blob: d253649360d4f4361646b5755a08f93a195795ef [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
Takuto Ikuta33e2ff32019-09-30 12:44:03 +0000270 just inherit whatever account is already set. This may happen if users invoke
vadimsh9c54b2c2017-07-25 14:08:29 -0700271 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()
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000372 for k, v in env.items():
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -0500373 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
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000383 for key, paths in env_prefixes.items():
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(
tikuta2d678212019-09-23 23:12:08 +0000486 '<The executable does not exist, a dependent library is missing or '
487 'the command line is too long>\n'
488 '<Check for missing .so/.dll in the .isolate or GN file or length of '
489 'command line args>\n'
maruela72f46e2016-02-24 11:05:45 -0800490 '<Command: %s>\n' % command)
491 if os.environ.get('SWARMING_TASK_ID'):
492 # Give an additional hint when running as a swarming task.
493 sys.stderr.write(
494 '<See the task\'s page for commands to help diagnose this issue '
495 'by reproducing the task locally>\n')
maruela9cfd6f2015-09-15 11:03:15 -0700496 exit_code = 1
497 logging.info(
498 'Command finished with exit code %d (%s)',
499 exit_code, hex(0xffffffff & exit_code))
maruel6be7f9e2015-10-01 12:25:30 -0700500 return exit_code, had_hard_timeout
maruela9cfd6f2015-09-15 11:03:15 -0700501
502
maruel4409e302016-07-19 14:25:51 -0700503def fetch_and_map(isolated_hash, storage, cache, outdir, use_symlinks):
504 """Fetches an isolated tree, create the tree and returns (bundle, stats)."""
nodir6f801882016-04-29 14:41:50 -0700505 start = time.time()
506 bundle = isolateserver.fetch_isolated(
507 isolated_hash=isolated_hash,
508 storage=storage,
509 cache=cache,
maruel4409e302016-07-19 14:25:51 -0700510 outdir=outdir,
511 use_symlinks=use_symlinks)
Takuto Ikuta2b9640e2019-06-19 00:53:23 +0000512 hot = (collections.Counter(cache.used) -
513 collections.Counter(cache.added)).elements()
nodir6f801882016-04-29 14:41:50 -0700514 return bundle, {
515 'duration': time.time() - start,
nodir6f801882016-04-29 14:41:50 -0700516 'items_cold': base64.b64encode(large.pack(sorted(cache.added))),
Takuto Ikuta2b9640e2019-06-19 00:53:23 +0000517 'items_hot': base64.b64encode(large.pack(sorted(hot))),
nodir6f801882016-04-29 14:41:50 -0700518 }
519
520
aludwin0a8e17d2016-10-27 15:57:39 -0700521def link_outputs_to_outdir(run_dir, out_dir, outputs):
522 """Links any named outputs to out_dir so they can be uploaded.
523
524 Raises an error if the file already exists in that directory.
525 """
526 if not outputs:
527 return
528 isolateserver.create_directories(out_dir, outputs)
529 for o in outputs:
Sadaf Matinkhoo10743a62018-03-29 16:28:58 -0400530 copy_recursively(os.path.join(run_dir, o), os.path.join(out_dir, o))
531
532
533def copy_recursively(src, dst):
534 """Efficiently copies a file or directory from src_dir to dst_dir.
535
536 `item` may be a file, directory, or a symlink to a file or directory.
537 All symlinks are replaced with their targets, so the resulting
538 directory structure in dst_dir will never have any symlinks.
539
540 To increase speed, copy_recursively hardlinks individual files into the
541 (newly created) directory structure if possible, unlike Python's
542 shutil.copytree().
543 """
544 orig_src = src
545 try:
546 # Replace symlinks with their final target.
547 while fs.islink(src):
548 res = fs.readlink(src)
549 src = os.path.join(os.path.dirname(src), res)
550 # TODO(sadafm): Explicitly handle cyclic symlinks.
551
552 # Note that fs.isfile (which is a wrapper around os.path.isfile) throws
553 # an exception if src does not exist. A warning will be logged in that case.
554 if fs.isfile(src):
555 file_path.link_file(dst, src, file_path.HARDLINK_WITH_FALLBACK)
556 return
557
558 if not fs.exists(dst):
559 os.makedirs(dst)
560
561 for child in fs.listdir(src):
562 copy_recursively(os.path.join(src, child), os.path.join(dst, child))
563
564 except OSError as e:
565 if e.errno == errno.ENOENT:
566 logging.warning('Path %s does not exist or %s is a broken symlink',
567 src, orig_src)
568 else:
569 logging.info("Couldn't collect output file %s: %s", src, e)
aludwin0a8e17d2016-10-27 15:57:39 -0700570
571
Takuto Ikutab4aa8662019-09-17 05:54:36 +0000572def upload_then_delete(storage, out_dir, leak_temp_dir):
maruela9cfd6f2015-09-15 11:03:15 -0700573 """Deletes the temporary run directory and uploads results back.
574
575 Returns:
nodir6f801882016-04-29 14:41:50 -0700576 tuple(outputs_ref, success, stats)
maruel064c0a32016-04-05 11:47:15 -0700577 - outputs_ref: a dict referring to the results archived back to the isolated
578 server, if applicable.
579 - success: False if something occurred that means that the task must
580 forcibly be considered a failure, e.g. zombie processes were left
581 behind.
nodir6f801882016-04-29 14:41:50 -0700582 - stats: uploading stats.
maruela9cfd6f2015-09-15 11:03:15 -0700583 """
maruela9cfd6f2015-09-15 11:03:15 -0700584 # Upload out_dir and generate a .isolated file out of this directory. It is
585 # only done if files were written in the directory.
586 outputs_ref = None
maruel064c0a32016-04-05 11:47:15 -0700587 cold = []
588 hot = []
nodir6f801882016-04-29 14:41:50 -0700589 start = time.time()
590
maruel12e30012015-10-09 11:55:35 -0700591 if fs.isdir(out_dir) and fs.listdir(out_dir):
maruela9cfd6f2015-09-15 11:03:15 -0700592 with tools.Profiler('ArchiveOutput'):
593 try:
maruel064c0a32016-04-05 11:47:15 -0700594 results, f_cold, f_hot = isolateserver.archive_files_to_storage(
maruela9cfd6f2015-09-15 11:03:15 -0700595 storage, [out_dir], None)
596 outputs_ref = {
Marc-Antoine Rueld0868ec2018-11-28 20:47:29 +0000597 'isolated': results.values()[0],
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +0000598 'isolatedserver': storage.server_ref.url,
599 'namespace': storage.server_ref.namespace,
maruela9cfd6f2015-09-15 11:03:15 -0700600 }
maruel064c0a32016-04-05 11:47:15 -0700601 cold = sorted(i.size for i in f_cold)
602 hot = sorted(i.size for i in f_hot)
maruela9cfd6f2015-09-15 11:03:15 -0700603 except isolateserver.Aborted:
604 # This happens when a signal SIGTERM was received while uploading data.
