J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 1 | #!/usr/bin/env python |
| 2 | # Copyright 2015 The Chromium OS Authors. All rights reserved. |
| 3 | # Use of this source code is governed by a BSD-style license that can be |
| 4 | # found in the LICENSE file. |
| 5 | |
| 6 | """Create e-mail reports of the Lab's DUT inventory. |
| 7 | |
| 8 | Gathers a list of all DUTs of interest in the Lab, segregated by |
| 9 | board and pool, and determines whether each DUT is working or |
| 10 | broken. Then, send one or more e-mail reports summarizing the |
| 11 | status to e-mail addresses provided on the command line. |
| 12 | |
| 13 | usage: lab_inventory.py [ options ] [ board ... ] |
| 14 | |
| 15 | Options: |
| 16 | --duration / -d <hours> |
| 17 | How far back in time to search job history to determine DUT |
| 18 | status. |
| 19 | |
| 20 | --board-notify <address>[,<address>] |
| 21 | Send the "board status" e-mail to all the specified e-mail |
| 22 | addresses. |
| 23 | |
| 24 | --pool-notify <address>[,<address>] |
| 25 | Send the "pool status" e-mail to all the specified e-mail |
| 26 | addresses. |
| 27 | |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 28 | --recommend <number> |
| 29 | When generating the "board status" e-mail, included a list of |
| 30 | <number> specific DUTs to be recommended for repair. |
| 31 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 32 | --logdir <directory> |
| 33 | Log progress and actions in a file under this directory. Text |
| 34 | of any e-mail sent will also be logged in a timestamped file in |
| 35 | this directory. |
| 36 | |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 37 | --debug |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 38 | Suppress all logging and sending e-mail. Instead, write the |
| 39 | output that would be generated onto stdout. |
| 40 | |
| 41 | <board> arguments: |
| 42 | With no arguments, gathers the status for all boards in the lab. |
| 43 | With one or more named boards on the command line, restricts |
| 44 | reporting to just those boards. |
| 45 | |
| 46 | """ |
| 47 | |
| 48 | |
| 49 | import argparse |
| 50 | import logging |
| 51 | import logging.handlers |
| 52 | import os |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 53 | import re |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 54 | import sys |
| 55 | import time |
| 56 | |
| 57 | import common |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 58 | from autotest_lib.client.bin import utils |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 59 | from autotest_lib.client.common_lib import time_utils |
J. Richard Barnette | a7c514e | 2015-09-15 11:13:23 -0700 | [diff] [blame] | 60 | from autotest_lib.server.cros.dynamic_suite import frontend_wrappers |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 61 | from autotest_lib.server.hosts import servo_host |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 62 | from autotest_lib.site_utils import gmail_lib |
| 63 | from autotest_lib.site_utils import status_history |
| 64 | from autotest_lib.site_utils.suite_scheduler import constants |
| 65 | |
| 66 | |
| 67 | # The pools in the Lab that are actually of interest. |
| 68 | # |
| 69 | # These are general purpose pools of DUTs that are considered |
| 70 | # identical for purposes of testing. That is, a device in one of |
| 71 | # these pools can be shifted to another pool at will for purposes |
| 72 | # of supplying test demand. |
| 73 | # |
| 74 | # Devices in these pools are not allowed to have special-purpose |
| 75 | # attachments, or to be part of in any kind of custom fixture. |
| 76 | # Devices in these pools are also required to reside in areas |
| 77 | # managed by the Platforms team (i.e. at the time of this writing, |
| 78 | # only in "Atlantis" or "Destiny"). |
| 79 | # |
| 80 | # _CRITICAL_POOLS - Pools that must be kept fully supplied in order |
| 81 | # to guarantee timely completion of tests from builders. |
| 82 | # _SPARE_POOL - A low priority pool that is allowed to provide |
| 83 | # spares to replace broken devices in the critical pools. |
| 84 | # _MANAGED_POOLS - The set of all the general purpose pools |
| 85 | # monitored by this script. |
| 86 | |
J. Richard Barnette | 222d7f4 | 2015-12-14 17:22:51 -0800 | [diff] [blame^] | 87 | _CRITICAL_POOLS = ['bvt', 'cq', 'continuous'] |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 88 | _SPARE_POOL = 'suites' |
| 89 | _MANAGED_POOLS = _CRITICAL_POOLS + [_SPARE_POOL] |
| 90 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 91 | # _DEFAULT_DURATION: |
| 92 | # Default value used for the --duration command line option. |
| 93 | # Specifies how far back in time to search in order to determine |
| 94 | # DUT status. |
| 95 | |
| 96 | _DEFAULT_DURATION = 24 |
| 97 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 98 | # _LOGDIR: |
| 99 | # Relative path used in the calculation of the default setting |
| 100 | # for the --logdir option. The full path path is relative to |
| 101 | # the root of the autotest directory, as determined from |
| 102 | # sys.argv[0]. |
| 103 | # _LOGFILE: |
| 104 | # Basename of a file to which general log information will be |
| 105 | # written. |
| 106 | # _LOG_FORMAT: |
| 107 | # Format string for log messages. |
| 108 | |
| 109 | _LOGDIR = os.path.join('logs', 'dut-data') |
| 110 | _LOGFILE = 'lab-inventory.log' |
| 111 | _LOG_FORMAT = '%(asctime)s | %(levelname)-10s | %(message)s' |
| 112 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 113 | # Pattern describing location-based host names in the Chrome OS test |
| 114 | # labs. Each DUT hostname designates the DUT's location: |
| 115 | # * A lab (room) that's physically separated from other labs |
| 116 | # (i.e. there's a door). |
| 117 | # * A row (or aisle) of DUTs within the lab. |
| 118 | # * A vertical rack of shelves on the row. |
| 119 | # * A specific host on one shelf of the rack. |
| 120 | |
| 121 | _HOSTNAME_PATTERN = re.compile( |
| 122 | r'(chromeos\d+)-row(\d+)-rack(\d+)-host(\d+)') |
| 123 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 124 | |
| 125 | class _PoolCounts(object): |
