
891 'win': [
892 # ( Whether to stop on hit, reason tuple, needle text. )
893 ( True, ktReason_Host_InstallationWantReboot, 'ERROR_SUCCESS_REBOOT_REQUIRED' ),
894 ( False, ktReason_Host_InstallationFailed, 'Installation error.' ),
895 ( True, ktReason_Host_InvalidPackage, 'Uninstaller failed, exit code: 1620' ),
896 ( True, ktReason_Host_InstallSourceAbsent, 'Uninstaller failed, exit code: 1612' ),
897 ],
898 };
899
900
901 def investigateInstallUninstallFailure(self, oCaseFile, oFailedResult, sResultLog, fInstall):
902 """
903 Investigates an install or uninstall failure.
904
905 We lump the two together since the installation typically also performs
906 an uninstall first and will be seeing similar issues to the uninstall.
907 """
908 self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
909
910 if fInstall and oFailedResult.enmStatus == TestSetData.ksTestStatus_TimedOut:
911 oCaseFile.noteReasonForId(self.ktReason_Host_Install_Hang, oFailedResult.idTestResult)
912 return True;
913
914 atSimple = self.katSimpleInstallUninstallMainLogReasons;
915 if oCaseFile.oTestBox.sOs in self.kdatSimpleInstallUninstallMainLogReasonsPerOs:
916 atSimple = self.kdatSimpleInstallUninstallMainLogReasonsPerOs[oCaseFile.oTestBox.sOs] + atSimple;
917
918 fFoundSomething = False;
919 for fStopOnHit, tReason, sNeedle in atSimple:
920 if sResultLog.find(sNeedle) > 0:
921 oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
922 if fStopOnHit:
923 return True;
924 fFoundSomething = True;
925
926 return fFoundSomething if fFoundSomething else None;
927
928
929 def investigateBadTestBox(self, oCaseFile):
930 """
931 Checks out bad-testbox statuses.
932 """
933 _ = oCaseFile;
934 return False;
935
936
937 def investigateVBoxUnitTest(self, oCaseFile):
938 """
939 Checks out a VBox unittest problem.
940 """
941
942 #
943 # Process simple test case failures first, using their name as reason.
944 # We do the reason management just like for BSODs.
945 #
946 cRelevantOnes = 0;
947 sMainLog = oCaseFile.getMainLog();
948 aoFailedResults = oCaseFile.oTree.getListOfFailures();
949 for oFailedResult in aoFailedResults:
950 if oFailedResult is oCaseFile.oTree:
951 self.vprint('TODO: toplevel failure');
952 cRelevantOnes += 1
953
954 elif oFailedResult.sName == 'Installing VirtualBox':
955 sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
956 self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
957 cRelevantOnes += 1
958
959 elif oFailedResult.sName == 'Uninstalling VirtualBox':
960 sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
961 self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
962 cRelevantOnes += 1
963
964 elif oFailedResult.oParent is not None:
965 # Get the 2nd level node because that's where we'll find the unit test name.
966 while oFailedResult.oParent.oParent is not None:
967 oFailedResult = oFailedResult.oParent;
968
969 # Only report a failure once.
970 if oFailedResult.idTestResult not in oCaseFile.dReasonForResultId:
971 sKey = oFailedResult.sName;
972 if sKey.startswith('testcase/'):
973 sKey = sKey[9:];
974 if sKey in self.asUnitTestReasons:
975 tReason = ( self.ksUnitTestCategory, sKey );
976 oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
977 else:
978 self.dprint(u'Unit test failure "%s" not found in %s;' % (sKey, self.asUnitTestReasons));
979 tReason = ( self.ksUnitTestCategory, self.ksUnitTestAddNew );
980 oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sKey);
981 cRelevantOnes += 1
982 else:
983 self.vprint(u'Internal error: expected oParent to NOT be None for %s' % (oFailedResult,));
984
985 #
986 # If we've caught all the relevant ones by now, report the result.
987 #
988 if len(oCaseFile.dReasonForResultId) >= cRelevantOnes:
989 return self.caseClosed(oCaseFile);
990 return False;
991
992 def extractGuestCpuStack(self, sInfoText):
993 """
994 Extracts the guest CPU stacks from the input file.
995
996 Returns a dictionary keyed by the CPU number, value being a list of
997 raw stack lines (no header).
998 Returns empty dictionary if no stacks where found.
999 """
1000 dRet = {};
1001 off = 0;
1002 while True:
1003 # Find the stack.
1004 offStart = sInfoText.find('=== start guest stack VCPU ', off);
1005 if offStart < 0:
1006 break;
1007 offEnd = sInfoText.find('=== end guest stack', offStart + 20);
1008 if offEnd >= 0:
1009 offEnd += 3;
1010 else:
1011 offEnd = sInfoText.find('=== start guest stack VCPU', offStart + 20);
1012 if offEnd < 0:
1013 offEnd = len(sInfoText);
1014
1015 sStack = sInfoText[offStart : offEnd];
1016 sStack = sStack.replace('\r',''); # paranoia
1017 asLines = sStack.split('\n');
1018
1019 # Figure the CPU.
1020 asWords = asLines[0].split();
1021 if len(asWords) < 6 or not asWords[5].isdigit():
1022 break;
1023 iCpu = int(asWords[5]);
1024
1025 # Add it and advance.
1026 dRet[iCpu] = [sLine.rstrip() for sLine in asLines[2:-1]]
1027 off = offEnd;
1028 return dRet;
1029
1030 def investigateInfoKvmLockSpinning(self, oCaseFile, sInfoText, dLogs):
1031 """ Investigates kvm_lock_spinning deadlocks """
1032 #
1033 # Extract the stacks. We need more than one CPU to create a deadlock.
1034 #
1035 dStacks = self.extractGuestCpuStack(sInfoText);
1036 self.dprint('kvm_lock_spinning: found %s stacks' % (len(dStacks),));
1037 if len(dStacks) >= 2:
1038 #
1039 # Examin each of the stacks. Each must have kvm_lock_spinning in
1040 # one of the first three entries.