605 # There is 2 causes:
606 # - The task was too slow and was about to be killed anyway due to
607 # exceeding the hard timeout.
608 # - The amount of data uploaded back is very large and took too much
609 # time to archive.
610 sys.stderr.write('Received SIGTERM while uploading')
611 # Re-raise, so it will be treated as an internal failure.
612 raise
nodir6f801882016-04-29 14:41:50 -0700613
614 success = False
maruela9cfd6f2015-09-15 11:03:15 -0700615 try:
maruel12e30012015-10-09 11:55:35 -0700616 if (not leak_temp_dir and fs.isdir(out_dir) and
maruel6eeea7d2015-09-16 12:17:42 -0700617 not file_path.rmtree(out_dir)):
maruela9cfd6f2015-09-15 11:03:15 -0700618 logging.error('Had difficulties removing out_dir %s', out_dir)
nodir6f801882016-04-29 14:41:50 -0700619 else:
620 success = True
maruela9cfd6f2015-09-15 11:03:15 -0700621 except OSError as e:
622 # When this happens, it means there's a process error.
maruel12e30012015-10-09 11:55:35 -0700623 logging.exception('Had difficulties removing out_dir %s: %s', out_dir, e)
nodir6f801882016-04-29 14:41:50 -0700624 stats = {
625 'duration': time.time() - start,
626 'items_cold': base64.b64encode(large.pack(cold)),
627 'items_hot': base64.b64encode(large.pack(hot)),
628 }
629 return outputs_ref, success, stats
maruela9cfd6f2015-09-15 11:03:15 -0700630
631
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500632def map_and_run(data, constant_run_path):
nodir55be77b2016-05-03 09:39:57 -0700633 """Runs a command with optional isolated input/output.
634
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500635 Arguments:
636 - data: TaskData instance.
637 - constant_run_path: TODO
nodir55be77b2016-05-03 09:39:57 -0700638
639 Returns metadata about the result.
640 """
maruela9cfd6f2015-09-15 11:03:15 -0700641 result = {
Takuto Ikuta5ed62ad2019-09-26 09:16:00 +0000642 'duration': None,
643 'exit_code': None,
644 'had_hard_timeout': False,
645 'internal_failure': 'run_isolated did not complete properly',
646 'stats': {
647 #'cipd': {
648 # 'duration': 0.,
649 # 'get_client_duration': 0.,
650 #},
651 'isolated': {
652 'download': {
653 #'duration': 0.,
654 'initial_number_items': len(data.isolate_cache),
655 'initial_size': data.isolate_cache.total_size,
656 #'items_cold': '<large.pack()>',
657 #'items_hot': '<large.pack()>',
658 },
659 #'upload': {
660 # 'duration': 0.,
661 # 'items_cold': '<large.pack()>',
662 # 'items_hot': '<large.pack()>',
663 #},
664 },
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000665 },
Takuto Ikuta5ed62ad2019-09-26 09:16:00 +0000666 #'cipd_pins': {
667 # 'packages': [
668 # {'package_name': ..., 'version': ..., 'path': ...},
669 # ...
670 # ],
671 # 'client_package': {'package_name': ..., 'version': ...},
672 #},
673 'outputs_ref': None,
674 'version': 5,
maruela9cfd6f2015-09-15 11:03:15 -0700675 }
nodirbe642ff2016-06-09 15:51:51 -0700676
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500677 if data.root_dir:
Lei Leife202df2019-06-11 17:33:34 +0000678 file_path.ensure_tree(data.root_dir, 0o700)
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500679 elif data.isolate_cache.cache_dir:
680 data = data._replace(
681 root_dir=os.path.dirname(data.isolate_cache.cache_dir))
maruele2f2cb82016-07-13 14:41:03 -0700682 # See comment for these constants.
maruelcffa0542017-04-07 08:39:20 -0700683 # If root_dir is not specified, it is not constant.
684 # TODO(maruel): This is not obvious. Change this to become an error once we
685 # make the constant_run_path an exposed flag.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500686 if constant_run_path and data.root_dir:
687 run_dir = os.path.join(data.root_dir, ISOLATED_RUN_DIR)
maruel5c4eed82017-05-26 05:33:40 -0700688 if os.path.isdir(run_dir):
689 file_path.rmtree(run_dir)
Lei Leife202df2019-06-11 17:33:34 +0000690 os.mkdir(run_dir, 0o700)
maruelcffa0542017-04-07 08:39:20 -0700691 else:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500692 run_dir = make_temp_dir(ISOLATED_RUN_DIR, data.root_dir)
maruel03e11842016-07-14 10:50:16 -0700693 # storage should be normally set but don't crash if it is not. This can happen
694 # as Swarming task can run without an isolate server.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500695 out_dir = make_temp_dir(
696 ISOLATED_OUT_DIR, data.root_dir) if data.storage else None
697 tmp_dir = make_temp_dir(ISOLATED_TMP_DIR, data.root_dir)
nodir55be77b2016-05-03 09:39:57 -0700698 cwd = run_dir
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500699 if data.relative_cwd:
700 cwd = os.path.normpath(os.path.join(cwd, data.relative_cwd))
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500701 command = data.command
nodir55be77b2016-05-03 09:39:57 -0700702 try:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500703 with data.install_packages_fn(run_dir) as cipd_info:
vadimsh232f5a82017-01-20 19:23:44 -0800704 if cipd_info:
705 result['stats']['cipd'] = cipd_info.stats
706 result['cipd_pins'] = cipd_info.pins
nodir90bc8dc2016-06-15 13:35:21 -0700707
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500708 if data.isolated_hash:
vadimsh232f5a82017-01-20 19:23:44 -0800709 isolated_stats = result['stats'].setdefault('isolated', {})
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000710 bundle, stats = fetch_and_map(
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500711 isolated_hash=data.isolated_hash,
712 storage=data.storage,
713 cache=data.isolate_cache,
vadimsh232f5a82017-01-20 19:23:44 -0800714 outdir=run_dir,
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500715 use_symlinks=data.use_symlinks)
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +0000716 isolated_stats['download'].update(stats)
vadimsh232f5a82017-01-20 19:23:44 -0800717 change_tree_read_only(run_dir, bundle.read_only)
maruelabec63c2017-04-26 11:53:24 -0700718 # Inject the command
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500719 if not command and bundle.command:
720 command = bundle.command + data.extra_args
Marc-Antoine Rueld704a1f2017-10-31 10:51:23 -0400721 # Only set the relative directory if the isolated file specified a
722 # command, and no raw command was specified.
723 if bundle.relative_cwd:
724 cwd = os.path.normpath(os.path.join(cwd, bundle.relative_cwd))
maruelabec63c2017-04-26 11:53:24 -0700725
726 if not command:
727 # Handle this as a task failure, not an internal failure.