| 126 | """Maintains a set of `HostJobHistory` objects for a pool. |
| 127 | |
| 128 | The collected history objects are nominally all part of a single |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 129 | scheduling pool of DUTs. The collection maintains a list of |
| 130 | working DUTs, a list of broken DUTs, and a list of all DUTs. |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 131 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 132 | Performance note: Certain methods in this class are potentially |
| 133 | expensive: |
| 134 | * `get_working()` |
| 135 | * `get_working_list()` |
| 136 | * `get_broken()` |
| 137 | * `get_broken_list()` |
| 138 | The first time any one of these methods is called, it causes |
| 139 | multiple RPC calls with a relatively expensive set of database |
| 140 | queries. However, the results of the queries are cached in the |
| 141 | individual `HostJobHistory` objects, so only the first call |
| 142 | actually pays the full cost. |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 143 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 144 | Additionally, `get_working_list()` and `get_broken_list()` both |
| 145 | cache their return values to avoid recalculating lists at every |
| 146 | call; this caching is separate from the caching of RPC results |
| 147 | described above. |
| 148 | |
| 149 | This class is deliberately constructed to delay the RPC cost |
| 150 | until the accessor methods are called (rather than to query in |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 151 | `record_host()`) so that it's possible to construct a complete |
| 152 | `_LabInventory` without making the expensive queries at creation |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 153 | time. `_populate_board_counts()`, below, assumes this behavior. |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 154 | |
| 155 | """ |
| 156 | |
| 157 | def __init__(self): |
| 158 | self._histories = [] |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 159 | self._working_list = None |
| 160 | self._broken_list = None |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 161 | |
| 162 | |
| 163 | def record_host(self, host_history): |
| 164 | """Add one `HostJobHistory` object to the collection. |
| 165 | |
| 166 | @param host_history The `HostJobHistory` object to be |
| 167 | remembered. |
| 168 | |
| 169 | """ |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 170 | self._working_list = None |
| 171 | self._broken_list = None |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 172 | self._histories.append(host_history) |
| 173 | |
| 174 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 175 | def get_working_list(self): |
| 176 | """Return a list of all working DUTs in the pool. |
| 177 | |
| 178 | Filter `self._histories` for histories where the last |
| 179 | diagnosis is `WORKING`. |
| 180 | |
| 181 | Cache the result so that we only cacluate it once. |
| 182 | |
| 183 | @return A list of HostJobHistory objects. |
| 184 | |
| 185 | """ |
| 186 | if self._working_list is None: |
| 187 | self._working_list = [h for h in self._histories |
| 188 | if h.last_diagnosis()[0] == status_history.WORKING] |
| 189 | return self._working_list |
| 190 | |
| 191 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 192 | def get_working(self): |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 193 | """Return the number of working DUTs in the pool.""" |
| 194 | return len(self.get_working_list()) |
| 195 | |
| 196 | |
| 197 | def get_broken_list(self): |
| 198 | """Return a list of all broken DUTs in the pool. |
| 199 | |
| 200 | Filter `self._histories` for histories where the last |
| 201 | diagnosis is not `WORKING`. |
| 202 | |
| 203 | Cache the result so that we only cacluate it once. |
| 204 | |
| 205 | @return A list of HostJobHistory objects. |
| 206 | |
| 207 | """ |
| 208 | if self._broken_list is None: |
| 209 | self._broken_list = [h for h in self._histories |
| 210 | if h.last_diagnosis()[0] != status_history.WORKING] |
| 211 | return self._broken_list |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 212 | |
| 213 | |
| 214 | def get_broken(self): |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 215 | """Return the number of broken DUTs in the pool.""" |
| 216 | return len(self.get_broken_list()) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 217 | |
| 218 | |
| 219 | def get_total(self): |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 220 | """Return the total number of DUTs in the pool.""" |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 221 | return len(self._histories) |
| 222 | |
| 223 | |
| 224 | class _BoardCounts(object): |
| 225 | """Maintains a set of `HostJobHistory` objects for a board. |
| 226 | |
| 227 | The collected history objects are nominally all of the same |
| 228 | board. The collection maintains a count of working DUTs, a |
| 229 | count of broken DUTs, and a total count. The counts can be |
| 230 | obtained either for a single pool, or as a total across all |
| 231 | pools. |
| 232 | |
| 233 | DUTs in the collection must be assigned to one of the pools |
| 234 | in `_MANAGED_POOLS`. |
| 235 | |
| 236 | The `get_working()` and `get_broken()` methods rely on the |
| 237 | methods of the same name in _PoolCounts, so the performance |
| 238 | note in _PoolCounts applies here as well. |
| 239 | |
| 240 | """ |
| 241 | |
| 242 | def __init__(self): |
| 243 | self._pools = { |
| 244 | pool: _PoolCounts() for pool in _MANAGED_POOLS |
| 245 | } |
| 246 | |
| 247 | def record_host(self, host_history): |
| 248 | """Add one `HostJobHistory` object to the collection. |
| 249 | |
| 250 | @param host_history The `HostJobHistory` object to be |
| 251 | remembered. |
| 252 | |
| 253 | """ |
J. Richard Barnette | 3d0590a | 2015-04-29 12:56:12 -0700 | [diff] [blame] | 254 | pool = host_history.host_pool |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 255 | self._pools[pool].record_host(host_history) |
| 256 | |
| 257 | |
| 258 | def _count_pool(self, get_pool_count, pool=None): |
| 259 | """Internal helper to count hosts in a given pool. |
| 260 | |
