blob: e2cd52c550d33f5c756c8aa9a9d2edf6541f9192 [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 Rueleed2f3a2019-03-14 00:00:40 +000042__version__ = '1.0.0'
maruel@chromium.orgdedbf492013-09-12 20:42:11 +000043
aludwin7556e0c2016-10-26 08:46:10 -070044import argparse
maruel064c0a32016-04-05 11:47:15 -070045import base64
iannucci96fcccc2016-08-30 15:52:22 -070046import collections
vadimsh232f5a82017-01-20 19:23:44 -080047import contextlib
Sadaf Matinkhoo10743a62018-03-29 16:28:58 -040048import errno
aludwin7556e0c2016-10-26 08:46:10 -070049import json
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000050import logging
51import optparse
52import os
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -040053import re
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000054import sys
55import tempfile
maruel064c0a32016-04-05 11:47:15 -070056import time
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +000057
Marc-Antoine Ruel016c7602019-04-02 18:31:13 +000058from utils import tools
59tools.force_local_third_party()
vadimsh@chromium.orga4326472013-08-24 02:05:41 +000060
Marc-Antoine Ruel016c7602019-04-02 18:31:13 +000061# third_party/
62from depot_tools import fix_encoding
63
64# pylint: disable=ungrouped-imports
65import auth
66import cipd
67import isolate_storage
68import isolateserver
69import local_caching
70from libs import luci_context
Vadim Shtayura6b555c12014-07-23 16:22:18 -070071from utils import file_path
maruel12e30012015-10-09 11:55:35 -070072from utils import fs
maruel064c0a32016-04-05 11:47:15 -070073from utils import large
Marc-Antoine Ruelf74cffe2015-07-15 15:21:34 -040074from utils import logging_utils
Marc-Antoine Ruelcfb60852014-07-02 15:22:00 -040075from utils import on_error
Marc-Antoine Ruelc44f5722015-01-08 16:10:01 -050076from utils import subprocess42
maruel@chromium.orgdedbf492013-09-12 20:42:11 +000077
vadimsh@chromium.orga4326472013-08-24 02:05:41 +000078
maruele2f2cb82016-07-13 14:41:03 -070079# Magic variables that can be found in the isolate task command line.
80ISOLATED_OUTDIR_PARAMETER = '${ISOLATED_OUTDIR}'
81EXECUTABLE_SUFFIX_PARAMETER = '${EXECUTABLE_SUFFIX}'
82SWARMING_BOT_FILE_PARAMETER = '${SWARMING_BOT_FILE}'
83
84
csharp@chromium.orgff2a4662012-11-21 20:49:32 +000085# The name of the log file to use.
86RUN_ISOLATED_LOG_FILE = 'run_isolated.log'
87
maruele2f2cb82016-07-13 14:41:03 -070088
csharp@chromium.orge217f302012-11-22 16:51:53 +000089# The name of the log to use for the run_test_cases.py command
vadimsh@chromium.org8b9d56b2013-08-21 22:24:35 +000090RUN_TEST_CASES_LOG = 'run_test_cases.log'
csharp@chromium.orge217f302012-11-22 16:51:53 +000091
vadimsh@chromium.org87d63262013-04-04 19:34:21 +000092
maruele2f2cb82016-07-13 14:41:03 -070093# Use short names for temporary directories. This is driven by Windows, which
94# imposes a relatively short maximum path length of 260 characters, often
95# referred to as MAX_PATH. It is relatively easy to create files with longer
Marc-Antoine Ruel793bff32019-04-18 17:50:48 +000096# path length. A use case is with recursive dependency trees like npm packages.
maruele2f2cb82016-07-13 14:41:03 -070097#
98# It is recommended to start the script with a `root_dir` as short as
99# possible.
100# - ir stands for isolated_run
101# - io stands for isolated_out
102# - it stands for isolated_tmp
103ISOLATED_RUN_DIR = u'ir'
104ISOLATED_OUT_DIR = u'io'
105ISOLATED_TMP_DIR = u'it'
106
107
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -0400108# Keep synced with task_request.py
Lei Leife202df2019-06-11 17:33:34 +0000109CACHE_NAME_RE = re.compile(r'^[a-z0-9_]{1,4096}$')
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -0400110
111
marueld928c862017-06-08 08:20:04 -0700112OUTLIVING_ZOMBIE_MSG = """\
113*** Swarming tried multiple times to delete the %s directory and failed ***
114*** Hard failing the task ***
115
116Swarming detected that your testing script ran an executable, which may have
117started a child executable, and the main script returned early, leaving the
118children executables playing around unguided.
119
120You don't want to leave children processes outliving the task on the Swarming
121bot, do you? The Swarming bot doesn't.
122
123How to fix?
124- For any process that starts children processes, make sure all children
125 processes terminated properly before each parent process exits. This is
126 especially important in very deep process trees.
127 - This must be done properly both in normal successful task and in case of
128 task failure. Cleanup is very important.
129- The Swarming bot sends a SIGTERM in case of timeout.
130 - You have %s seconds to comply after the signal was sent to the process
131 before the process is forcibly killed.
132- To achieve not leaking children processes in case of signals on timeout, you
133 MUST handle signals in each executable / python script and propagate them to
134 children processes.
135 - When your test script (python or binary) receives a signal like SIGTERM or
136 CTRL_BREAK_EVENT on Windows), send it to all children processes and wait for
137 them to terminate before quitting.
138
139See
Marc-Antoine Ruelc7243592018-05-24 17:04:04 -0400140https://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 -0700141for more information.
142
143*** May the SIGKILL force be with you ***
144"""
145
146
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000147# Currently hardcoded. Eventually could be exposed as a flag once there's value.
148# 3 weeks
149MAX_AGE_SECS = 21*24*60*60
150
151
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500152TaskData = collections.namedtuple(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000153 'TaskData',
154 [
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500155 # List of strings; the command line to use, independent of what was
156 # specified in the isolated file.
157 'command',
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500158 # Relative directory to start command into.
159 'relative_cwd',
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500160 # List of strings; the arguments to add to the command specified in the
161 # isolated file.
162 'extra_args',
163 # Hash of the .isolated file that must be retrieved to recreate the tree
164 # of files to run the target executable. The command specified in the
165 # .isolated is executed. Mutually exclusive with command argument.
166 'isolated_hash',
167 # isolateserver.Storage instance to retrieve remote objects. This object
168 # has a reference to an isolateserver.StorageApi, which does the actual
169 # I/O.
170 'storage',
171 # isolateserver.LocalCache instance to keep from retrieving the same
172 # objects constantly by caching the objects retrieved. Can be on-disk or
173 # in-memory.
174 'isolate_cache',
175 # List of paths relative to root_dir to put into the output isolated
176 # bundle upon task completion (see link_outputs_to_outdir).
177 'outputs',
178 # Function (run_dir) => context manager that installs named caches into
179 # |run_dir|.
180 'install_named_caches',
181 # If True, the temporary directory will be deliberately leaked for later
182 # examination.
183 'leak_temp_dir',
184 # Path to the directory to use to create the temporary directory. If not
185 # specified, a random temporary directory is created.
186 'root_dir',
187 # Kills the process if it lasts more than this amount of seconds.
188 'hard_timeout',
189 # Number of seconds to wait between SIGTERM and SIGKILL.
190 'grace_period',
191 # Path to a file with bot state, used in place of ${SWARMING_BOT_FILE}
192 # task command line argument.
193 'bot_file',
194 # Logical account to switch LUCI_CONTEXT into.
195 'switch_to_account',
196 # Context manager dir => CipdInfo, see install_client_and_packages.
197 'install_packages_fn',
198 # Create tree with symlinks instead of hardlinks.
199 'use_symlinks',
200 # Environment variables to set.
201 'env',
202 # Environment variables to mutate with relative directories.
203 # Example: {"ENV_KEY": ['relative', 'paths', 'to', 'prepend']}
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000204 'env_prefix',
205 # Lowers the task process priority.
206 'lower_priority',
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000207 # subprocess42.Containment instance. Can be None.
208 'containment',
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000209 ])
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500210
211
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -0500212def _to_str(s):
213 """Downgrades a unicode instance to str. Pass str through as-is."""
214 if isinstance(s, str):
215 return s
216 # This is technically incorrect, especially on Windows. In theory
217 # sys.getfilesystemencoding() should be used to use the right 'ANSI code
218 # page' on Windows, but that causes other problems, as the character set
219 # is very limited.
220 return s.encode('utf-8')
221
222
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -0500223def _to_unicode(s):
224 """Upgrades a str instance to unicode. Pass unicode through as-is."""
225 if isinstance(s, unicode) or s is None:
226 return s
227 return s.decode('utf-8')
228
229
maruel03e11842016-07-14 10:50:16 -0700230def make_temp_dir(prefix, root_dir):
231 """Returns a new unique temporary directory."""
232 return unicode(tempfile.mkdtemp(prefix=prefix, dir=root_dir))
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000233
234
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500235def change_tree_read_only(rootdir, read_only):
236 """Changes the tree read-only bits according to the read_only specification.
237
238 The flag can be 0, 1 or 2, which will affect the possibility to modify files
239 and create or delete files.
240 """
241 if read_only == 2:
242 # Files and directories (except on Windows) are marked read only. This
243 # inhibits modifying, creating or deleting files in the test directory,
244 # except on Windows where creating and deleting files is still possible.