1041 #
1042 cHits = 0;
1043 for asBacktrace in dStacks.values():
1044 for iFrame in xrange(min(3, len(asBacktrace))):
1045 if asBacktrace[iFrame].find('kvm_lock_spinning') >= 0:
1046 cHits += 1;
1047 break;
1048 self.dprint('kvm_lock_spinning: %s/%s hits' % (cHits, len(dStacks),));
1049 if cHits == len(dStacks):
1050 return (True, self.ktReason_VMM_kvm_lock_spinning);
1051
1052 _ = dLogs; _ = oCaseFile;
1053 return (False, None);
1054
1055 def investigateInfoHalReturnToFirmware(self, oCaseFile, sInfoText, dLogs):
1056 """ Investigates HalReturnToFirmware hangs """
1057 del oCaseFile
1058 del sInfoText
1059 del dLogs
1060 # hope that's sufficient
1061 return (True, self.ktReason_Unknown_HalReturnToFirmware);
1062
1063 ## Things we search a main or VM log for to figure out why something went bust.
1064 ## @note DO NOT ADD MORE STUFF HERE!
1065 ## Please use katSimpleMainLogReasons and katSimpleVmLogReasons instead!
1066 katSimpleMainAndVmLogReasonsDeprecated = [
1067 # ( Whether to stop on hit, reason tuple, needle text. )
1068 ( False, ktReason_Guru_Generic, 'GuruMeditation' ),
1069 ( False, ktReason_Guru_Generic, 'Guru Meditation' ),
1070 ( True, ktReason_Guru_VERR_IEM_INSTR_NOT_IMPLEMENTED, 'VERR_IEM_INSTR_NOT_IMPLEMENTED' ),
1071 ( True, ktReason_Guru_VERR_IEM_ASPECT_NOT_IMPLEMENTED, 'VERR_IEM_ASPECT_NOT_IMPLEMENTED' ),
1072 ( True, ktReason_Guru_VERR_TRPM_DONT_PANIC, 'VERR_TRPM_DONT_PANIC' ),
1073 ( True, ktReason_Guru_VERR_PGM_PHYS_PAGE_RESERVED, 'VERR_PGM_PHYS_PAGE_RESERVED' ),
1074 ( True, ktReason_Guru_VERR_VMX_INVALID_GUEST_STATE, 'VERR_VMX_INVALID_GUEST_STATE' ),
1075 ( True, ktReason_Guru_VINF_EM_TRIPLE_FAULT, 'VINF_EM_TRIPLE_FAULT' ),
1076 ( True, ktReason_Networking_Nonexistent_host_nic,
1077 'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
1078 ( True, ktReason_Networking_VERR_INTNET_FLT_IF_NOT_FOUND,
1079 'Failed to attach the network LUN (VERR_INTNET_FLT_IF_NOT_FOUND)' ),
1080 ( True, ktReason_Host_Reboot_OSX_Watchdog_Timeout, ': "OSX Watchdog Timeout: ' ),
1081 ( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
1082 'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
1083 ( True, ktReason_API_std_bad_alloc, 'Unexpected exception: std::bad_alloc' ),
1084 ( True, ktReason_Host_HostMemoryLow, 'HostMemoryLow' ),
1085 ( True, ktReason_Host_HostMemoryLow, 'Failed to procure handy pages; rc=VERR_NO_MEMORY' ),
1086 ( True, ktReason_Unknown_File_Not_Found,
1087 'Error: failed to start machine. Error message: File not found. (VERR_FILE_NOT_FOUND)' ),
1088 ( True, ktReason_Unknown_File_Not_Found, # lump it in with file-not-found for now.
1089 'Error: failed to start machine. Error message: Not supported. (VERR_NOT_SUPPORTED)' ),
1090 ( False, ktReason_Unknown_VM_Crash, 'txsDoConnectViaTcp: Machine state: Aborted' ),
1091 ( True, ktReason_Host_Modprobe_Failed, 'Kernel driver not installed' ),
1092 ( True, ktReason_OSInstall_Sata_no_BM, 'PCHS=14128/14134/8224' ),
1093 ( True, ktReason_Host_DoubleFreeHeap, 'double free or corruption' ),
1094 #( False, ktReason_Unknown_VM_Start_Error, 'VMSetError: ' ), - false positives for stuff like:
1095 # "VMSetError: VD: Backend 'VBoxIsoMaker' does not support async I/O"
1096 ( False, ktReason_Unknown_VM_Start_Error, 'error: failed to open session for' ),
1097 ( False, ktReason_Unknown_VM_Runtime_Error, 'Console: VM runtime error: fatal=true' ),
1098 ];
1099
1100 ## This we search a main log for to figure out why something went bust.
1101 katSimpleMainLogReasons = [
1102 # ( Whether to stop on hit, reason tuple, needle text. )
1103 ( False, ktReason_Host_win32com_gen_py, 'ModuleNotFoundError: No module named \'win32com.gen_py' ),
1104
1105 ];
1106
1107 ## This we search a VM log for to figure out why something went bust.
1108 katSimpleVmLogReasons = [
1109 # ( Whether to stop on hit, reason tuple, needle text. )
1110 # Note: Works for ATA and VD drivers.
1111 ( False, ktReason_Host_DiskFull, '_DISKFULL' ),
1112 ];
1113
1114 ## Things we search a VBoxHardening.log file for to figure out why something went bust.
1115 katSimpleVBoxHardeningLogReasons = [
1116 # ( Whether to stop on hit, reason tuple, needle text. )
1117 ( True, ktReason_Host_DriverNotLoaded, 'Error opening VBoxDrvStub: STATUS_OBJECT_NAME_NOT_FOUND' ),
1118 ( True, ktReason_Host_NotSignedWithBuildCert, 'Not signed with the build certificate' ),
1119 ( True, ktReason_Host_TSTInfo_Accuracy_OOR, 'RTCRTSPTSTINFO::Accuracy::Millis: Out of range' ),
1120 ( False, ktReason_Unknown_VM_Crash, 'Quitting: ExitCode=0xc0000005 (rcNtWait=' ),
1121 ];
1122
1123 ## Things we search a kernel.log file for to figure out why something went bust.