| 261 | The `get_pool_count` parameter is a function to calculate |
| 262 | the exact count of interest for the pool. |
| 263 | |
| 264 | @param get_pool_count Function to return a count from a |
| 265 | _PoolCount object. |
| 266 | @param pool The pool to be counted. If `None`, |
| 267 | return the total across all pools. |
| 268 | |
| 269 | """ |
| 270 | if pool is None: |
| 271 | return sum([get_pool_count(counts) |
| 272 | for counts in self._pools.values()]) |
| 273 | else: |
| 274 | return get_pool_count(self._pools[pool]) |
| 275 | |
| 276 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 277 | def get_working_list(self): |
| 278 | """Return a list of all working DUTs for the board. |
| 279 | |
| 280 | Go through all HostJobHistory objects in the board's pools, |
| 281 | selecting the ones where the last diagnosis is `WORKING`. |
| 282 | |
| 283 | @return A list of HostJobHistory objects. |
| 284 | |
| 285 | """ |
| 286 | l = [] |
| 287 | for p in self._pools.values(): |
| 288 | l.extend(p.get_working_list()) |
| 289 | return l |
| 290 | |
| 291 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 292 | def get_working(self, pool=None): |
| 293 | """Return the number of working DUTs in a pool. |
| 294 | |
| 295 | @param pool The pool to be counted. If `None`, return the |
| 296 | total across all pools. |
| 297 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 298 | @return The total number of working DUTs in the selected |
| 299 | pool(s). |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 300 | """ |
| 301 | return self._count_pool(_PoolCounts.get_working, pool) |
| 302 | |
| 303 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 304 | def get_broken_list(self): |
| 305 | """Return a list of all broken DUTs for the board. |
| 306 | |
| 307 | Go through all HostJobHistory objects in the board's pools, |
| 308 | selecting the ones where the last diagnosis is not |
| 309 | `WORKING`. |
| 310 | |
| 311 | @return A list of HostJobHistory objects. |
| 312 | |
| 313 | """ |
| 314 | l = [] |
| 315 | for p in self._pools.values(): |
| 316 | l.extend(p.get_broken_list()) |
| 317 | return l |
| 318 | |
| 319 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 320 | def get_broken(self, pool=None): |
| 321 | """Return the number of broken DUTs in a pool. |
| 322 | |
| 323 | @param pool The pool to be counted. If `None`, return the |
| 324 | total across all pools. |
| 325 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 326 | @return The total number of broken DUTs in the selected pool(s). |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 327 | """ |
| 328 | return self._count_pool(_PoolCounts.get_broken, pool) |
| 329 | |
| 330 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 331 | def get_spares_buffer(self): |
| 332 | """Return the the nominal number of working spares. |
| 333 | |
| 334 | Calculates and returns how many working spares there would |
| 335 | be in the spares pool if all broken DUTs were in the spares |
| 336 | pool. This number may be negative, indicating a shortfall |
| 337 | in the critical pools. |
| 338 | |
| 339 | @return The total number DUTs in the spares pool, less the total |
| 340 | number of broken DUTs in all pools. |
| 341 | """ |
| 342 | return self.get_total(_SPARE_POOL) - self.get_broken() |
| 343 | |
| 344 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 345 | def get_total(self, pool=None): |
| 346 | """Return the total number of DUTs in a pool. |
| 347 | |
| 348 | @param pool The pool to be counted. If `None`, return the |
| 349 | total across all pools. |
| 350 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 351 | @return The total number of DUTs in the selected pool(s). |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 352 | """ |
| 353 | return self._count_pool(_PoolCounts.get_total, pool) |
| 354 | |
| 355 | |
| 356 | class _LabInventory(dict): |
| 357 | """Collection of `HostJobHistory` objects for the Lab's inventory. |
| 358 | |
| 359 | The collection is indexed by board. Indexing returns the |
| 360 | _BoardCounts object associated with the board. |
| 361 | |
| 362 | The collection is also iterable. The iterator returns all the |
| 363 | boards in the inventory, in unspecified order. |
| 364 | |
| 365 | """ |
| 366 | |
| 367 | @classmethod |
| 368 | def create_inventory(cls, afe, start_time, end_time, boardlist=[]): |
| 369 | """Return a Lab inventory with specified parameters. |
| 370 | |
| 371 | By default, gathers inventory from `HostJobHistory` objects |
| 372 | for all DUTs in the `_MANAGED_POOLS` list. If `boardlist` |
| 373 | is supplied, the inventory will be restricted to only the |
| 374 | given boards. |
| 375 | |
| 376 | @param afe AFE object for constructing the |
| 377 | `HostJobHistory` objects. |
| 378 | @param start_time Start time for the `HostJobHistory` |
| 379 | objects. |
| 380 | @param end_time End time for the `HostJobHistory` |
| 381 | objects. |
| 382 | @param boardlist List of boards to include. If empty, |
| 383 | include all available boards. |
| 384 | @return A `_LabInventory` object for the specified boards. |
| 385 | |
| 386 | """ |
| 387 | label_list = [constants.Labels.POOL_PREFIX + l |
| 388 | for l in _MANAGED_POOLS] |
| 389 | afehosts = afe.get_hosts(labels__name__in=label_list) |
| 390 | if boardlist: |
| 391 | boardhosts = [] |
| 392 | for board in boardlist: |
| 393 | board_label = constants.Labels.BOARD_PREFIX + board |
| 394 | host_list = [h for h in afehosts |
| 395 | if board_label in h.labels] |
| 396 | boardhosts.extend(host_list) |
| 397 | afehosts = boardhosts |
| 398 | create = lambda host: ( |
| 399 | status_history.HostJobHistory(afe, host, |
| 400 | start_time, end_time)) |
| 401 | return cls([create(host) for host in afehosts]) |
| 402 | |
| 403 | |
| 404 | def __init__(self, histories): |
J. Richard Barnette | 6948ed3 | 2015-05-06 08:57:10 -0700 | [diff] [blame] | 405 | # N.B. The query that finds our hosts is restricted to those |
| 406 | # with a valid pool: label, but doesn't check for a valid |
| 407 | # board: label. In some (insufficiently) rare cases, the |
| 408 | # AFE hosts table has been known to (incorrectly) have DUTs |