728 sys.stderr.write(
729 '<No command was specified!>\n'
730 '<Please secify a command when triggering your Swarming task>\n')
731 result['exit_code'] = 1
732 return result
nodirbe642ff2016-06-09 15:51:51 -0700733
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500734 if not cwd.startswith(run_dir):
735 # Handle this as a task failure, not an internal failure. This is a
736 # 'last chance' way to gate against directory escape.
737 sys.stderr.write('<Relative CWD is outside of run directory!>\n')
738 result['exit_code'] = 1
739 return result
740
741 if not os.path.isdir(cwd):
742 # Accepts relative_cwd that does not exist.
Lei Leife202df2019-06-11 17:33:34 +0000743 os.makedirs(cwd, 0o700)
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -0500744
vadimsh232f5a82017-01-20 19:23:44 -0800745 # If we have an explicit list of files to return, make sure their
746 # directories exist now.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500747 if data.storage and data.outputs:
748 isolateserver.create_directories(run_dir, data.outputs)
aludwin0a8e17d2016-10-27 15:57:39 -0700749
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500750 with data.install_named_caches(run_dir):
nodird6160682017-02-02 13:03:35 -0800751 sys.stdout.flush()
752 start = time.time()
753 try:
vadimsh9c54b2c2017-07-25 14:08:29 -0700754 # Need to switch the default account before 'get_command_env' call,
755 # so it can grab correct value of LUCI_CONTEXT env var.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500756 with set_luci_context_account(data.switch_to_account, tmp_dir):
757 env = get_command_env(
Roberto Carrillo71ade6d2018-10-08 22:30:24 +0000758 tmp_dir, cipd_info, run_dir, data.env, data.env_prefix, out_dir,
759 data.bot_file)
Brian Sheedy7a761172019-08-30 22:55:14 +0000760 command = tools.find_executable(command, env)
Robert Iannucci24ae76a2018-02-26 12:51:18 -0800761 command = process_command(command, out_dir, data.bot_file)
762 file_path.ensure_command_has_abs_path(command, cwd)
763
vadimsh9c54b2c2017-07-25 14:08:29 -0700764 result['exit_code'], result['had_hard_timeout'] = run_command(
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +0000765 command, cwd, env, data.hard_timeout, data.grace_period,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +0000766 data.lower_priority, data.containment)
nodird6160682017-02-02 13:03:35 -0800767 finally:
768 result['duration'] = max(time.time() - start, 0)
Seth Koehler49139812017-12-19 13:59:33 -0500769
770 # We successfully ran the command, set internal_failure back to
771 # None (even if the command failed, it's not an internal error).
772 result['internal_failure'] = None
maruela9cfd6f2015-09-15 11:03:15 -0700773 except Exception as e:
nodir90bc8dc2016-06-15 13:35:21 -0700774 # An internal error occurred. Report accordingly so the swarming task will
775 # be retried automatically.
maruel12e30012015-10-09 11:55:35 -0700776 logging.exception('internal failure: %s', e)
maruela9cfd6f2015-09-15 11:03:15 -0700777 result['internal_failure'] = str(e)
778 on_error.report(None)
aludwin0a8e17d2016-10-27 15:57:39 -0700779
780 # Clean up
maruela9cfd6f2015-09-15 11:03:15 -0700781 finally:
782 try:
aludwin0a8e17d2016-10-27 15:57:39 -0700783 # Try to link files to the output directory, if specified.
784 if out_dir:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500785 link_outputs_to_outdir(run_dir, out_dir, data.outputs)
aludwin0a8e17d2016-10-27 15:57:39 -0700786
nodir32a1ec12016-10-26 18:34:07 -0700787 success = False
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500788 if data.leak_temp_dir:
nodir32a1ec12016-10-26 18:34:07 -0700789 success = True
maruela9cfd6f2015-09-15 11:03:15 -0700790 logging.warning(
791 'Deliberately leaking %s for later examination', run_dir)
marueleb5fbee2015-09-17 13:01:36 -0700792 else:
maruel84537cb2015-10-16 14:21:28 -0700793 # On Windows rmtree(run_dir) call above has a synchronization effect: it
794 # finishes only when all task child processes terminate (since a running
795 # process locks *.exe file). Examine out_dir only after that call
796 # completes (since child processes may write to out_dir too and we need
797 # to wait for them to finish).
798 if fs.isdir(run_dir):
799 try:
800 success = file_path.rmtree(run_dir)
801 except OSError as e:
Marc-Antoine Ruel44699b32018-09-24 23:31:50 +0000802 logging.error('rmtree(%r) failed: %s', run_dir, e)
maruel84537cb2015-10-16 14:21:28 -0700803 success = False
804 if not success:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500805 sys.stderr.write(OUTLIVING_ZOMBIE_MSG % ('run', data.grace_period))
maruel84537cb2015-10-16 14:21:28 -0700806 if result['exit_code'] == 0:
807 result['exit_code'] = 1
808 if fs.isdir(tmp_dir):
809 try:
810 success = file_path.rmtree(tmp_dir)
811 except OSError as e:
Marc-Antoine Ruel44699b32018-09-24 23:31:50 +0000812 logging.error('rmtree(%r) failed: %s', tmp_dir, e)
maruel84537cb2015-10-16 14:21:28 -0700813 success = False
814 if not success:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500815 sys.stderr.write(OUTLIVING_ZOMBIE_MSG % ('temp', data.grace_period))
maruel84537cb2015-10-16 14:21:28 -0700816 if result['exit_code'] == 0:
817 result['exit_code'] = 1
maruela9cfd6f2015-09-15 11:03:15 -0700818
marueleb5fbee2015-09-17 13:01:36 -0700819 # This deletes out_dir if leak_temp_dir is not set.
nodir9130f072016-05-27 13:59:08 -0700820 if out_dir:
nodir55715712016-06-03 12:28:19 -0700821 isolated_stats = result['stats'].setdefault('isolated', {})
822 result['outputs_ref'], success, isolated_stats['upload'] = (
Takuto Ikutab4aa8662019-09-17 05:54:36 +0000823 upload_then_delete(data.storage, out_dir, data.leak_temp_dir))
maruela9cfd6f2015-09-15 11:03:15 -0700824 if not success and result['exit_code'] == 0:
825 result['exit_code'] = 1
826 except Exception as e:
827 # Swallow any exception in the main finally clause.
nodir9130f072016-05-27 13:59:08 -0700828 if out_dir:
829 logging.exception('Leaking out_dir %s: %s', out_dir, e)
maruela9cfd6f2015-09-15 11:03:15 -0700830 result['internal_failure'] = str(e)
831 return result
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -0500832
833
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500834def run_tha_test(data, result_json):
nodir55be77b2016-05-03 09:39:57 -0700835 """Runs an executable and records execution metadata.
836
nodir55be77b2016-05-03 09:39:57 -0700837 If isolated_hash is specified, downloads the dependencies in the cache,
838 hardlinks them into a temporary directory and runs the command specified in
839 the .isolated.