1124 katSimpleKernelLogReasons = [
1125 # ( Whether to stop on hit, reason tuple, needle text. )
1126 ( True, ktReason_Panic_HugeMemory, 'mm/huge_memory.c:1988' ),
1127 ( True, ktReason_Panic_IOAPICDoesntWork, 'IO-APIC + timer doesn\'t work' ),
1128 ( True, ktReason_Panic_TxUnitHang, 'Detected Tx Unit Hang' ),
1129 ( True, ktReason_GuestBug_CompizVBoxQt, 'error 4 in libQt5CoreVBox' ),
1130 ( True, ktReason_GuestBug_CompizVBoxQt, 'error 4 in libgtk-3' ),
1131 ];
1132
1133 ## Things we search the _RIGHT_ _STRIPPED_ vgatext for.
1134 katSimpleVgaTextReasons = [
1135 # ( Whether to stop on hit, reason tuple, needle text. )
1136 ( True, ktReason_Panic_MP_BIOS_IO_APIC,
1137 "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n\n" ),
1138 ( True, ktReason_Panic_MP_BIOS_IO_APIC,
1139 "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n"
1140 "...trying to set up timer (IRQ0) through the 8259A ... failed.\n"
1141 "...trying to set up timer as Virtual Wire IRQ... failed.\n"
1142 "...trying to set up timer as ExtINT IRQ... failed :(.\n"
1143 "Kernel panic - not syncing: IO-APIC + timer doesn't work! Boot with apic=debug\n"
1144 "and send a report. Then try booting with the 'noapic' option\n"
1145 "\n" ),
1146 ( True, ktReason_OSInstall_GRUB_hang,
1147 "-----\nGRUB Loading stage2..\n\n\n\n" ),
1148 ( True, ktReason_OSInstall_GRUB_hang,
1149 "-----\nGRUB Loading stage2...\n\n\n\n" ), # the 3 dot hang appears to be less frequent
1150 ( True, ktReason_OSInstall_GRUB_hang,
1151 "-----\nGRUB Loading stage2....\n\n\n\n" ), # the 4 dot hang appears to be very infrequent
1152 ( True, ktReason_OSInstall_GRUB_hang,
1153 "-----\nGRUB Loading stage2.....\n\n\n\n" ), # the 5 dot hang appears to be more frequent again
1154 ( True, ktReason_OSInstall_Udev_hang,
1155 "\nStarting udev:\n\n\n\n" ),
1156 ( True, ktReason_OSInstall_Udev_hang,
1157 "\nStarting udev:\n------" ),
1158 ( True, ktReason_Panic_BootManagerC000000F,
1159 "Windows failed to start. A recent hardware or software change might be the" ),
1160 ( True, ktReason_BootManager_Image_corrupt,
1161 "BOOTMGR image is corrupt. The system cannot boot." ),
1162 ];
1163
1164 ## Things we search for in the info.txt file. Require handlers for now.
1165 katInfoTextHandlers = [
1166 # ( Trigger text, handler method )
1167 ( "kvm_lock_spinning", investigateInfoKvmLockSpinning ),
1168 ( "HalReturnToFirmware", investigateInfoHalReturnToFirmware ),
1169 ];
1170
1171 ## Mapping screenshot/failure SHA-256 hashes to failure reasons.
1172 katSimpleScreenshotHashReasons = [
1173 # ( Whether to stop on hit, reason tuple, lowercased sha-256 of PIL.Image.tostring output )
1174 ( True, ktReason_BSOD_Recovery, '576f8e38d62b311cac7e3dc3436a0d0b9bd8cfd7fa9c43aafa95631520a45eac' ),
1175 ( True, ktReason_BSOD_Automatic_Repair, 'c6a72076cc619937a7a39cfe9915b36d94cee0d4e3ce5ce061485792dcee2749' ),
1176 ( True, ktReason_BSOD_Automatic_Repair, '26c4d8a724ff2c5e1051f3d5b650dbda7b5fdee0aa3e3c6059797f7484a515df' ),
1177 ( True, ktReason_BSOD_0000007F, '57e1880619e13042a87100e7a38c8974b85ce3866501be621bea0cc696bb2c63' ),
1178 ( True, ktReason_BSOD_000000D1, '134621281f00a3f8aeeb7660064bffbf6187ed56d5852142328d0bcb18ef0ede' ),
1179 ( True, ktReason_BSOD_000000D1, '279f11258150c9d2fef041eca65501f3141da8df39256d8f6377e897e3b45a93' ),
1180 ( True, ktReason_BSOD_C0000225, 'bd13a144be9dcdfb16bc863ff4c8f02a86e263c174f2cd5ffd27ca5f3aa31789' ),
1181 ( True, ktReason_BSOD_C0000225, '8348b465e7ee9e59dd4e785880c57fd8677de05d11ac21e786bfde935307b42f' ),
1182 ( True, ktReason_BSOD_C0000225, '1316e1fc818a73348412788e6910b8c016f237d8b4e15b20caf4a866f7a7840e' ),
1183 ( True, ktReason_BSOD_C0000225, '54e0acbff365ce20a85abbe42bcd53647b8b9e80c68e45b2cd30e86bf177a0b5' ),
1184 ( True, ktReason_BSOD_C0000225, '50fec50b5199923fa48b3f3e782687cc381e1c8a788ebda14e6a355fbe3bb1b3' ),
1185 ];
1186
1187
1188 def scanLog(self, asLogs, atNeedles, oCaseFile, idTestResult):
1189 """
1190 Scans for atNeedles in sLog.
1191
1192 Returns True if a stop-on-hit neelde was found.
1193 Returns None if a no-stop reason was found.
1194 Returns False if no hit.
1195 """
1196 fRet = False;
1197 for fStopOnHit, tReason, oNeedle in atNeedles:
1198 fMatch = False;
1199 if utils.isString(oNeedle):
1200 for sLog in asLogs:
1201 if sLog:
1202 fMatch |= sLog.find(oNeedle) > 0;
1203 else:
1204 for sLog in asLogs:
1205 if sLog:
1206 fMatch |= oNeedle.search(sLog) is not None;
1207 if fMatch:
1208 oCaseFile.noteReasonForId(tReason, idTestResult);
1209 if fStopOnHit:
1210 return True;
1211 fRet = None;
1212 return fRet;
1213
1214
1215 def investigateGATest(self, oCaseFile, oFailedResult, sResultLog):
1216 """
