blob: 250a196fc3337008a20ef26c1104d632f09c2f23 [file] [log] [blame]
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001#!/usr/bin/env python
maruelea586f32016-04-05 11:11:33 -07002# Copyright 2012 The LUCI Authors. All rights reserved.
maruelf1f5e2a2016-05-25 17:10:39 -07003# Use of this source code is governed under the Apache License, Version 2.0
4# that can be found in the LICENSE file.
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00005
nodir55be77b2016-05-03 09:39:57 -07006"""Runs a command with optional isolated input/output.
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00007
Marc-Antoine Rueleed2f3a2019-03-14 00:00:40 +00008run_isolated takes cares of setting up a temporary environment, running a
9command, and tearing it down.
nodir55be77b2016-05-03 09:39:57 -070010
Marc-Antoine Rueleed2f3a2019-03-14 00:00:40 +000011It handles downloading and uploading isolated files, mapping CIPD packages and
12reusing stateful named caches.
13
14The isolated files, CIPD packages and named caches are kept as a global LRU
15cache.
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -050016
Roberto Carrillo71ade6d2018-10-08 22:30:24 +000017Any ${EXECUTABLE_SUFFIX} on the command line or the environment variables passed
18with the --env option will be replaced with ".exe" string on Windows and "" on
19other platforms.
nodirbe642ff2016-06-09 15:51:51 -070020
Roberto Carrillo71ade6d2018-10-08 22:30:24 +000021Any ${ISOLATED_OUTDIR} on the command line or the environment variables passed
22with the --env option will be replaced by the location of a temporary directory
23upon execution of the command specified in the .isolated file. All content
24written to this directory will be uploaded upon termination and the .isolated
25file describing this directory will be printed to stdout.
bpastene447c1992016-06-20 15:21:47 -070026
Marc-Antoine Rueleed2f3a2019-03-14 00:00:40 +000027Any ${SWARMING_BOT_FILE} on the command line or the environment variables passed
28with the --env option will be replaced by the value of the --bot-file parameter.
29This file is used by a swarming bot to communicate state of the host to tasks.
30It is written to by the swarming bot's on_before_task() hook in the swarming
31server's custom bot_config.py.
32
33See
34https://chromium.googlesource.com/infra/luci/luci-py.git/+/master/appengine/swarming/doc/Magic-Values.md
35for all the variables.
36
37See
38https://chromium.googlesource.com/infra/luci/luci-py.git/+/master/appengine/swarming/swarming_bot/config/bot_config.py
39for more information about bot_config.py.
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000040"""
41
Marc-Antoine Ruelf899c482019-10-10 23:32:06 +000042from __future__ import print_function
43
44__version__ = '1.0.1'
maruel@chromium.orgdedbf492013-09-12 20:42:11 +000045
aludwin7556e0c2016-10-26 08:46:10 -070046import argparse
maruel064c0a32016-04-05 11:47:15 -070047import base64
iannucci96fcccc2016-08-30 15:52:22 -070048import collections
vadimsh232f5a82017-01-20 19:23:44 -080049import contextlib
Sadaf Matinkhoo10743a62018-03-29 16:28:58 -040050import errno
aludwin7556e0c2016-10-26 08:46:10 -070051import json
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000052import logging
53import optparse
54import os
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -040055import re
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000056import sys
57import tempfile
maruel064c0a32016-04-05 11:47:15 -070058import time
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000059
Marc-Antoine Ruel016c7602019-04-02 18:31:13 +000060from utils import tools
61tools.force_local_third_party()
vadimsh@chromium.orga4326472013-08-24 02:05:41 +000062
Marc-Antoine Ruel016c7602019-04-02 18:31:13 +000063# third_party/
64from depot_tools import fix_encoding
Takuto Ikuta6e2ff962019-10-29 12:35:27 +000065import six
Marc-Antoine Ruel016c7602019-04-02 18:31:13 +000066
67# pylint: disable=ungrouped-imports
68import auth
69import cipd
70import isolate_storage
71import isolateserver
72import local_caching
73from libs import luci_context
Vadim Shtayura6b555c12014-07-23 16:22:18 -070074from utils import file_path
maruel12e30012015-10-09 11:55:35 -070075from utils import fs
maruel064c0a32016-04-05 11:47:15 -070076from utils import large
Marc-Antoine Ruelf74cffe2015-07-15 15:21:34 -040077from utils import logging_utils
Marc-Antoine Ruelcfb60852014-07-02 15:22:00 -040078from utils import on_error
Marc-Antoine Ruelc44f5722015-01-08 16:10:01 -050079from utils import subprocess42
maruel@chromium.orgdedbf492013-09-12 20:42:11 +000080
vadimsh@chromium.orga4326472013-08-24 02:05:41 +000081
maruele2f2cb82016-07-13 14:41:03 -070082# Magic variables that can be found in the isolate task command line.
83ISOLATED_OUTDIR_PARAMETER = '${ISOLATED_OUTDIR}'
84EXECUTABLE_SUFFIX_PARAMETER = '${EXECUTABLE_SUFFIX}'
85SWARMING_BOT_FILE_PARAMETER = '${SWARMING_BOT_FILE}'
86
87
csharp@chromium.orgff2a4662012-11-21 20:49:32 +000088# The name of the log file to use.
89RUN_ISOLATED_LOG_FILE = 'run_isolated.log'
90
maruele2f2cb82016-07-13 14:41:03 -070091
csharp@chromium.orge217f302012-11-22 16:51:53 +000092# The name of the log to use for the run_test_cases.py command
vadimsh@chromium.org8b9d56b2013-08-21 22:24:35 +000093RUN_TEST_CASES_LOG = 'run_test_cases.log'
csharp@chromium.orge217f302012-11-22 16:51:53 +000094
vadimsh@chromium.org87d63262013-04-04 19:34:21 +000095
maruele2f2cb82016-07-13 14:41:03 -070096# Use short names for temporary directories. This is driven by Windows, which
97# imposes a relatively short maximum path length of 260 characters, often
98# referred to as MAX_PATH. It is relatively easy to create files with longer
Marc-Antoine Ruel793bff32019-04-18 17:50:48 +000099# path length. A use case is with recursive dependency trees like npm packages.
maruele2f2cb82016-07-13 14:41:03 -0700100#
101# It is recommended to start the script with a `root_dir` as short as
102# possible.
103# - ir stands for isolated_run
104# - io stands for isolated_out
105# - it stands for isolated_tmp
106ISOLATED_RUN_DIR = u'ir'
107ISOLATED_OUT_DIR = u'io'
108ISOLATED_TMP_DIR = u'it'
109
110
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -0400111# Keep synced with task_request.py
Lei Leife202df2019-06-11 17:33:34 +0000112CACHE_NAME_RE = re.compile(r'^[a-z0-9_]{1,4096}$')
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -0400113
114
marueld928c862017-06-08 08:20:04 -0700115OUTLIVING_ZOMBIE_MSG = """\
116*** Swarming tried multiple times to delete the %s directory and failed ***
117*** Hard failing the task ***
118
119Swarming detected that your testing script ran an executable, which may have
120started a child executable, and the main script returned early, leaving the
121children executables playing around unguided.
122
123You don't want to leave children processes outliving the task on the Swarming
124bot, do you? The Swarming bot doesn't.
125
126How to fix?
127- For any process that starts children processes, make sure all children
128 processes terminated properly before each parent process exits. This is
129 especially important in very deep process trees.
130 - This must be done properly both in normal successful task and in case of
131 task failure. Cleanup is very important.
132- The Swarming bot sends a SIGTERM in case of timeout.
133 - You have %s seconds to comply after the signal was sent to the process
134 before the process is forcibly killed.
135- To achieve not leaking children processes in case of signals on timeout, you
136 MUST handle signals in each executable / python script and propagate them to
137 children processes.
138 - When your test script (python or binary) receives a signal like SIGTERM or
139 CTRL_BREAK_EVENT on Windows), send it to all children processes and wait for
140 them to terminate before quitting.
141
142See
Marc-Antoine Ruelc7243592018-05-24 17:04:04 -0400143https://chromium.googlesource.com/infra/luci/luci-py.git/+/master/appengine/swarming/doc/Bot.md#Graceful-termination_aka-the-SIGTERM-and-SIGKILL-dance
marueld928c862017-06-08 08:20:04 -0700144for more information.
145
146*** May the SIGKILL force be with you ***
147"""
148
149
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000150# Currently hardcoded. Eventually could be exposed as a flag once there's value.
151# 3 weeks
152MAX_AGE_SECS = 21*24*60*60
153
154
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500155TaskData = collections.namedtuple(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000156 'TaskData',
157 [
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500158 # List of strings; the command line to use, independent of what was
159 # specified in the isolated file.
160 'command',
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500161 # Relative directory to start command into.
162 'relative_cwd',
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500163 # List of strings; the arguments to add to the command specified in the
164 # isolated file.
165 'extra_args',
166 # Hash of the .isolated file that must be retrieved to recreate the tree
167 # of files to run the target executable. The command specified in the
168 # .isolated is executed. Mutually exclusive with command argument.
169 'isolated_hash',
170 # isolateserver.Storage instance to retrieve remote objects. This object
171 # has a reference to an isolateserver.StorageApi, which does the actual
172 # I/O.
173 'storage',
174 # isolateserver.LocalCache instance to keep from retrieving the same
175 # objects constantly by caching the objects retrieved. Can be on-disk or
176 # in-memory.
177 'isolate_cache',
178 # List of paths relative to root_dir to put into the output isolated
179 # bundle upon task completion (see link_outputs_to_outdir).
180 'outputs',
181 # Function (run_dir) => context manager that installs named caches into
182 # |run_dir|.
183 'install_named_caches',
184 # If True, the temporary directory will be deliberately leaked for later
185 # examination.
186 'leak_temp_dir',
187 # Path to the directory to use to create the temporary directory. If not
188 # specified, a random temporary directory is created.