Marc-Antoine Ruel2283ad12014-02-09 11:14:57 -0500840
841 A temporary directory is created to hold the output files. The content inside
842 this directory will be uploaded back to |storage| packaged as a .isolated
843 file.
844
845 Arguments:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500846 - data: TaskData instance.
847 - result_json: File path to dump result metadata into. If set, the process
848 exit code is always 0 unless an internal error occurred.
maruela9cfd6f2015-09-15 11:03:15 -0700849
850 Returns:
851 Process exit code that should be used.
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000852 """
maruela76b9ee2015-12-15 06:18:08 -0800853 if result_json:
854 # Write a json output file right away in case we get killed.
855 result = {
856 'exit_code': None,
857 'had_hard_timeout': False,
858 'internal_failure': 'Was terminated before completion',
859 'outputs_ref': None,
nodirbe642ff2016-06-09 15:51:51 -0700860 'version': 5,
maruela76b9ee2015-12-15 06:18:08 -0800861 }
862 tools.write_json(result_json, result, dense=True)
863
maruela9cfd6f2015-09-15 11:03:15 -0700864 # run_isolated exit code. Depends on if result_json is used or not.
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -0500865 result = map_and_run(data, True)
maruela9cfd6f2015-09-15 11:03:15 -0700866 logging.info('Result:\n%s', tools.format_json(result, dense=True))
bpastene3ae09522016-06-10 17:12:59 -0700867
maruela9cfd6f2015-09-15 11:03:15 -0700868 if result_json:
maruel05d5a882015-09-21 13:59:02 -0700869 # We've found tests to delete 'work' when quitting, causing an exception
870 # here. Try to recreate the directory if necessary.
nodire5028a92016-04-29 14:38:21 -0700871 file_path.ensure_tree(os.path.dirname(result_json))
maruela9cfd6f2015-09-15 11:03:15 -0700872 tools.write_json(result_json, result, dense=True)
873 # Only return 1 if there was an internal error.
874 return int(bool(result['internal_failure']))
maruel@chromium.org781ccf62013-09-17 19:39:47 +0000875
maruela9cfd6f2015-09-15 11:03:15 -0700876 # Marshall into old-style inline output.
877 if result['outputs_ref']:
Marc-Antoine Ruel793bff32019-04-18 17:50:48 +0000878 # pylint: disable=unsubscriptable-object
maruela9cfd6f2015-09-15 11:03:15 -0700879 data = {
880 'hash': result['outputs_ref']['isolated'],
881 'namespace': result['outputs_ref']['namespace'],
882 'storage': result['outputs_ref']['isolatedserver'],
883 }
Marc-Antoine Ruelc44f5722015-01-08 16:10:01 -0500884 sys.stdout.flush()
maruela9cfd6f2015-09-15 11:03:15 -0700885 print(
886 '[run_isolated_out_hack]%s[/run_isolated_out_hack]' %
887 tools.format_json(data, dense=True))
maruelb76604c2015-11-11 11:53:44 -0800888 sys.stdout.flush()
maruela9cfd6f2015-09-15 11:03:15 -0700889 return result['exit_code'] or int(bool(result['internal_failure']))
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +0000890
891
iannuccib58d10d2017-03-18 02:00:25 -0700892# Yielded by 'install_client_and_packages'.
vadimsh232f5a82017-01-20 19:23:44 -0800893CipdInfo = collections.namedtuple('CipdInfo', [
894 'client', # cipd.CipdClient object
895 'cache_dir', # absolute path to bot-global cipd tag and instance cache
896 'stats', # dict with stats to return to the server
897 'pins', # dict with installed cipd pins to return to the server
898])
899
900
901@contextlib.contextmanager
902def noop_install_packages(_run_dir):
iannuccib58d10d2017-03-18 02:00:25 -0700903 """Placeholder for 'install_client_and_packages' if cipd is disabled."""
vadimsh232f5a82017-01-20 19:23:44 -0800904 yield None
905
906
iannuccib58d10d2017-03-18 02:00:25 -0700907def _install_packages(run_dir, cipd_cache_dir, client, packages, timeout):
908 """Calls 'cipd ensure' for packages.
909
910 Args:
911 run_dir (str): root of installation.
912 cipd_cache_dir (str): the directory to use for the cipd package cache.
913 client (CipdClient): the cipd client to use
914 packages: packages to install, list [(path, package_name, version), ...].
915 timeout: max duration in seconds that this function can take.
916
917 Returns: list of pinned packages. Looks like [
918 {
919 'path': 'subdirectory',
920 'package_name': 'resolved/package/name',
921 'version': 'deadbeef...',
922 },
923 ...
924 ]
925 """
926 package_pins = [None]*len(packages)
927 def insert_pin(path, name, version, idx):
928 package_pins[idx] = {
929 'package_name': name,
930 # swarming deals with 'root' as '.'
931 'path': path or '.',
932 'version': version,
933 }
934
935 by_path = collections.defaultdict(list)
936 for i, (path, name, version) in enumerate(packages):
937 # cipd deals with 'root' as ''
938 if path == '.':
939 path = ''
940 by_path[path].append((name, version, i))
941
942 pins = client.ensure(
943 run_dir,
944 {
945 subdir: [(name, vers) for name, vers, _ in pkgs]
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000946 for subdir, pkgs in by_path.items()
iannuccib58d10d2017-03-18 02:00:25 -0700947 },
948 cache_dir=cipd_cache_dir,
949 timeout=timeout,
950 )
951
Marc-Antoine Ruel04903a32019-10-09 21:09:25 +0000952 for subdir, pin_list in sorted(pins.items()):
iannuccib58d10d2017-03-18 02:00:25 -0700953 this_subdir = by_path[subdir]
954 for i, (name, version) in enumerate(pin_list):
955 insert_pin(subdir, name, version, this_subdir[i][2])
956
Robert Iannucci461b30d2017-12-13 11:34:03 -0800957 assert None not in package_pins, (packages, pins, package_pins)
iannuccib58d10d2017-03-18 02:00:25 -0700958
959 return package_pins
960
961
vadimsh232f5a82017-01-20 19:23:44 -0800962@contextlib.contextmanager
iannuccib58d10d2017-03-18 02:00:25 -0700963def install_client_and_packages(
nodirff531b42016-06-23 13:05:06 -0700964 run_dir, packages, service_url, client_package_name,
vadimsh232f5a82017-01-20 19:23:44 -0800965 client_version, cache_dir, timeout=None):
vadimsh902948e2017-01-20 15:57:32 -0800966 """Bootstraps CIPD client and installs CIPD packages.
iannucci96fcccc2016-08-30 15:52:22 -0700967
vadimsh232f5a82017-01-20 19:23:44 -0800968 Yields CipdClient, stats, client info and pins (as single CipdInfo object).
969
970 Pins and the CIPD client info are in the form of:
iannucci96fcccc2016-08-30 15:52:22 -0700971 [
972 {
973 "path": path, "package_name": package_name, "version": version,
974 },
975 ...