Marc-Antoine Ruele4ad07e2014-10-15 20:22:29 -0400245 file_path.make_tree_read_only(rootdir)
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500246 elif read_only == 1:
247 # Files are marked read only but not the directories. This inhibits
248 # modifying files but creating or deleting files is still possible.
Marc-Antoine Ruele4ad07e2014-10-15 20:22:29 -0400249 file_path.make_tree_files_read_only(rootdir)
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500250 elif read_only in (0, None):
Marc-Antoine Ruelf1d827c2014-11-24 15:22:25 -0500251 # Anything can be modified.
Marc-Antoine Ruel2666d9c2018-05-18 13:52:02 -0400252 # TODO(maruel): This is currently dangerous as long as
253 # DiskContentAddressedCache.touch() is not yet changed to verify the hash of
254 # the content of the files it is looking at, so that if a test modifies an
255 # input file, the file must be deleted.
Marc-Antoine Ruele4ad07e2014-10-15 20:22:29 -0400256 file_path.make_tree_writeable(rootdir)
Marc-Antoine Ruel7124e392014-01-09 11:49:21 -0500257 else:
258 raise ValueError(
259 'change_tree_read_only(%s, %s): Unknown flag %s' %
260 (rootdir, read_only, read_only))
261
262
vadimsh9c54b2c2017-07-25 14:08:29 -0700263@contextlib.contextmanager
264def set_luci_context_account(account, tmp_dir):
265 """Sets LUCI_CONTEXT account to be used by the task.
266
267 If 'account' is None or '', does nothing at all. This happens when
268 run_isolated.py is called without '--switch-to-account' flag. In this case,
269 if run_isolated.py is running in some LUCI_CONTEXT environment, the task will
270 just inherit whatever account is already set. This may happen is users invoke
271 run_isolated.py explicitly from their code.
272
273 If the requested account is not defined in the context, switches to
274 non-authenticated access. This happens for Swarming tasks that don't use
275 'task' service accounts.
276
277 If not using LUCI_CONTEXT-based auth, does nothing.
278 If already running as requested account, does nothing.
279 """
280 if not account:
281 # Not actually switching.
282 yield
283 return
284
285 local_auth = luci_context.read('local_auth')
286 if not local_auth:
287 # Not using LUCI_CONTEXT auth at all.
288 yield
289 return
290
291 # See LUCI_CONTEXT.md for the format of 'local_auth'.
292 if local_auth.get('default_account_id') == account:
293 # Already set, no need to switch.
294 yield
295 return
296
297 available = {a['id'] for a in local_auth.get('accounts') or []}
298 if account in available:
299 logging.info('Switching default LUCI_CONTEXT account to %r', account)
300 local_auth['default_account_id'] = account
301 else:
302 logging.warning(
303 'Requested LUCI_CONTEXT account %r is not available (have only %r), '
304 'disabling authentication', account, sorted(available))
305 local_auth.pop('default_account_id', None)
306
307 with luci_context.write(_tmpdir=tmp_dir, local_auth=local_auth):
308 yield
309
310
nodir90bc8dc2016-06-15 13:35:21 -0700311def process_command(command, out_dir, bot_file):
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000312 """Replaces parameters in a command line.
nodirbe642ff2016-06-09 15:51:51 -0700313
314 Raises:
315 ValueError if a parameter is requested in |command| but its value is not
316 provided.
317 """
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000318 return [replace_parameters(arg, out_dir, bot_file) for arg in command]
319
320
321def replace_parameters(arg, out_dir, bot_file):
322 """Replaces parameter tokens with appropriate values in a string.
323
324 Raises:
325 ValueError if a parameter is requested in |arg| but its value is not
326 provided.
327 """
328 arg = arg.replace(EXECUTABLE_SUFFIX_PARAMETER, cipd.EXECUTABLE_SUFFIX)
329 replace_slash = False
330 if ISOLATED_OUTDIR_PARAMETER in arg:
331 if not out_dir:
332 raise ValueError(
333 'output directory is requested in command or env var, but not '
334 'provided; please specify one')
335 arg = arg.replace(ISOLATED_OUTDIR_PARAMETER, out_dir)
336 replace_slash = True
337 if SWARMING_BOT_FILE_PARAMETER in arg:
338 if bot_file:
339 arg = arg.replace(SWARMING_BOT_FILE_PARAMETER, bot_file)
nodirbe642ff2016-06-09 15:51:51 -0700340 replace_slash = True
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000341 else:
342 logging.warning('SWARMING_BOT_FILE_PARAMETER found in command or env '
343 'var, but no bot_file specified. Leaving parameter '
344 'unchanged.')
345 if replace_slash:
346 # Replace slashes only if parameters are present
347 # because of arguments like '${ISOLATED_OUTDIR}/foo/bar'
348 arg = arg.replace('/', os.sep)
349 return arg
maruela9cfd6f2015-09-15 11:03:15 -0700350
351
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000352
353def get_command_env(tmp_dir, cipd_info, run_dir, env, env_prefixes, out_dir,
354 bot_file):
vadimsh232f5a82017-01-20 19:23:44 -0800355 """Returns full OS environment to run a command in.
356
Robert Iannuccibf5f84c2017-11-22 12:56:50 -0800357 Sets up TEMP, puts directory with cipd binary in front of PATH, exposes
358 CIPD_CACHE_DIR env var, and installs all env_prefixes.
vadimsh232f5a82017-01-20 19:23:44 -0800359
360 Args:
361 tmp_dir: temp directory.
362 cipd_info: CipdInfo object is cipd client is used, None if not.
Marc-Antoine Ruel9ec1e9f2017-12-20 16:36:54 -0500363 run_dir: The root directory the isolated tree is mapped in.
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500364 env: environment variables to use
Robert Iannuccibf5f84c2017-11-22 12:56:50 -0800365 env_prefixes: {"ENV_KEY": ['cwd', 'relative', 'paths', 'to', 'prepend']}
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000366 out_dir: Isolated output directory. Required to be != None if any of the
367 env vars contain ISOLATED_OUTDIR_PARAMETER.
368 bot_file: Required to be != None if any of the env vars contain
369 SWARMING_BOT_FILE_PARAMETER.
vadimsh232f5a82017-01-20 19:23:44 -0800370 """
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500371 out = os.environ.copy()
372 for k, v in env.iteritems():
373 if not v:
Marc-Antoine Ruel9ec1e9f2017-12-20 16:36:54 -0500374 out.pop(k, None)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500375 else:
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000376 out[k] = replace_parameters(v, out_dir, bot_file)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500377
378 if cipd_info:
379 bin_dir = os.path.dirname(cipd_info.client.binary_path)
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -0500380 out['PATH'] = '%s%s%s' % (_to_str(bin_dir), os.pathsep, out['PATH'])
381 out['CIPD_CACHE_DIR'] = _to_str(cipd_info.cache_dir)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500382
383 for key, paths in env_prefixes.iteritems():
Marc-Antoine Ruel9ec1e9f2017-12-20 16:36:54 -0500384 assert isinstance(paths, list), paths
385 paths = [os.path.normpath(os.path.join(run_dir, p)) for p in paths]
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500386 cur = out.get(key)
387 if cur:
388 paths.append(cur)
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -0500389 out[key] = _to_str(os.path.pathsep.join(paths))
vadimsh232f5a82017-01-20 19:23:44 -0800390
Marc-Antoine Ruelefb30b12018-07-25 18:34:36 +0000391 tmp_dir = _to_str(tmp_dir)
392 # pylint: disable=line-too-long
393 # * python respects $TMPDIR, $TEMP, and $TMP in this order, regardless of
394 # platform. So $TMPDIR must be set on all platforms.
395 # https://github.com/python/cpython/blob/2.7/Lib/tempfile.py#L155
396 out['TMPDIR'] = tmp_dir
397 if sys.platform == 'win32':
398 # * chromium's base utils uses GetTempPath().
399 # https://cs.chromium.org/chromium/src/base/files/file_util_win.cc?q=GetTempPath
400 # * Go uses GetTempPath().
401 # * GetTempDir() uses %TMP%, then %TEMP%, then other stuff. So %TMP% must be
402 # set.
403 # https://docs.microsoft.com/en-us/windows/desktop/api/fileapi/nf-fileapi-gettemppathw
404 out['TMP'] = tmp_dir
405 # https://blogs.msdn.microsoft.com/oldnewthing/20150417-00/?p=44213
406 out['TEMP'] = tmp_dir
407 elif sys.platform == 'darwin':
408 # * Chromium uses an hack on macOS before calling into
409 # NSTemporaryDirectory().
410 # https://cs.chromium.org/chromium/src/base/files/file_util_mac.mm?q=GetTempDir
411 # https://developer.apple.com/documentation/foundation/1409211-nstemporarydirectory
412 out['MAC_CHROMIUM_TMPDIR'] = tmp_dir
413 else:
414 # TMPDIR is specified as the POSIX standard envvar for the temp directory.
415 # http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html
416 # * mktemp on linux respects $TMPDIR.
417 # * Chromium respects $TMPDIR on linux.
418 # https://cs.chromium.org/chromium/src/base/files/file_util_posix.cc?q=GetTempDir
419 # * Go uses $TMPDIR.
420 # https://go.googlesource.com/go/+/go1.10.3/src/os/file_unix.go#307
421 pass
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500422 return out
vadimsh232f5a82017-01-20 19:23:44 -0800423
424
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000425def run_command(
426 command, cwd, env, hard_timeout, grace_period, lower_priority, containment):
maruel6be7f9e2015-10-01 12:25:30 -0700427 """Runs the command.