189 'root_dir',
190 # Kills the process if it lasts more than this amount of seconds.
191 'hard_timeout',
192 # Number of seconds to wait between SIGTERM and SIGKILL.
193 'grace_period',
194 # Path to a file with bot state, used in place of ${SWARMING_BOT_FILE}
195 # task command line argument.
196 'bot_file',
197 # Logical account to switch LUCI_CONTEXT into.
198 'switch_to_account',
199 # Context manager dir => CipdInfo, see install_client_and_packages.
200 'install_packages_fn',
201 # Create tree with symlinks instead of hardlinks.
202 'use_symlinks',
203 # Environment variables to set.
204 'env',
205 # Environment variables to mutate with relative directories.
206 # Example: {"ENV_KEY": ['relative', 'paths', 'to', 'prepend']}
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000207 'env_prefix',
208 # Lowers the task process priority.
209 'lower_priority',
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000210 # subprocess42.Containment instance. Can be None.
211 'containment',
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000212 ])
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500213
214
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -0500215def _to_str(s):
216 """Downgrades a unicode instance to str. Pass str through as-is."""
217 if isinstance(s, str):
218 return s
219 # This is technically incorrect, especially on Windows. In theory
220 # sys.getfilesystemencoding() should be used to use the right 'ANSI code
221 # page' on Windows, but that causes other problems, as the character set
222 # is very limited.
223 return s.encode('utf-8')
224
225
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -0500226def _to_unicode(s):
227 """Upgrades a str instance to unicode. Pass unicode through as-is."""
228 if isinstance(s, unicode) or s is None:
229 return s
230 return s.decode('utf-8')
231
232
maruel03e11842016-07-14 10:50:16 -0700233def make_temp_dir(prefix, root_dir):
234 """Returns a new unique temporary directory."""
Takuto Ikuta6e2ff962019-10-29 12:35:27 +0000235 return six.text_type(tempfile.mkdtemp(prefix=prefix, dir=root_dir))
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000236
237
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500238def change_tree_read_only(rootdir, read_only):
239 """Changes the tree read-only bits according to the read_only specification.
240
241 The flag can be 0, 1 or 2, which will affect the possibility to modify files
242 and create or delete files.
243 """
244 if read_only == 2:
245 # Files and directories (except on Windows) are marked read only. This
246 # inhibits modifying, creating or deleting files in the test directory,
247 # except on Windows where creating and deleting files is still possible.
Marc-Antoine Ruele4ad07e2014-10-15 20:22:29 -0400248 file_path.make_tree_read_only(rootdir)
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500249 elif read_only == 1:
250 # Files are marked read only but not the directories. This inhibits
251 # modifying files but creating or deleting files is still possible.
Marc-Antoine Ruele4ad07e2014-10-15 20:22:29 -0400252 file_path.make_tree_files_read_only(rootdir)
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500253 elif read_only in (0, None):
Marc-Antoine Ruelf1d827c2014-11-24 15:22:25 -0500254 # Anything can be modified.
Marc-Antoine Ruel2666d9c2018-05-18 13:52:02 -0400255 # TODO(maruel): This is currently dangerous as long as
256 # DiskContentAddressedCache.touch() is not yet changed to verify the hash of
257 # the content of the files it is looking at, so that if a test modifies an
258 # input file, the file must be deleted.
Marc-Antoine Ruele4ad07e2014-10-15 20:22:29 -0400259 file_path.make_tree_writeable(rootdir)
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500260 else:
261 raise ValueError(
262 'change_tree_read_only(%s, %s): Unknown flag %s' %
263 (rootdir, read_only, read_only))
264
265
vadimsh9c54b2c2017-07-25 14:08:29 -0700266@contextlib.contextmanager
267def set_luci_context_account(account, tmp_dir):
268 """Sets LUCI_CONTEXT account to be used by the task.
269
270 If 'account' is None or '', does nothing at all. This happens when
271 run_isolated.py is called without '--switch-to-account' flag. In this case,
272 if run_isolated.py is running in some LUCI_CONTEXT environment, the task will
Takuto Ikuta33e2ff32019-09-30 12:44:03 +0000273 just inherit whatever account is already set. This may happen if users invoke
vadimsh9c54b2c2017-07-25 14:08:29 -0700274 run_isolated.py explicitly from their code.
275
276 If the requested account is not defined in the context, switches to
277 non-authenticated access. This happens for Swarming tasks that don't use
278 'task' service accounts.
279
280 If not using LUCI_CONTEXT-based auth, does nothing.
281 If already running as requested account, does nothing.
282 """
283 if not account:
284 # Not actually switching.
285 yield
286 return
287
288 local_auth = luci_context.read('local_auth')
289 if not local_auth:
290 # Not using LUCI_CONTEXT auth at all.
291 yield
292 return
293
294 # See LUCI_CONTEXT.md for the format of 'local_auth'.
295 if local_auth.get('default_account_id') == account:
296 # Already set, no need to switch.
297 yield
298 return
299
300 available = {a['id'] for a in local_auth.get('accounts') or []}
301 if account in available:
302 logging.info('Switching default LUCI_CONTEXT account to %r', account)
303 local_auth['default_account_id'] = account
304 else:
305 logging.warning(
306 'Requested LUCI_CONTEXT account %r is not available (have only %r), '
307 'disabling authentication', account, sorted(available))
308 local_auth.pop('default_account_id', None)
309
310 with luci_context.write(_tmpdir=tmp_dir, local_auth=local_auth):
311 yield
312
313
nodir90bc8dc2016-06-15 13:35:21 -0700314def process_command(command, out_dir, bot_file):
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000315 """Replaces parameters in a command line.
nodirbe642ff2016-06-09 15:51:51 -0700316
317 Raises:
318 ValueError if a parameter is requested in |command| but its value is not
319 provided.
320 """
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000321 return [replace_parameters(arg, out_dir, bot_file) for arg in command]
322
323
324def replace_parameters(arg, out_dir, bot_file):
325 """Replaces parameter tokens with appropriate values in a string.
326
327 Raises:
328 ValueError if a parameter is requested in |arg| but its value is not
329 provided.
330 """
331 arg = arg.replace(EXECUTABLE_SUFFIX_PARAMETER, cipd.EXECUTABLE_SUFFIX)
332 replace_slash = False
333 if ISOLATED_OUTDIR_PARAMETER in arg:
334 if not out_dir:
335 raise ValueError(
336 'output directory is requested in command or env var, but not '
337 'provided; please specify one')
338 arg = arg.replace(ISOLATED_OUTDIR_PARAMETER, out_dir)
339 replace_slash = True
340 if SWARMING_BOT_FILE_PARAMETER in arg:
341 if bot_file:
342 arg = arg.replace(SWARMING_BOT_FILE_PARAMETER, bot_file)
nodirbe642ff2016-06-09 15:51:51 -0700343 replace_slash = True
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000344 else:
345 logging.warning('SWARMING_BOT_FILE_PARAMETER found in command or env '
346 'var, but no bot_file specified. Leaving parameter '
347 'unchanged.')
348 if replace_slash:
349 # Replace slashes only if parameters are present
350 # because of arguments like '${ISOLATED_OUTDIR}/foo/bar'
351 arg = arg.replace('/', os.sep)
352 return arg
maruela9cfd6f2015-09-15 11:03:15 -0700353
354
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000355
356def get_command_env(tmp_dir, cipd_info, run_dir, env, env_prefixes, out_dir,
357 bot_file):
vadimsh232f5a82017-01-20 19:23:44 -0800358 """Returns full OS environment to run a command in.
359
Robert Iannuccibf5f84c2017-11-22 12:56:50 -0800360 Sets up TEMP, puts directory with cipd binary in front of PATH, exposes
361 CIPD_CACHE_DIR env var, and installs all env_prefixes.
vadimsh232f5a82017-01-20 19:23:44 -0800362
363 Args:
364 tmp_dir: temp directory.
365 cipd_info: CipdInfo object is cipd client is used, None if not.
Marc-Antoine Ruel9ec1e9f2017-12-20 16:36:54 -0500366 run_dir: The root directory the isolated tree is mapped in.
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500367 env: environment variables to use
Robert Iannuccibf5f84c2017-11-22 12:56:50 -0800368 env_prefixes: {"ENV_KEY": ['cwd', 'relative', 'paths', 'to', 'prepend']}
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000369 out_dir: Isolated output directory. Required to be != None if any of the
370 env vars contain ISOLATED_OUTDIR_PARAMETER.
371 bot_file: Required to be != None if any of the env vars contain
372 SWARMING_BOT_FILE_PARAMETER.
vadimsh232f5a82017-01-20 19:23:44 -0800373 """
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500374 out = os.environ.copy()
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000375 for k, v in env.items():
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500376 if not v:
Marc-Antoine Ruel9ec1e9f2017-12-20 16:36:54 -0500377 out.pop(k, None)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500378 else:
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000379 out[k] = replace_parameters(v, out_dir, bot_file)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500380
381 if cipd_info:
382 bin_dir = os.path.dirname(cipd_info.client.binary_path)
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -0500383 out['PATH'] = '%s%s%s' % (_to_str(bin_dir), os.pathsep, out['PATH'])
384 out['CIPD_CACHE_DIR'] = _to_str(cipd_info.cache_dir)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500385
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000386 for key, paths in env_prefixes.items():
Marc-Antoine Ruel9ec1e9f2017-12-20 16:36:54 -0500387 assert isinstance(paths, list), paths
388 paths = [os.path.normpath(os.path.join(run_dir, p)) for p in paths]
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500389 cur = out.get(key)
390 if cur:
391 paths.append(cur)
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -0500392 out[key] = _to_str(os.path.pathsep.join(paths))
vadimsh232f5a82017-01-20 19:23:44 -0800393
Marc-Antoine Ruelefb30b12018-07-25 18:34:36 +0000394 tmp_dir = _to_str(tmp_dir)
395 # pylint: disable=line-too-long
396 # * python respects $TMPDIR, $TEMP, and $TMP in this order, regardless of
397 # platform. So $TMPDIR must be set on all platforms.