1217 Investigates a failed VM run.
1218 """
1219 enmReason = None;
1220 sParentName = oFailedResult.oParent.sName if oFailedResult.oParent else '';
1221 if oFailedResult.sName == 'VBoxWindowsAdditions.exe' or sResultLog.find('VBoxWindowsAdditions.exe" failed with') > 0:
1222 enmReason = self.ktReason_Add_Installer_Win_Failed;
1223 # guest control:
1224 elif sParentName == 'Guest Control' and oFailedResult.sName == 'Preparations':
1225 enmReason = self.ktReason_Add_GstCtl_Preparations;
1226 elif oFailedResult.sName == 'Session Basics':
1227 enmReason = self.ktReason_Add_GstCtl_SessionBasics;
1228 elif oFailedResult.sName == 'Session Process References':
1229 enmReason = self.ktReason_Add_GstCtl_SessionProcRefs;
1230 elif oFailedResult.sName == 'Copy from guest':
1231 if sResultLog.find('*** abort action ***') >= 0:
1232 enmReason = self.ktReason_Add_GstCtl_CopyFromGuest_Timeout;
1233 elif oFailedResult.sName == 'Copy to guest':
1234 off = sResultLog.find('"Guest directory "');
1235 if off > 0 and sResultLog.find('" already exists"', off, off + 80):
1236 enmReason = self.ktReason_Add_GstCtl_CopyToGuest_DstExists;
1237 elif sResultLog.find('Guest destination must not be empty') >= 0:
1238 enmReason = self.ktReason_Add_GstCtl_CopyToGuest_DstEmpty;
1239 elif sResultLog.find('*** abort action ***') >= 0:
1240 enmReason = self.ktReason_Add_GstCtl_CopyToGuest_Timeout;
1241 elif oFailedResult.sName.find('Session w/ Guest Reboot') >= 0:
1242 enmReason = self.ktReason_Add_GstCtl_Session_Reboot;
1243 # shared folders:
1244 elif sParentName == 'Shared Folders' and oFailedResult.sName == 'Automounting':
1245 enmReason = self.ktReason_Add_ShFl_Automount;
1246 elif oFailedResult.sName == 'mmap':
1247 if sResultLog.find('FsPerf: Flush issue at offset ') >= 0:
1248 enmReason = self.ktReason_Add_Mmap_Coherency;
1249 elif sResultLog.find('FlushViewOfFile') >= 0:
1250 enmReason = self.ktReason_Add_FlushViewOfFile;
1251 elif sParentName == 'Shared Folders' and oFailedResult.sName == 'Running FsPerf':
1252 enmReason = self.ktReason_Add_ShFl_FsPerf; ## Maybe it would be better to be more specific...
1253
1254 if enmReason is not None:
1255 return oCaseFile.noteReasonForId(enmReason, oFailedResult.idTestResult);
1256
1257 self.vprint(u'TODO: Cannot place GA failure idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1258 self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1259 return False;
1260
1261 def isResultFromGATest(self, oCaseFile, oFailedResult):
1262 """
1263 Checks if this result and corresponding log snippet looks like a GA test run.
1264 """
1265 while oFailedResult is not None:
1266 if oFailedResult.sName in [ 'Guest Control', 'Shared Folders', 'FsPerf', 'VBoxWindowsAdditions.exe' ]:
1267 return True;
1268 if oCaseFile.oTestCase.sName == 'Guest Additions' and oFailedResult.sName in [ 'Install', ]:
1269 return True;
1270 oFailedResult = oFailedResult.oParent;
1271 return False;
1272
1273
1274 def investigateVMResult(self, oCaseFile, oFailedResult, sResultLog):
1275 """
1276 Investigates a failed VM run.
1277 """
1278
1279 def investigateLogSet():
1280 """
1281 Investigates the current set of VM related logs.
1282 """
1283 self.dprint('investigateLogSet: log lengths: result %u, VM %u, kernel %u, vga text %u, info text %u, hard %u'
1284 % ( len(sResultLog if sResultLog else ''),
1285 len(sVMLog if sVMLog else ''),
1286 len(sKrnlLog if sKrnlLog else ''),
1287 len(sVgaText if sVgaText else ''),
1288 len(sInfoText if sInfoText else ''),
1289 len(sNtHardLog if sNtHardLog else ''),));
1290
1291 #self.dprint(u'main.log<<<\n%s\n<<<\n' % (sResultLog,));
1292 #self.dprint(u'vbox.log<<<\n%s\n<<<\n' % (sVMLog,));
1293 #self.dprint(u'krnl.log<<<\n%s\n<<<\n' % (sKrnlLog,));
1294 #self.dprint(u'vgatext.txt<<<\n%s\n<<<\n' % (sVgaText,));
1295 #self.dprint(u'info.txt<<<\n%s\n<<<\n' % (sInfoText,));
1296 #self.dprint(u'hard.txt<<<\n%s\n<<<\n' % (sNtHardLog,));
1297
1298 # TODO: more
1299
1300 #
1301 # Look for BSODs. Some stupid stupid inconsistencies in reason and log messages here, so don't try prettify this.
1302 #
1303 sDetails = self.findInAnyAndReturnRestOfLine([ sVMLog, sResultLog ],
1304 'GIM: HyperV: Guest indicates a fatal condition! P0=');
1305 if sDetails is not None:
1306 # P0=%#RX64 P1=%#RX64 P2=%#RX64 P3=%#RX64 P4=%#RX64 "
1307 sKey = sDetails.split(' ', 1)[0];
1308 try: sKey = '0x%08X' % (int(sKey, 16),);
1309 except: pass;
1310 if sKey in self.asBsodReasons:
1311 tReason = ( self.ksBsodCategory, sKey );
1312 elif sKey.lower() in self.asBsodReasons: # just in case.
1313 tReason = ( self.ksBsodCategory, sKey.lower() );
1314 else:
1315 self.dprint(u'BSOD "%s" not found in %s;' % (sKey, self.asBsodReasons));
1316 tReason = ( self.ksBsodCategory, self.ksBsodAddNew );
1317 return oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult, sComment = sDetails.strip());
1318
1319 fFoundSomething = False;
1320
1321 #
1322 # Look for linux panic.
