Brad Bishop | c342db3 | 2019-05-15 21:57:59 -0400 | [diff] [blame] | 1 | # |
| 2 | # SPDX-License-Identifier: GPL-2.0-only |
| 3 | # |
| 4 | |
Brad Bishop | d7bf8c1 | 2018-02-25 22:55:05 -0500 | [diff] [blame] | 5 | from django.core.management.base import BaseCommand |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 6 | from django.db import transaction |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 7 | from django.db.models import Q |
| 8 | |
| 9 | from bldcontrol.bbcontroller import getBuildEnvironmentController |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 10 | from bldcontrol.models import BuildRequest, BuildEnvironment |
| 11 | from bldcontrol.models import BRError, BRVariable |
| 12 | |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 13 | from orm.models import Build, LogMessage, Target |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 14 | |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 15 | import logging |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 16 | import traceback |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 17 | import signal |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 18 | import os |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 19 | |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 20 | logger = logging.getLogger("toaster") |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 21 | |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 22 | |
Brad Bishop | d7bf8c1 | 2018-02-25 22:55:05 -0500 | [diff] [blame] | 23 | class Command(BaseCommand): |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 24 | args = "" |
| 25 | help = "Schedules and executes build requests as possible. "\ |
| 26 | "Does not return (interrupt with Ctrl-C)" |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 27 | |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 28 | @transaction.atomic |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 29 | def _selectBuildEnvironment(self): |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 30 | bec = getBuildEnvironmentController(lock=BuildEnvironment.LOCK_FREE) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 31 | bec.be.lock = BuildEnvironment.LOCK_LOCK |
| 32 | bec.be.save() |
| 33 | return bec |
| 34 | |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 35 | @transaction.atomic |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 36 | def _selectBuildRequest(self): |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 37 | br = BuildRequest.objects.filter(state=BuildRequest.REQ_QUEUED).first() |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 38 | return br |
| 39 | |
| 40 | def schedule(self): |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 41 | try: |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 42 | # select the build environment and the request to build |
| 43 | br = self._selectBuildRequest() |
| 44 | if br: |
| 45 | br.state = BuildRequest.REQ_INPROGRESS |
| 46 | br.save() |
| 47 | else: |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 48 | return |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 49 | |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 50 | try: |
| 51 | bec = self._selectBuildEnvironment() |
| 52 | except IndexError as e: |
| 53 | # we could not find a BEC; postpone the BR |
| 54 | br.state = BuildRequest.REQ_QUEUED |
| 55 | br.save() |
Brad Bishop | 1a4b7ee | 2018-12-16 17:11:34 -0800 | [diff] [blame] | 56 | logger.debug("runbuilds: No build env (%s)" % e) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 57 | return |
| 58 | |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 59 | logger.info("runbuilds: starting build %s, environment %s" % |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 60 | (br, bec.be)) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 61 | |
| 62 | # let the build request know where it is being executed |
| 63 | br.environment = bec.be |
| 64 | br.save() |
| 65 | |
| 66 | # this triggers an async build |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 67 | bec.triggerBuild(br.brbitbake, br.brlayer_set.all(), |
| 68 | br.brvariable_set.all(), br.brtarget_set.all(), |
| 69 | "%d:%d" % (br.pk, bec.be.pk)) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 70 | |
| 71 | except Exception as e: |
| 72 | logger.error("runbuilds: Error launching build %s" % e) |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 73 | traceback.print_exc() |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 74 | if "[Errno 111] Connection refused" in str(e): |
| 75 | # Connection refused, read toaster_server.out |
| 76 | errmsg = bec.readServerLogFile() |
| 77 | else: |
| 78 | errmsg = str(e) |
| 79 | |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 80 | BRError.objects.create(req=br, errtype=str(type(e)), errmsg=errmsg, |
| 81 | traceback=traceback.format_exc()) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 82 | br.state = BuildRequest.REQ_FAILED |
| 83 | br.save() |