398 # https://github.com/python/cpython/blob/2.7/Lib/tempfile.py#L155
399 out['TMPDIR'] = tmp_dir
400 if sys.platform == 'win32':
401 # * chromium's base utils uses GetTempPath().
402 # https://cs.chromium.org/chromium/src/base/files/file_util_win.cc?q=GetTempPath
403 # * Go uses GetTempPath().
404 # * GetTempDir() uses %TMP%, then %TEMP%, then other stuff. So %TMP% must be
405 # set.
406 # https://docs.microsoft.com/en-us/windows/desktop/api/fileapi/nf-fileapi-gettemppathw
407 out['TMP'] = tmp_dir
408 # https://blogs.msdn.microsoft.com/oldnewthing/20150417-00/?p=44213
409 out['TEMP'] = tmp_dir
410 elif sys.platform == 'darwin':
411 # * Chromium uses an hack on macOS before calling into
412 # NSTemporaryDirectory().
413 # https://cs.chromium.org/chromium/src/base/files/file_util_mac.mm?q=GetTempDir
414 # https://developer.apple.com/documentation/foundation/1409211-nstemporarydirectory
415 out['MAC_CHROMIUM_TMPDIR'] = tmp_dir
416 else:
417 # TMPDIR is specified as the POSIX standard envvar for the temp directory.
418 # http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html
419 # * mktemp on linux respects $TMPDIR.
420 # * Chromium respects $TMPDIR on linux.
421 # https://cs.chromium.org/chromium/src/base/files/file_util_posix.cc?q=GetTempDir
422 # * Go uses $TMPDIR.
423 # https://go.googlesource.com/go/+/go1.10.3/src/os/file_unix.go#307
424 pass
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500425 return out
vadimsh232f5a82017-01-20 19:23:44 -0800426
427
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000428def run_command(
429 command, cwd, env, hard_timeout, grace_period, lower_priority, containment):
maruel6be7f9e2015-10-01 12:25:30 -0700430 """Runs the command.
431
432 Returns:
433 tuple(process exit code, bool if had a hard timeout)
434 """
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000435 logging.info(
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000436 'run_command(%s, %s, %s, %s, %s, %s)',
437 command, cwd, hard_timeout, grace_period, lower_priority, containment)
marueleb5fbee2015-09-17 13:01:36 -0700438
maruel6be7f9e2015-10-01 12:25:30 -0700439 exit_code = None
440 had_hard_timeout = False
maruela9cfd6f2015-09-15 11:03:15 -0700441 with tools.Profiler('RunTest'):
maruel6be7f9e2015-10-01 12:25:30 -0700442 proc = None
443 had_signal = []
maruela9cfd6f2015-09-15 11:03:15 -0700444 try:
maruel6be7f9e2015-10-01 12:25:30 -0700445 # TODO(maruel): This code is imperfect. It doesn't handle well signals
446 # during the download phase and there's short windows were things can go
447 # wrong.
448 def handler(signum, _frame):
449 if proc and not had_signal:
450 logging.info('Received signal %d', signum)
451 had_signal.append(True)
maruel556d9052015-10-05 11:12:44 -0700452 raise subprocess42.TimeoutExpired(command, None)
maruel6be7f9e2015-10-01 12:25:30 -0700453
Marc-Antoine Ruel30b80fe2019-02-08 13:51:31 +0000454 proc = subprocess42.Popen(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000455 command, cwd=cwd, env=env, detached=True, close_fds=True,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000456 lower_priority=lower_priority, containment=containment)
maruel6be7f9e2015-10-01 12:25:30 -0700457 with subprocess42.set_signal_handler(subprocess42.STOP_SIGNALS, handler):
458 try:
John Budorickc398f092019-06-10 22:49:44 +0000459 exit_code = proc.wait(hard_timeout or None)
maruel6be7f9e2015-10-01 12:25:30 -0700460 except subprocess42.TimeoutExpired:
461 if not had_signal:
462 logging.warning('Hard timeout')
463 had_hard_timeout = True
464 logging.warning('Sending SIGTERM')
465 proc.terminate()
466
467 # Ignore signals in grace period. Forcibly give the grace period to the
468 # child process.
469 if exit_code is None:
470 ignore = lambda *_: None
471 with subprocess42.set_signal_handler(subprocess42.STOP_SIGNALS, ignore):
472 try:
473 exit_code = proc.wait(grace_period or None)
474 except subprocess42.TimeoutExpired:
475 # Now kill for real. The user can distinguish between the
476 # following states:
477 # - signal but process exited within grace period,
478 # hard_timed_out will be set but the process exit code will be
479 # script provided.
480 # - processed exited late, exit code will be -9 on posix.
481 logging.warning('Grace exhausted; sending SIGKILL')
482 proc.kill()
martiniss5c8043e2017-08-01 17:09:43 -0700483 logging.info('Waiting for process exit')
maruel6be7f9e2015-10-01 12:25:30 -0700484 exit_code = proc.wait()
maruela9cfd6f2015-09-15 11:03:15 -0700485 except OSError:
486 # This is not considered to be an internal error. The executable simply
487 # does not exit.
maruela72f46e2016-02-24 11:05:45 -0800488 sys.stderr.write(
tikuta2d678212019-09-23 23:12:08 +0000489 '<The executable does not exist, a dependent library is missing or '
490 'the command line is too long>\n'
491 '<Check for missing .so/.dll in the .isolate or GN file or length of '
492 'command line args>\n'
maruela72f46e2016-02-24 11:05:45 -0800493 '<Command: %s>\n' % command)
494 if os.environ.get('SWARMING_TASK_ID'):
495 # Give an additional hint when running as a swarming task.
496 sys.stderr.write(
497 '<See the task\'s page for commands to help diagnose this issue '
498 'by reproducing the task locally>\n')
maruela9cfd6f2015-09-15 11:03:15 -0700499 exit_code = 1
500 logging.info(
501 'Command finished with exit code %d (%s)',
502 exit_code, hex(0xffffffff & exit_code))
maruel6be7f9e2015-10-01 12:25:30 -0700503 return exit_code, had_hard_timeout
maruela9cfd6f2015-09-15 11:03:15 -0700504
505
maruel4409e302016-07-19 14:25:51 -0700506def fetch_and_map(isolated_hash, storage, cache, outdir, use_symlinks):
507 """Fetches an isolated tree, create the tree and returns (bundle, stats)."""
nodir6f801882016-04-29 14:41:50 -0700508 start = time.time()
509 bundle = isolateserver.fetch_isolated(
510 isolated_hash=isolated_hash,
511 storage=storage,
512 cache=cache,
maruel4409e302016-07-19 14:25:51 -0700513 outdir=outdir,
514 use_symlinks=use_symlinks)
Takuto Ikuta2b9640e2019-06-19 00:53:23 +0000515 hot = (collections.Counter(cache.used) -
516 collections.Counter(cache.added)).elements()
nodir6f801882016-04-29 14:41:50 -0700517 return bundle, {
518 'duration': time.time() - start,
nodir6f801882016-04-29 14:41:50 -0700519 'items_cold': base64.b64encode(large.pack(sorted(cache.added))),
Takuto Ikuta2b9640e2019-06-19 00:53:23 +0000520 'items_hot': base64.b64encode(large.pack(sorted(hot))),
nodir6f801882016-04-29 14:41:50 -0700521 }
522
523
aludwin0a8e17d2016-10-27 15:57:39 -0700524def link_outputs_to_outdir(run_dir, out_dir, outputs):
525 """Links any named outputs to out_dir so they can be uploaded.
526
527 Raises an error if the file already exists in that directory.
528 """
529 if not outputs:
530 return
531 isolateserver.create_directories(out_dir, outputs)
532 for o in outputs:
Sadaf Matinkhoo10743a62018-03-29 16:28:58 -0400533 copy_recursively(os.path.join(run_dir, o), os.path.join(out_dir, o))
534
535
536def copy_recursively(src, dst):
537 """Efficiently copies a file or directory from src_dir to dst_dir.
538
539 `item` may be a file, directory, or a symlink to a file or directory.
540 All symlinks are replaced with their targets, so the resulting
541 directory structure in dst_dir will never have any symlinks.
542
543 To increase speed, copy_recursively hardlinks individual files into the
544 (newly created) directory structure if possible, unlike Python's
545 shutil.copytree().
546 """
547 orig_src = src
548 try:
549 # Replace symlinks with their final target.
550 while fs.islink(src):
551 res = fs.readlink(src)
552 src = os.path.join(os.path.dirname(src), res)
553 # TODO(sadafm): Explicitly handle cyclic symlinks.
554
555 # Note that fs.isfile (which is a wrapper around os.path.isfile) throws
556 # an exception if src does not exist. A warning will be logged in that case.
557 if fs.isfile(src):
558 file_path.link_file(dst, src, file_path.HARDLINK_WITH_FALLBACK)
559 return
560
561 if not fs.exists(dst):
562 os.makedirs(dst)
563
564 for child in fs.listdir(src):
565 copy_recursively(os.path.join(src, child), os.path.join(dst, child))
566
567 except OSError as e:
568 if e.errno == errno.ENOENT:
569 logging.warning('Path %s does not exist or %s is a broken symlink',
570 src, orig_src)
571 else:
572 logging.info("Couldn't collect output file %s: %s", src, e)
aludwin0a8e17d2016-10-27 15:57:39 -0700573
574
Takuto Ikutab4aa8662019-09-17 05:54:36 +0000575def upload_then_delete(storage, out_dir, leak_temp_dir):
maruela9cfd6f2015-09-15 11:03:15 -0700576 """Deletes the temporary run directory and uploads results back.