1323 #
1324 if sKrnlLog is not None:
1325 fRet = self.scanLog([sKrnlLog,], self.katSimpleKernelLogReasons, oCaseFile, oFailedResult.idTestResult);
1326 if fRet is True:
1327 return fRet;
1328 fFoundSomething |= fRet is None;
1329
1330 #
1331 # Loop thru the simple stuff.
1332 #
1333
1334 # Main log.
1335 fRet = self.scanLog([sResultLog,], self.katSimpleMainLogReasons, oCaseFile, oFailedResult.idTestResult);
1336 if fRet is True:
1337 return fRet;
1338 fFoundSomething |= fRet is None;
1339
1340 # VM log.
1341 fRet = self.scanLog([sVMLog,], self.katSimpleVmLogReasons, oCaseFile, oFailedResult.idTestResult);
1342 if fRet is True:
1343 return fRet;
1344 fFoundSomething |= fRet is None;
1345
1346 # Old main + vm log.
1347 fRet = self.scanLog([sResultLog, sVMLog], self.katSimpleMainAndVmLogReasonsDeprecated,
1348 oCaseFile, oFailedResult.idTestResult);
1349 if fRet is True:
1350 return fRet;
1351 fFoundSomething |= fRet is None;
1352
1353 # Continue with vga text.
1354 if sVgaText:
1355 fRet = self.scanLog([sVgaText,], self.katSimpleVgaTextReasons, oCaseFile, oFailedResult.idTestResult);
1356 if fRet is True:
1357 return fRet;
1358 fFoundSomething |= fRet is None;
1359
1360 # Continue with screen hashes.
1361 if sScreenHash is not None:
1362 for fStopOnHit, tReason, sHash in self.katSimpleScreenshotHashReasons:
1363 if sScreenHash == sHash:
1364 oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1365 if fStopOnHit:
1366 return True;
1367 fFoundSomething = True;
1368
1369 # Check VBoxHardening.log.
1370 if sNtHardLog is not None:
1371 fRet = self.scanLog([sNtHardLog,], self.katSimpleVBoxHardeningLogReasons, oCaseFile, oFailedResult.idTestResult);
1372 if fRet is True:
1373 return fRet;
1374 fFoundSomething |= fRet is None;
1375
1376 #
1377 # Complicated stuff.
1378 #
1379 dLogs = {
1380 'sVMLog': sVMLog,
1381 'sNtHardLog': sNtHardLog,
1382 'sScreenHash': sScreenHash,
1383 'sKrnlLog': sKrnlLog,
1384 'sVgaText': sVgaText,
1385 'sInfoText': sInfoText,
1386 };
1387
1388 # info.txt.
1389 if sInfoText:
1390 for sNeedle, fnHandler in self.katInfoTextHandlers:
1391 if sInfoText.find(sNeedle) > 0:
1392 (fStop, tReason) = fnHandler(self, oCaseFile, sInfoText, dLogs);
1393 if tReason is not None:
1394 oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1395 if fStop:
1396 return True;
1397 fFoundSomething = True;
1398
1399 #
1400 # Check for repeated reboots...
1401 #
1402 if sVMLog is not None:
1403 cResets = sVMLog.count('Changing the VM state from \'RUNNING\' to \'RESETTING\'');
1404 if cResets > 10:
1405 return oCaseFile.noteReasonForId(self.ktReason_Unknown_Reboot_Loop, oFailedResult.idTestResult,
1406 sComment = 'Counted %s reboots' % (cResets,));
1407
1408 return fFoundSomething;
1409
1410 #
1411 # Check if we got any VM or/and kernel logs. Treat them as sets in
1412 # case we run multiple VMs here (this is of course ASSUMING they
1413 # appear in the order that terminateVmBySession uploads them).
1414 #
1415 cTimes = 0;
1416 sVMLog = None;
1417 sNtHardLog = None;
1418 sScreenHash = None;
1419 sKrnlLog = None;
1420 sVgaText = None;
1421 sInfoText = None;
1422 for oFile in oFailedResult.aoFiles:
1423 if oFile.sKind == TestResultFileData.ksKind_LogReleaseVm:
1424 if 'VBoxHardening.log' not in oFile.sFile:
1425 if sVMLog is not None:
1426 if investigateLogSet() is True:
1427 return True;
1428 cTimes += 1;
1429 sInfoText = None;
1430 sVgaText = None;
1431 sKrnlLog = None;
1432 sScreenHash = None;
1433 sNtHardLog = None;
1434 sVMLog = oCaseFile.getLogFile(oFile);
1435 else:
1436 sNtHardLog = oCaseFile.getLogFile(oFile);
1437 elif oFile.sKind == TestResultFileData.ksKind_LogGuestKernel:
1438 sKrnlLog = oCaseFile.getLogFile(oFile);
1439 elif oFile.sKind == TestResultFileData.ksKind_InfoVgaText:
1440 sVgaText = '\n'.join([sLine.rstrip() for sLine in oCaseFile.getLogFile(oFile).split('\n')]);
1441 elif oFile.sKind == TestResultFileData.ksKind_InfoCollection:
1442 sInfoText = oCaseFile.getLogFile(oFile);
1443 elif oFile.sKind == TestResultFileData.ksKind_ScreenshotFailure:
1444 sScreenHash = oCaseFile.getScreenshotSha256(oFile);
1445 if sScreenHash is not None:
1446 sScreenHash = sScreenHash.lower();
1447 self.vprint(u'%s %s' % ( sScreenHash, oFile.sFile,));
1448
1449 if ( sVMLog is not None \
1450 or sNtHardLog is not None \
1451 or cTimes == 0) \
1452 and investigateLogSet() is True:
1453 return True;
1454
1455 return None;
1456
1457 def isResultFromVMRun(self, oFailedResult, sResultLog):
1458 """
1459 Checks if this result and corresponding log snippet looks like a VM run.
1460 """
1461
1462 # Look for startVmEx/ startVmAndConnectToTxsViaTcp and similar output in the log.
1463 if sResultLog.find(' startVm') > 0:
1464 return True;
1465
1466 # Any other indicators? No?
1467 _ = oFailedResult;
1468 return False;
1469
1470
1471 ## Things we search a VBoxSVC log for to figure out why something went bust.