428
429 Returns:
430 tuple(process exit code, bool if had a hard timeout)
431 """
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000432 logging.info(
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000433 'run_command(%s, %s, %s, %s, %s, %s)',
434 command, cwd, hard_timeout, grace_period, lower_priority, containment)
marueleb5fbee2015-09-17 13:01:36 -0700435
maruel6be7f9e2015-10-01 12:25:30 -0700436 exit_code = None
437 had_hard_timeout = False
maruela9cfd6f2015-09-15 11:03:15 -0700438 with tools.Profiler('RunTest'):
maruel6be7f9e2015-10-01 12:25:30 -0700439 proc = None
440 had_signal = []
maruela9cfd6f2015-09-15 11:03:15 -0700441 try:
maruel6be7f9e2015-10-01 12:25:30 -0700442 # TODO(maruel): This code is imperfect. It doesn't handle well signals
443 # during the download phase and there's short windows were things can go
444 # wrong.
445 def handler(signum, _frame):
446 if proc and not had_signal:
447 logging.info('Received signal %d', signum)
448 had_signal.append(True)
maruel556d9052015-10-05 11:12:44 -0700449 raise subprocess42.TimeoutExpired(command, None)
maruel6be7f9e2015-10-01 12:25:30 -0700450
Marc-Antoine Ruel30b80fe2019-02-08 13:51:31 +0000451 proc = subprocess42.Popen(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000452 command, cwd=cwd, env=env, detached=True, close_fds=True,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000453 lower_priority=lower_priority, containment=containment)
maruel6be7f9e2015-10-01 12:25:30 -0700454 with subprocess42.set_signal_handler(subprocess42.STOP_SIGNALS, handler):
455 try:
John Budorickc398f092019-06-10 22:49:44 +0000456 exit_code = proc.wait(hard_timeout or None)
maruel6be7f9e2015-10-01 12:25:30 -0700457 except subprocess42.TimeoutExpired:
458 if not had_signal:
459 logging.warning('Hard timeout')
460 had_hard_timeout = True
461 logging.warning('Sending SIGTERM')
462 proc.terminate()
463
464 # Ignore signals in grace period. Forcibly give the grace period to the
465 # child process.
466 if exit_code is None:
467 ignore = lambda *_: None
468 with subprocess42.set_signal_handler(subprocess42.STOP_SIGNALS, ignore):
469 try:
470 exit_code = proc.wait(grace_period or None)
471 except subprocess42.TimeoutExpired:
472 # Now kill for real. The user can distinguish between the
473 # following states:
474 # - signal but process exited within grace period,
475 # hard_timed_out will be set but the process exit code will be
476 # script provided.
477 # - processed exited late, exit code will be -9 on posix.
478 logging.warning('Grace exhausted; sending SIGKILL')
479 proc.kill()
martiniss5c8043e2017-08-01 17:09:43 -0700480 logging.info('Waiting for process exit')
maruel6be7f9e2015-10-01 12:25:30 -0700481 exit_code = proc.wait()
maruela9cfd6f2015-09-15 11:03:15 -0700482 except OSError:
483 # This is not considered to be an internal error. The executable simply
484 # does not exit.
maruela72f46e2016-02-24 11:05:45 -0800485 sys.stderr.write(
486 '<The executable does not exist or a dependent library is missing>\n'
487 '<Check for missing .so/.dll in the .isolate or GN file>\n'
488 '<Command: %s>\n' % command)
489 if os.environ.get('SWARMING_TASK_ID'):
490 # Give an additional hint when running as a swarming task.
491 sys.stderr.write(
492 '<See the task\'s page for commands to help diagnose this issue '
493 'by reproducing the task locally>\n')
maruela9cfd6f2015-09-15 11:03:15 -0700494 exit_code = 1
495 logging.info(
496 'Command finished with exit code %d (%s)',
497 exit_code, hex(0xffffffff & exit_code))
maruel6be7f9e2015-10-01 12:25:30 -0700498 return exit_code, had_hard_timeout
maruela9cfd6f2015-09-15 11:03:15 -0700499
500
maruel4409e302016-07-19 14:25:51 -0700501def fetch_and_map(isolated_hash, storage, cache, outdir, use_symlinks):
502 """Fetches an isolated tree, create the tree and returns (bundle, stats)."""
nodir6f801882016-04-29 14:41:50 -0700503 start = time.time()
504 bundle = isolateserver.fetch_isolated(
505 isolated_hash=isolated_hash,
506 storage=storage,
507 cache=cache,
maruel4409e302016-07-19 14:25:51 -0700508 outdir=outdir,
509 use_symlinks=use_symlinks)
nodir6f801882016-04-29 14:41:50 -0700510 return bundle, {
511 'duration': time.time() - start,
nodir6f801882016-04-29 14:41:50 -0700512 'items_cold': base64.b64encode(large.pack(sorted(cache.added))),
513 'items_hot': base64.b64encode(
tansell9e04a8d2016-07-28 09:31:59 -0700514 large.pack(sorted(set(cache.used) - set(cache.added)))),
nodir6f801882016-04-29 14:41:50 -0700515 }
516
517
aludwin0a8e17d2016-10-27 15:57:39 -0700518def link_outputs_to_outdir(run_dir, out_dir, outputs):
519 """Links any named outputs to out_dir so they can be uploaded.
520
521 Raises an error if the file already exists in that directory.
522 """
523 if not outputs:
524 return
525 isolateserver.create_directories(out_dir, outputs)
526 for o in outputs:
Sadaf Matinkhoo10743a62018-03-29 16:28:58 -0400527 copy_recursively(os.path.join(run_dir, o), os.path.join(out_dir, o))
528
529
530def copy_recursively(src, dst):
531 """Efficiently copies a file or directory from src_dir to dst_dir.
532
533 `item` may be a file, directory, or a symlink to a file or directory.
534 All symlinks are replaced with their targets, so the resulting
535 directory structure in dst_dir will never have any symlinks.
536
537 To increase speed, copy_recursively hardlinks individual files into the
538 (newly created) directory structure if possible, unlike Python's
539 shutil.copytree().
540 """
541 orig_src = src
542 try:
543 # Replace symlinks with their final target.
544 while fs.islink(src):
545 res = fs.readlink(src)
546 src = os.path.join(os.path.dirname(src), res)
547 # TODO(sadafm): Explicitly handle cyclic symlinks.
548
549 # Note that fs.isfile (which is a wrapper around os.path.isfile) throws
550 # an exception if src does not exist. A warning will be logged in that case.
551 if fs.isfile(src):
552 file_path.link_file(dst, src, file_path.HARDLINK_WITH_FALLBACK)
553 return
554
555 if not fs.exists(dst):
556 os.makedirs(dst)
557
558 for child in fs.listdir(src):
559 copy_recursively(os.path.join(src, child), os.path.join(dst, child))
560
561 except OSError as e:
562 if e.errno == errno.ENOENT:
563 logging.warning('Path %s does not exist or %s is a broken symlink',
564 src, orig_src)
565 else:
566 logging.info("Couldn't collect output file %s: %s", src, e)
aludwin0a8e17d2016-10-27 15:57:39 -0700567
568
maruela9cfd6f2015-09-15 11:03:15 -0700569def delete_and_upload(storage, out_dir, leak_temp_dir):
570 """Deletes the temporary run directory and uploads results back.
571
572 Returns:
nodir6f801882016-04-29 14:41:50 -0700573 tuple(outputs_ref, success, stats)
maruel064c0a32016-04-05 11:47:15 -0700574 - outputs_ref: a dict referring to the results archived back to the isolated
575 server, if applicable.
576 - success: False if something occurred that means that the task must
577 forcibly be considered a failure, e.g. zombie processes were left
578 behind.
nodir6f801882016-04-29 14:41:50 -0700579 - stats: uploading stats.
maruela9cfd6f2015-09-15 11:03:15 -0700580 """
maruela9cfd6f2015-09-15 11:03:15 -0700581 # Upload out_dir and generate a .isolated file out of this directory. It is
582 # only done if files were written in the directory.
583 outputs_ref = None
maruel064c0a32016-04-05 11:47:15 -0700584 cold = []
585 hot = []
nodir6f801882016-04-29 14:41:50 -0700586 start = time.time()
587
maruel12e30012015-10-09 11:55:35 -0700588 if fs.isdir(out_dir) and fs.listdir(out_dir):
maruela9cfd6f2015-09-15 11:03:15 -0700589 with tools.Profiler('ArchiveOutput'):
590 try:
maruel064c0a32016-04-05 11:47:15 -0700591 results, f_cold, f_hot = isolateserver.archive_files_to_storage(
maruela9cfd6f2015-09-15 11:03:15 -0700592 storage, [out_dir], None)
593 outputs_ref = {
Marc-Antoine Rueld0868ec2018-11-28 20:47:29 +0000594 'isolated': results.values()[0],
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +0000595 'isolatedserver': storage.server_ref.url,
596 'namespace': storage.server_ref.namespace,
maruela9cfd6f2015-09-15 11:03:15 -0700597 }
maruel064c0a32016-04-05 11:47:15 -0700598 cold = sorted(i.size for i in f_cold)
599 hot = sorted(i.size for i in f_hot)
maruela9cfd6f2015-09-15 11:03:15 -0700600 except isolateserver.Aborted:
601 # This happens when a signal SIGTERM was received while uploading data.