976 ]
vadimsh902948e2017-01-20 15:57:32 -0800977 (the CIPD client info is a single dictionary instead of a list)
iannucci96fcccc2016-08-30 15:52:22 -0700978
979 such that they correspond 1:1 to all input package arguments from the command
980 line. These dictionaries make their all the way back to swarming, where they
981 become the arguments of CipdPackage.
nodirbe642ff2016-06-09 15:51:51 -0700982
vadimsh902948e2017-01-20 15:57:32 -0800983 If 'packages' list is empty, will bootstrap CIPD client, but won't install
984 any packages.
985
986 The bootstrapped client (regardless whether 'packages' list is empty or not),
vadimsh232f5a82017-01-20 19:23:44 -0800987 will be made available to the task via $PATH.
vadimsh902948e2017-01-20 15:57:32 -0800988
nodirbe642ff2016-06-09 15:51:51 -0700989 Args:
nodir90bc8dc2016-06-15 13:35:21 -0700990 run_dir (str): root of installation.
vadimsh902948e2017-01-20 15:57:32 -0800991 packages: packages to install, list [(path, package_name, version), ...].
nodirbe642ff2016-06-09 15:51:51 -0700992 service_url (str): CIPD server url, e.g.
993 "https://chrome-infra-packages.appspot.com."
nodir90bc8dc2016-06-15 13:35:21 -0700994 client_package_name (str): CIPD package name of CIPD client.
995 client_version (str): Version of CIPD client.
nodirbe642ff2016-06-09 15:51:51 -0700996 cache_dir (str): where to keep cache of cipd clients, packages and tags.
997 timeout: max duration in seconds that this function can take.
nodirbe642ff2016-06-09 15:51:51 -0700998 """
999 assert cache_dir
nodir90bc8dc2016-06-15 13:35:21 -07001000
nodirbe642ff2016-06-09 15:51:51 -07001001 timeoutfn = tools.sliding_timeout(timeout)
nodirbe642ff2016-06-09 15:51:51 -07001002 start = time.time()
nodirbe642ff2016-06-09 15:51:51 -07001003
vadimsh902948e2017-01-20 15:57:32 -08001004 cache_dir = os.path.abspath(cache_dir)
vadimsh232f5a82017-01-20 19:23:44 -08001005 cipd_cache_dir = os.path.join(cache_dir, 'cache') # tag and instance caches
nodir90bc8dc2016-06-15 13:35:21 -07001006 run_dir = os.path.abspath(run_dir)
vadimsh902948e2017-01-20 15:57:32 -08001007 packages = packages or []
nodir90bc8dc2016-06-15 13:35:21 -07001008
nodirbe642ff2016-06-09 15:51:51 -07001009 get_client_start = time.time()
1010 client_manager = cipd.get_client(
1011 service_url, client_package_name, client_version, cache_dir,
1012 timeout=timeoutfn())
iannucci96fcccc2016-08-30 15:52:22 -07001013
nodirbe642ff2016-06-09 15:51:51 -07001014 with client_manager as client:
1015 get_client_duration = time.time() - get_client_start
nodir90bc8dc2016-06-15 13:35:21 -07001016
iannuccib58d10d2017-03-18 02:00:25 -07001017 package_pins = []
1018 if packages:
1019 package_pins = _install_packages(
1020 run_dir, cipd_cache_dir, client, packages, timeoutfn())
1021
1022 file_path.make_tree_files_read_only(run_dir)
nodir90bc8dc2016-06-15 13:35:21 -07001023
vadimsh232f5a82017-01-20 19:23:44 -08001024 total_duration = time.time() - start
1025 logging.info(
1026 'Installing CIPD client and packages took %d seconds', total_duration)
nodir90bc8dc2016-06-15 13:35:21 -07001027
vadimsh232f5a82017-01-20 19:23:44 -08001028 yield CipdInfo(
1029 client=client,
1030 cache_dir=cipd_cache_dir,
1031 stats={
1032 'duration': total_duration,
1033 'get_client_duration': get_client_duration,
1034 },
1035 pins={
iannuccib58d10d2017-03-18 02:00:25 -07001036 'client_package': {
1037 'package_name': client.package_name,
1038 'version': client.instance_id,
1039 },
vadimsh232f5a82017-01-20 19:23:44 -08001040 'packages': package_pins,
1041 })
nodirbe642ff2016-06-09 15:51:51 -07001042
1043
1044def create_option_parser():
Marc-Antoine Ruelf74cffe2015-07-15 15:21:34 -04001045 parser = logging_utils.OptionParserWithLogging(
nodir55be77b2016-05-03 09:39:57 -07001046 usage='%prog <options> [command to run or extra args]',
maruel@chromium.orgdedbf492013-09-12 20:42:11 +00001047 version=__version__,
1048 log_file=RUN_ISOLATED_LOG_FILE)
maruela9cfd6f2015-09-15 11:03:15 -07001049 parser.add_option(
maruel36a963d2016-04-08 17:15:49 -07001050 '--clean', action='store_true',
1051 help='Cleans the cache, trimming it necessary and remove corrupted items '
1052 'and returns without executing anything; use with -v to know what '
1053 'was done')
1054 parser.add_option(
maruel4409e302016-07-19 14:25:51 -07001055 '--use-symlinks', action='store_true',
1056 help='Use symlinks instead of hardlinks')
1057 parser.add_option(
maruela9cfd6f2015-09-15 11:03:15 -07001058 '--json',
1059 help='dump output metadata to json file. When used, run_isolated returns '
1060 'non-zero only on internal failure')
maruel6be7f9e2015-10-01 12:25:30 -07001061 parser.add_option(
maruel5c9e47b2015-12-18 13:02:30 -08001062 '--hard-timeout', type='float', help='Enforce hard timeout in execution')
maruel6be7f9e2015-10-01 12:25:30 -07001063 parser.add_option(
maruel5c9e47b2015-12-18 13:02:30 -08001064 '--grace-period', type='float',
maruel6be7f9e2015-10-01 12:25:30 -07001065 help='Grace period between SIGTERM and SIGKILL')
bpastene3ae09522016-06-10 17:12:59 -07001066 parser.add_option(
Marc-Antoine Ruel49e347d2017-10-24 16:52:02 -07001067 '--raw-cmd', action='store_true',
1068 help='Ignore the isolated command, use the one supplied at the command '
1069 'line')
1070 parser.add_option(
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001071 '--relative-cwd',
1072 help='Ignore the isolated \'relative_cwd\' and use this one instead; '
1073 'requires --raw-cmd')
1074 parser.add_option(
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001075 '--env', default=[], action='append',
1076 help='Environment variables to set for the child process')
1077 parser.add_option(
1078 '--env-prefix', default=[], action='append',
Robert Iannuccibf5f84c2017-11-22 12:56:50 -08001079 help='Specify a VAR=./path/fragment to put in the environment variable '
1080 'before executing the command. The path fragment must be relative '
1081 'to the isolated run directory, and must not contain a `..` token. '
1082 'The path will be made absolute and prepended to the indicated '
1083 '$VAR using the OS\'s path separator. Multiple items for the same '
1084 '$VAR will be prepended in order.')