1472 katSimpleSvcLogReasons = [
1473 # ( Whether to stop on hit, reason tuple, needle text. )
1474 ( False, ktReason_Unknown_VM_Crash, re.compile(r'Reaper.* exited normally: -1073741819 \(0xc0000005\)') ),
1475 ( False, ktReason_Unknown_VM_Crash, re.compile(r'Reaper.* was signalled: 11 \(0xb\)') ), # For VBox < 6.1.
1476 ( False, ktReason_Unknown_VM_Crash, re.compile(r'Reaper.* was signalled: SIGABRT.*') ), # Since VBox 7.0.
1477 ( False, ktReason_Unknown_VM_Crash, re.compile(r'Reaper.* was signalled: SIGSEGV.*') ),
1478 ( False, ktReason_Unknown_VM_Terminated, re.compile(r'Reaper.* was signalled: SIGTERM.*') ),
1479 ( False, ktReason_Unknown_VM_Terminated, re.compile(r'Reaper.* was signalled: SIGKILL.*') ),
1480 ];
1481
1482 def investigateSvcLogForVMRun(self, oCaseFile, sSvcLog):
1483 """
1484 Check the VBoxSVC log for a single VM run.
1485 """
1486 if sSvcLog:
1487 fRet = self.scanLog([sSvcLog,], self.katSimpleSvcLogReasons, oCaseFile, oCaseFile.oTree.idTestResult);
1488 if fRet is True or fRet is None:
1489 return True;
1490 return False;
1491
1492 def investigateNtHardLogForVMRun(self, oCaseFile):
1493 """
1494 Check if the hardening log for a single VM run contains VM crash indications.
1495 """
1496 aoLogFiles = oCaseFile.oTree.getListOfLogFilesByKind(TestResultFileData.ksKind_LogReleaseVm);
1497 for oLogFile in aoLogFiles:
1498 if oLogFile.sFile.find('VBoxHardening.log') >= 0:
1499 sLog = oCaseFile.getLogFile(oLogFile);
1500 if sLog.find('Quitting: ExitCode=0xc0000005') >= 0:
1501 return oCaseFile.noteReasonForId(self.ktReason_Unknown_VM_Crash, oCaseFile.oTree.idTestResult);
1502 return False;
1503
1504
1505 def investigateVBoxVMTest(self, oCaseFile, fSingleVM):
1506 """
1507 Checks out a VBox VM test.
1508
1509 This is generic investigation of a test running one or more VMs, like
1510 for example a smoke test or a guest installation test.
1511
1512 The fSingleVM parameter is a hint, which probably won't come in useful.
1513 """
1514 _ = fSingleVM;
1515
1516 #
1517 # Get a list of test result failures we should be looking into and the main log.
1518 #
1519 aoFailedResults = oCaseFile.oTree.getListOfFailures();
1520 sMainLog = oCaseFile.getMainLog();
1521
1522 #
1523 # There are a set of errors ending up on the top level result record.
1524 # Should deal with these first.
1525 #
1526 if len(aoFailedResults) == 1 and aoFailedResults[0] == oCaseFile.oTree:
1527 # Check if we've just got that XPCOM client smoke test shutdown issue. This will currently always
1528 # be reported on the top result because vboxinstall.py doesn't add an error for it. It is easy to
1529 # ignore other failures in the test if we're not a little bit careful here.
1530 if sMainLog.find('vboxinstaller: Exit code: -11 (') > 0:
1531 oCaseFile.noteReason(self.ktReason_XPCOM_Exit_Minus_11);
1532 return self.caseClosed(oCaseFile);
1533
1534 # Hang after starting VBoxSVC (e.g. idTestSet=136307258)
1535 if self.isThisFollowedByTheseLines(sMainLog, 'oVBoxMgr=<vboxapi.VirtualBoxManager object at',
1536 (' Timeout: ', ' Attempting to abort child...',) ):
1537 if sMainLog.find('*** glibc detected *** /') > 0:
1538 oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang_Plus_Heap_Corruption);
1539 else:
1540 oCaseFile.noteReason(self.ktReason_XPCOM_VBoxSVC_Hang);
1541 return self.caseClosed(oCaseFile);
1542
1543 # Look for heap corruption without visible hang.
1544 if sMainLog.find('*** glibc detected *** /') > 0 \
1545 or sMainLog.find("-1073740940") > 0: # STATUS_HEAP_CORRUPTION / 0xc0000374
1546 oCaseFile.noteReason(self.ktReason_Unknown_Heap_Corruption);
1547 return self.caseClosed(oCaseFile);
1548
1549 # Out of memory w/ timeout.
1550 if sMainLog.find('sErrId=HostMemoryLow') > 0:
1551 oCaseFile.noteReason(self.ktReason_Host_HostMemoryLow);
1552 return self.caseClosed(oCaseFile);
1553
1554 # Stale files like vts_rm.exe (windows).
1555 offEnd = sMainLog.rfind('*** The test driver exits successfully. ***');
1556 if offEnd > 0 and sMainLog.find('[Error 145] The directory is not empty: ', offEnd) > 0:
1557 oCaseFile.noteReason(self.ktReason_Ignore_Stale_Files);
1558 return self.caseClosed(oCaseFile);
1559
1560 #
1561 # XPCOM screwup
1562 #
1563 if sMainLog.find('AttributeError: \'NoneType\' object has no attribute \'addObserver\'') > 0:
1564 oCaseFile.noteReason(self.ktReason_Buggy_Build_Broken_Build);
1565 return self.caseClosed(oCaseFile);
1566
1567 #
1568 # Go thru each failed result.