| 84 | bec.be.lock = BuildEnvironment.LOCK_FREE |
| 85 | bec.be.save() |
Brad Bishop | d7bf8c1 | 2018-02-25 22:55:05 -0500 | [diff] [blame] | 86 | # Cancel the pending build and report the exception to the UI |
| 87 | log_object = LogMessage.objects.create( |
| 88 | build = br.build, |
| 89 | level = LogMessage.EXCEPTION, |
| 90 | message = errmsg) |
| 91 | log_object.save() |
| 92 | br.build.outcome = Build.FAILED |
| 93 | br.build.save() |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 94 | |
| 95 | def archive(self): |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 96 | for br in BuildRequest.objects.filter(state=BuildRequest.REQ_ARCHIVE): |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 97 | if br.build is None: |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 98 | br.state = BuildRequest.REQ_FAILED |
Patrick Williams | d7e9631 | 2015-09-22 08:09:05 -0500 | [diff] [blame] | 99 | else: |
| 100 | br.state = BuildRequest.REQ_COMPLETED |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 101 | br.save() |
| 102 | |
| 103 | def cleanup(self): |
| 104 | from django.utils import timezone |
| 105 | from datetime import timedelta |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 106 | # environments locked for more than 30 seconds |
| 107 | # they should be unlocked |
| 108 | BuildEnvironment.objects.filter( |
| 109 | Q(buildrequest__state__in=[BuildRequest.REQ_FAILED, |
| 110 | BuildRequest.REQ_COMPLETED, |
| 111 | BuildRequest.REQ_CANCELLING]) & |
| 112 | Q(lock=BuildEnvironment.LOCK_LOCK) & |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 113 | Q(updated__lt=timezone.now() - timedelta(seconds=30)) |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 114 | ).update(lock=BuildEnvironment.LOCK_FREE) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 115 | |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 116 | # update all Builds that were in progress and failed to start |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 117 | for br in BuildRequest.objects.filter( |
| 118 | state=BuildRequest.REQ_FAILED, |
| 119 | build__outcome=Build.IN_PROGRESS): |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 120 | # transpose the launch errors in ToasterExceptions |
| 121 | br.build.outcome = Build.FAILED |
| 122 | for brerror in br.brerror_set.all(): |
| 123 | logger.debug("Saving error %s" % brerror) |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 124 | LogMessage.objects.create(build=br.build, |
| 125 | level=LogMessage.EXCEPTION, |
| 126 | message=brerror.errmsg) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 127 | br.build.save() |
| 128 | |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 129 | # we don't have a true build object here; hence, toasterui |
| 130 | # didn't have a change to release the BE lock |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 131 | br.environment.lock = BuildEnvironment.LOCK_FREE |
| 132 | br.environment.save() |
| 133 | |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 134 | # update all BuildRequests without a build created |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 135 | for br in BuildRequest.objects.filter(build=None): |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 136 | br.build = Build.objects.create(project=br.project, |
| 137 | completed_on=br.updated, |
| 138 | started_on=br.created) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 139 | br.build.outcome = Build.FAILED |
| 140 | try: |
| 141 | br.build.machine = br.brvariable_set.get(name='MACHINE').value |
| 142 | except BRVariable.DoesNotExist: |
| 143 | pass |
| 144 | br.save() |
| 145 | # transpose target information |
| 146 | for brtarget in br.brtarget_set.all(): |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 147 | Target.objects.create(build=br.build, |
| 148 | target=brtarget.target, |
| 149 | task=brtarget.task) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 150 | # transpose the launch errors in ToasterExceptions |
| 151 | for brerror in br.brerror_set.all(): |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 152 | LogMessage.objects.create(build=br.build, |
| 153 | level=LogMessage.EXCEPTION, |
| 154 | message=brerror.errmsg) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 155 | |
| 156 | br.build.save() |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 157 | |
| 158 | # Make sure the LOCK is removed for builds which have been fully |
| 159 | # cancelled |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 160 | for br in BuildRequest.objects.filter( |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 161 | Q(build__outcome=Build.CANCELLED) & |
| 162 | Q(state=BuildRequest.REQ_CANCELLING) & |
| 163 | ~Q(environment=None)): |
Patrick Williams | d8c66bc | 2016-06-20 12:57:21 -0500 | [diff] [blame] | 164 | br.environment.lock = BuildEnvironment.LOCK_FREE |
| 165 | br.environment.save() |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 166 | |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 167 | def runbuild(self): |