1085 parser.add_option(
bpastene3ae09522016-06-10 17:12:59 -07001086 '--bot-file',
1087 help='Path to a file describing the state of the host. The content is '
1088 'defined by on_before_task() in bot_config.')
aludwin7556e0c2016-10-26 08:46:10 -07001089 parser.add_option(
vadimsh9c54b2c2017-07-25 14:08:29 -07001090 '--switch-to-account',
1091 help='If given, switches LUCI_CONTEXT to given logical service account '
1092 '(e.g. "task" or "system") before launching the isolated process.')
1093 parser.add_option(
aludwin0a8e17d2016-10-27 15:57:39 -07001094 '--output', action='append',
1095 help='Specifies an output to return. If no outputs are specified, all '
1096 'files located in $(ISOLATED_OUTDIR) will be returned; '
1097 'otherwise, outputs in both $(ISOLATED_OUTDIR) and those '
1098 'specified by --output option (there can be multiple) will be '
1099 'returned. Note that if a file in OUT_DIR has the same path '
1100 'as an --output option, the --output version will be returned.')
1101 parser.add_option(
aludwin7556e0c2016-10-26 08:46:10 -07001102 '-a', '--argsfile',
1103 # This is actually handled in parse_args; it's included here purely so it
1104 # can make it into the help text.
1105 help='Specify a file containing a JSON array of arguments to this '
1106 'script. If --argsfile is provided, no other argument may be '
1107 'provided on the command line.')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001108
1109 group = optparse.OptionGroup(parser, 'Data source')
1110 group.add_option(
Marc-Antoine Ruel185ded42015-01-28 20:49:18 -05001111 '-s', '--isolated',
nodir55be77b2016-05-03 09:39:57 -07001112 help='Hash of the .isolated to grab from the isolate server.')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001113 isolateserver.add_isolate_server_options(group)
1114 parser.add_option_group(group)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001115
Marc-Antoine Ruela57d7db2014-10-15 20:31:19 -04001116 isolateserver.add_cache_options(parser)
nodirbe642ff2016-06-09 15:51:51 -07001117
1118 cipd.add_cipd_options(parser)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001119
1120 group = optparse.OptionGroup(parser, 'Named caches')
1121 group.add_option(
1122 '--named-cache',
1123 dest='named_caches',
1124 action='append',
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001125 nargs=3,
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001126 default=[],
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001127 help='A named cache to request. Accepts 3 arguments: name, path, hint. '
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001128 'name identifies the cache, must match regex [a-z0-9_]{1,4096}. '
1129 'path is a path relative to the run dir where the cache directory '
1130 'must be put to. '
1131 'This option can be specified more than once.')
1132 group.add_option(
1133 '--named-cache-root', default='named_caches',
1134 help='Cache root directory. Default=%default')
1135 parser.add_option_group(group)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001136
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001137 group = optparse.OptionGroup(parser, 'Process containment')
1138 parser.add_option(
1139 '--lower-priority', action='store_true',
1140 help='Lowers the child process priority')
1141 parser.add_option(
1142 '--containment-type', choices=('NONE', 'AUTO', 'JOB_OBJECT'),
1143 default='NONE',
1144 help='Type of container to use')
1145 parser.add_option(
1146 '--limit-processes', type='int', default=0,
1147 help='Maximum number of active processes in the containment')
1148 parser.add_option(
1149 '--limit-total-committed-memory', type='int', default=0,
1150 help='Maximum sum of committed memory in the containment')
1151 parser.add_option_group(group)
1152
1153 group = optparse.OptionGroup(parser, 'Debugging')
1154 group.add_option(
Kenneth Russell61d42352014-09-15 11:41:16 -07001155 '--leak-temp-dir',
1156 action='store_true',
nodirbe642ff2016-06-09 15:51:51 -07001157 help='Deliberately leak isolate\'s temp dir for later examination. '
1158 'Default: %default')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001159 group.add_option(
marueleb5fbee2015-09-17 13:01:36 -07001160 '--root-dir', help='Use a directory instead of a random one')
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001161 parser.add_option_group(group)
Kenneth Russell61d42352014-09-15 11:41:16 -07001162
Vadim Shtayurae34e13a2014-02-02 11:23:26 -08001163 auth.add_auth_options(parser)
nodirbe642ff2016-06-09 15:51:51 -07001164
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001165 parser.set_defaults(cache='cache', cipd_cache='cipd_cache')
nodirbe642ff2016-06-09 15:51:51 -07001166 return parser
1167
1168
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001169def process_named_cache_options(parser, options, time_fn=None):
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001170 """Validates named cache options and returns a CacheManager."""
1171 if options.named_caches and not options.named_cache_root:
1172 parser.error('--named-cache is specified, but --named-cache-root is empty')
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001173 for name, path, hint in options.named_caches:
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001174 if not CACHE_NAME_RE.match(name):
1175 parser.error(
1176 'cache name %r does not match %r' % (name, CACHE_NAME_RE.pattern))
1177 if not path:
1178 parser.error('cache path cannot be empty')
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001179 try:
1180 long(hint)
1181 except ValueError:
1182 parser.error('cache hint must be a number')
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001183 if options.named_cache_root:
1184 # Make these configurable later if there is use case but for now it's fairly
1185 # safe values.
1186 # In practice, a fair chunk of bots are already recycled on a daily schedule
1187 # so this code doesn't have any effect to them, unless they are preloaded
1188 # with a really old cache.
1189 policies = local_caching.CachePolicies(
1190 # 1TiB.
1191 max_cache_size=1024*1024*1024*1024,
1192 min_free_space=options.min_free_space,
1193 max_items=50,
Marc-Antoine Ruel5d7606b2018-06-15 19:06:12 +00001194 max_age_secs=MAX_AGE_SECS)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001195 root_dir = unicode(os.path.abspath(options.named_cache_root))
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001196 return local_caching.NamedCache(root_dir, policies, time_fn=time_fn)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001197 return None
1198
1199
aludwin7556e0c2016-10-26 08:46:10 -07001200def parse_args(args):
1201 # Create a fake mini-parser just to get out the "-a" command. Note that
1202 # it's not documented here; instead, it's documented in create_option_parser
1203 # even though that parser will never actually get to parse it. This is
1204 # because --argsfile is exclusive with all other options and arguments.
1205 file_argparse = argparse.ArgumentParser(add_help=False)
1206 file_argparse.add_argument('-a', '--argsfile')
1207 (file_args, nonfile_args) = file_argparse.parse_known_args(args)
1208 if file_args.argsfile:
1209 if nonfile_args:
1210 file_argparse.error('Can\'t specify --argsfile with'
1211 'any other arguments (%s)' % nonfile_args)
1212 try:
1213 with open(file_args.argsfile, 'r') as f:
1214 args = json.load(f)
1215 except (IOError, OSError, ValueError) as e:
1216 # We don't need to error out here - "args" is now empty,
1217 # so the call below to parser.parse_args(args) will fail
1218 # and print the full help text.