| 409 | # with a pool: but no board: label. We explicitly exclude |
| 410 | # those here. |
| 411 | histories = [h for h in histories |
| 412 | if h.host_board is not None] |
J. Richard Barnette | 3d0590a | 2015-04-29 12:56:12 -0700 | [diff] [blame] | 413 | boards = set([h.host_board for h in histories]) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 414 | initval = { board: _BoardCounts() for board in boards } |
| 415 | super(_LabInventory, self).__init__(initval) |
| 416 | self._dut_count = len(histories) |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 417 | self._managed_boards = None |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 418 | for h in histories: |
J. Richard Barnette | 3d0590a | 2015-04-29 12:56:12 -0700 | [diff] [blame] | 419 | self[h.host_board].record_host(h) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 420 | |
| 421 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 422 | def get_managed_boards(self): |
| 423 | """Return the set of "managed" boards. |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 424 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 425 | Operationally, saying a board is "managed" means that the |
| 426 | board will be included in the "board" and "repair |
| 427 | recommendations" reports. That is, if there are failures in |
| 428 | the board's inventory then lab techs will be asked to fix |
| 429 | them without a separate ticket. |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 430 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 431 | For purposes of implementation, a board is "managed" if it |
| 432 | has DUTs in both the spare and a non-spare (i.e. critical) |
| 433 | pool. |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 434 | |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 435 | @return A set of all the boards that have both spare and |
| 436 | non-spare pools. |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 437 | """ |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 438 | if self._managed_boards is None: |
| 439 | self._managed_boards = set() |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 440 | for board, counts in self.items(): |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 441 | spares = counts.get_total(_SPARE_POOL) |
| 442 | total = counts.get_total() |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 443 | if spares != 0 and spares != total: |
| 444 | self._managed_boards.add(board) |
| 445 | return self._managed_boards |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 446 | |
| 447 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 448 | def get_num_duts(self): |
| 449 | """Return the total number of DUTs in the inventory.""" |
| 450 | return self._dut_count |
| 451 | |
| 452 | |
| 453 | def get_num_boards(self): |
| 454 | """Return the total number of boards in the inventory.""" |
| 455 | return len(self) |
| 456 | |
| 457 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 458 | def _sort_by_location(inventory_list): |
| 459 | """Return a list of DUTs, organized by location. |
| 460 | |
| 461 | Take the given list of `HostJobHistory` objects, separate it |
| 462 | into a list per lab, and sort each lab's list by location. The |
| 463 | order of sorting within a lab is |
| 464 | * By row number within the lab, |
| 465 | * then by rack number within the row, |
| 466 | * then by host shelf number within the rack. |
| 467 | |
| 468 | Return a list of the sorted lists. |
| 469 | |
| 470 | Implementation note: host locations are sorted by converting |
| 471 | each location into a base 100 number. If row, rack or |
| 472 | host numbers exceed the range [0..99], then sorting will |
| 473 | break down. |
| 474 | |
| 475 | @return A list of sorted lists of DUTs. |
| 476 | |
| 477 | """ |
| 478 | BASE = 100 |
| 479 | lab_lists = {} |
| 480 | for history in inventory_list: |
| 481 | location = _HOSTNAME_PATTERN.match(history.host.hostname) |
| 482 | if location: |
| 483 | lab = location.group(1) |
| 484 | key = 0 |
| 485 | for idx in location.group(2, 3, 4): |
| 486 | key = BASE * key + int(idx) |
| 487 | lab_lists.setdefault(lab, []).append((key, history)) |
| 488 | return_list = [] |
| 489 | for dut_list in lab_lists.values(): |
| 490 | dut_list.sort(key=lambda t: t[0]) |
| 491 | return_list.append([t[1] for t in dut_list]) |
| 492 | return return_list |
| 493 | |
| 494 | |
| 495 | def _score_repair_set(buffer_counts, repair_list): |
| 496 | """Return a numeric score rating a set of DUTs to be repaired. |
| 497 | |
| 498 | `buffer_counts` is a dictionary mapping board names to the |
| 499 | size of the board's spares buffer. |
| 500 | |
| 501 | `repair_list` is a list of DUTs to be repaired. |
| 502 | |
| 503 | This function calculates the new set of buffer counts that would |
| 504 | result from the proposed repairs, and scores the new set using |
| 505 | two numbers: |
| 506 | * Worst case buffer count for any board (higher is better). |
| 507 | This is the more siginficant number for comparison. |
| 508 | * Number of boards at the worst case (lower is better). This |
| 509 | is the less significant number. |
| 510 | |
| 511 | Implementation note: The score could fail to reflect the |
| 512 | intended criteria if there are more than 1000 boards in the |
| 513 | inventory. |
| 514 | |
| 515 | @param spare_counts A dictionary mapping boards to buffer counts. |
| 516 | @param repair_list A list of boards to be repaired. |
| 517 | @return A numeric score. |
| 518 | |
| 519 | """ |
| 520 | # Go through `buffer_counts`, and create a list of new counts |
| 521 | # that records the buffer count for each board after repair. |
| 522 | # The new list of counts discards the board names, as they don't |
| 523 | # contribute to the final score. |
| 524 | _NBOARDS = 1000 |
| 525 | repair_inventory = _LabInventory(repair_list) |
| 526 | new_counts = [] |
| 527 | for b, c in buffer_counts.items(): |
| 528 | if b in repair_inventory: |
| 529 | newcount = repair_inventory[b].get_total() |
| 530 | else: |
| 531 | newcount = 0 |
| 532 | new_counts.append(c + newcount) |
| 533 | # Go through the new list of counts. Find the worst available |