577
578 Returns:
nodir6f801882016-04-29 14:41:50 -0700579 tuple(outputs_ref, success, stats)
maruel064c0a32016-04-05 11:47:15 -0700580 - outputs_ref: a dict referring to the results archived back to the isolated
581 server, if applicable.
582 - success: False if something occurred that means that the task must
583 forcibly be considered a failure, e.g. zombie processes were left
584 behind.
nodir6f801882016-04-29 14:41:50 -0700585 - stats: uploading stats.
maruela9cfd6f2015-09-15 11:03:15 -0700586 """
maruela9cfd6f2015-09-15 11:03:15 -0700587 # Upload out_dir and generate a .isolated file out of this directory. It is
588 # only done if files were written in the directory.
589 outputs_ref = None
maruel064c0a32016-04-05 11:47:15 -0700590 cold = []
591 hot = []
nodir6f801882016-04-29 14:41:50 -0700592 start = time.time()
593
maruel12e30012015-10-09 11:55:35 -0700594 if fs.isdir(out_dir) and fs.listdir(out_dir):
maruela9cfd6f2015-09-15 11:03:15 -0700595 with tools.Profiler('ArchiveOutput'):
596 try:
maruel064c0a32016-04-05 11:47:15 -0700597 results, f_cold, f_hot = isolateserver.archive_files_to_storage(
maruela9cfd6f2015-09-15 11:03:15 -0700598 storage, [out_dir], None)
599 outputs_ref = {
Marc-Antoine Rueld0868ec2018-11-28 20:47:29 +0000600 'isolated': results.values()[0],
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +0000601 'isolatedserver': storage.server_ref.url,
602 'namespace': storage.server_ref.namespace,
maruela9cfd6f2015-09-15 11:03:15 -0700603 }
maruel064c0a32016-04-05 11:47:15 -0700604 cold = sorted(i.size for i in f_cold)
605 hot = sorted(i.size for i in f_hot)
maruela9cfd6f2015-09-15 11:03:15 -0700606 except isolateserver.Aborted:
607 # This happens when a signal SIGTERM was received while uploading data.
608 # There is 2 causes:
609 # - The task was too slow and was about to be killed anyway due to
610 # exceeding the hard timeout.
611 # - The amount of data uploaded back is very large and took too much
612 # time to archive.
613 sys.stderr.write('Received SIGTERM while uploading')
614 # Re-raise, so it will be treated as an internal failure.
615 raise
nodir6f801882016-04-29 14:41:50 -0700616
617 success = False
maruela9cfd6f2015-09-15 11:03:15 -0700618 try:
maruel12e30012015-10-09 11:55:35 -0700619 if (not leak_temp_dir and fs.isdir(out_dir) and
maruel6eeea7d2015-09-16 12:17:42 -0700620 not file_path.rmtree(out_dir)):
maruela9cfd6f2015-09-15 11:03:15 -0700621 logging.error('Had difficulties removing out_dir %s', out_dir)
nodir6f801882016-04-29 14:41:50 -0700622 else:
623 success = True
maruela9cfd6f2015-09-15 11:03:15 -0700624 except OSError as e:
625 # When this happens, it means there's a process error.
maruel12e30012015-10-09 11:55:35 -0700626 logging.exception('Had difficulties removing out_dir %s: %s', out_dir, e)
nodir6f801882016-04-29 14:41:50 -0700627 stats = {
628 'duration': time.time() - start,
629 'items_cold': base64.b64encode(large.pack(cold)),
630 'items_hot': base64.b64encode(large.pack(hot)),
631 }
632 return outputs_ref, success, stats
maruela9cfd6f2015-09-15 11:03:15 -0700633
634
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500635def map_and_run(data, constant_run_path):
nodir55be77b2016-05-03 09:39:57 -0700636 """Runs a command with optional isolated input/output.
637
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500638 Arguments:
639 - data: TaskData instance.
640 - constant_run_path: TODO
nodir55be77b2016-05-03 09:39:57 -0700641
642 Returns metadata about the result.
643 """
maruela9cfd6f2015-09-15 11:03:15 -0700644 result = {
Takuto Ikuta5ed62ad2019-09-26 09:16:00 +0000645 'duration': None,
646 'exit_code': None,
647 'had_hard_timeout': False,
648 'internal_failure': 'run_isolated did not complete properly',
649 'stats': {
650 #'cipd': {
651 # 'duration': 0.,
652 # 'get_client_duration': 0.,
653 #},
654 'isolated': {
655 'download': {
656 #'duration': 0.,
657 'initial_number_items': len(data.isolate_cache),
658 'initial_size': data.isolate_cache.total_size,
659 #'items_cold': '<large.pack()>',
660 #'items_hot': '<large.pack()>',
661 },
662 #'upload': {
663 # 'duration': 0.,
664 # 'items_cold': '<large.pack()>',
665 # 'items_hot': '<large.pack()>',
666 #},
667 },
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000668 },
Takuto Ikuta5ed62ad2019-09-26 09:16:00 +0000669 #'cipd_pins': {
670 # 'packages': [
671 # {'package_name': ..., 'version': ..., 'path': ...},
672 # ...
673 # ],
674 # 'client_package': {'package_name': ..., 'version': ...},
675 #},
676 'outputs_ref': None,
677 'version': 5,
maruela9cfd6f2015-09-15 11:03:15 -0700678 }
nodirbe642ff2016-06-09 15:51:51 -0700679
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500680 if data.root_dir:
Lei Leife202df2019-06-11 17:33:34 +0000681 file_path.ensure_tree(data.root_dir, 0o700)
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500682 elif data.isolate_cache.cache_dir:
683 data = data._replace(
684 root_dir=os.path.dirname(data.isolate_cache.cache_dir))
maruele2f2cb82016-07-13 14:41:03 -0700685 # See comment for these constants.
maruelcffa0542017-04-07 08:39:20 -0700686 # If root_dir is not specified, it is not constant.
687 # TODO(maruel): This is not obvious. Change this to become an error once we
688 # make the constant_run_path an exposed flag.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500689 if constant_run_path and data.root_dir:
690 run_dir = os.path.join(data.root_dir, ISOLATED_RUN_DIR)
maruel5c4eed82017-05-26 05:33:40 -0700691 if os.path.isdir(run_dir):
692 file_path.rmtree(run_dir)
Lei Leife202df2019-06-11 17:33:34 +0000693 os.mkdir(run_dir, 0o700)
maruelcffa0542017-04-07 08:39:20 -0700694 else:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500695 run_dir = make_temp_dir(ISOLATED_RUN_DIR, data.root_dir)
maruel03e11842016-07-14 10:50:16 -0700696 # storage should be normally set but don't crash if it is not. This can happen
697 # as Swarming task can run without an isolate server.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500698 out_dir = make_temp_dir(
699 ISOLATED_OUT_DIR, data.root_dir) if data.storage else None
700 tmp_dir = make_temp_dir(ISOLATED_TMP_DIR, data.root_dir)
nodir55be77b2016-05-03 09:39:57 -0700701 cwd = run_dir
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500702 if data.relative_cwd:
703 cwd = os.path.normpath(os.path.join(cwd, data.relative_cwd))
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500704 command = data.command
nodir55be77b2016-05-03 09:39:57 -0700705 try:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500706 with data.install_packages_fn(run_dir) as cipd_info:
vadimsh232f5a82017-01-20 19:23:44 -0800707 if cipd_info:
708 result['stats']['cipd'] = cipd_info.stats
709 result['cipd_pins'] = cipd_info.pins
nodir90bc8dc2016-06-15 13:35:21 -0700710
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500711 if data.isolated_hash:
vadimsh232f5a82017-01-20 19:23:44 -0800712 isolated_stats = result['stats'].setdefault('isolated', {})
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000713 bundle, stats = fetch_and_map(
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500714 isolated_hash=data.isolated_hash,
715 storage=data.storage,
716 cache=data.isolate_cache,
vadimsh232f5a82017-01-20 19:23:44 -0800717 outdir=run_dir,
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500718 use_symlinks=data.use_symlinks)
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000719 isolated_stats['download'].update(stats)
vadimsh232f5a82017-01-20 19:23:44 -0800720 change_tree_read_only(run_dir, bundle.read_only)
maruelabec63c2017-04-26 11:53:24 -0700721 # Inject the command
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500722 if not command and bundle.command:
723 command = bundle.command + data.extra_args
Marc-Antoine Rueld704a1f2017-10-31 10:51:23 -0400724 # Only set the relative directory if the isolated file specified a
725 # command, and no raw command was specified.
726 if bundle.relative_cwd:
727 cwd = os.path.normpath(os.path.join(cwd, bundle.relative_cwd))
maruelabec63c2017-04-26 11:53:24 -0700728
729 if not command:
730 # Handle this as a task failure, not an internal failure.
731 sys.stderr.write(
732 '<No command was specified!>\n'
733 '<Please secify a command when triggering your Swarming task>\n')
734 result['exit_code'] = 1
735 return result
nodirbe642ff2016-06-09 15:51:51 -0700736
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500737 if not cwd.startswith(run_dir):
738 # Handle this as a task failure, not an internal failure. This is a
739 # 'last chance' way to gate against directory escape.
740 sys.stderr.write('<Relative CWD is outside of run directory!>\n')
741 result['exit_code'] = 1
742 return result
743
744 if not os.path.isdir(cwd):
745 # Accepts relative_cwd that does not exist.