1219 print >> sys.stderr, 'Couldn\'t read arguments: %s' % e
1220
1221 # Even if we failed to read the args, just call the normal parser now since it
1222 # will print the correct help message.
nodirbe642ff2016-06-09 15:51:51 -07001223 parser = create_option_parser()
Marc-Antoine Ruel90c98162013-12-18 15:11:57 -05001224 options, args = parser.parse_args(args)
aludwin7556e0c2016-10-26 08:46:10 -07001225 return (parser, options, args)
1226
1227
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001228def _calc_named_cache_hint(named_cache, named_caches):
1229 """Returns the expected size of the missing named caches."""
1230 present = named_cache.available
1231 size = 0
1232 for name, _, hint in named_caches:
1233 if name not in present:
1234 hint = long(hint)
1235 if hint > 0:
1236 size += hint
1237 return size
1238
1239
aludwin7556e0c2016-10-26 08:46:10 -07001240def main(args):
Marc-Antoine Ruelee6ca622017-11-29 11:19:16 -05001241 # Warning: when --argsfile is used, the strings are unicode instances, when
1242 # parsed normally, the strings are str instances.
aludwin7556e0c2016-10-26 08:46:10 -07001243 (parser, options, args) = parse_args(args)
maruel36a963d2016-04-08 17:15:49 -07001244
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001245 if not file_path.enable_symlink():
Marc-Antoine Ruel5a024272019-01-15 20:11:16 +00001246 logging.warning('Symlink support is not enabled')
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001247
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001248 named_cache = process_named_cache_options(parser, options)
Marc-Antoine Ruel0d8b0f62018-09-10 14:40:35 +00001249 # hint is 0 if there's no named cache.
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001250 hint = _calc_named_cache_hint(named_cache, options.named_caches)
1251 if hint:
1252 # Increase the --min-free-space value by the hint, and recreate the
1253 # NamedCache instance so it gets the updated CachePolicy.
1254 options.min_free_space += hint
1255 named_cache = process_named_cache_options(parser, options)
1256
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001257 # TODO(maruel): CIPD caches should be defined at an higher level here too, so
1258 # they can be cleaned the same way.
nodirf33b8d62016-10-26 22:34:58 -07001259 isolate_cache = isolateserver.process_cache_options(options, trim=False)
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001260 caches = []
1261 if isolate_cache:
1262 caches.append(isolate_cache)
1263 if named_cache:
1264 caches.append(named_cache)
1265 root = caches[0].cache_dir if caches else unicode(os.getcwd())
maruel36a963d2016-04-08 17:15:49 -07001266 if options.clean:
1267 if options.isolated:
1268 parser.error('Can\'t use --isolated with --clean.')
1269 if options.isolate_server:
1270 parser.error('Can\'t use --isolate-server with --clean.')
1271 if options.json:
1272 parser.error('Can\'t use --json with --clean.')
nodirf33b8d62016-10-26 22:34:58 -07001273 if options.named_caches:
1274 parser.error('Can\t use --named-cache with --clean.')
Marc-Antoine Ruel7139d912018-06-15 20:04:42 +00001275 # Trim first, then clean.
1276 local_caching.trim_caches(
1277 caches,
1278 root,
1279 min_free_space=options.min_free_space,
1280 max_age_secs=MAX_AGE_SECS)
1281 for c in caches:
Marc-Antoine Ruel87fc2222018-06-18 13:09:24 +00001282 c.cleanup()
maruel36a963d2016-04-08 17:15:49 -07001283 return 0
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001284
1285 # Trim must still be done for the following case:
1286 # - named-cache was used
1287 # - some entries, with a large hint, where missing
1288 # - --min-free-space was increased accordingly, thus trimming is needed
1289 # Otherwise, this will have no effect, as bot_main calls run_isolated with
1290 # --clean after each task.
1291 if hint:
1292 logging.info('Additional trimming of %d bytes', hint)
Marc-Antoine Ruel0d8b0f62018-09-10 14:40:35 +00001293 local_caching.trim_caches(
1294 caches,
1295 root,
1296 min_free_space=options.min_free_space,
1297 max_age_secs=MAX_AGE_SECS)
maruel36a963d2016-04-08 17:15:49 -07001298
nodir55be77b2016-05-03 09:39:57 -07001299 if not options.isolated and not args:
1300 parser.error('--isolated or command to run is required.')
1301
Vadim Shtayura5d1efce2014-02-04 10:55:43 -08001302 auth.process_auth_options(parser, options)
nodir55be77b2016-05-03 09:39:57 -07001303
1304 isolateserver.process_isolate_server_options(
Marc-Antoine Ruel5028ba22017-08-25 17:37:51 -04001305 parser, options, True, False)
nodir55be77b2016-05-03 09:39:57 -07001306 if not options.isolate_server:
1307 if options.isolated:
1308 parser.error('--isolated requires --isolate-server')
1309 if ISOLATED_OUTDIR_PARAMETER in args:
1310 parser.error(
1311 '%s in args requires --isolate-server' % ISOLATED_OUTDIR_PARAMETER)
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001312
nodir90bc8dc2016-06-15 13:35:21 -07001313 if options.root_dir:
1314 options.root_dir = unicode(os.path.abspath(options.root_dir))
maruel12e30012015-10-09 11:55:35 -07001315 if options.json:
1316 options.json = unicode(os.path.abspath(options.json))
nodir55be77b2016-05-03 09:39:57 -07001317
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001318 if any('=' not in i for i in options.env):
1319 parser.error(
1320 '--env required key=value form. value can be skipped to delete '
1321 'the variable')
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -05001322 options.env = dict(i.split('=', 1) for i in options.env)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001323
1324 prefixes = {}
1325 cwd = os.path.realpath(os.getcwd())
1326 for item in options.env_prefix:
1327 if '=' not in item:
1328 parser.error(
1329 '--env-prefix %r is malformed, must be in the form `VAR=./path`'
1330 % item)
Marc-Antoine Ruel7a68f712017-12-01 18:45:18 -05001331 key, opath = item.split('=', 1)
Marc-Antoine Ruel19dd8872017-11-28 18:33:39 -05001332 if os.path.isabs(opath):
1333 parser.error('--env-prefix %r path is bad, must be relative.' % opath)
1334 opath = os.path.normpath(opath)
1335 if not os.path.realpath(os.path.join(cwd, opath)).startswith(cwd):
1336 parser.error(
1337 '--env-prefix %r path is bad, must be relative and not contain `..`.'