| 534 | # spares count, and count how many times that worst case occurs. |
| 535 | worst_count = new_counts[0] |
| 536 | num_worst = 1 |
| 537 | for c in new_counts[1:]: |
| 538 | if c == worst_count: |
| 539 | num_worst += 1 |
| 540 | elif c < worst_count: |
| 541 | worst_count = c |
| 542 | num_worst = 1 |
| 543 | # Return the calculated score |
| 544 | return _NBOARDS * worst_count - num_worst |
| 545 | |
| 546 | |
| 547 | def _generate_repair_recommendation(inventory, num_recommend): |
| 548 | """Return a summary of selected DUTs needing repair. |
| 549 | |
| 550 | Returns a message recommending a list of broken DUTs to be |
| 551 | repaired. The list of DUTs is selected based on these |
| 552 | criteria: |
| 553 | * No more than `num_recommend` DUTs will be listed. |
| 554 | * All DUTs must be in the same lab. |
| 555 | * DUTs should be selected for some degree of physical |
| 556 | proximity. |
| 557 | * DUTs for boards with a low spares buffer are more important |
| 558 | than DUTs with larger buffers. |
| 559 | |
| 560 | The algorithm used will guarantee that at least one DUT from a |
| 561 | board with the smallest spares buffer will be recommended. If |
| 562 | the worst spares buffer number is shared by more than one board, |
| 563 | the algorithm will tend to prefer repair sets that include more |
| 564 | of those boards over sets that cover fewer boards. |
| 565 | |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 566 | @param inventory Inventory for generating recommendations. |
| 567 | @param num_recommend Number of DUTs to recommend for repair. |
| 568 | |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 569 | """ |
| 570 | logging.debug('Creating DUT repair recommendations') |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 571 | board_buffer_counts = {} |
| 572 | broken_list = [] |
| 573 | for board in inventory.get_managed_boards(): |
| 574 | logging.debug('Listing failed DUTs for %s', board) |
| 575 | counts = inventory[board] |
| 576 | if counts.get_broken() != 0: |
| 577 | board_buffer_counts[board] = counts.get_spares_buffer() |
| 578 | broken_list.extend(counts.get_broken_list()) |
J. Richard Barnette | 5512743 | 2015-10-13 17:01:56 -0700 | [diff] [blame] | 579 | # N.B. The logic inside this loop may seem complicated, but |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 580 | # simplification is hard: |
| 581 | # * Calculating an initial recommendation outside of |
| 582 | # the loop likely would make things more complicated, |
| 583 | # not less. |
| 584 | # * It's necessary to calculate an initial lab slice once per |
| 585 | # lab _before_ the while loop, in case the number of broken |
| 586 | # DUTs in a lab is less than `num_recommend`. |
J. Richard Barnette | 5512743 | 2015-10-13 17:01:56 -0700 | [diff] [blame] | 587 | recommendation = None |
| 588 | best_score = None |
| 589 | for lab_duts in _sort_by_location(broken_list): |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 590 | start = 0 |
| 591 | end = num_recommend |
| 592 | lab_slice = lab_duts[start : end] |
| 593 | lab_score = _score_repair_set(board_buffer_counts, |
| 594 | lab_slice) |
| 595 | while end < len(lab_duts): |
| 596 | start += 1 |
| 597 | end += 1 |
| 598 | new_slice = lab_duts[start : end] |
| 599 | new_score = _score_repair_set(board_buffer_counts, |
| 600 | new_slice) |
| 601 | if new_score > lab_score: |
| 602 | lab_slice = new_slice |
| 603 | lab_score = new_score |
| 604 | if recommendation is None or lab_score > best_score: |
| 605 | recommendation = lab_slice |
| 606 | best_score = lab_score |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 607 | message = ['Repair recommendations:\n', |
| 608 | '%-30s %-16s %s' % ( |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 609 | 'Hostname', 'Board', 'Servo instructions')] |
| 610 | for h in recommendation: |
| 611 | servo_name = servo_host.make_servo_hostname(h.host.hostname) |
| 612 | if utils.host_is_in_lab_zone(servo_name): |
| 613 | servo_message = 'Repair servo first' |
| 614 | else: |
| 615 | servo_message = 'No servo present' |
| 616 | line = '%-30s %-16s %s' % ( |
| 617 | h.host.hostname, h.host_board, servo_message) |
| 618 | message.append(line) |
| 619 | return '\n'.join(message) |
| 620 | |
| 621 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 622 | def _generate_board_inventory_message(inventory): |
| 623 | """Generate the "board inventory" e-mail message. |
| 624 | |
| 625 | The board inventory is a list by board summarizing the number |
| 626 | of working and broken DUTs, and the total shortfall or surplus |
| 627 | of working devices relative to the minimum critical pool |
| 628 | requirement. |
| 629 | |
| 630 | The report omits boards with no DUTs in the spare pool or with |
| 631 | no DUTs in a critical pool. |
| 632 | |
| 633 | N.B. For sample output text formattted as users can expect to |
| 634 | see it in e-mail and log files, refer to the unit tests. |
| 635 | |
| 636 | @param inventory _LabInventory object with the inventory to |
| 637 | be reported on. |
| 638 | @return String with the inventory message to be sent. |
| 639 | |
| 640 | """ |
| 641 | logging.debug('Creating board inventory') |
J. Richard Barnette | d3ba33a | 2015-10-14 11:20:49 -0700 | [diff] [blame] | 642 | nworking = 0 |
| 643 | nbroken = 0 |
| 644 | nbroken_boards = 0 |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 645 | summaries = [] |
| 646 | for board in inventory.get_managed_boards(): |
| 647 | logging.debug('Counting board inventory for %s', board) |
| 648 | counts = inventory[board] |
| 649 | # Summary elements laid out in the same order as the text |
| 650 | # headers: |
| 651 | # Board Avail Bad Good Spare Total |
| 652 | # e[0] e[1] e[2] e[3] e[4] e[5] |
| 653 | element = (board, |
| 654 | counts.get_spares_buffer(), |
| 655 | counts.get_broken(), |
| 656 | counts.get_working(), |
| 657 | counts.get_total(_SPARE_POOL), |
| 658 | counts.get_total()) |
| 659 | summaries.append(element) |
| 660 | nbroken += element[2] |
| 661 | nworking += element[3] |
| 662 | if element[2]: |
J. Richard Barnette | d3ba33a | 2015-10-14 11:20:49 -0700 | [diff] [blame] | 663 | nbroken_boards += 1 |