Lei Leife202df2019-06-11 17:33:34 +0000746 os.makedirs(cwd, 0o700)
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500747
vadimsh232f5a82017-01-20 19:23:44 -0800748 # If we have an explicit list of files to return, make sure their
749 # directories exist now.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500750 if data.storage and data.outputs:
751 isolateserver.create_directories(run_dir, data.outputs)
aludwin0a8e17d2016-10-27 15:57:39 -0700752
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500753 with data.install_named_caches(run_dir):
nodird6160682017-02-02 13:03:35 -0800754 sys.stdout.flush()
755 start = time.time()
756 try:
vadimsh9c54b2c2017-07-25 14:08:29 -0700757 # Need to switch the default account before 'get_command_env' call,
758 # so it can grab correct value of LUCI_CONTEXT env var.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500759 with set_luci_context_account(data.switch_to_account, tmp_dir):
760 env = get_command_env(
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000761 tmp_dir, cipd_info, run_dir, data.env, data.env_prefix, out_dir,
762 data.bot_file)
Brian Sheedy7a761172019-08-30 22:55:14 +0000763 command = tools.find_executable(command, env)
Robert Iannucci24ae76a2018-02-26 12:51:18 -0800764 command = process_command(command, out_dir, data.bot_file)
765 file_path.ensure_command_has_abs_path(command, cwd)
766
vadimsh9c54b2c2017-07-25 14:08:29 -0700767 result['exit_code'], result['had_hard_timeout'] = run_command(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000768 command, cwd, env, data.hard_timeout, data.grace_period,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000769 data.lower_priority, data.containment)
nodird6160682017-02-02 13:03:35 -0800770 finally:
771 result['duration'] = max(time.time() - start, 0)
Seth Koehler49139812017-12-19 13:59:33 -0500772
773 # We successfully ran the command, set internal_failure back to
774 # None (even if the command failed, it's not an internal error).
775 result['internal_failure'] = None
maruela9cfd6f2015-09-15 11:03:15 -0700776 except Exception as e:
nodir90bc8dc2016-06-15 13:35:21 -0700777 # An internal error occurred. Report accordingly so the swarming task will
778 # be retried automatically.
maruel12e30012015-10-09 11:55:35 -0700779 logging.exception('internal failure: %s', e)
maruela9cfd6f2015-09-15 11:03:15 -0700780 result['internal_failure'] = str(e)
781 on_error.report(None)
aludwin0a8e17d2016-10-27 15:57:39 -0700782
783 # Clean up
maruela9cfd6f2015-09-15 11:03:15 -0700784 finally:
785 try:
aludwin0a8e17d2016-10-27 15:57:39 -0700786 # Try to link files to the output directory, if specified.
787 if out_dir:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500788 link_outputs_to_outdir(run_dir, out_dir, data.outputs)
aludwin0a8e17d2016-10-27 15:57:39 -0700789
nodir32a1ec12016-10-26 18:34:07 -0700790 success = False
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500791 if data.leak_temp_dir:
nodir32a1ec12016-10-26 18:34:07 -0700792 success = True
maruela9cfd6f2015-09-15 11:03:15 -0700793 logging.warning(
794 'Deliberately leaking %s for later examination', run_dir)
marueleb5fbee2015-09-17 13:01:36 -0700795 else:
maruel84537cb2015-10-16 14:21:28 -0700796 # On Windows rmtree(run_dir) call above has a synchronization effect: it
797 # finishes only when all task child processes terminate (since a running
798 # process locks *.exe file). Examine out_dir only after that call
799 # completes (since child processes may write to out_dir too and we need
800 # to wait for them to finish).
801 if fs.isdir(run_dir):
802 try:
803 success = file_path.rmtree(run_dir)
804 except OSError as e:
Marc-Antoine Ruel44699b32018-09-24 23:31:50 +0000805 logging.error('rmtree(%r) failed: %s', run_dir, e)
maruel84537cb2015-10-16 14:21:28 -0700806 success = False
807 if not success:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500808 sys.stderr.write(OUTLIVING_ZOMBIE_MSG % ('run', data.grace_period))
maruel84537cb2015-10-16 14:21:28 -0700809 if result['exit_code'] == 0:
810 result['exit_code'] = 1
811 if fs.isdir(tmp_dir):
812 try:
813 success = file_path.rmtree(tmp_dir)
814 except OSError as e:
Marc-Antoine Ruel44699b32018-09-24 23:31:50 +0000815 logging.error('rmtree(%r) failed: %s', tmp_dir, e)
maruel84537cb2015-10-16 14:21:28 -0700816 success = False
817 if not success:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500818 sys.stderr.write(OUTLIVING_ZOMBIE_MSG % ('temp', data.grace_period))
maruel84537cb2015-10-16 14:21:28 -0700819 if result['exit_code'] == 0:
820 result['exit_code'] = 1
maruela9cfd6f2015-09-15 11:03:15 -0700821
marueleb5fbee2015-09-17 13:01:36 -0700822 # This deletes out_dir if leak_temp_dir is not set.
nodir9130f072016-05-27 13:59:08 -0700823 if out_dir:
nodir55715712016-06-03 12:28:19 -0700824 isolated_stats = result['stats'].setdefault('isolated', {})
825 result['outputs_ref'], success, isolated_stats['upload'] = (
Takuto Ikutab4aa8662019-09-17 05:54:36 +0000826 upload_then_delete(data.storage, out_dir, data.leak_temp_dir))
maruela9cfd6f2015-09-15 11:03:15 -0700827 if not success and result['exit_code'] == 0:
828 result['exit_code'] = 1
829 except Exception as e:
830 # Swallow any exception in the main finally clause.
nodir9130f072016-05-27 13:59:08 -0700831 if out_dir:
832 logging.exception('Leaking out_dir %s: %s', out_dir, e)
maruela9cfd6f2015-09-15 11:03:15 -0700833 result['internal_failure'] = str(e)
834 return result
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -0500835
836
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500837def run_tha_test(data, result_json):
nodir55be77b2016-05-03 09:39:57 -0700838 """Runs an executable and records execution metadata.
839
nodir55be77b2016-05-03 09:39:57 -0700840 If isolated_hash is specified, downloads the dependencies in the cache,
841 hardlinks them into a temporary directory and runs the command specified in
842 the .isolated.
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -0500843
844 A temporary directory is created to hold the output files. The content inside
845 this directory will be uploaded back to |storage| packaged as a .isolated
846 file.
847
848 Arguments:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500849 - data: TaskData instance.
850 - result_json: File path to dump result metadata into. If set, the process
851 exit code is always 0 unless an internal error occurred.
maruela9cfd6f2015-09-15 11:03:15 -0700852
853 Returns:
854 Process exit code that should be used.
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000855 """
maruela76b9ee2015-12-15 06:18:08 -0800856 if result_json:
857 # Write a json output file right away in case we get killed.
858 result = {
859 'exit_code': None,
860 'had_hard_timeout': False,
861 'internal_failure': 'Was terminated before completion',
862 'outputs_ref': None,
nodirbe642ff2016-06-09 15:51:51 -0700863 'version': 5,
maruela76b9ee2015-12-15 06:18:08 -0800864 }
865 tools.write_json(result_json, result, dense=True)
866
maruela9cfd6f2015-09-15 11:03:15 -0700867 # run_isolated exit code. Depends on if result_json is used or not.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500868 result = map_and_run(data, True)
maruela9cfd6f2015-09-15 11:03:15 -0700869 logging.info('Result:\n%s', tools.format_json(result, dense=True))
bpastene3ae09522016-06-10 17:12:59 -0700870
maruela9cfd6f2015-09-15 11:03:15 -0700871 if result_json:
maruel05d5a882015-09-21 13:59:02 -0700872 # We've found tests to delete 'work' when quitting, causing an exception
873 # here. Try to recreate the directory if necessary.
nodire5028a92016-04-29 14:38:21 -0700874 file_path.ensure_tree(os.path.dirname(result_json))
maruela9cfd6f2015-09-15 11:03:15 -0700875 tools.write_json(result_json, result, dense=True)
876 # Only return 1 if there was an internal error.
877 return int(bool(result['internal_failure']))
maruel@chromium.org781ccf62013-09-17 19:39:47 +0000878
maruela9cfd6f2015-09-15 11:03:15 -0700879 # Marshall into old-style inline output.
880 if result['outputs_ref']:
Marc-Antoine Ruel793bff32019-04-18 17:50:48 +0000881 # pylint: disable=unsubscriptable-object
maruela9cfd6f2015-09-15 11:03:15 -0700882 data = {
883 'hash': result['outputs_ref']['isolated'],
884 'namespace': result['outputs_ref']['namespace'],
885 'storage': result['outputs_ref']['isolatedserver'],
886 }
Marc-Antoine Ruelc44f5722015-01-08 16:10:01 -0500887 sys.stdout.flush()
maruela9cfd6f2015-09-15 11:03:15 -0700888 print(
889 '[run_isolated_out_hack]%s[/run_isolated_out_hack]' %
890 tools.format_json(data, dense=True))
maruelb76604c2015-11-11 11:53:44 -0800891 sys.stdout.flush()
maruela9cfd6f2015-09-15 11:03:15 -0700892 return result['exit_code'] or int(bool(result['internal_failure']))
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000893
894
iannuccib58d10d2017-03-18 02:00:25 -0700895# Yielded by 'install_client_and_packages'.
vadimsh232f5a82017-01-20 19:23:44 -0800896CipdInfo = collections.namedtuple('CipdInfo', [
897 'client', # cipd.CipdClient object
898 'cache_dir', # absolute path to bot-global cipd tag and instance cache
899 'stats', # dict with stats to return to the server
900 'pins', # dict with installed cipd pins to return to the server
901])
902
903
904@contextlib.contextmanager
905def noop_install_packages(_run_dir):
iannuccib58d10d2017-03-18 02:00:25 -0700906 """Placeholder for 'install_client_and_packages' if cipd is disabled."""
vadimsh232f5a82017-01-20 19:23:44 -0800907 yield None
908
909
iannuccib58d10d2017-03-18 02:00:25 -0700910def _install_packages(run_dir, cipd_cache_dir, client, packages, timeout):
911 """Calls 'cipd ensure' for packages.
912
913 Args:
914 run_dir (str): root of installation.