602 # There is 2 causes:
603 # - The task was too slow and was about to be killed anyway due to
604 # exceeding the hard timeout.
605 # - The amount of data uploaded back is very large and took too much
606 # time to archive.
607 sys.stderr.write('Received SIGTERM while uploading')
608 # Re-raise, so it will be treated as an internal failure.
609 raise
nodir6f801882016-04-29 14:41:50 -0700610
611 success = False
maruela9cfd6f2015-09-15 11:03:15 -0700612 try:
maruel12e30012015-10-09 11:55:35 -0700613 if (not leak_temp_dir and fs.isdir(out_dir) and
maruel6eeea7d2015-09-16 12:17:42 -0700614 not file_path.rmtree(out_dir)):
maruela9cfd6f2015-09-15 11:03:15 -0700615 logging.error('Had difficulties removing out_dir %s', out_dir)
nodir6f801882016-04-29 14:41:50 -0700616 else:
617 success = True
maruela9cfd6f2015-09-15 11:03:15 -0700618 except OSError as e:
619 # When this happens, it means there's a process error.
maruel12e30012015-10-09 11:55:35 -0700620 logging.exception('Had difficulties removing out_dir %s: %s', out_dir, e)
nodir6f801882016-04-29 14:41:50 -0700621 stats = {
622 'duration': time.time() - start,
623 'items_cold': base64.b64encode(large.pack(cold)),
624 'items_hot': base64.b64encode(large.pack(hot)),
625 }
626 return outputs_ref, success, stats
maruela9cfd6f2015-09-15 11:03:15 -0700627
628
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500629def map_and_run(data, constant_run_path):
nodir55be77b2016-05-03 09:39:57 -0700630 """Runs a command with optional isolated input/output.
631
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500632 Arguments:
633 - data: TaskData instance.
634 - constant_run_path: TODO
nodir55be77b2016-05-03 09:39:57 -0700635
636 Returns metadata about the result.
637 """
maruela9cfd6f2015-09-15 11:03:15 -0700638 result = {
maruel064c0a32016-04-05 11:47:15 -0700639 'duration': None,
maruela9cfd6f2015-09-15 11:03:15 -0700640 'exit_code': None,
maruel6be7f9e2015-10-01 12:25:30 -0700641 'had_hard_timeout': False,
Seth Koehler49139812017-12-19 13:59:33 -0500642 'internal_failure': 'run_isolated did not complete properly',
maruel064c0a32016-04-05 11:47:15 -0700643 'stats': {
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000644 'isolated': {
645 #'cipd': {
646 # 'duration': 0.,
647 # 'get_client_duration': 0.,
648 #},
649 'download': {
650 #'duration': 0.,
651 'initial_number_items': len(data.isolate_cache),
652 'initial_size': data.isolate_cache.total_size,
653 #'items_cold': '<large.pack()>',
654 #'items_hot': '<large.pack()>',
655 },
656 #'upload': {
657 # 'duration': 0.,
658 # 'items_cold': '<large.pack()>',
659 # 'items_hot': '<large.pack()>',
660 #},
661 },
maruel064c0a32016-04-05 11:47:15 -0700662 },
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000663 #'cipd_pins': {
664 # 'packages': [
665 # {'package_name': ..., 'version': ..., 'path': ...},
666 # ...
667 # ],
668 # 'client_package': {'package_name': ..., 'version': ...},
669 #},
maruela9cfd6f2015-09-15 11:03:15 -0700670 'outputs_ref': None,
nodirbe642ff2016-06-09 15:51:51 -0700671 'version': 5,
maruela9cfd6f2015-09-15 11:03:15 -0700672 }
nodirbe642ff2016-06-09 15:51:51 -0700673
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500674 if data.root_dir:
Lei Leife202df2019-06-11 17:33:34 +0000675 file_path.ensure_tree(data.root_dir, 0o700)
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500676 elif data.isolate_cache.cache_dir:
677 data = data._replace(
678 root_dir=os.path.dirname(data.isolate_cache.cache_dir))
maruele2f2cb82016-07-13 14:41:03 -0700679 # See comment for these constants.
maruelcffa0542017-04-07 08:39:20 -0700680 # If root_dir is not specified, it is not constant.
681 # TODO(maruel): This is not obvious. Change this to become an error once we
682 # make the constant_run_path an exposed flag.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500683 if constant_run_path and data.root_dir:
684 run_dir = os.path.join(data.root_dir, ISOLATED_RUN_DIR)
maruel5c4eed82017-05-26 05:33:40 -0700685 if os.path.isdir(run_dir):
686 file_path.rmtree(run_dir)
Lei Leife202df2019-06-11 17:33:34 +0000687 os.mkdir(run_dir, 0o700)
maruelcffa0542017-04-07 08:39:20 -0700688 else:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500689 run_dir = make_temp_dir(ISOLATED_RUN_DIR, data.root_dir)
maruel03e11842016-07-14 10:50:16 -0700690 # storage should be normally set but don't crash if it is not. This can happen
691 # as Swarming task can run without an isolate server.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500692 out_dir = make_temp_dir(
693 ISOLATED_OUT_DIR, data.root_dir) if data.storage else None
694 tmp_dir = make_temp_dir(ISOLATED_TMP_DIR, data.root_dir)
nodir55be77b2016-05-03 09:39:57 -0700695 cwd = run_dir
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500696 if data.relative_cwd:
697 cwd = os.path.normpath(os.path.join(cwd, data.relative_cwd))
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500698 command = data.command
nodir55be77b2016-05-03 09:39:57 -0700699 try:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500700 with data.install_packages_fn(run_dir) as cipd_info:
vadimsh232f5a82017-01-20 19:23:44 -0800701 if cipd_info:
702 result['stats']['cipd'] = cipd_info.stats
703 result['cipd_pins'] = cipd_info.pins
nodir90bc8dc2016-06-15 13:35:21 -0700704
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500705 if data.isolated_hash:
vadimsh232f5a82017-01-20 19:23:44 -0800706 isolated_stats = result['stats'].setdefault('isolated', {})
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000707 bundle, stats = fetch_and_map(
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500708 isolated_hash=data.isolated_hash,
709 storage=data.storage,
710 cache=data.isolate_cache,
vadimsh232f5a82017-01-20 19:23:44 -0800711 outdir=run_dir,
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500712 use_symlinks=data.use_symlinks)
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000713 isolated_stats['download'].update(stats)
vadimsh232f5a82017-01-20 19:23:44 -0800714 change_tree_read_only(run_dir, bundle.read_only)
maruelabec63c2017-04-26 11:53:24 -0700715 # Inject the command
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500716 if not command and bundle.command:
717 command = bundle.command + data.extra_args
Marc-Antoine Rueld704a1f2017-10-31 10:51:23 -0400718 # Only set the relative directory if the isolated file specified a
719 # command, and no raw command was specified.
720 if bundle.relative_cwd:
721 cwd = os.path.normpath(os.path.join(cwd, bundle.relative_cwd))
maruelabec63c2017-04-26 11:53:24 -0700722
723 if not command:
724 # Handle this as a task failure, not an internal failure.
725 sys.stderr.write(
726 '<No command was specified!>\n'
727 '<Please secify a command when triggering your Swarming task>\n')
728 result['exit_code'] = 1
729 return result
nodirbe642ff2016-06-09 15:51:51 -0700730
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500731 if not cwd.startswith(run_dir):
732 # Handle this as a task failure, not an internal failure. This is a
733 # 'last chance' way to gate against directory escape.
734 sys.stderr.write('<Relative CWD is outside of run directory!>\n')
735 result['exit_code'] = 1
736 return result
737
738 if not os.path.isdir(cwd):
739 # Accepts relative_cwd that does not exist.
Lei Leife202df2019-06-11 17:33:34 +0000740 os.makedirs(cwd, 0o700)
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500741
vadimsh232f5a82017-01-20 19:23:44 -0800742 # If we have an explicit list of files to return, make sure their
743 # directories exist now.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500744 if data.storage and data.outputs:
745 isolateserver.create_directories(run_dir, data.outputs)
aludwin0a8e17d2016-10-27 15:57:39 -0700746
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500747 with data.install_named_caches(run_dir):
nodird6160682017-02-02 13:03:35 -0800748 sys.stdout.flush()
749 start = time.time()
750 try:
vadimsh9c54b2c2017-07-25 14:08:29 -0700751 # Need to switch the default account before 'get_command_env' call,
752 # so it can grab correct value of LUCI_CONTEXT env var.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500753 with set_luci_context_account(data.switch_to_account, tmp_dir):
754 env = get_command_env(
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000755 tmp_dir, cipd_info, run_dir, data.env, data.env_prefix, out_dir,
756 data.bot_file)
Robert Iannucci24ae76a2018-02-26 12:51:18 -0800757 command = tools.fix_python_cmd(command, env)
758 command = process_command(command, out_dir, data.bot_file)
759 file_path.ensure_command_has_abs_path(command, cwd)
760
vadimsh9c54b2c2017-07-25 14:08:29 -0700761 result['exit_code'], result['had_hard_timeout'] = run_command(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000762 command, cwd, env, data.hard_timeout, data.grace_period,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000763 data.lower_priority, data.containment)
nodird6160682017-02-02 13:03:35 -0800764 finally:
765 result['duration'] = max(time.time() - start, 0)
Seth Koehler49139812017-12-19 13:59:33 -0500766
767 # We successfully ran the command, set internal_failure back to
768 # None (even if the command failed, it's not an internal error).