| 664 | ntotal = nworking + nbroken |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 665 | summaries = sorted(summaries, key=lambda e: (e[1], -e[2])) |
J. Richard Barnette | d3ba33a | 2015-10-14 11:20:49 -0700 | [diff] [blame] | 666 | broken_percent = int(round(100.0 * nbroken / ntotal)) |
| 667 | working_percent = 100 - broken_percent |
| 668 | message = ['Summary of DUTs in inventory:', |
| 669 | '%10s %10s %6s' % ('Bad', 'Good', 'Total'), |
| 670 | '%5d %3d%% %5d %3d%% %6d' % ( |
| 671 | nbroken, broken_percent, |
| 672 | nworking, working_percent, |
| 673 | ntotal), |
| 674 | '', |
| 675 | 'Boards with failures: %d' % nbroken_boards, |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 676 | 'Boards in inventory: %d' % len(summaries), |
J. Richard Barnette | d3ba33a | 2015-10-14 11:20:49 -0700 | [diff] [blame] | 677 | '', '', |
| 678 | 'Full board inventory:\n', |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 679 | '%-22s %5s %5s %5s %5s %5s' % ( |
| 680 | 'Board', 'Avail', 'Bad', 'Good', |
| 681 | 'Spare', 'Total')] |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 682 | message.extend( |
J. Richard Barnette | e39c827 | 2015-10-20 17:58:30 -0700 | [diff] [blame] | 683 | ['%-22s %5d %5d %5d %5d %5d' % e for e in summaries]) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 684 | return '\n'.join(message) |
| 685 | |
| 686 | |
J. Richard Barnette | 4845fcf | 2015-04-20 14:26:25 -0700 | [diff] [blame] | 687 | _POOL_INVENTORY_HEADER = '''\ |
Aviv Keshet | 056d74c | 2015-07-14 09:18:43 -0700 | [diff] [blame] | 688 | Notice to Infrastructure deputies: All boards shown below are at |
J. Richard Barnette | c9a143c | 2015-06-04 11:11:19 -0700 | [diff] [blame] | 689 | less than full strength, please take action to resolve the issues. |
| 690 | Once you're satisified that failures won't recur, failed DUTs can |
| 691 | be replaced with spares by running `balance_pool`. Detailed |
| 692 | instructions can be found here: |
J. Richard Barnette | 4845fcf | 2015-04-20 14:26:25 -0700 | [diff] [blame] | 693 | http://go/cros-manage-duts |
| 694 | ''' |
| 695 | |
| 696 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 697 | def _generate_pool_inventory_message(inventory): |
| 698 | """Generate the "pool inventory" e-mail message. |
| 699 | |
| 700 | The pool inventory is a list by pool and board summarizing the |
| 701 | number of working and broken DUTs in the pool. Only boards with |
| 702 | at least one broken DUT are included in the list. |
| 703 | |
| 704 | N.B. For sample output text formattted as users can expect to |
| 705 | see it in e-mail and log files, refer to the unit tests. |
| 706 | |
| 707 | @param inventory _LabInventory object with the inventory to |
| 708 | be reported on. |
| 709 | @return String with the inventory message to be sent. |
| 710 | |
| 711 | """ |
| 712 | logging.debug('Creating pool inventory') |
J. Richard Barnette | 4845fcf | 2015-04-20 14:26:25 -0700 | [diff] [blame] | 713 | message = [_POOL_INVENTORY_HEADER] |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 714 | newline = '' |
| 715 | for pool in _CRITICAL_POOLS: |
| 716 | message.append( |
| 717 | '%sStatus for pool:%s, by board:' % (newline, pool)) |
| 718 | message.append( |
| 719 | '%-20s %5s %5s %5s' % ( |
| 720 | 'Board', 'Bad', 'Good', 'Total')) |
| 721 | data_list = [] |
| 722 | for board, counts in inventory.items(): |
| 723 | logging.debug('Counting inventory for %s, %s', |
| 724 | board, pool) |
| 725 | broken = counts.get_broken(pool) |
| 726 | if broken == 0: |
| 727 | continue |
| 728 | working = counts.get_working(pool) |
| 729 | total = counts.get_total(pool) |
| 730 | data_list.append((board, broken, working, total)) |
| 731 | if data_list: |
| 732 | data_list = sorted(data_list, key=lambda d: -d[1]) |
| 733 | message.extend( |
| 734 | ['%-20s %5d %5d %5d' % t for t in data_list]) |
| 735 | else: |
| 736 | message.append('(All boards at full strength)') |
| 737 | newline = '\n' |
| 738 | return '\n'.join(message) |
| 739 | |
| 740 | |
| 741 | def _send_email(arguments, tag, subject, recipients, body): |
| 742 | """Send an inventory e-mail message. |
| 743 | |
| 744 | The message is logged in the selected log directory using `tag` |
| 745 | for the file name. |
| 746 | |
| 747 | If the --print option was requested, the message is neither |
| 748 | logged nor sent, but merely printed on stdout. |
| 749 | |
| 750 | @param arguments Parsed command-line options. |
| 751 | @param tag Tag identifying the inventory for logging |
| 752 | purposes. |
| 753 | @param subject E-mail Subject: header line. |
| 754 | @param recipients E-mail addresses for the To: header line. |
| 755 | @param body E-mail message body. |
| 756 | |
| 757 | """ |
| 758 | logging.debug('Generating email: "%s"', subject) |
| 759 | all_recipients = ', '.join(recipients) |
| 760 | report_body = '\n'.join([ |
| 761 | 'To: %s' % all_recipients, |
| 762 | 'Subject: %s' % subject, |
| 763 | '', body, '']) |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 764 | if arguments.debug: |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 765 | print report_body |
| 766 | else: |
| 767 | filename = os.path.join(arguments.logdir, tag) |
| 768 | try: |
| 769 | report_file = open(filename, 'w') |
| 770 | report_file.write(report_body) |
| 771 | report_file.close() |
| 772 | except EnvironmentError as e: |
| 773 | logging.error('Failed to write %s: %s', filename, e) |
| 774 | try: |
| 775 | gmail_lib.send_email(all_recipients, subject, body) |
| 776 | except Exception as e: |
| 777 | logging.error('Failed to send e-mail to %s: %s', |
| 778 | all_recipients, e) |
| 779 | |
| 780 | |
| 781 | def _separate_email_addresses(address_list): |
| 782 | """Parse a list of comma-separated lists of e-mail addresses. |
| 783 | |
| 784 | @param address_list A list of strings containing comma |
| 785 | separate e-mail addresses. |
| 786 | @return A list of the individual e-mail addresses. |
| 787 | |
| 788 | """ |
| 789 | newlist = [] |
| 790 | for arg in address_list: |
| 791 | newlist.extend([email.strip() for email in arg.split(',')]) |
| 792 | return newlist |
| 793 | |