915 cipd_cache_dir (str): the directory to use for the cipd package cache.
916 client (CipdClient): the cipd client to use
917 packages: packages to install, list [(path, package_name, version), ...].
918 timeout: max duration in seconds that this function can take.
919
920 Returns: list of pinned packages. Looks like [
921 {
922 'path': 'subdirectory',
923 'package_name': 'resolved/package/name',
924 'version': 'deadbeef...',
925 },
926 ...
927 ]
928 """
929 package_pins = [None]*len(packages)
930 def insert_pin(path, name, version, idx):
931 package_pins[idx] = {
932 'package_name': name,
933 # swarming deals with 'root' as '.'
934 'path': path or '.',
935 'version': version,
936 }
937
938 by_path = collections.defaultdict(list)
939 for i, (path, name, version) in enumerate(packages):
940 # cipd deals with 'root' as ''
941 if path == '.':
942 path = ''
943 by_path[path].append((name, version, i))
944
945 pins = client.ensure(
946 run_dir,
947 {
948 subdir: [(name, vers) for name, vers, _ in pkgs]
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000949 for subdir, pkgs in by_path.items()
iannuccib58d10d2017-03-18 02:00:25 -0700950 },
951 cache_dir=cipd_cache_dir,
952 timeout=timeout,
953 )
954
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000955 for subdir, pin_list in sorted(pins.items()):
iannuccib58d10d2017-03-18 02:00:25 -0700956 this_subdir = by_path[subdir]
957 for i, (name, version) in enumerate(pin_list):
958 insert_pin(subdir, name, version, this_subdir[i][2])
959
Robert Iannucci461b30d2017-12-13 11:34:03 -0800960 assert None not in package_pins, (packages, pins, package_pins)
iannuccib58d10d2017-03-18 02:00:25 -0700961
962 return package_pins
963
964
vadimsh232f5a82017-01-20 19:23:44 -0800965@contextlib.contextmanager
iannuccib58d10d2017-03-18 02:00:25 -0700966def install_client_and_packages(
nodirff531b42016-06-23 13:05:06 -0700967 run_dir, packages, service_url, client_package_name,
vadimsh232f5a82017-01-20 19:23:44 -0800968 client_version, cache_dir, timeout=None):
vadimsh902948e2017-01-20 15:57:32 -0800969 """Bootstraps CIPD client and installs CIPD packages.
iannucci96fcccc2016-08-30 15:52:22 -0700970
vadimsh232f5a82017-01-20 19:23:44 -0800971 Yields CipdClient, stats, client info and pins (as single CipdInfo object).
972
973 Pins and the CIPD client info are in the form of:
iannucci96fcccc2016-08-30 15:52:22 -0700974 [
975 {
976 "path": path, "package_name": package_name, "version": version,
977 },
978 ...
979 ]
vadimsh902948e2017-01-20 15:57:32 -0800980 (the CIPD client info is a single dictionary instead of a list)
iannucci96fcccc2016-08-30 15:52:22 -0700981
982 such that they correspond 1:1 to all input package arguments from the command
983 line. These dictionaries make their all the way back to swarming, where they
984 become the arguments of CipdPackage.
nodirbe642ff2016-06-09 15:51:51 -0700985
vadimsh902948e2017-01-20 15:57:32 -0800986 If 'packages' list is empty, will bootstrap CIPD client, but won't install
987 any packages.
988
989 The bootstrapped client (regardless whether 'packages' list is empty or not),
vadimsh232f5a82017-01-20 19:23:44 -0800990 will be made available to the task via $PATH.
vadimsh902948e2017-01-20 15:57:32 -0800991
nodirbe642ff2016-06-09 15:51:51 -0700992 Args:
nodir90bc8dc2016-06-15 13:35:21 -0700993 run_dir (str): root of installation.
vadimsh902948e2017-01-20 15:57:32 -0800994 packages: packages to install, list [(path, package_name, version), ...].
nodirbe642ff2016-06-09 15:51:51 -0700995 service_url (str): CIPD server url, e.g.
996 "https://chrome-infra-packages.appspot.com."
nodir90bc8dc2016-06-15 13:35:21 -0700997 client_package_name (str): CIPD package name of CIPD client.
998 client_version (str): Version of CIPD client.
nodirbe642ff2016-06-09 15:51:51 -0700999 cache_dir (str): where to keep cache of cipd clients, packages and tags.
1000 timeout: max duration in seconds that this function can take.
nodirbe642ff2016-06-09 15:51:51 -07001001 """
1002 assert cache_dir
nodir90bc8dc2016-06-15 13:35:21 -07001003
nodirbe642ff2016-06-09 15:51:51 -07001004 timeoutfn = tools.sliding_timeout(timeout)
nodirbe642ff2016-06-09 15:51:51 -07001005 start = time.time()
nodirbe642ff2016-06-09 15:51:51 -07001006
vadimsh902948e2017-01-20 15:57:32 -08001007 cache_dir = os.path.abspath(cache_dir)
vadimsh232f5a82017-01-20 19:23:44 -08001008 cipd_cache_dir = os.path.join(cache_dir, 'cache') # tag and instance caches
nodir90bc8dc2016-06-15 13:35:21 -07001009 run_dir = os.path.abspath(run_dir)
vadimsh902948e2017-01-20 15:57:32 -08001010 packages = packages or []
nodir90bc8dc2016-06-15 13:35:21 -07001011
nodirbe642ff2016-06-09 15:51:51 -07001012 get_client_start = time.time()
1013 client_manager = cipd.get_client(
1014 service_url, client_package_name, client_version, cache_dir,
1015 timeout=timeoutfn())
iannucci96fcccc2016-08-30 15:52:22 -07001016
nodirbe642ff2016-06-09 15:51:51 -07001017 with client_manager as client:
1018 get_client_duration = time.time() - get_client_start
nodir90bc8dc2016-06-15 13:35:21 -07001019
iannuccib58d10d2017-03-18 02:00:25 -07001020 package_pins = []
1021 if packages:
1022 package_pins = _install_packages(
1023 run_dir, cipd_cache_dir, client, packages, timeoutfn())
1024
1025 file_path.make_tree_files_read_only(run_dir)
nodir90bc8dc2016-06-15 13:35:21 -07001026
vadimsh232f5a82017-01-20 19:23:44 -08001027 total_duration = time.time() - start
1028 logging.info(
1029 'Installing CIPD client and packages took %d seconds', total_duration)
nodir90bc8dc2016-06-15 13:35:21 -07001030
vadimsh232f5a82017-01-20 19:23:44 -08001031 yield CipdInfo(
1032 client=client,
1033 cache_dir=cipd_cache_dir,
1034 stats={
1035 'duration': total_duration,
1036 'get_client_duration': get_client_duration,
1037 },
1038 pins={
iannuccib58d10d2017-03-18 02:00:25 -07001039 'client_package': {
1040 'package_name': client.package_name,
1041 'version': client.instance_id,
1042 },
vadimsh232f5a82017-01-20 19:23:44 -08001043 'packages': package_pins,
1044 })
nodirbe642ff2016-06-09 15:51:51 -07001045
1046
1047def create_option_parser():
Marc-Antoine Ruelf74cffe2015-07-15 15:21:34 -04001048 parser = logging_utils.OptionParserWithLogging(
nodir55be77b2016-05-03 09:39:57 -07001049 usage='%prog <options> [command to run or extra args]',
maruel@chromium.orgdedbf492013-09-12 20:42:11 +00001050 version=__version__,
1051 log_file=RUN_ISOLATED_LOG_FILE)
maruela9cfd6f2015-09-15 11:03:15 -07001052 parser.add_option(
maruel36a963d2016-04-08 17:15:49 -07001053 '--clean', action='store_true',
1054 help='Cleans the cache, trimming it necessary and remove corrupted items '
1055 'and returns without executing anything; use with -v to know what '
1056 'was done')
1057 parser.add_option(
maruel4409e302016-07-19 14:25:51 -07001058 '--use-symlinks', action='store_true',
1059 help='Use symlinks instead of hardlinks')
1060 parser.add_option(
maruela9cfd6f2015-09-15 11:03:15 -07001061 '--json',
1062 help='dump output metadata to json file. When used, run_isolated returns '
1063 'non-zero only on internal failure')
maruel6be7f9e2015-10-01 12:25:30 -07001064 parser.add_option(
maruel5c9e47b2015-12-18 13:02:30 -08001065 '--hard-timeout', type='float', help='Enforce hard timeout in execution')
maruel6be7f9e2015-10-01 12:25:30 -07001066 parser.add_option(
maruel5c9e47b2015-12-18 13:02:30 -08001067 '--grace-period', type='float',
maruel6be7f9e2015-10-01 12:25:30 -07001068 help='Grace period between SIGTERM and SIGKILL')
bpastene3ae09522016-06-10 17:12:59 -07001069 parser.add_option(
Marc-Antoine Ruel49e347d2017-10-24 16:52:02 -07001070 '--raw-cmd', action='store_true',
1071 help='Ignore the isolated command, use the one supplied at the command '
1072 'line')
1073 parser.add_option(
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001074 '--relative-cwd',
1075 help='Ignore the isolated \'relative_cwd\' and use this one instead; '
1076 'requires --raw-cmd')
1077 parser.add_option(
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001078 '--env', default=[], action='append',
1079 help='Environment variables to set for the child process')
1080 parser.add_option(
1081 '--env-prefix', default=[], action='append',
Robert Iannuccibf5f84c2017-11-22 12:56:50 -08001082 help='Specify a VAR=./path/fragment to put in the environment variable '
1083 'before executing the command. The path fragment must be relative '
1084 'to the isolated run directory, and must not contain a `..` token. '
1085 'The path will be made absolute and prepended to the indicated '
1086 '$VAR using the OS\'s path separator. Multiple items for the same '
1087 '$VAR will be prepended in order.')