1338 % opath)
1339 prefixes.setdefault(key, []).append(opath)
1340 options.env_prefix = prefixes
Robert Iannuccibf5f84c2017-11-22 12:56:50 -08001341
nodirbe642ff2016-06-09 15:51:51 -07001342 cipd.validate_cipd_options(parser, options)
1343
vadimsh232f5a82017-01-20 19:23:44 -08001344 install_packages_fn = noop_install_packages
vadimsh902948e2017-01-20 15:57:32 -08001345 if options.cipd_enabled:
iannuccib58d10d2017-03-18 02:00:25 -07001346 install_packages_fn = lambda run_dir: install_client_and_packages(
vadimsh902948e2017-01-20 15:57:32 -08001347 run_dir, cipd.parse_package_args(options.cipd_packages),
1348 options.cipd_server, options.cipd_client_package,
1349 options.cipd_client_version, cache_dir=options.cipd_cache)
nodirbe642ff2016-06-09 15:51:51 -07001350
nodird6160682017-02-02 13:03:35 -08001351 @contextlib.contextmanager
nodir0ae98b32017-05-11 13:21:53 -07001352 def install_named_caches(run_dir):
nodird6160682017-02-02 13:03:35 -08001353 # WARNING: this function depends on "options" variable defined in the outer
1354 # function.
Marc-Antoine Ruel49f9f8d2018-05-24 15:57:06 -04001355 assert unicode(run_dir), repr(run_dir)
1356 assert os.path.isabs(run_dir), run_dir
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001357 named_caches = [
nodir0ae98b32017-05-11 13:21:53 -07001358 (os.path.join(run_dir, unicode(relpath)), name)
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001359 for name, relpath, _ in options.named_caches
nodir0ae98b32017-05-11 13:21:53 -07001360 ]
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001361 for path, name in named_caches:
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001362 named_cache.install(path, name)
nodird6160682017-02-02 13:03:35 -08001363 try:
1364 yield
1365 finally:
dnje289d132017-07-07 11:16:44 -07001366 # Uninstall each named cache, returning it to the cache pool. If an
1367 # uninstall fails for a given cache, it will remain in the task's
1368 # temporary space, get cleaned up by the Swarming bot, and be lost.
1369 #
1370 # If the Swarming bot cannot clean up the cache, it will handle it like
1371 # any other bot file that could not be removed.
Marc-Antoine Ruelc7a704b2018-08-29 19:02:23 +00001372 for path, name in reversed(named_caches):
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001373 try:
Marc-Antoine Ruele9558372018-08-03 03:41:22 +00001374 # uninstall() doesn't trim but does call save() implicitly. Trimming
1375 # *must* be done manually via periodic 'run_isolated.py --clean'.
Marc-Antoine Ruele79ddbf2018-06-13 18:33:07 +00001376 named_cache.uninstall(path, name)
1377 except local_caching.NamedCacheError:
1378 logging.exception('Error while removing named cache %r at %r. '
1379 'The cache will be lost.', path, name)
nodirf33b8d62016-10-26 22:34:58 -07001380
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001381 extra_args = []
1382 command = []
1383 if options.raw_cmd:
1384 command = args
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001385 if options.relative_cwd:
1386 a = os.path.normpath(os.path.abspath(options.relative_cwd))
1387 if not a.startswith(os.getcwd()):
1388 parser.error(
1389 '--relative-cwd must not try to escape the working directory')
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001390 else:
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001391 if options.relative_cwd:
1392 parser.error('--relative-cwd requires --raw-cmd')
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001393 extra_args = args
1394
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001395 containment_type = subprocess42.Containment.NONE
1396 if options.containment_type == 'AUTO':
1397 containment_type = subprocess42.Containment.AUTO
1398 if options.containment_type == 'JOB_OBJECT':
1399 containment_type = subprocess42.Containment.JOB_OBJECT
1400 containment = subprocess42.Containment(
1401 containment_type=containment_type,
1402 limit_processes=options.limit_processes,
1403 limit_total_committed_memory=options.limit_total_committed_memory)
1404
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001405 data = TaskData(
1406 command=command,
Marc-Antoine Ruel95068cf2017-12-07 21:35:05 -05001407 relative_cwd=options.relative_cwd,
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001408 extra_args=extra_args,
1409 isolated_hash=options.isolated,
1410 storage=None,
1411 isolate_cache=isolate_cache,
1412 outputs=options.output,
1413 install_named_caches=install_named_caches,
1414 leak_temp_dir=options.leak_temp_dir,
1415 root_dir=_to_unicode(options.root_dir),
1416 hard_timeout=options.hard_timeout,
1417 grace_period=options.grace_period,
1418 bot_file=options.bot_file,
1419 switch_to_account=options.switch_to_account,
1420 install_packages_fn=install_packages_fn,
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +00001421 use_symlinks=bool(options.use_symlinks),
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001422 env=options.env,
Marc-Antoine Ruel03c6fd12019-04-30 12:12:55 +00001423 env_prefix=options.env_prefix,
Marc-Antoine Ruel1b65f4e2019-05-02 21:56:58 +00001424 lower_priority=bool(options.lower_priority),
1425 containment=containment)
nodirbe642ff2016-06-09 15:51:51 -07001426 try:
nodir90bc8dc2016-06-15 13:35:21 -07001427 if options.isolate_server:
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001428 server_ref = isolate_storage.ServerRef(
nodir90bc8dc2016-06-15 13:35:21 -07001429 options.isolate_server, options.namespace)
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001430 storage = isolateserver.get_storage(server_ref)
nodir90bc8dc2016-06-15 13:35:21 -07001431 with storage:
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001432 data = data._replace(storage=storage)
nodirf33b8d62016-10-26 22:34:58 -07001433 # Hashing schemes used by |storage| and |isolate_cache| MUST match.
Marc-Antoine Ruelb8513132018-11-20 19:48:53 +00001434 assert storage.server_ref.hash_algo == server_ref.hash_algo
Marc-Antoine Ruel7de52592017-12-07 10:41:12 -05001435 return run_tha_test(data, options.json)
1436 return run_tha_test(data, options.json)
Marc-Antoine Ruel8b11dbd2018-05-18 14:31:22 -04001437 except (
1438 cipd.Error,
1439 local_caching.NamedCacheError,
Marc-Antoine Ruelb6e9e232018-11-20 00:12:33 +00001440 local_caching.NoMoreSpace) as ex:
nodirbe642ff2016-06-09 15:51:51 -07001441 print >> sys.stderr, ex.message
1442 return 1
maruel@chromium.org9c72d4e2012-09-28 19:20:25 +00001443
1444
1445if __name__ == '__main__':
maruel8e4e40c2016-05-30 06:21:07 -07001446 subprocess42.inhibit_os_error_reporting()
csharp@chromium.orgbfb98742013-03-26 20:28:36 +00001447 # Ensure that we are always running with the correct encoding.
vadimsh@chromium.orga4326472013-08-24 02:05:41 +00001448 fix_encoding.fix_encoding()
Marc-Antoine Ruel90c98162013-12-18 15:11:57 -05001449 sys.exit(main(sys.argv[1:]))