| 794 | |
| 795 | def _verify_arguments(arguments): |
| 796 | """Validate command-line arguments. |
| 797 | |
| 798 | Join comma separated e-mail addresses for `--board-notify` and |
| 799 | `--pool-notify` in separate option arguments into a single list. |
| 800 | |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 801 | For non-debug uses, require that notification be requested for |
| 802 | at least one report. For debug, if notification isn't specified, |
| 803 | treat it as "run all the reports." |
| 804 | |
| 805 | The return value indicates success or failure; in the case of |
| 806 | failure, we also write an error message to stderr. |
| 807 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 808 | @param arguments Command-line arguments as returned by |
| 809 | `ArgumentParser` |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 810 | @return True if the arguments are semantically good, or False |
| 811 | if the arguments don't meet requirements. |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 812 | |
| 813 | """ |
| 814 | arguments.board_notify = _separate_email_addresses( |
| 815 | arguments.board_notify) |
| 816 | arguments.pool_notify = _separate_email_addresses( |
| 817 | arguments.pool_notify) |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 818 | if not arguments.board_notify and not arguments.pool_notify: |
| 819 | if not arguments.debug: |
| 820 | sys.stderr.write('Must specify at least one of ' |
| 821 | '--board-notify or --pool-notify\n') |
| 822 | return False |
| 823 | else: |
| 824 | # We want to run all the reports. An empty notify list |
| 825 | # will cause a report to be skipped, so make sure the |
| 826 | # lists are non-empty. |
| 827 | arguments.board_notify = [''] |
| 828 | arguments.pool_notify = [''] |
| 829 | return True |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 830 | |
| 831 | |
| 832 | def _get_logdir(script): |
| 833 | """Get the default directory for the `--logdir` option. |
| 834 | |
| 835 | The default log directory is based on the parent directory |
| 836 | containing this script. |
| 837 | |
| 838 | @param script Path to this script file. |
| 839 | @return A path to a directory. |
| 840 | |
| 841 | """ |
| 842 | basedir = os.path.dirname(os.path.abspath(script)) |
| 843 | basedir = os.path.dirname(basedir) |
| 844 | return os.path.join(basedir, _LOGDIR) |
| 845 | |
| 846 | |
| 847 | def _parse_command(argv): |
| 848 | """Parse the command line arguments. |
| 849 | |
| 850 | Create an argument parser for this command's syntax, parse the |
| 851 | command line, and return the result of the ArgumentParser |
| 852 | parse_args() method. |
| 853 | |
| 854 | @param argv Standard command line argument vector; argv[0] is |
| 855 | assumed to be the command name. |
| 856 | @return Result returned by ArgumentParser.parse_args(). |
| 857 | |
| 858 | """ |
| 859 | parser = argparse.ArgumentParser( |
| 860 | prog=argv[0], |
| 861 | description='Gather and report lab inventory statistics') |
| 862 | parser.add_argument('-d', '--duration', type=int, |
| 863 | default=_DEFAULT_DURATION, metavar='HOURS', |
| 864 | help='number of hours back to search for status' |
| 865 | ' (default: %d)' % _DEFAULT_DURATION) |
| 866 | parser.add_argument('--board-notify', action='append', |
| 867 | default=[], metavar='ADDRESS', |
| 868 | help='Generate board inventory message, ' |
| 869 | 'and send it to the given e-mail address(es)') |
| 870 | parser.add_argument('--pool-notify', action='append', |
| 871 | default=[], metavar='ADDRESS', |
| 872 | help='Generate pool inventory message, ' |
| 873 | 'and send it to the given address(es)') |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 874 | parser.add_argument('-r', '--recommend', type=int, default=None, |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 875 | help=('Specify how many DUTs should be ' |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 876 | 'recommended for repair (default: no ' |
| 877 | 'recommendation)')) |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 878 | parser.add_argument('--debug', action='store_true', |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 879 | help='Print e-mail messages on stdout ' |
| 880 | 'without sending them.') |
| 881 | parser.add_argument('--logdir', default=_get_logdir(argv[0]), |
| 882 | help='Directory where logs will be written.') |
| 883 | parser.add_argument('boardnames', nargs='*', |
| 884 | metavar='BOARD', |
| 885 | help='names of boards to report on ' |
| 886 | '(default: all boards)') |
| 887 | arguments = parser.parse_args(argv[1:]) |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 888 | if not _verify_arguments(arguments): |
| 889 | return None |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 890 | return arguments |
| 891 | |
| 892 | |
| 893 | def _configure_logging(arguments): |
| 894 | """Configure the `logging` module for our needs. |
| 895 | |
| 896 | How we log depends on whether the `--print` option was |
| 897 | provided on the command line. Without the option, we log all |
| 898 | messages at DEBUG level or above, and write them to a file in |
| 899 | the directory specified by the `--logdir` option. With the |
| 900 | option, we write log messages to stdout; messages below INFO |
| 901 | level are discarded. |
| 902 | |
| 903 | The log file is configured to rotate once a week on Friday |
| 904 | evening, preserving ~3 months worth of history. |
| 905 | |
| 906 | @param arguments Command-line arguments as returned by |
| 907 | `ArgumentParser` |
| 908 | |
| 909 | """ |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 910 | root_logger = logging.getLogger() |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 911 | if arguments.debug: |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 912 | root_logger.setLevel(logging.INFO) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 913 | handler = logging.StreamHandler(sys.stdout) |
| 914 | handler.setFormatter(logging.Formatter()) |
| 915 | else: |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 916 | root_logger.setLevel(logging.DEBUG) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 917 | logfile = os.path.join(arguments.logdir, _LOGFILE) |