1088 parser.add_option(
bpastene3ae09522016-06-10 17:12:59 -07001089 '--bot-file',
1090 help='Path to a file describing the state of the host. The content is '
1091 'defined by on_before_task() in bot_config.')
aludwin7556e0c2016-10-26 08:46:10 -07001092 parser.add_option(
vadimsh9c54b2c2017-07-25 14:08:29 -07001093 '--switch-to-account',
1094 help='If given, switches LUCI_CONTEXT to given logical service account '
1095 '(e.g. "task" or "system") before launching the isolated process.')
1096 parser.add_option(
aludwin0a8e17d2016-10-27 15:57:39 -07001097 '--output', action='append',
1098 help='Specifies an output to return. If no outputs are specified, all '
1099 'files located in $(ISOLATED_OUTDIR) will be returned; '
1100 'otherwise, outputs in both $(ISOLATED_OUTDIR) and those '
1101 'specified by --output option (there can be multiple) will be '
1102 'returned. Note that if a file in OUT_DIR has the same path '
1103 'as an --output option, the --output version will be returned.')
1104 parser.add_option(
aludwin7556e0c2016-10-26 08:46:10 -07001105 '-a', '--argsfile',
1106 # This is actually handled in parse_args; it's included here purely so it
1107 # can make it into the help text.
1108 help='Specify a file containing a JSON array of arguments to this '
1109 'script. If --argsfile is provided, no other argument may be '
1110 'provided on the command line.')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001111
1112 group = optparse.OptionGroup(parser, 'Data source')
1113 group.add_option(
Marc-Antoine Ruel185ded42015-01-28 20:49:18 -05001114 '-s', '--isolated',
nodir55be77b2016-05-03 09:39:57 -07001115 help='Hash of the .isolated to grab from the isolate server.')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001116 isolateserver.add_isolate_server_options(group)
1117 parser.add_option_group(group)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001118
Marc-Antoine Ruela57d7db2014-10-15 20:31:19 -04001119 isolateserver.add_cache_options(parser)
nodirbe642ff2016-06-09 15:51:51 -07001120
1121 cipd.add_cipd_options(parser)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001122
1123 group = optparse.OptionGroup(parser, 'Named caches')
1124 group.add_option(
1125 '--named-cache',
1126 dest='named_caches',
1127 action='append',
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001128 nargs=3,
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001129 default=[],
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001130 help='A named cache to request. Accepts 3 arguments: name, path, hint. '
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001131 'name identifies the cache, must match regex [a-z0-9_]{1,4096}. '
1132 'path is a path relative to the run dir where the cache directory '
1133 'must be put to. '
1134 'This option can be specified more than once.')
1135 group.add_option(
1136 '--named-cache-root', default='named_caches',
1137 help='Cache root directory. Default=%default')
1138 parser.add_option_group(group)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001139
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001140 group = optparse.OptionGroup(parser, 'Process containment')
1141 parser.add_option(
1142 '--lower-priority', action='store_true',
1143 help='Lowers the child process priority')
1144 parser.add_option(
1145 '--containment-type', choices=('NONE', 'AUTO', 'JOB_OBJECT'),
1146 default='NONE',
1147 help='Type of container to use')
1148 parser.add_option(
1149 '--limit-processes', type='int', default=0,
1150 help='Maximum number of active processes in the containment')
1151 parser.add_option(
1152 '--limit-total-committed-memory', type='int', default=0,
1153 help='Maximum sum of committed memory in the containment')
1154 parser.add_option_group(group)
1155
1156 group = optparse.OptionGroup(parser, 'Debugging')
1157 group.add_option(
Kenneth Russell61d42352014-09-15 11:41:16 -07001158 '--leak-temp-dir',
1159 action='store_true',
nodirbe642ff2016-06-09 15:51:51 -07001160 help='Deliberately leak isolate\'s temp dir for later examination. '
1161 'Default: %default')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001162 group.add_option(
marueleb5fbee2015-09-17 13:01:36 -07001163 '--root-dir', help='Use a directory instead of a random one')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001164 parser.add_option_group(group)
Kenneth Russell61d42352014-09-15 11:41:16 -07001165
Vadim Shtayurae34e13a2014-02-02 11:23:26 -08001166 auth.add_auth_options(parser)
nodirbe642ff2016-06-09 15:51:51 -07001167
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001168 parser.set_defaults(cache='cache', cipd_cache='cipd_cache')
nodirbe642ff2016-06-09 15:51:51 -07001169 return parser
1170
1171
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001172def process_named_cache_options(parser, options, time_fn=None):
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001173 """Validates named cache options and returns a CacheManager."""
1174 if options.named_caches and not options.named_cache_root:
1175 parser.error('--named-cache is specified, but --named-cache-root is empty')
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001176 for name, path, hint in options.named_caches:
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001177 if not CACHE_NAME_RE.match(name):
1178 parser.error(
1179 'cache name %r does not match %r' % (name, CACHE_NAME_RE.pattern))
1180 if not path:
1181 parser.error('cache path cannot be empty')
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001182 try:
1183 long(hint)
1184 except ValueError:
1185 parser.error('cache hint must be a number')
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001186 if options.named_cache_root:
1187 # Make these configurable later if there is use case but for now it's fairly
1188 # safe values.
1189 # In practice, a fair chunk of bots are already recycled on a daily schedule
1190 # so this code doesn't have any effect to them, unless they are preloaded
1191 # with a really old cache.
1192 policies = local_caching.CachePolicies(
1193 # 1TiB.
1194 max_cache_size=1024*1024*1024*1024,
1195 min_free_space=options.min_free_space,
1196 max_items=50,
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +00001197 max_age_secs=MAX_AGE_SECS)
Takuto Ikuta6e2ff962019-10-29 12:35:27 +00001198 root_dir = six.text_type(os.path.abspath(options.named_cache_root))
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001199 return local_caching.NamedCache(root_dir, policies, time_fn=time_fn)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001200 return None
1201
1202
aludwin7556e0c2016-10-26 08:46:10 -07001203def parse_args(args):
1204 # Create a fake mini-parser just to get out the "-a" command. Note that
1205 # it's not documented here; instead, it's documented in create_option_parser
1206 # even though that parser will never actually get to parse it. This is
1207 # because --argsfile is exclusive with all other options and arguments.
1208 file_argparse = argparse.ArgumentParser(add_help=False)
1209 file_argparse.add_argument('-a', '--argsfile')
1210 (file_args, nonfile_args) = file_argparse.parse_known_args(args)
1211 if file_args.argsfile:
1212 if nonfile_args:
1213 file_argparse.error('Can\'t specify --argsfile with'
1214 'any other arguments (%s)' % nonfile_args)
1215 try:
1216 with open(file_args.argsfile, 'r') as f:
1217 args = json.load(f)
1218 except (IOError, OSError, ValueError) as e:
1219 # We don't need to error out here - "args" is now empty,
1220 # so the call below to parser.parse_args(args) will fail
1221 # and print the full help text.
Marc-Antoine Ruelf899c482019-10-10 23:32:06 +00001222 print('Couldn\'t read arguments: %s' % e, file=sys.stderr)
aludwin7556e0c2016-10-26 08:46:10 -07001223
1224 # Even if we failed to read the args, just call the normal parser now since it
1225 # will print the correct help message.
nodirbe642ff2016-06-09 15:51:51 -07001226 parser = create_option_parser()
Marc-Antoine Ruel90c98162013-12-18 15:11:57 -05001227 options, args = parser.parse_args(args)
aludwin7556e0c2016-10-26 08:46:10 -07001228 return (parser, options, args)
1229
1230
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001231def _calc_named_cache_hint(named_cache, named_caches):
1232 """Returns the expected size of the missing named caches."""
1233 present = named_cache.available
1234 size = 0
1235 for name, _, hint in named_caches:
1236 if name not in present:
1237 hint = long(hint)
1238 if hint > 0:
1239 size += hint
1240 return size
1241
1242
aludwin7556e0c2016-10-26 08:46:10 -07001243def main(args):
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -05001244 # Warning: when --argsfile is used, the strings are unicode instances, when
1245 # parsed normally, the strings are str instances.
aludwin7556e0c2016-10-26 08:46:10 -07001246 (parser, options, args) = parse_args(args)
maruel36a963d2016-04-08 17:15:49 -07001247
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001248 if not file_path.enable_symlink():
Marc-Antoine Ruel5a024272019-01-15 20:11:16 +00001249 logging.warning('Symlink support is not enabled')
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001250
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001251 named_cache = process_named_cache_options(parser, options)
Marc-Antoine Ruel0d8b0f62018-09-10 14:40:35 +00001252 # hint is 0 if there's no named cache.
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001253 hint = _calc_named_cache_hint(named_cache, options.named_caches)
1254 if hint:
1255 # Increase the --min-free-space value by the hint, and recreate the
1256 # NamedCache instance so it gets the updated CachePolicy.
1257 options.min_free_space += hint
1258 named_cache = process_named_cache_options(parser, options)
1259
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001260 # TODO(maruel): CIPD caches should be defined at an higher level here too, so
1261 # they can be cleaned the same way.
nodirf33b8d62016-10-26 22:34:58 -07001262 isolate_cache = isolateserver.process_cache_options(options, trim=False)
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001263 caches = []
1264 if isolate_cache:
1265 caches.append(isolate_cache)
1266 if named_cache:
1267 caches.append(named_cache)
Takuto Ikuta6e2ff962019-10-29 12:35:27 +00001268 root = caches[0].cache_dir if caches else six.text_type(os.getcwd())
maruel36a963d2016-04-08 17:15:49 -07001269 if options.clean:
1270 if options.isolated:
1271 parser.error('Can\'t use --isolated with --clean.')
1272 if options.isolate_server:
1273 parser.error('Can\'t use --isolate-server with --clean.')