769 result['internal_failure'] = None
maruela9cfd6f2015-09-15 11:03:15 -0700770 except Exception as e:
nodir90bc8dc2016-06-15 13:35:21 -0700771 # An internal error occurred. Report accordingly so the swarming task will
772 # be retried automatically.
maruel12e30012015-10-09 11:55:35 -0700773 logging.exception('internal failure: %s', e)
maruela9cfd6f2015-09-15 11:03:15 -0700774 result['internal_failure'] = str(e)
775 on_error.report(None)
aludwin0a8e17d2016-10-27 15:57:39 -0700776
777 # Clean up
maruela9cfd6f2015-09-15 11:03:15 -0700778 finally:
779 try:
aludwin0a8e17d2016-10-27 15:57:39 -0700780 # Try to link files to the output directory, if specified.
781 if out_dir:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500782 link_outputs_to_outdir(run_dir, out_dir, data.outputs)
aludwin0a8e17d2016-10-27 15:57:39 -0700783
nodir32a1ec12016-10-26 18:34:07 -0700784 success = False
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500785 if data.leak_temp_dir:
nodir32a1ec12016-10-26 18:34:07 -0700786 success = True
maruela9cfd6f2015-09-15 11:03:15 -0700787 logging.warning(
788 'Deliberately leaking %s for later examination', run_dir)
marueleb5fbee2015-09-17 13:01:36 -0700789 else:
maruel84537cb2015-10-16 14:21:28 -0700790 # On Windows rmtree(run_dir) call above has a synchronization effect: it
791 # finishes only when all task child processes terminate (since a running
792 # process locks *.exe file). Examine out_dir only after that call
793 # completes (since child processes may write to out_dir too and we need
794 # to wait for them to finish).
795 if fs.isdir(run_dir):
796 try:
797 success = file_path.rmtree(run_dir)
798 except OSError as e:
Marc-Antoine Ruel44699b32018-09-24 23:31:50 +0000799 logging.error('rmtree(%r) failed: %s', run_dir, e)
maruel84537cb2015-10-16 14:21:28 -0700800 success = False
801 if not success:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500802 sys.stderr.write(OUTLIVING_ZOMBIE_MSG % ('run', data.grace_period))
maruel84537cb2015-10-16 14:21:28 -0700803 if result['exit_code'] == 0:
804 result['exit_code'] = 1
805 if fs.isdir(tmp_dir):
806 try:
807 success = file_path.rmtree(tmp_dir)
808 except OSError as e:
Marc-Antoine Ruel44699b32018-09-24 23:31:50 +0000809 logging.error('rmtree(%r) failed: %s', tmp_dir, e)
maruel84537cb2015-10-16 14:21:28 -0700810 success = False
811 if not success:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500812 sys.stderr.write(OUTLIVING_ZOMBIE_MSG % ('temp', data.grace_period))
maruel84537cb2015-10-16 14:21:28 -0700813 if result['exit_code'] == 0:
814 result['exit_code'] = 1
maruela9cfd6f2015-09-15 11:03:15 -0700815
marueleb5fbee2015-09-17 13:01:36 -0700816 # This deletes out_dir if leak_temp_dir is not set.
nodir9130f072016-05-27 13:59:08 -0700817 if out_dir:
nodir55715712016-06-03 12:28:19 -0700818 isolated_stats = result['stats'].setdefault('isolated', {})
819 result['outputs_ref'], success, isolated_stats['upload'] = (
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500820 delete_and_upload(data.storage, out_dir, data.leak_temp_dir))
maruela9cfd6f2015-09-15 11:03:15 -0700821 if not success and result['exit_code'] == 0:
822 result['exit_code'] = 1
823 except Exception as e:
824 # Swallow any exception in the main finally clause.
nodir9130f072016-05-27 13:59:08 -0700825 if out_dir:
826 logging.exception('Leaking out_dir %s: %s', out_dir, e)
maruela9cfd6f2015-09-15 11:03:15 -0700827 result['internal_failure'] = str(e)
828 return result
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -0500829
830
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500831def run_tha_test(data, result_json):
nodir55be77b2016-05-03 09:39:57 -0700832 """Runs an executable and records execution metadata.
833
nodir55be77b2016-05-03 09:39:57 -0700834 If isolated_hash is specified, downloads the dependencies in the cache,
835 hardlinks them into a temporary directory and runs the command specified in
836 the .isolated.
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -0500837
838 A temporary directory is created to hold the output files. The content inside
839 this directory will be uploaded back to |storage| packaged as a .isolated
840 file.
841
842 Arguments:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500843 - data: TaskData instance.
844 - result_json: File path to dump result metadata into. If set, the process
845 exit code is always 0 unless an internal error occurred.
maruela9cfd6f2015-09-15 11:03:15 -0700846
847 Returns:
848 Process exit code that should be used.
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000849 """
maruela76b9ee2015-12-15 06:18:08 -0800850 if result_json:
851 # Write a json output file right away in case we get killed.
852 result = {
853 'exit_code': None,
854 'had_hard_timeout': False,
855 'internal_failure': 'Was terminated before completion',
856 'outputs_ref': None,
nodirbe642ff2016-06-09 15:51:51 -0700857 'version': 5,
maruela76b9ee2015-12-15 06:18:08 -0800858 }
859 tools.write_json(result_json, result, dense=True)
860
maruela9cfd6f2015-09-15 11:03:15 -0700861 # run_isolated exit code. Depends on if result_json is used or not.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500862 result = map_and_run(data, True)
maruela9cfd6f2015-09-15 11:03:15 -0700863 logging.info('Result:\n%s', tools.format_json(result, dense=True))
bpastene3ae09522016-06-10 17:12:59 -0700864
maruela9cfd6f2015-09-15 11:03:15 -0700865 if result_json:
maruel05d5a882015-09-21 13:59:02 -0700866 # We've found tests to delete 'work' when quitting, causing an exception
867 # here. Try to recreate the directory if necessary.
nodire5028a92016-04-29 14:38:21 -0700868 file_path.ensure_tree(os.path.dirname(result_json))
maruela9cfd6f2015-09-15 11:03:15 -0700869 tools.write_json(result_json, result, dense=True)
870 # Only return 1 if there was an internal error.
871 return int(bool(result['internal_failure']))
maruel@chromium.org781ccf62013-09-17 19:39:47 +0000872
maruela9cfd6f2015-09-15 11:03:15 -0700873 # Marshall into old-style inline output.
874 if result['outputs_ref']:
Marc-Antoine Ruel793bff32019-04-18 17:50:48 +0000875 # pylint: disable=unsubscriptable-object
maruela9cfd6f2015-09-15 11:03:15 -0700876 data = {
877 'hash': result['outputs_ref']['isolated'],
878 'namespace': result['outputs_ref']['namespace'],
879 'storage': result['outputs_ref']['isolatedserver'],
880 }
Marc-Antoine Ruelc44f5722015-01-08 16:10:01 -0500881 sys.stdout.flush()
maruela9cfd6f2015-09-15 11:03:15 -0700882 print(
883 '[run_isolated_out_hack]%s[/run_isolated_out_hack]' %
884 tools.format_json(data, dense=True))
maruelb76604c2015-11-11 11:53:44 -0800885 sys.stdout.flush()
maruela9cfd6f2015-09-15 11:03:15 -0700886 return result['exit_code'] or int(bool(result['internal_failure']))
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000887
888
iannuccib58d10d2017-03-18 02:00:25 -0700889# Yielded by 'install_client_and_packages'.
vadimsh232f5a82017-01-20 19:23:44 -0800890CipdInfo = collections.namedtuple('CipdInfo', [
891 'client', # cipd.CipdClient object
892 'cache_dir', # absolute path to bot-global cipd tag and instance cache
893 'stats', # dict with stats to return to the server
894 'pins', # dict with installed cipd pins to return to the server
895])
896
897
898@contextlib.contextmanager
899def noop_install_packages(_run_dir):
iannuccib58d10d2017-03-18 02:00:25 -0700900 """Placeholder for 'install_client_and_packages' if cipd is disabled."""
vadimsh232f5a82017-01-20 19:23:44 -0800901 yield None
902
903
iannuccib58d10d2017-03-18 02:00:25 -0700904def _install_packages(run_dir, cipd_cache_dir, client, packages, timeout):
905 """Calls 'cipd ensure' for packages.
906
907 Args:
908 run_dir (str): root of installation.
909 cipd_cache_dir (str): the directory to use for the cipd package cache.
910 client (CipdClient): the cipd client to use
911 packages: packages to install, list [(path, package_name, version), ...].
912 timeout: max duration in seconds that this function can take.
913
914 Returns: list of pinned packages. Looks like [
915 {
916 'path': 'subdirectory',
917 'package_name': 'resolved/package/name',
918 'version': 'deadbeef...',
919 },
920 ...
921 ]
922 """
923 package_pins = [None]*len(packages)
924 def insert_pin(path, name, version, idx):
925 package_pins[idx] = {
926 'package_name': name,
927 # swarming deals with 'root' as '.'