| 918 | handler = logging.handlers.TimedRotatingFileHandler( |
| 919 | logfile, when='W4', backupCount=13) |
| 920 | formatter = logging.Formatter(_LOG_FORMAT, |
| 921 | time_utils.TIME_FMT) |
| 922 | handler.setFormatter(formatter) |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 923 | # TODO(jrbarnette) This is gross. Importing client.bin.utils |
| 924 | # implicitly imported logging_config, which calls |
| 925 | # logging.basicConfig() *at module level*. That gives us an |
| 926 | # extra logging handler that we don't want. So, clear out all |
| 927 | # the handlers here. |
| 928 | for h in root_logger.handlers: |
| 929 | root_logger.removeHandler(h) |
| 930 | root_logger.addHandler(handler) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 931 | |
| 932 | |
| 933 | def _populate_board_counts(inventory): |
| 934 | """Gather board counts while providing interactive feedback. |
| 935 | |
| 936 | Gathering the status of all individual DUTs in the lab can take |
| 937 | considerable time (~30 minutes at the time of this writing). |
| 938 | |
| 939 | Normally, we pay that cost by querying as we go. However, with |
| 940 | the `--print` option, a human being may be watching the |
| 941 | progress. So, we force the first (expensive) queries to happen |
| 942 | up front, and provide a small ASCII progress bar to give an |
| 943 | indicator of how many boards have been processed. |
| 944 | |
| 945 | @param inventory _LabInventory object with the inventory to |
| 946 | be gathered. |
| 947 | |
| 948 | """ |
| 949 | n = 0 |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 950 | total_broken = 0 |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 951 | for counts in inventory.values(): |
| 952 | n += 1 |
| 953 | if n % 10 == 5: |
| 954 | c = '+' |
| 955 | elif n % 10 == 0: |
| 956 | c = '%d' % ((n / 10) % 10) |
| 957 | else: |
| 958 | c = '.' |
| 959 | sys.stdout.write(c) |
| 960 | sys.stdout.flush() |
| 961 | # This next call is where all the time goes - it forces all |
| 962 | # of a board's HostJobHistory objects to query the database |
| 963 | # and cache their results. |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 964 | total_broken += counts.get_broken() |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 965 | sys.stdout.write('\n') |
J. Richard Barnette | f683928 | 2015-06-01 16:00:35 -0700 | [diff] [blame] | 966 | sys.stdout.write('Found %d broken DUTs\n' % total_broken) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 967 | |
| 968 | |
| 969 | def main(argv): |
| 970 | """Standard main routine. |
| 971 | @param argv Command line arguments including `sys.argv[0]`. |
| 972 | """ |
| 973 | arguments = _parse_command(argv) |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 974 | if not arguments: |
| 975 | sys.exit(1) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 976 | _configure_logging(arguments) |
| 977 | try: |
| 978 | end_time = int(time.time()) |
| 979 | start_time = end_time - arguments.duration * 60 * 60 |
| 980 | timestamp = time.strftime('%Y-%m-%d.%H', |
| 981 | time.localtime(end_time)) |
| 982 | logging.debug('Starting lab inventory for %s', timestamp) |
| 983 | if arguments.board_notify: |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 984 | if arguments.recommend: |
| 985 | logging.debug('Will include repair recommendations') |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 986 | logging.debug('Will include board inventory') |
| 987 | if arguments.pool_notify: |
| 988 | logging.debug('Will include pool inventory') |
| 989 | |
J. Richard Barnette | a7c514e | 2015-09-15 11:13:23 -0700 | [diff] [blame] | 990 | afe = frontend_wrappers.RetryingAFE(server=None) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 991 | inventory = _LabInventory.create_inventory( |
| 992 | afe, start_time, end_time, arguments.boardnames) |
| 993 | logging.info('Found %d hosts across %d boards', |
| 994 | inventory.get_num_duts(), |
| 995 | inventory.get_num_boards()) |
| 996 | |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 997 | if arguments.debug: |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 998 | _populate_board_counts(inventory) |
| 999 | |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 1000 | if arguments.board_notify: |
J. Richard Barnette | 1df6a56 | 2015-06-09 10:06:17 -0700 | [diff] [blame] | 1001 | if arguments.recommend: |
| 1002 | recommend_message = _generate_repair_recommendation( |
| 1003 | inventory, arguments.recommend) + '\n\n\n' |
| 1004 | else: |
| 1005 | recommend_message = '' |
| 1006 | board_message = _generate_board_inventory_message(inventory) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 1007 | _send_email(arguments, |
| 1008 | 'boards-%s.txt' % timestamp, |
| 1009 | 'DUT board inventory %s' % timestamp, |
| 1010 | arguments.board_notify, |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 1011 | recommend_message + board_message) |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 1012 | |
J. Richard Barnette | 02e8243 | 2015-10-13 16:02:47 -0700 | [diff] [blame] | 1013 | if arguments.pool_notify: |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 1014 | _send_email(arguments, |
| 1015 | 'pools-%s.txt' % timestamp, |
| 1016 | 'DUT pool inventory %s' % timestamp, |
| 1017 | arguments.pool_notify, |
| 1018 | _generate_pool_inventory_message(inventory)) |
| 1019 | except KeyboardInterrupt: |
| 1020 | pass |
| 1021 | except EnvironmentError as e: |
| 1022 | logging.exception('Unexpected OS error: %s', e) |
| 1023 | except Exception as e: |
| 1024 | logging.exception('Unexpected exception: %s', e) |
| 1025 | |
| 1026 | |
J. Richard Barnette | aa86893 | 2015-10-23 13:28:59 -0700 | [diff] [blame] | 1027 | def get_managed_boards(afe): |
| 1028 | end_time = int(time.time()) |
| 1029 | start_time = end_time - 24 * 60 * 60 |
| 1030 | inventory = _LabInventory.create_inventory( |
| 1031 | afe, start_time, end_time) |
| 1032 | return inventory.get_managed_boards() |
| 1033 | |
| 1034 | |
J. Richard Barnette | 96db349 | 2015-03-27 17:23:52 -0700 | [diff] [blame] | 1035 | if __name__ == '__main__': |
| 1036 | main(sys.argv) |