1274 if options.json:
1275 parser.error('Can\'t use --json with --clean.')
nodirf33b8d62016-10-26 22:34:58 -07001276 if options.named_caches:
1277 parser.error('Can\t use --named-cache with --clean.')
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001278 # Trim first, then clean.
1279 local_caching.trim_caches(
1280 caches,
1281 root,
1282 min_free_space=options.min_free_space,
1283 max_age_secs=MAX_AGE_SECS)
1284 for c in caches:
Marc-Antoine Ruel87fc2222018-06-18 13:09:24 +00001285 c.cleanup()
maruel36a963d2016-04-08 17:15:49 -07001286 return 0
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001287
1288 # Trim must still be done for the following case:
1289 # - named-cache was used
1290 # - some entries, with a large hint, where missing
1291 # - --min-free-space was increased accordingly, thus trimming is needed
1292 # Otherwise, this will have no effect, as bot_main calls run_isolated with
1293 # --clean after each task.
1294 if hint:
1295 logging.info('Additional trimming of %d bytes', hint)
Marc-Antoine Ruel0d8b0f62018-09-10 14:40:35 +00001296 local_caching.trim_caches(
1297 caches,
1298 root,
1299 min_free_space=options.min_free_space,
1300 max_age_secs=MAX_AGE_SECS)
maruel36a963d2016-04-08 17:15:49 -07001301
nodir55be77b2016-05-03 09:39:57 -07001302 if not options.isolated and not args:
1303 parser.error('--isolated or command to run is required.')
1304
Vadim Shtayura5d1efce2014-02-04 10:55:43 -08001305 auth.process_auth_options(parser, options)
nodir55be77b2016-05-03 09:39:57 -07001306
1307 isolateserver.process_isolate_server_options(
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001308 parser, options, True, False)
nodir55be77b2016-05-03 09:39:57 -07001309 if not options.isolate_server:
1310 if options.isolated:
1311 parser.error('--isolated requires --isolate-server')
1312 if ISOLATED_OUTDIR_PARAMETER in args:
1313 parser.error(
1314 '%s in args requires --isolate-server' % ISOLATED_OUTDIR_PARAMETER)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001315
nodir90bc8dc2016-06-15 13:35:21 -07001316 if options.root_dir:
Takuto Ikuta6e2ff962019-10-29 12:35:27 +00001317 options.root_dir = six.text_type(os.path.abspath(options.root_dir))
maruel12e30012015-10-09 11:55:35 -07001318 if options.json:
Takuto Ikuta6e2ff962019-10-29 12:35:27 +00001319 options.json = six.text_type(os.path.abspath(options.json))
nodir55be77b2016-05-03 09:39:57 -07001320
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001321 if any('=' not in i for i in options.env):
1322 parser.error(
1323 '--env required key=value form. value can be skipped to delete '
1324 'the variable')
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -05001325 options.env = dict(i.split('=', 1) for i in options.env)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001326
1327 prefixes = {}
1328 cwd = os.path.realpath(os.getcwd())
1329 for item in options.env_prefix:
1330 if '=' not in item:
1331 parser.error(
1332 '--env-prefix %r is malformed, must be in the form `VAR=./path`'
1333 % item)
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -05001334 key, opath = item.split('=', 1)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001335 if os.path.isabs(opath):
1336 parser.error('--env-prefix %r path is bad, must be relative.' % opath)
1337 opath = os.path.normpath(opath)
1338 if not os.path.realpath(os.path.join(cwd, opath)).startswith(cwd):
1339 parser.error(
1340 '--env-prefix %r path is bad, must be relative and not contain `..`.'
1341 % opath)
1342 prefixes.setdefault(key, []).append(opath)
1343 options.env_prefix = prefixes
Robert Iannuccibf5f84c2017-11-22 12:56:50 -08001344
nodirbe642ff2016-06-09 15:51:51 -07001345 cipd.validate_cipd_options(parser, options)
1346
vadimsh232f5a82017-01-20 19:23:44 -08001347 install_packages_fn = noop_install_packages
vadimsh902948e2017-01-20 15:57:32 -08001348 if options.cipd_enabled:
iannuccib58d10d2017-03-18 02:00:25 -07001349 install_packages_fn = lambda run_dir: install_client_and_packages(
vadimsh902948e2017-01-20 15:57:32 -08001350 run_dir, cipd.parse_package_args(options.cipd_packages),
1351 options.cipd_server, options.cipd_client_package,
1352 options.cipd_client_version, cache_dir=options.cipd_cache)
nodirbe642ff2016-06-09 15:51:51 -07001353
nodird6160682017-02-02 13:03:35 -08001354 @contextlib.contextmanager
nodir0ae98b32017-05-11 13:21:53 -07001355 def install_named_caches(run_dir):
nodird6160682017-02-02 13:03:35 -08001356 # WARNING: this function depends on "options" variable defined in the outer
1357 # function.
Takuto Ikuta6e2ff962019-10-29 12:35:27 +00001358 assert six.text_type(run_dir), repr(run_dir)
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001359 assert os.path.isabs(run_dir), run_dir
Takuto Ikuta6e2ff962019-10-29 12:35:27 +00001360 named_caches = [(os.path.join(run_dir, six.text_type(relpath)), name)
1361 for name, relpath, _ in options.named_caches]
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001362 for path, name in named_caches:
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001363 named_cache.install(path, name)
nodird6160682017-02-02 13:03:35 -08001364 try:
1365 yield
1366 finally:
dnje289d132017-07-07 11:16:44 -07001367 # Uninstall each named cache, returning it to the cache pool. If an
1368 # uninstall fails for a given cache, it will remain in the task's
1369 # temporary space, get cleaned up by the Swarming bot, and be lost.
1370 #
1371 # If the Swarming bot cannot clean up the cache, it will handle it like
1372 # any other bot file that could not be removed.
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001373 for path, name in reversed(named_caches):
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001374 try:
Marc-Antoine Ruele9558372018-08-03 03:41:22 +00001375 # uninstall() doesn't trim but does call save() implicitly. Trimming
1376 # *must* be done manually via periodic 'run_isolated.py --clean'.
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001377 named_cache.uninstall(path, name)
1378 except local_caching.NamedCacheError:
1379 logging.exception('Error while removing named cache %r at %r. '
1380 'The cache will be lost.', path, name)
nodirf33b8d62016-10-26 22:34:58 -07001381
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001382 extra_args = []
1383 command = []
1384 if options.raw_cmd:
1385 command = args
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001386 if options.relative_cwd:
1387 a = os.path.normpath(os.path.abspath(options.relative_cwd))
1388 if not a.startswith(os.getcwd()):
1389 parser.error(
1390 '--relative-cwd must not try to escape the working directory')
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001391 else:
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001392 if options.relative_cwd:
1393 parser.error('--relative-cwd requires --raw-cmd')
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001394 extra_args = args
1395
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001396 containment_type = subprocess42.Containment.NONE
1397 if options.containment_type == 'AUTO':
1398 containment_type = subprocess42.Containment.AUTO
1399 if options.containment_type == 'JOB_OBJECT':
1400 containment_type = subprocess42.Containment.JOB_OBJECT
1401 containment = subprocess42.Containment(
1402 containment_type=containment_type,
1403 limit_processes=options.limit_processes,
1404 limit_total_committed_memory=options.limit_total_committed_memory)
1405
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001406 data = TaskData(
1407 command=command,
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001408 relative_cwd=options.relative_cwd,
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001409 extra_args=extra_args,
1410 isolated_hash=options.isolated,
1411 storage=None,
1412 isolate_cache=isolate_cache,
1413 outputs=options.output,
1414 install_named_caches=install_named_caches,
1415 leak_temp_dir=options.leak_temp_dir,
1416 root_dir=_to_unicode(options.root_dir),
1417 hard_timeout=options.hard_timeout,
1418 grace_period=options.grace_period,
1419 bot_file=options.bot_file,
1420 switch_to_account=options.switch_to_account,
1421 install_packages_fn=install_packages_fn,
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +00001422 use_symlinks=bool(options.use_symlinks),
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001423 env=options.env,
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +00001424 env_prefix=options.env_prefix,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001425 lower_priority=bool(options.lower_priority),
1426 containment=containment)
nodirbe642ff2016-06-09 15:51:51 -07001427 try:
nodir90bc8dc2016-06-15 13:35:21 -07001428 if options.isolate_server:
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001429 server_ref = isolate_storage.ServerRef(
nodir90bc8dc2016-06-15 13:35:21 -07001430 options.isolate_server, options.namespace)
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001431 storage = isolateserver.get_storage(server_ref)
nodir90bc8dc2016-06-15 13:35:21 -07001432 with storage:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001433 data = data._replace(storage=storage)
nodirf33b8d62016-10-26 22:34:58 -07001434 # Hashing schemes used by |storage| and |isolate_cache| MUST match.
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001435 assert storage.server_ref.hash_algo == server_ref.hash_algo
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001436 return run_tha_test(data, options.json)
1437 return run_tha_test(data, options.json)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001438 except (
1439 cipd.Error,
1440 local_caching.NamedCacheError,
Marc-Antoine Ruelb6e9e232018-11-20 00:12:33 +00001441 local_caching.NoMoreSpace) as ex:
Marc-Antoine Ruelf899c482019-10-10 23:32:06 +00001442 print(ex.message, file=sys.stderr)
nodirbe642ff2016-06-09 15:51:51 -07001443 return 1
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001444
1445
1446if __name__ == '__main__':
maruel8e4e40c2016-05-30 06:21:07 -07001447 subprocess42.inhibit_os_error_reporting()
csharp@chromium.orgbfb98742013-03-26 20:28:36 +00001448 # Ensure that we are always running with the correct encoding.
vadimsh@chromium.orga4326472013-08-24 02:05:41 +00001449 fix_encoding.fix_encoding()
Marc-Antoine Ruel90c98162013-12-18 15:11:57 -05001450 sys.exit(main(sys.argv[1:]))