1569 #
1570 for oFailedResult in aoFailedResults:
1571 self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1572 sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1573 if oFailedResult.sName == 'Installing VirtualBox':
1574 self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1575
1576 elif oFailedResult.sName == 'Uninstalling VirtualBox':
1577 self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1578
1579 elif self.isResultFromVMRun(oFailedResult, sResultLog):
1580 self.investigateVMResult(oCaseFile, oFailedResult, sResultLog);
1581
1582 elif self.isResultFromGATest(oCaseFile, oFailedResult):
1583 self.investigateGATest(oCaseFile, oFailedResult, sResultLog);
1584
1585 elif sResultLog.find('most likely not unique') > 0:
1586 oCaseFile.noteReasonForId(self.ktReason_Host_NetworkMisconfiguration, oFailedResult.idTestResult)
1587 elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1588 oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1589
1590 elif sResultLog.find('The machine is not mutable (state is ') > 0:
1591 self.vprint('Ignoring "machine not mutable" error as it is probably due to an earlier problem');
1592 oCaseFile.noteReasonForId(self.ktHarmless, oFailedResult.idTestResult);
1593
1594 elif sResultLog.find('** error: no action was specified') > 0 \
1595 or sResultLog.find('(len(self._asXml, asText))') > 0:
1596 oCaseFile.noteReasonForId(self.ktReason_Ignore_Buggy_Test_Driver, oFailedResult.idTestResult);
1597
1598 else:
1599 self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1600 self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1601
1602 #
1603 # Windows python/com screwup.
1604 #
1605 if sMainLog.find('ModuleNotFoundError: No module named \'win32com.gen_py') > 0:
1606 oCaseFile.noteReason(self.ktReason_Host_win32com_gen_py);
1607 return self.caseClosed(oCaseFile);
1608
1609 #
1610 # Check VBoxSVC.log and VBoxHardening.log for VM crashes if inconclusive on single VM runs.
1611 #
1612 if fSingleVM and len(oCaseFile.dReasonForResultId) < len(aoFailedResults):
1613 self.dprint(u'Got %u out of %u - checking VBoxSVC.log...'
1614 % (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1615 if self.investigateSvcLogForVMRun(oCaseFile, oCaseFile.getSvcLog()):
1616 return self.caseClosed(oCaseFile);
1617 if self.investigateNtHardLogForVMRun(oCaseFile):
1618 return self.caseClosed(oCaseFile);
1619
1620 #
1621 # Report home and close the case if we got them all, otherwise log it.
1622 #
1623 if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1624 return self.caseClosed(oCaseFile);
1625
1626 if oCaseFile.dReasonForResultId:
1627 self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1628 % (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1629 else:
1630 self.vprint(u'XXX: Could not figure out anything at all! :-(');
1631 return False;
1632
1633
1634 ## Things we search a main log for to figure out why something in the API test went bust.
1635 katSimpleApiMainLogReasons = [
1636 # ( Whether to stop on hit, reason tuple, needle text. )
1637 ( True, ktReason_Networking_Nonexistent_host_nic,
1638 'rc=E_FAIL text="Nonexistent host networking interface, name \'eth0\' (VERR_INTERNAL_ERROR)"' ),
1639 ( False, ktReason_XPCOM_NS_ERROR_CALL_FAILED,
1640 'Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))' ),
1641 ( True, ktReason_API_std_bad_alloc, 'Unexpected exception: std::bad_alloc' ),
1642 ( True, ktReason_API_Digest_Mismatch, 'Digest mismatch (VERR_NOT_EQUAL)' ),
1643 ( True, ktReason_API_MoveVM_SharingViolation, 'rc=VBOX_E_IPRT_ERROR text="Could not copy the log file ' ),
1644 ( True, ktReason_API_MoveVM_InvalidParameter,
1645 'rc=VBOX_E_IPRT_ERROR text="Could not copy the setting file ' ),
1646 ( True, ktReason_API_Open_Session_Failed, 'error: failed to open session for' ),
1647 ];
1648
1649 def investigateVBoxApiTest(self, oCaseFile):
1650 """
1651 Checks out a VBox API test.
1652 """
1653
1654 #
1655 # Get a list of test result failures we should be looking into and the main log.
1656 #
1657 aoFailedResults = oCaseFile.oTree.getListOfFailures();
1658 sMainLog = oCaseFile.getMainLog();
1659
1660 #
1661 # Go thru each failed result.
1662 #
1663 for oFailedResult in aoFailedResults:
1664 self.dprint(u'Looking at test result #%u - %s' % (oFailedResult.idTestResult, oFailedResult.getFullName(),));
1665 sResultLog = TestSetData.extractLogSectionElapsed(sMainLog, oFailedResult.tsCreated, oFailedResult.tsElapsed);
1666 if oFailedResult.sName == 'Installing VirtualBox':
1667 self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = True)
1668
1669 elif oFailedResult.sName == 'Uninstalling VirtualBox':
1670 self.investigateInstallUninstallFailure(oCaseFile, oFailedResult, sResultLog, fInstall = False)
1671
1672 elif sResultLog.find('Exception: 0x800706be (Call to remote object failed (NS_ERROR_CALL_FAILED))') > 0:
1673 oCaseFile.noteReasonForId(self.ktReason_XPCOM_NS_ERROR_CALL_FAILED, oFailedResult.idTestResult);
1674
1675 else:
1676 fFoundSomething = False;
1677 for fStopOnHit, tReason, sNeedle in self.katSimpleApiMainLogReasons:
1678 if sResultLog.find(sNeedle) > 0:
1679 oCaseFile.noteReasonForId(tReason, oFailedResult.idTestResult);
1680 fFoundSomething = True;
1681 if fStopOnHit:
1682 break;
1683 if fFoundSomething:
1684 self.vprint(u'TODO: Cannot place idTestResult=%u - %s' % (oFailedResult.idTestResult, oFailedResult.sName,));
1685 self.dprint(u'%s + %s <<\n%s\n<<' % (oFailedResult.tsCreated, oFailedResult.tsElapsed, sResultLog,));
1686
1687 #
1688 # Report home and close the case if we got them all, otherwise log it.
1689 #
1690 if len(oCaseFile.dReasonForResultId) >= len(aoFailedResults):
1691 return self.caseClosed(oCaseFile);
1692
1693 if oCaseFile.dReasonForResultId:
1694 self.vprint(u'TODO: Got %u out of %u - close, but no cigar. :-/'
1695 % (len(oCaseFile.dReasonForResultId), len(aoFailedResults)));
1696 else:
1697 self.vprint(u'XXX: Could not figure out anything at all! :-(');
1698 return False;
1699
1700
1701 def reasoningFailures(self):
1702 """
1703 Guess the reason for failures.
1704 """
1705 #
1706 # Get a list of failed test sets without any assigned failure reason.