| 168 | try: |
| 169 | self.cleanup() |
| 170 | except Exception as e: |
Andrew Geissler | 82c905d | 2020-04-13 13:39:40 -0500 | [diff] [blame] | 171 | logger.warning("runbuilds: cleanup exception %s" % str(e)) |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 172 | |
| 173 | try: |
| 174 | self.archive() |
| 175 | except Exception as e: |
Andrew Geissler | 82c905d | 2020-04-13 13:39:40 -0500 | [diff] [blame] | 176 | logger.warning("runbuilds: archive exception %s" % str(e)) |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 177 | |
| 178 | try: |
| 179 | self.schedule() |
| 180 | except Exception as e: |
Andrew Geissler | 82c905d | 2020-04-13 13:39:40 -0500 | [diff] [blame] | 181 | logger.warning("runbuilds: schedule exception %s" % str(e)) |
Patrick Williams | c124f4f | 2015-09-15 14:41:29 -0500 | [diff] [blame] | 182 | |
Andrew Geissler | 9aee500 | 2022-03-30 16:27:02 +0000 | [diff] [blame^] | 183 | # Test to see if a build pre-maturely died due to a bitbake crash |
| 184 | def check_dead_builds(self): |
| 185 | do_cleanup = False |
| 186 | try: |
| 187 | for br in BuildRequest.objects.filter(state=BuildRequest.REQ_INPROGRESS): |
| 188 | # Get the build directory |
| 189 | if br.project.builddir: |
| 190 | builddir = br.project.builddir |
| 191 | else: |
| 192 | builddir = '%s-toaster-%d' % (br.environment.builddir,br.project.id) |
| 193 | # Check log to see if there is a recent traceback |
| 194 | toaster_ui_log = os.path.join(builddir, 'toaster_ui.log') |
| 195 | test_file = os.path.join(builddir, '._toaster_check.txt') |
| 196 | os.system("tail -n 50 %s > %s" % (os.path.join(builddir, 'toaster_ui.log'),test_file)) |
| 197 | traceback_text = '' |
| 198 | is_traceback = False |
| 199 | with open(test_file,'r') as test_file_fd: |
| 200 | test_file_tail = test_file_fd.readlines() |
| 201 | for line in test_file_tail: |
| 202 | if line.startswith('Traceback (most recent call last):'): |
| 203 | traceback_text = line |
| 204 | is_traceback = True |
| 205 | elif line.startswith('NOTE: ToasterUI waiting for events'): |
| 206 | # Ignore any traceback before new build start |
| 207 | traceback_text = '' |
| 208 | is_traceback = False |
| 209 | elif line.startswith('Note: Toaster traceback auto-stop'): |
| 210 | # Ignore any traceback before this previous traceback catch |
| 211 | traceback_text = '' |
| 212 | is_traceback = False |
| 213 | elif is_traceback: |
| 214 | traceback_text += line |
| 215 | # Test the results |
| 216 | is_stop = False |
| 217 | if is_traceback: |
| 218 | # Found a traceback |
| 219 | errtype = 'Bitbake crash' |
| 220 | errmsg = 'Bitbake crash\n' + traceback_text |
| 221 | state = BuildRequest.REQ_FAILED |
| 222 | # Clean up bitbake files |
| 223 | bitbake_lock = os.path.join(builddir, 'bitbake.lock') |
| 224 | if os.path.isfile(bitbake_lock): |
| 225 | os.remove(bitbake_lock) |
| 226 | bitbake_sock = os.path.join(builddir, 'bitbake.sock') |
| 227 | if os.path.isfile(bitbake_sock): |
| 228 | os.remove(bitbake_sock) |
| 229 | if os.path.isfile(test_file): |
| 230 | os.remove(test_file) |
| 231 | # Add note to ignore this traceback on next check |
| 232 | os.system('echo "Note: Toaster traceback auto-stop" >> %s' % toaster_ui_log) |
| 233 | is_stop = True |
| 234 | # Add more tests here |
| 235 | #elif ... |
| 236 | # Stop the build request? |
| 237 | if is_stop: |
| 238 | brerror = BRError( |
| 239 | req = br, |
| 240 | errtype = errtype, |
| 241 | errmsg = errmsg, |
| 242 | traceback = traceback_text, |
| 243 | ) |
| 244 | brerror.save() |
| 245 | br.state = state |
| 246 | br.save() |
| 247 | do_cleanup = True |
| 248 | # Do cleanup |
| 249 | if do_cleanup: |
| 250 | self.cleanup() |
| 251 | except Exception as e: |
| 252 | logger.error("runbuilds: Error in check_dead_builds %s" % e) |
| 253 | |
Brad Bishop | d7bf8c1 | 2018-02-25 22:55:05 -0500 | [diff] [blame] | 254 | def handle(self, **options): |
Brad Bishop | 6e60e8b | 2018-02-01 10:27:11 -0500 | [diff] [blame] | 255 | pidfile_path = os.path.join(os.environ.get("BUILDDIR", "."), |
| 256 | ".runbuilds.pid") |
| 257 | |
| 258 | with open(pidfile_path, 'w') as pidfile: |
| 259 | pidfile.write("%s" % os.getpid()) |
| 260 | |
Andrew Geissler | 9aee500 | 2022-03-30 16:27:02 +0000 | [diff] [blame^] | 261 | # Clean up any stale/failed builds from previous Toaster run |
Patrick Williams | c0f7c04 | 2017-02-23 20:41:17 -0600 | [diff] [blame] | 262 | self.runbuild() |
| 263 | |
| 264 | signal.signal(signal.SIGUSR1, lambda sig, frame: None) |
| 265 | |
Patrick Williams | f1e5d69 | 2016-03-30 15:21:19 -0500 | [diff] [blame] | 266 | while True: |
Andrew Geissler | 9aee500 | 2022-03-30 16:27:02 +0000 | [diff] [blame^] | 267 | sigset = signal.sigtimedwait([signal.SIGUSR1], 5) |
| 268 | if sigset: |
| 269 | for sig in sigset: |
| 270 | # Consume each captured pending event |
| 271 | self.runbuild() |
| 272 | else: |
| 273 | # Check for build exceptions |
| 274 | self.check_dead_builds() |
| 275 | |