928 'path': path or '.',
929 'version': version,
930 }
931
932 by_path = collections.defaultdict(list)
933 for i, (path, name, version) in enumerate(packages):
934 # cipd deals with 'root' as ''
935 if path == '.':
936 path = ''
937 by_path[path].append((name, version, i))
938
939 pins = client.ensure(
940 run_dir,
941 {
942 subdir: [(name, vers) for name, vers, _ in pkgs]
943 for subdir, pkgs in by_path.iteritems()
944 },
945 cache_dir=cipd_cache_dir,
946 timeout=timeout,
947 )
948
949 for subdir, pin_list in sorted(pins.iteritems()):
950 this_subdir = by_path[subdir]
951 for i, (name, version) in enumerate(pin_list):
952 insert_pin(subdir, name, version, this_subdir[i][2])
953
Robert Iannucci461b30d2017-12-13 11:34:03 -0800954 assert None not in package_pins, (packages, pins, package_pins)
iannuccib58d10d2017-03-18 02:00:25 -0700955
956 return package_pins
957
958
vadimsh232f5a82017-01-20 19:23:44 -0800959@contextlib.contextmanager
iannuccib58d10d2017-03-18 02:00:25 -0700960def install_client_and_packages(
nodirff531b42016-06-23 13:05:06 -0700961 run_dir, packages, service_url, client_package_name,
vadimsh232f5a82017-01-20 19:23:44 -0800962 client_version, cache_dir, timeout=None):
vadimsh902948e2017-01-20 15:57:32 -0800963 """Bootstraps CIPD client and installs CIPD packages.
iannucci96fcccc2016-08-30 15:52:22 -0700964
vadimsh232f5a82017-01-20 19:23:44 -0800965 Yields CipdClient, stats, client info and pins (as single CipdInfo object).
966
967 Pins and the CIPD client info are in the form of:
iannucci96fcccc2016-08-30 15:52:22 -0700968 [
969 {
970 "path": path, "package_name": package_name, "version": version,
971 },
972 ...
973 ]
vadimsh902948e2017-01-20 15:57:32 -0800974 (the CIPD client info is a single dictionary instead of a list)
iannucci96fcccc2016-08-30 15:52:22 -0700975
976 such that they correspond 1:1 to all input package arguments from the command
977 line. These dictionaries make their all the way back to swarming, where they
978 become the arguments of CipdPackage.
nodirbe642ff2016-06-09 15:51:51 -0700979
vadimsh902948e2017-01-20 15:57:32 -0800980 If 'packages' list is empty, will bootstrap CIPD client, but won't install
981 any packages.
982
983 The bootstrapped client (regardless whether 'packages' list is empty or not),
vadimsh232f5a82017-01-20 19:23:44 -0800984 will be made available to the task via $PATH.
vadimsh902948e2017-01-20 15:57:32 -0800985
nodirbe642ff2016-06-09 15:51:51 -0700986 Args:
nodir90bc8dc2016-06-15 13:35:21 -0700987 run_dir (str): root of installation.
vadimsh902948e2017-01-20 15:57:32 -0800988 packages: packages to install, list [(path, package_name, version), ...].
nodirbe642ff2016-06-09 15:51:51 -0700989 service_url (str): CIPD server url, e.g.
990 "https://chrome-infra-packages.appspot.com."
nodir90bc8dc2016-06-15 13:35:21 -0700991 client_package_name (str): CIPD package name of CIPD client.
992 client_version (str): Version of CIPD client.
nodirbe642ff2016-06-09 15:51:51 -0700993 cache_dir (str): where to keep cache of cipd clients, packages and tags.
994 timeout: max duration in seconds that this function can take.
nodirbe642ff2016-06-09 15:51:51 -0700995 """
996 assert cache_dir
nodir90bc8dc2016-06-15 13:35:21 -0700997
nodirbe642ff2016-06-09 15:51:51 -0700998 timeoutfn = tools.sliding_timeout(timeout)
nodirbe642ff2016-06-09 15:51:51 -0700999 start = time.time()
nodirbe642ff2016-06-09 15:51:51 -07001000
vadimsh902948e2017-01-20 15:57:32 -08001001 cache_dir = os.path.abspath(cache_dir)
vadimsh232f5a82017-01-20 19:23:44 -08001002 cipd_cache_dir = os.path.join(cache_dir, 'cache') # tag and instance caches
nodir90bc8dc2016-06-15 13:35:21 -07001003 run_dir = os.path.abspath(run_dir)
vadimsh902948e2017-01-20 15:57:32 -08001004 packages = packages or []
nodir90bc8dc2016-06-15 13:35:21 -07001005
nodirbe642ff2016-06-09 15:51:51 -07001006 get_client_start = time.time()
1007 client_manager = cipd.get_client(
1008 service_url, client_package_name, client_version, cache_dir,
1009 timeout=timeoutfn())
iannucci96fcccc2016-08-30 15:52:22 -07001010
nodirbe642ff2016-06-09 15:51:51 -07001011 with client_manager as client:
1012 get_client_duration = time.time() - get_client_start
nodir90bc8dc2016-06-15 13:35:21 -07001013
iannuccib58d10d2017-03-18 02:00:25 -07001014 package_pins = []
1015 if packages:
1016 package_pins = _install_packages(
1017 run_dir, cipd_cache_dir, client, packages, timeoutfn())
1018
1019 file_path.make_tree_files_read_only(run_dir)
nodir90bc8dc2016-06-15 13:35:21 -07001020
vadimsh232f5a82017-01-20 19:23:44 -08001021 total_duration = time.time() - start
1022 logging.info(
1023 'Installing CIPD client and packages took %d seconds', total_duration)
nodir90bc8dc2016-06-15 13:35:21 -07001024
vadimsh232f5a82017-01-20 19:23:44 -08001025 yield CipdInfo(
1026 client=client,
1027 cache_dir=cipd_cache_dir,
1028 stats={
1029 'duration': total_duration,
1030 'get_client_duration': get_client_duration,
1031 },
1032 pins={
iannuccib58d10d2017-03-18 02:00:25 -07001033 'client_package': {
1034 'package_name': client.package_name,
1035 'version': client.instance_id,
1036 },
vadimsh232f5a82017-01-20 19:23:44 -08001037 'packages': package_pins,
1038 })
nodirbe642ff2016-06-09 15:51:51 -07001039
1040
1041def create_option_parser():
Marc-Antoine Ruelf74cffe2015-07-15 15:21:34 -04001042 parser = logging_utils.OptionParserWithLogging(
nodir55be77b2016-05-03 09:39:57 -07001043 usage='%prog <options> [command to run or extra args]',
maruel@chromium.orgdedbf492013-09-12 20:42:11 +00001044 version=__version__,
1045 log_file=RUN_ISOLATED_LOG_FILE)
maruela9cfd6f2015-09-15 11:03:15 -07001046 parser.add_option(
maruel36a963d2016-04-08 17:15:49 -07001047 '--clean', action='store_true',
1048 help='Cleans the cache, trimming it necessary and remove corrupted items '
1049 'and returns without executing anything; use with -v to know what '
1050 'was done')
1051 parser.add_option(
maruel4409e302016-07-19 14:25:51 -07001052 '--use-symlinks', action='store_true',
1053 help='Use symlinks instead of hardlinks')
1054 parser.add_option(
maruela9cfd6f2015-09-15 11:03:15 -07001055 '--json',
1056 help='dump output metadata to json file. When used, run_isolated returns '
1057 'non-zero only on internal failure')
maruel6be7f9e2015-10-01 12:25:30 -07001058 parser.add_option(
maruel5c9e47b2015-12-18 13:02:30 -08001059 '--hard-timeout', type='float', help='Enforce hard timeout in execution')
maruel6be7f9e2015-10-01 12:25:30 -07001060 parser.add_option(
maruel5c9e47b2015-12-18 13:02:30 -08001061 '--grace-period', type='float',
maruel6be7f9e2015-10-01 12:25:30 -07001062 help='Grace period between SIGTERM and SIGKILL')
bpastene3ae09522016-06-10 17:12:59 -07001063 parser.add_option(
Marc-Antoine Ruel49e347d2017-10-24 16:52:02 -07001064 '--raw-cmd', action='store_true',
1065 help='Ignore the isolated command, use the one supplied at the command '
1066 'line')
1067 parser.add_option(
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001068 '--relative-cwd',
1069 help='Ignore the isolated \'relative_cwd\' and use this one instead; '
1070 'requires --raw-cmd')
1071 parser.add_option(
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001072 '--env', default=[], action='append',
1073 help='Environment variables to set for the child process')
1074 parser.add_option(
1075 '--env-prefix', default=[], action='append',
Robert Iannuccibf5f84c2017-11-22 12:56:50 -08001076 help='Specify a VAR=./path/fragment to put in the environment variable '
1077 'before executing the command. The path fragment must be relative '
1078 'to the isolated run directory, and must not contain a `..` token. '
1079 'The path will be made absolute and prepended to the indicated '
1080 '$VAR using the OS\'s path separator. Multiple items for the same '
1081 '$VAR will be prepended in order.')
1082 parser.add_option(
bpastene3ae09522016-06-10 17:12:59 -07001083 '--bot-file',
1084 help='Path to a file describing the state of the host. The content is '
1085 'defined by on_before_task() in bot_config.')
aludwin7556e0c2016-10-26 08:46:10 -07001086 parser.add_option(
vadimsh9c54b2c2017-07-25 14:08:29 -07001087 '--switch-to-account',
1088 help='If given, switches LUCI_CONTEXT to given logical service account '
1089 '(e.g. "task" or "system") before launching the isolated process.')
1090 parser.add_option(
aludwin0a8e17d2016-10-27 15:57:39 -07001091 '--output', action='append',
1092 help='Specifies an output to return. If no outputs are specified, all '
1093 'files located in $(ISOLATED_OUTDIR) will be returned; '
1094 'otherwise, outputs in both $(ISOLATED_OUTDIR) and those '
1095 'specified by --output option (there can be multiple) will be '
1096 'returned. Note that if a file in OUT_DIR has the same path '
1097 'as an --output option, the --output version will be returned.')
1098 parser.add_option(
aludwin7556e0c2016-10-26 08:46:10 -07001099 '-a', '--argsfile',
1100 # This is actually handled in parse_args; it's included here purely so it
1101 # can make it into the help text.
1102 help='Specify a file containing a JSON array of arguments to this '
1103 'script. If --argsfile is provided, no other argument may be '
1104 'provided on the command line.')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001105
1106 group = optparse.OptionGroup(parser, 'Data source')
1107 group.add_option(
Marc-Antoine Ruel185ded42015-01-28 20:49:18 -05001108 '-s', '--isolated',
nodir55be77b2016-05-03 09:39:57 -07001109 help='Hash of the .isolated to grab from the isolate server.')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001110 isolateserver.add_isolate_server_options(group)
1111 parser.add_option_group(group)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001112
Marc-Antoine Ruela57d7db2014-10-15 20:31:19 -04001113 isolateserver.add_cache_options(parser)
nodirbe642ff2016-06-09 15:51:51 -07001114
1115 cipd.add_cipd_options(parser)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001116
1117 group = optparse.OptionGroup(parser, 'Named caches')
1118 group.add_option(
1119 '--named-cache',
1120 dest='named_caches',
1121 action='append',
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001122 nargs=3,
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001123 default=[],
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001124 help='A named cache to request. Accepts 3 arguments: name, path, hint. '
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001125 'name identifies the cache, must match regex [a-z0-9_]{1,4096}. '
1126 'path is a path relative to the run dir where the cache directory '
1127 'must be put to. '
1128 'This option can be specified more than once.')
1129 group.add_option(
1130 '--named-cache-root', default='named_caches',
1131 help='Cache root directory. Default=%default')
1132 parser.add_option_group(group)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001133
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001134 group = optparse.OptionGroup(parser, 'Process containment')
1135 parser.add_option(
1136 '--lower-priority', action='store_true',
1137 help='Lowers the child process priority')
1138 parser.add_option(
1139 '--containment-type', choices=('NONE', 'AUTO', 'JOB_OBJECT'),
1140 default='NONE',
1141 help='Type of container to use')
1142 parser.add_option(
1143 '--limit-processes', type='int', default=0,
1144 help='Maximum number of active processes in the containment')
1145 parser.add_option(
1146 '--limit-total-committed-memory', type='int', default=0,
1147 help='Maximum sum of committed memory in the containment')
1148 parser.add_option_group(group)
1149
1150 group = optparse.OptionGroup(parser, 'Debugging')
1151 group.add_option(
Kenneth Russell61d42352014-09-15 11:41:16 -07001152 '--leak-temp-dir',
1153 action='store_true',
nodirbe642ff2016-06-09 15:51:51 -07001154 help='Deliberately leak isolate\'s temp dir for later examination. '
1155 'Default: %default')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001156 group.add_option(
marueleb5fbee2015-09-17 13:01:36 -07001157 '--root-dir', help='Use a directory instead of a random one')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001158 parser.add_option_group(group)
Kenneth Russell61d42352014-09-15 11:41:16 -07001159
Vadim Shtayurae34e13a2014-02-02 11:23:26 -08001160 auth.add_auth_options(parser)
nodirbe642ff2016-06-09 15:51:51 -07001161
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001162 parser.set_defaults(cache='cache', cipd_cache='cipd_cache')
nodirbe642ff2016-06-09 15:51:51 -07001163 return parser
1164
1165
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001166def process_named_cache_options(parser, options, time_fn=None):
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001167 """Validates named cache options and returns a CacheManager."""
1168 if options.named_caches and not options.named_cache_root:
1169 parser.error('--named-cache is specified, but --named-cache-root is empty')
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001170 for name, path, hint in options.named_caches:
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001171 if not CACHE_NAME_RE.match(name):
1172 parser.error(
1173 'cache name %r does not match %r' % (name, CACHE_NAME_RE.pattern))
1174 if not path:
1175 parser.error('cache path cannot be empty')
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001176 try:
1177 long(hint)
1178 except ValueError:
1179 parser.error('cache hint must be a number')
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001180 if options.named_cache_root:
1181 # Make these configurable later if there is use case but for now it's fairly
1182 # safe values.
1183 # In practice, a fair chunk of bots are already recycled on a daily schedule
1184 # so this code doesn't have any effect to them, unless they are preloaded
1185 # with a really old cache.
1186 policies = local_caching.CachePolicies(
1187 # 1TiB.
1188 max_cache_size=1024*1024*1024*1024,
1189 min_free_space=options.min_free_space,
1190 max_items=50,
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +00001191 max_age_secs=MAX_AGE_SECS)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001192 root_dir = unicode(os.path.abspath(options.named_cache_root))
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001193 return local_caching.NamedCache(root_dir, policies, time_fn=time_fn)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001194 return None
1195
1196
aludwin7556e0c2016-10-26 08:46:10 -07001197def parse_args(args):
1198 # Create a fake mini-parser just to get out the "-a" command. Note that
1199 # it's not documented here; instead, it's documented in create_option_parser
1200 # even though that parser will never actually get to parse it. This is
1201 # because --argsfile is exclusive with all other options and arguments.
1202 file_argparse = argparse.ArgumentParser(add_help=False)
1203 file_argparse.add_argument('-a', '--argsfile')
1204 (file_args, nonfile_args) = file_argparse.parse_known_args(args)
1205 if file_args.argsfile:
1206 if nonfile_args:
1207 file_argparse.error('Can\'t specify --argsfile with'
1208 'any other arguments (%s)' % nonfile_args)
1209 try:
1210 with open(file_args.argsfile, 'r') as f:
1211 args = json.load(f)
1212 except (IOError, OSError, ValueError) as e:
1213 # We don't need to error out here - "args" is now empty,
1214 # so the call below to parser.parse_args(args) will fail
1215 # and print the full help text.
1216 print >> sys.stderr, 'Couldn\'t read arguments: %s' % e
1217
1218 # Even if we failed to read the args, just call the normal parser now since it
1219 # will print the correct help message.
nodirbe642ff2016-06-09 15:51:51 -07001220 parser = create_option_parser()
Marc-Antoine Ruel90c98162013-12-18 15:11:57 -05001221 options, args = parser.parse_args(args)
aludwin7556e0c2016-10-26 08:46:10 -07001222 return (parser, options, args)
1223
1224
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001225def _calc_named_cache_hint(named_cache, named_caches):
1226 """Returns the expected size of the missing named caches."""
1227 present = named_cache.available
1228 size = 0
1229 for name, _, hint in named_caches:
1230 if name not in present:
1231 hint = long(hint)
1232 if hint > 0:
1233 size += hint
1234 return size
1235
1236
aludwin7556e0c2016-10-26 08:46:10 -07001237def main(args):
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -05001238 # Warning: when --argsfile is used, the strings are unicode instances, when
1239 # parsed normally, the strings are str instances.
aludwin7556e0c2016-10-26 08:46:10 -07001240 (parser, options, args) = parse_args(args)
maruel36a963d2016-04-08 17:15:49 -07001241
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001242 if not file_path.enable_symlink():
Marc-Antoine Ruel5a024272019-01-15 20:11:16 +00001243 logging.warning('Symlink support is not enabled')
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001244
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001245 named_cache = process_named_cache_options(parser, options)
Marc-Antoine Ruel0d8b0f62018-09-10 14:40:35 +00001246 # hint is 0 if there's no named cache.
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001247 hint = _calc_named_cache_hint(named_cache, options.named_caches)
1248 if hint:
1249 # Increase the --min-free-space value by the hint, and recreate the
1250 # NamedCache instance so it gets the updated CachePolicy.
1251 options.min_free_space += hint
1252 named_cache = process_named_cache_options(parser, options)
1253
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001254 # TODO(maruel): CIPD caches should be defined at an higher level here too, so
1255 # they can be cleaned the same way.
nodirf33b8d62016-10-26 22:34:58 -07001256 isolate_cache = isolateserver.process_cache_options(options, trim=False)
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001257 caches = []
1258 if isolate_cache:
1259 caches.append(isolate_cache)
1260 if named_cache:
1261 caches.append(named_cache)
1262 root = caches[0].cache_dir if caches else unicode(os.getcwd())
maruel36a963d2016-04-08 17:15:49 -07001263 if options.clean:
1264 if options.isolated:
1265 parser.error('Can\'t use --isolated with --clean.')
1266 if options.isolate_server:
1267 parser.error('Can\'t use --isolate-server with --clean.')
1268 if options.json:
1269 parser.error('Can\'t use --json with --clean.')
nodirf33b8d62016-10-26 22:34:58 -07001270 if options.named_caches:
1271 parser.error('Can\t use --named-cache with --clean.')
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001272 # Trim first, then clean.
1273 local_caching.trim_caches(
1274 caches,
1275 root,
1276 min_free_space=options.min_free_space,
1277 max_age_secs=MAX_AGE_SECS)
1278 for c in caches:
Marc-Antoine Ruel87fc2222018-06-18 13:09:24 +00001279 c.cleanup()
maruel36a963d2016-04-08 17:15:49 -07001280 return 0
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001281
1282 # Trim must still be done for the following case:
1283 # - named-cache was used
1284 # - some entries, with a large hint, where missing
1285 # - --min-free-space was increased accordingly, thus trimming is needed
1286 # Otherwise, this will have no effect, as bot_main calls run_isolated with
1287 # --clean after each task.
1288 if hint:
1289 logging.info('Additional trimming of %d bytes', hint)
Marc-Antoine Ruel0d8b0f62018-09-10 14:40:35 +00001290 local_caching.trim_caches(
1291 caches,
1292 root,
1293 min_free_space=options.min_free_space,
1294 max_age_secs=MAX_AGE_SECS)
maruel36a963d2016-04-08 17:15:49 -07001295
nodir55be77b2016-05-03 09:39:57 -07001296 if not options.isolated and not args:
1297 parser.error('--isolated or command to run is required.')
1298
Vadim Shtayura5d1efce2014-02-04 10:55:43 -08001299 auth.process_auth_options(parser, options)
nodir55be77b2016-05-03 09:39:57 -07001300
1301 isolateserver.process_isolate_server_options(
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001302 parser, options, True, False)
nodir55be77b2016-05-03 09:39:57 -07001303 if not options.isolate_server:
1304 if options.isolated:
1305 parser.error('--isolated requires --isolate-server')
1306 if ISOLATED_OUTDIR_PARAMETER in args:
1307 parser.error(
1308 '%s in args requires --isolate-server' % ISOLATED_OUTDIR_PARAMETER)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001309
nodir90bc8dc2016-06-15 13:35:21 -07001310 if options.root_dir:
1311 options.root_dir = unicode(os.path.abspath(options.root_dir))
maruel12e30012015-10-09 11:55:35 -07001312 if options.json:
1313 options.json = unicode(os.path.abspath(options.json))
nodir55be77b2016-05-03 09:39:57 -07001314
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001315 if any('=' not in i for i in options.env):
1316 parser.error(
1317 '--env required key=value form. value can be skipped to delete '
1318 'the variable')
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -05001319 options.env = dict(i.split('=', 1) for i in options.env)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001320
1321 prefixes = {}
1322 cwd = os.path.realpath(os.getcwd())
1323 for item in options.env_prefix:
1324 if '=' not in item:
1325 parser.error(
1326 '--env-prefix %r is malformed, must be in the form `VAR=./path`'
1327 % item)
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -05001328 key, opath = item.split('=', 1)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001329 if os.path.isabs(opath):
1330 parser.error('--env-prefix %r path is bad, must be relative.' % opath)
1331 opath = os.path.normpath(opath)
1332 if not os.path.realpath(os.path.join(cwd, opath)).startswith(cwd):
1333 parser.error(
1334 '--env-prefix %r path is bad, must be relative and not contain `..`.'
1335 % opath)
1336 prefixes.setdefault(key, []).append(opath)
1337 options.env_prefix = prefixes
Robert Iannuccibf5f84c2017-11-22 12:56:50 -08001338
nodirbe642ff2016-06-09 15:51:51 -07001339 cipd.validate_cipd_options(parser, options)
1340
vadimsh232f5a82017-01-20 19:23:44 -08001341 install_packages_fn = noop_install_packages
vadimsh902948e2017-01-20 15:57:32 -08001342 if options.cipd_enabled:
iannuccib58d10d2017-03-18 02:00:25 -07001343 install_packages_fn = lambda run_dir: install_client_and_packages(
vadimsh902948e2017-01-20 15:57:32 -08001344 run_dir, cipd.parse_package_args(options.cipd_packages),
1345 options.cipd_server, options.cipd_client_package,
1346 options.cipd_client_version, cache_dir=options.cipd_cache)
nodirbe642ff2016-06-09 15:51:51 -07001347
nodird6160682017-02-02 13:03:35 -08001348 @contextlib.contextmanager
nodir0ae98b32017-05-11 13:21:53 -07001349 def install_named_caches(run_dir):
nodird6160682017-02-02 13:03:35 -08001350 # WARNING: this function depends on "options" variable defined in the outer
1351 # function.
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001352 assert unicode(run_dir), repr(run_dir)
1353 assert os.path.isabs(run_dir), run_dir
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001354 named_caches = [
nodir0ae98b32017-05-11 13:21:53 -07001355 (os.path.join(run_dir, unicode(relpath)), name)
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001356 for name, relpath, _ in options.named_caches
nodir0ae98b32017-05-11 13:21:53 -07001357 ]
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001358 for path, name in named_caches:
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001359 named_cache.install(path, name)
nodird6160682017-02-02 13:03:35 -08001360 try:
1361 yield
1362 finally:
dnje289d132017-07-07 11:16:44 -07001363 # Uninstall each named cache, returning it to the cache pool. If an
1364 # uninstall fails for a given cache, it will remain in the task's
1365 # temporary space, get cleaned up by the Swarming bot, and be lost.
1366 #
1367 # If the Swarming bot cannot clean up the cache, it will handle it like
1368 # any other bot file that could not be removed.
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001369 for path, name in reversed(named_caches):
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001370 try:
Marc-Antoine Ruele9558372018-08-03 03:41:22 +00001371 # uninstall() doesn't trim but does call save() implicitly. Trimming
1372 # *must* be done manually via periodic 'run_isolated.py --clean'.
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001373 named_cache.uninstall(path, name)
1374 except local_caching.NamedCacheError:
1375 logging.exception('Error while removing named cache %r at %r. '
1376 'The cache will be lost.', path, name)
nodirf33b8d62016-10-26 22:34:58 -07001377
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001378 extra_args = []
1379 command = []
1380 if options.raw_cmd:
1381 command = args
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001382 if options.relative_cwd:
1383 a = os.path.normpath(os.path.abspath(options.relative_cwd))
1384 if not a.startswith(os.getcwd()):
1385 parser.error(
1386 '--relative-cwd must not try to escape the working directory')
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001387 else:
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001388 if options.relative_cwd:
1389 parser.error('--relative-cwd requires --raw-cmd')
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001390 extra_args = args
1391
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001392 containment_type = subprocess42.Containment.NONE
1393 if options.containment_type == 'AUTO':
1394 containment_type = subprocess42.Containment.AUTO
1395 if options.containment_type == 'JOB_OBJECT':
1396 containment_type = subprocess42.Containment.JOB_OBJECT
1397 containment = subprocess42.Containment(
1398 containment_type=containment_type,
1399 limit_processes=options.limit_processes,
1400 limit_total_committed_memory=options.limit_total_committed_memory)
1401
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001402 data = TaskData(
1403 command=command,
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001404 relative_cwd=options.relative_cwd,
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001405 extra_args=extra_args,
1406 isolated_hash=options.isolated,
1407 storage=None,
1408 isolate_cache=isolate_cache,
1409 outputs=options.output,
1410 install_named_caches=install_named_caches,
1411 leak_temp_dir=options.leak_temp_dir,
1412 root_dir=_to_unicode(options.root_dir),
1413 hard_timeout=options.hard_timeout,
1414 grace_period=options.grace_period,
1415 bot_file=options.bot_file,
1416 switch_to_account=options.switch_to_account,
1417 install_packages_fn=install_packages_fn,
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +00001418 use_symlinks=bool(options.use_symlinks),
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001419 env=options.env,
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +00001420 env_prefix=options.env_prefix,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001421 lower_priority=bool(options.lower_priority),
1422 containment=containment)
nodirbe642ff2016-06-09 15:51:51 -07001423 try:
nodir90bc8dc2016-06-15 13:35:21 -07001424 if options.isolate_server:
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001425 server_ref = isolate_storage.ServerRef(
nodir90bc8dc2016-06-15 13:35:21 -07001426 options.isolate_server, options.namespace)
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001427 storage = isolateserver.get_storage(server_ref)
nodir90bc8dc2016-06-15 13:35:21 -07001428 with storage:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001429 data = data._replace(storage=storage)
nodirf33b8d62016-10-26 22:34:58 -07001430 # Hashing schemes used by |storage| and |isolate_cache| MUST match.
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001431 assert storage.server_ref.hash_algo == server_ref.hash_algo
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001432 return run_tha_test(data, options.json)
1433 return run_tha_test(data, options.json)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001434 except (
1435 cipd.Error,
1436 local_caching.NamedCacheError,
Marc-Antoine Ruelb6e9e232018-11-20 00:12:33 +00001437 local_caching.NoMoreSpace) as ex:
nodirbe642ff2016-06-09 15:51:51 -07001438 print >> sys.stderr, ex.message
1439 return 1
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001440
1441
1442if __name__ == '__main__':
maruel8e4e40c2016-05-30 06:21:07 -07001443 subprocess42.inhibit_os_error_reporting()
csharp@chromium.orgbfb98742013-03-26 20:28:36 +00001444 # Ensure that we are always running with the correct encoding.
vadimsh@chromium.orga4326472013-08-24 02:05:41 +00001445 fix_encoding.fix_encoding()
Marc-Antoine Ruel90c98162013-12-18 15:11:57 -05001446 sys.exit(main(sys.argv[1:]))