1707 #
1708 cGot = 0;
1709 if not self.oConfig.aidTestSets:
1710 aoTestSets = self.oTestSetLogic.fetchFailedSetsWithoutReason(cHoursBack = self.oConfig.cHoursBack,
1711 tsNow = self.tsNow);
1712 else:
1713 aoTestSets = [self.oTestSetLogic.getById(idTestSet) for idTestSet in self.oConfig.aidTestSets];
1714 for oTestSet in aoTestSets:
1715 self.dprint(u'----------------------------------- #%u, status %s -----------------------------------'
1716 % ( oTestSet.idTestSet, oTestSet.enmStatus,));
1717
1718 #
1719 # Open a case file and assign it to the right investigator.
1720 #
1721 (oTree, _ ) = self.oTestResultLogic.fetchResultTree(oTestSet.idTestSet);
1722 oBuild = BuildDataEx().initFromDbWithId( self.oDb, oTestSet.idBuild, oTestSet.tsCreated);
1723 oTestBox = TestBoxData().initFromDbWithGenId( self.oDb, oTestSet.idGenTestBox);
1724 oTestGroup = TestGroupData().initFromDbWithId( self.oDb, oTestSet.idTestGroup, oTestSet.tsCreated);
1725 oTestCase = TestCaseDataEx().initFromDbWithGenId( self.oDb, oTestSet.idGenTestCase, oTestSet.tsConfig);
1726
1727 oCaseFile = VirtualTestSheriffCaseFile(self, oTestSet, oTree, oBuild, oTestBox, oTestGroup, oTestCase);
1728
1729 if oTestSet.enmStatus == TestSetData.ksTestStatus_BadTestBox:
1730 self.dprint(u'investigateBadTestBox is taking over %s.' % (oCaseFile.sLongName,));
1731 fRc = self.investigateBadTestBox(oCaseFile);
1732
1733 elif oCaseFile.isVBoxUnitTest():
1734 self.dprint(u'investigateVBoxUnitTest is taking over %s.' % (oCaseFile.sLongName,));
1735 fRc = self.investigateVBoxUnitTest(oCaseFile);
1736
1737 elif oCaseFile.isVBoxInstallTest() or oCaseFile.isVBoxUnattendedInstallTest():
1738 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1739 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1740
1741 elif oCaseFile.isVBoxUSBTest():
1742 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1743 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1744
1745 elif oCaseFile.isVBoxStorageTest():
1746 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1747 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1748
1749 elif oCaseFile.isVBoxGAsTest():
1750 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1751 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = True);
1752
1753 elif oCaseFile.isVBoxAPITest():
1754 self.dprint(u'investigateVBoxApiTest is taking over %s.' % (oCaseFile.sLongName,));
1755 fRc = self.investigateVBoxApiTest(oCaseFile);
1756
1757 elif oCaseFile.isVBoxBenchmarkTest():
1758 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1759 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1760
1761 elif oCaseFile.isVBoxSmokeTest():
1762 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1763 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1764
1765 elif oCaseFile.isVBoxSerialTest():
1766 self.dprint(u'investigateVBoxVMTest is taking over %s.' % (oCaseFile.sLongName,));
1767 fRc = self.investigateVBoxVMTest(oCaseFile, fSingleVM = False);
1768
1769 else:
1770 self.vprint(u'reasoningFailures: Unable to classify test set: %s' % (oCaseFile.sLongName,));
1771 fRc = False;
1772 cGot += fRc is True;
1773
1774 self.vprint(u'reasoningFailures: Got %u out of %u' % (cGot, len(aoTestSets), ));
1775 return 0;
1776
1777
1778 def main(self):
1779 """
1780 The 'main' function.
1781 Return exit code (0, 1, etc).
1782 """
1783 # Database stuff.
1784 self.oDb = TMDatabaseConnection()
1785 self.oTestResultLogic = TestResultLogic(self.oDb);
1786 self.oTestSetLogic = TestSetLogic(self.oDb);
1787 self.oFailureReasonLogic = FailureReasonLogic(self.oDb);
1788 self.oTestResultFailureLogic = TestResultFailureLogic(self.oDb);
1789 self.asBsodReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksBsodCategory);
1790 self.asUnitTestReasons = self.oFailureReasonLogic.fetchForSheriffByNamedCategory(self.ksUnitTestCategory);
1791
1792 # Get a fix on our 'now' before we do anything..
1793 self.oDb.execute('SELECT CURRENT_TIMESTAMP - interval \'%s hours\'', (self.oConfig.cStartHoursAgo,));
1794 self.tsNow = self.oDb.fetchOne();
1795
1796 # If we're suppost to commit anything we need to get our user ID.
1797 rcExit = 0;
1798 if self.oConfig.fRealRun:
1799 self.oLogin = UserAccountLogic(self.oDb).tryFetchAccountByLoginName(VirtualTestSheriff.ksLoginName);
1800 if self.oLogin is None:
1801 rcExit = self.eprint('Cannot find my user account "%s"!' % (VirtualTestSheriff.ksLoginName,));
1802 else:
1803 self.uidSelf = self.oLogin.uid;
1804
1805 #
1806 # Do the stuff.
1807 #
1808 if rcExit == 0:
1809 rcExit = self.selfCheck();
1810 if rcExit == 0:
1811 rcExit = self.badTestBoxManagement();
1812 rcExit2 = self.reasoningFailures();
1813 if rcExit == 0:
1814 rcExit = rcExit2;
1815 # Redo the bad testbox management after failure reasons have been assigned (got timing issues).
1816 if rcExit == 0:
1817 rcExit = self.badTestBoxManagement();
1818
1819 # Cleanup.
1820 self.oFailureReasonLogic = None;
1821 self.oTestResultFailureLogic = None;
1822 self.oTestSetLogic = None;
1823 self.oTestResultLogic = None;
1824 self.oDb.close();
1825 self.oDb = None;
1826 if self.oLogFile is not None:
1827 self.oLogFile.close();
1828 self.oLogFile = None;
1829 return rcExit;
1830
1831if __name__ == '__main__':
1832 sys.exit(VirtualTestSheriff().main());
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette