| 1 |
<?php |
| 2 |
/** |
| 3 |
* Shared Import Task execution module. |
| 4 |
* |
| 5 |
* Manual actions, the setup wizard, and hourly cron will enter through this |
| 6 |
* class. The class owns the Import Run lifecycle while WordPress persistence, |
| 7 |
* the external MLS API, and theme-specific listing writes remain behind the |
| 8 |
* environment boundary. |
| 9 |
* |
| 10 |
* The module is being built in tested vertical slices. The first slice accepts |
| 11 |
* one manual run and completes the valid zero-listing case through the public |
| 12 |
* start() and execute() operations. |
| 13 |
* |
| 14 |
* @package MLSImport |
| 15 |
*/ |
| 16 |
|
| 17 |
if ( ! defined( 'ABSPATH' ) ) { |
| 18 |
exit; |
| 19 |
} |
| 20 |
|
| 21 |
require_once __DIR__ . '/interface-mlsimport-import-task-execution-environment.php'; |
| 22 |
|
| 23 |
/** |
| 24 |
* Coordinates one Import Run through the accepted public seam. |
| 25 |
*/ |
| 26 |
final class Mlsimport_Import_Task_Execution { |
| 27 |
|
| 28 |
/** Thirty minutes without activity makes an abandoned run replaceable. */ |
| 29 |
private const STALE_AFTER_SECONDS = 1800; |
| 30 |
|
| 31 |
/** Number of listings released together after each external request. */ |
| 32 |
private const BATCH_SIZE = 25; |
| 33 |
|
| 34 |
/** Maximum listings handled by one Import Run. */ |
| 35 |
private const MAX_LISTINGS = 10000; |
| 36 |
|
| 37 |
/** |
| 38 |
* Wall-clock seconds one manual worker may spend before handing off. |
| 39 |
* |
| 40 |
* Strict hosts kill long web requests at limits the plugin cannot see |
| 41 |
* (issue #199). Checked after every batch, so the worst request length is |
| 42 |
* this budget plus one final batch — safely inside a 60-second kill limit. |
| 43 |
*/ |
| 44 |
private const CHUNK_BUDGET_SECONDS = 45; |
| 45 |
|
| 46 |
/** |
| 47 |
* Seconds of worker silence before the watchdog revives a manual run. |
| 48 |
* |
| 49 |
* A live worker records activity after every listing, so this long a |
| 50 |
* silence means the worker chain died: a chunk was killed before it could |
| 51 |
* queue its follow-up, or a queued worker was never dispatched. |
| 52 |
*/ |
| 53 |
private const REVIVE_AFTER_SECONDS = 90; |
| 54 |
|
| 55 |
/** |
| 56 |
* Consecutive revivals at one unmoved position before the run fails. |
| 57 |
* |
| 58 |
* Progress between revivals resets the count: a run that keeps dying but |
| 59 |
* keeps advancing is allowed to grind to completion. Only a position the |
| 60 |
* server kills repeatedly is hopeless, and retrying it forever would loop. |
| 61 |
*/ |
| 62 |
private const MAX_STALLED_REVIVALS = 3; |
| 63 |
|
| 64 |
/** @var Mlsimport_Import_Task_Execution_Environment External operations. */ |
| 65 |
private $environment; |
| 66 |
|
| 67 |
/** |
| 68 |
* Receive the WordPress environment or a system-boundary test double. |
| 69 |
* |
| 70 |
* @param Mlsimport_Import_Task_Execution_Environment $environment External operations. |
| 71 |
*/ |
| 72 |
public function __construct( Mlsimport_Import_Task_Execution_Environment $environment ) { |
| 73 |
$this->environment = $environment; |
| 74 |
} |
| 75 |
|
| 76 |
/** |
| 77 |
* Request the one site-wide slot for a new Import Run. |
| 78 |
* |
| 79 |
* The accepted run is stored in waiting state so an HTTP manual caller can |
| 80 |
* return before Action Scheduler invokes execute(). The request is stored |
| 81 |
* with the run because background execution must use exactly the values the |
| 82 |
* caller supplied. |
| 83 |
* |
| 84 |
* @param array<string, mixed> $request Task id, source, found count, and limit. |
| 85 |
* @return array<string, bool|string> Public start response. |
| 86 |
*/ |
| 87 |
public function start( array $request ): array { |
| 88 |
$run_id = $this->environment->new_run_id(); |
| 89 |
$now = $this->environment->now(); |
| 90 |
$run = array( |
| 91 |
'run_id' => $run_id, |
| 92 |
'task_id' => (int) ( $request['task_id'] ?? 0 ), |
| 93 |
'source' => (string) ( $request['source'] ?? 'manual' ), |
| 94 |
'state' => 'waiting', |
| 95 |
'request' => $request, |
| 96 |
'started_at' => $now, |
| 97 |
'activity_at' => $now, |
| 98 |
); |
| 99 |
|
| 100 |
$stale_before = $now - self::STALE_AFTER_SECONDS; |
| 101 |
if ( ! $this->environment->claim_run( $run, $stale_before ) ) { |
| 102 |
return array( |
| 103 |
'accepted' => false, |
| 104 |
'reason' => 'already_running', |
| 105 |
); |
| 106 |
} |
| 107 |
|
| 108 |
return array( |
| 109 |
'accepted' => true, |
| 110 |
'run_id' => $run_id, |
| 111 |
'state' => 'waiting', |
| 112 |
); |
| 113 |
} |
| 114 |
|
| 115 |
/** |
| 116 |
* Request that the active Import Run for one task stop safely. |
| 117 |
* |
| 118 |
* Stop is final: storage records the stopped status and releases the |
| 119 |
* site-wide slot immediately, so a new import may start right away. |
| 120 |
* Execution still checks the persisted request before each listing, so a |
| 121 |
* listing already inside the theme-specific writer may finish, while no |
| 122 |
* new listing starts afterward and the stopped status is never overwritten. |
| 123 |
* |
| 124 |
* @param int $task_id Import Task identifier. |
| 125 |
* @return array<string, bool> Whether an active matching run was found. |
| 126 |
*/ |
| 127 |
public function stop( int $task_id ): array { |
| 128 |
return array( 'accepted' => $this->environment->request_stop( $task_id ) ); |
| 129 |
} |
| 130 |
|
| 131 |
/** |
| 132 |
* Revive a manual Import Run whose worker chain went silent (issue #199). |
| 133 |
* |
| 134 |
* Chunked execution depends on each worker queueing its follow-up. When a |
| 135 |
* host kills a worker before that hand-off, the chain is dead and the run |
| 136 |
* would sit unfinished. The watchdog is called from the polled admin |
| 137 |
* progress endpoint and from hourly cron; it queues a replacement worker |
| 138 |
* once activity has been silent long enough, and fails the run with an |
| 139 |
* explicit hosting error when revivals at one position keep dying. |
| 140 |
* |
| 141 |
* The worker generation advances with every revival: a presumed-dead |
| 142 |
* worker that is actually still alive sees the newer generation at its |
| 143 |
* next listing boundary and exits, so one run never has two live writers. |
| 144 |
* |
| 145 |
* @param int $task_id Import Task identifier. |
| 146 |
* @return array<string, bool|string> Whether a worker was queued, with reason. |
| 147 |
*/ |
| 148 |
public function revive( int $task_id ): array { |
| 149 |
$run = $this->environment->read_active_run( $task_id ); |
| 150 |
if ( empty( $run ) ) { |
| 151 |
return array( |
| 152 |
'revived' => false, |
| 153 |
'reason' => 'no_active_run', |
| 154 |
); |
| 155 |
} |
| 156 |
// Automatic runs chunk too since issue #330, and a resumed chunk reuses |
| 157 |
// the count persisted on the run record, so every source is revived. |
| 158 |
$now = $this->environment->now(); |
| 159 |
if ( $now - (int) ( $run['activity_at'] ?? 0 ) < self::REVIVE_AFTER_SECONDS ) { |
| 160 |
return array( |
| 161 |
'revived' => false, |
| 162 |
'reason' => 'recent_activity', |
| 163 |
); |
| 164 |
} |
| 165 |
|
| 166 |
$run_id = (string) $run['run_id']; |
| 167 |
$position = max( 0, (int) ( $run['handled'] ?? 0 ) ); |
| 168 |
// Progress since the last revival proves the run is advancing, so the |
| 169 |
// stall count restarts; the same position again means another death |
| 170 |
// with zero progress. |
| 171 |
$stalled_revivals = $position === (int) ( $run['revived_at_position'] ?? -1 ) |
| 172 |
? (int) ( $run['revive_count'] ?? 0 ) + 1 |
| 173 |
: 1; |
| 174 |
if ( $stalled_revivals >= self::MAX_STALLED_REVIVALS ) { |
| 175 |
$this->environment->finish_run( |
| 176 |
$run_id, |
| 177 |
array( |
| 178 |
'state' => 'failed', |
| 179 |
'found' => (int) ( $run['expected'] ?? 0 ), |
| 180 |
'saved' => (int) ( $run['saved'] ?? 0 ), |
| 181 |
'failed' => (int) ( $run['failed'] ?? 0 ), |
| 182 |
'error' => sprintf( |
| 183 |
'Import stopped at listing %1$d of %2$d: the server terminated the import worker %3$d times at this position. Ask your hosting provider about PHP execution limits.', |
| 184 |
$position, |
| 185 |
(int) ( $run['expected'] ?? 0 ), |
| 186 |
self::MAX_STALLED_REVIVALS |
| 187 |
), |
| 188 |
) |
| 189 |
); |
| 190 |
return array( |
| 191 |
'revived' => false, |
| 192 |
'reason' => 'stalled', |
| 193 |
); |
| 194 |
} |
| 195 |
|
| 196 |
// Refreshing activity_at here also arms a fresh 90-second window, so |
| 197 |
// repeated watchdog calls cannot queue a second replacement while the |
| 198 |
// first is still dispatching. |
| 199 |
$this->environment->update_run( |
| 200 |
$run_id, |
| 201 |
array( |
| 202 |
'worker_generation' => (int) ( $run['worker_generation'] ?? 0 ) + 1, |
| 203 |
'revive_count' => $stalled_revivals, |
| 204 |
'revived_at_position' => $position, |
| 205 |
'activity_at' => $now, |
| 206 |
) |
| 207 |
); |
| 208 |
$this->environment->revive_worker( $run_id ); |
| 209 |
return array( |
| 210 |
'revived' => true, |
| 211 |
'reason' => '', |
| 212 |
); |
| 213 |
} |
| 214 |
|
| 215 |
/** |
| 216 |
* Return the latest administrator-visible progress and final result. |
| 217 |
* |
| 218 |
* Storage retains the completed, stopped, or failed result until start() |
| 219 |
* accepts a later run for the same Import Task. |
| 220 |
* |
| 221 |
* @param int $task_id Import Task identifier. |
| 222 |
* @return array<string, mixed> Public Import Run Progress and Result. |
| 223 |
*/ |
| 224 |
public function status( int $task_id ): array { |
| 225 |
return $this->environment->read_task_status( $task_id ); |
| 226 |
} |
| 227 |
|
| 228 |
/** |
| 229 |
* Execute an accepted Import Run and return its final public result. |
| 230 |
* |
| 231 |
* The first vertical slice completes only the zero-listing path. A non-zero |
| 232 |
* run deliberately fails fast until the next red test introduces external |
| 233 |
* batch fetching and listing saves. |
| 234 |
* |
| 235 |
* @param string $run_id Accepted run identity. |
| 236 |
* @return array<string, int|string> Final Import Run Result. |
| 237 |
*/ |
| 238 |
public function execute( string $run_id ): array { |
| 239 |
$run = $this->environment->read_run( $run_id ); |
| 240 |
if ( empty( $run ) ) { |
| 241 |
throw new InvalidArgumentException( 'Import Run was not found.' ); |
| 242 |
} |
| 243 |
$request = isset( $run['request'] ) && is_array( $run['request'] ) ? $run['request'] : array(); |
| 244 |
$source = (string) ( $run['source'] ?? 'manual' ); |
| 245 |
// A resumed chunk reuses the plan persisted by the first worker: the |
| 246 |
// count is taken once per run (issue #330). Listings keep changing |
| 247 |
// during a long import, so counting again mid-run would move the |
| 248 |
// goal posts and end the run early or late. |
| 249 |
$found = max( 0, (int) ( $run['found'] ?? ( $request['found'] ?? 0 ) ) ); |
| 250 |
|
| 251 |
// A stale worker may wake after a replacement claimed the site-wide |
| 252 |
// slot. It must stop before changing state, fetching, or saving. |
| 253 |
if ( ! $this->environment->owns_run( $run_id ) ) { |
| 254 |
$result = array( |
| 255 |
'state' => 'stopped', |
| 256 |
'found' => $found, |
| 257 |
'saved' => 0, |
| 258 |
'failed' => 0, |
| 259 |
'error' => 'Import Run was replaced.', |
| 260 |
); |
| 261 |
$this->environment->finish_run( $run_id, $result ); |
| 262 |
return $result; |
| 263 |
} |
| 264 |
|
| 265 |
// Automatic runs count changed listings at execution time, on the first |
| 266 |
// worker only (a resumed chunk carries the count on the run record). |
| 267 |
// Manual runs intentionally retain the count already shown on the task |
| 268 |
// screen. |
| 269 |
if ( 'automatic' === $source && ! isset( $run['found'] ) ) { |
| 270 |
try { |
| 271 |
$count_response = $this->environment->count_listings( $run ); |
| 272 |
} catch ( Throwable $exception ) { |
| 273 |
$count_response = array( |
| 274 |
'success' => false, |
| 275 |
'error' => $exception->getMessage(), |
| 276 |
); |
| 277 |
} |
| 278 |
if ( true !== ( $count_response['success'] ?? false ) || ! isset( $count_response['found'] ) ) { |
| 279 |
$result = array( |
| 280 |
'state' => 'failed', |
| 281 |
'found' => 0, |
| 282 |
'saved' => 0, |
| 283 |
'failed' => 0, |
| 284 |
'error' => (string) ( $count_response['error'] ?? 'Listings count failed.' ), |
| 285 |
); |
| 286 |
$this->environment->finish_run( $run_id, $result ); |
| 287 |
return $result; |
| 288 |
} |
| 289 |
$found = max( 0, (int) $count_response['found'] ); |
| 290 |
if ( $found > self::MAX_LISTINGS ) { |
| 291 |
$result = array( |
| 292 |
'state' => 'failed', |
| 293 |
'found' => $found, |
| 294 |
'saved' => 0, |
| 295 |
'failed' => 0, |
| 296 |
'error' => 'Automatic Import Run exceeds the 10,000 listing limit.', |
| 297 |
); |
| 298 |
$this->environment->finish_run( $run_id, $result ); |
| 299 |
return $result; |
| 300 |
} |
| 301 |
} |
| 302 |
|
| 303 |
$limit = max( 0, (int) ( $request['limit'] ?? 0 ) ); |
| 304 |
$expected = 0 === $limit ? $found : min( $found, $limit ); |
| 305 |
$expected = min( self::MAX_LISTINGS, $expected ); |
| 306 |
|
| 307 |
// A resumed chunk continues exactly where the previous worker handed |
| 308 |
// off: position and totals come from the persisted run, so a fresh run |
| 309 |
// starts at zero and a continuation never repeats or skips a listing. |
| 310 |
$saved = max( 0, (int) ( $run['saved'] ?? 0 ) ); |
| 311 |
$failed = max( 0, (int) ( $run['failed'] ?? 0 ) ); |
| 312 |
$handled = max( 0, (int) ( $run['handled'] ?? 0 ) ); |
| 313 |
$error = (string) ( $run['error'] ?? '' ); |
| 314 |
$this->environment->update_run( |
| 315 |
$run_id, |
| 316 |
array( |
| 317 |
'state' => 'running', |
| 318 |
'found' => $found, |
| 319 |
'handled' => $handled, |
| 320 |
'expected' => $expected, |
| 321 |
'error' => $error, |
| 322 |
'activity_at' => $this->environment->now(), |
| 323 |
) |
| 324 |
); |
| 325 |
// A Stop request can arrive while an async manual run is still waiting. |
| 326 |
// Honor it before the first external request; during a listing save, the |
| 327 |
// existing per-listing check lets that current save finish safely. |
| 328 |
$stopped = $this->environment->stop_requested( $run_id ); |
| 329 |
$chunk_started_at = $this->environment->now(); |
| 330 |
// The generation this worker was started for. A watchdog revival |
| 331 |
// advances the stored generation, so an overtaken worker — presumed |
| 332 |
// dead but actually alive — recognizes its replacement per listing. |
| 333 |
$generation = (int) ( $run['worker_generation'] ?? 0 ); |
| 334 |
while ( $handled < $expected ) { |
| 335 |
if ( $stopped ) { |
| 336 |
break; |
| 337 |
} |
| 338 |
$batch_limit = min( self::BATCH_SIZE, $expected - $handled ); |
| 339 |
try { |
| 340 |
$response = $this->environment->fetch_listing_batch( $run, $handled, $batch_limit ); |
| 341 |
} catch ( Throwable $exception ) { |
| 342 |
$error = '' !== $exception->getMessage() ? $exception->getMessage() : 'Listings request failed.'; |
| 343 |
break; |
| 344 |
} |
| 345 |
$data = isset( $response['data'] ) && is_array( $response['data'] ) ? $response['data'] : array(); |
| 346 |
|
| 347 |
// A failed response is a real error: continuing would silently |
| 348 |
// leave a hole in the import. |
| 349 |
if ( true !== ( $response['success'] ?? false ) ) { |
| 350 |
$error = (string) ( $response['error'] ?? 'Listings request failed.' ); |
| 351 |
break; |
| 352 |
} |
| 353 |
|
| 354 |
// A valid but EMPTY page is not an error: the MLS holds fewer |
| 355 |
// matching listings than were counted when the run started |
| 356 |
// (listings change status or vanish during a long import — observed |
| 357 |
// as 1037 real listings for a count of 1039). Everything available |
| 358 |
// has been imported, so finish the run normally. |
| 359 |
if ( empty( $data ) ) { |
| 360 |
break; |
| 361 |
} |
| 362 |
|
| 363 |
$remaining = $expected - $handled; |
| 364 |
foreach ( array_slice( $data, 0, $remaining ) as $listing ) { |
| 365 |
// A final Stop releases the site-wide slot immediately, so an |
| 366 |
// explicit Stop must be recognized before the lost slot is |
| 367 |
// treated as a replacement by another run. |
| 368 |
if ( $this->environment->stop_requested( $run_id ) ) { |
| 369 |
$stopped = true; |
| 370 |
break; |
| 371 |
} |
| 372 |
if ( ! $this->environment->owns_run( $run_id ) ) { |
| 373 |
$stopped = true; |
| 374 |
$error = 'Import Run was replaced.'; |
| 375 |
break; |
| 376 |
} |
| 377 |
// A revival replaced this worker while it was presumed dead. |
| 378 |
// Unlike the lost-slot case above, the run itself is still |
| 379 |
// alive under the replacement worker, so this worker must |
| 380 |
// leave the run record, the slot, and the task status alone — |
| 381 |
// it exits without finishing anything. |
| 382 |
if ( (int) ( $this->environment->read_run( $run_id )['worker_generation'] ?? 0 ) !== $generation ) { |
| 383 |
return array( |
| 384 |
'state' => 'stopped', |
| 385 |
'found' => $found, |
| 386 |
'saved' => $saved, |
| 387 |
'failed' => $failed, |
| 388 |
'error' => 'Import Run was replaced.', |
| 389 |
); |
| 390 |
} |
| 391 |
++$handled; |
| 392 |
if ( ! is_array( $listing ) || empty( $listing['ListingKey'] ) ) { |
| 393 |
++$failed; |
| 394 |
$error = 'ListingKey is missing.'; |
| 395 |
} else { |
| 396 |
try { |
| 397 |
$save = $this->environment->save_listing( $run, $listing ); |
| 398 |
} catch ( Throwable $exception ) { |
| 399 |
$save = array( |
| 400 |
'success' => false, |
| 401 |
'error' => '' !== $exception->getMessage() |
| 402 |
? $exception->getMessage() |
| 403 |
: 'Listing could not be saved.', |
| 404 |
); |
| 405 |
} |
| 406 |
if ( true === ( $save['success'] ?? false ) ) { |
| 407 |
++$saved; |
| 408 |
} else { |
| 409 |
++$failed; |
| 410 |
$error = (string) ( $save['error'] ?? 'Listing could not be saved.' ); |
| 411 |
} |
| 412 |
} |
| 413 |
|
| 414 |
// Persist progress after every listing so the polled admin |
| 415 |
// progress bar advances in near real time, not once per batch. |
| 416 |
// Saved and failed totals are stored too, so a later chunk or a |
| 417 |
// watchdog revival can continue with correct final counts. |
| 418 |
$this->environment->update_run( |
| 419 |
$run_id, |
| 420 |
array( |
| 421 |
'handled' => $handled, |
| 422 |
'expected' => $expected, |
| 423 |
'saved' => $saved, |
| 424 |
'failed' => $failed, |
| 425 |
'error' => $error, |
| 426 |
'activity_at' => $this->environment->now(), |
| 427 |
) |
| 428 |
); |
| 429 |
} |
| 430 |
|
| 431 |
if ( $stopped ) { |
| 432 |
break; |
| 433 |
} |
| 434 |
|
| 435 |
// Resumable chunking (issue #199): a worker whose time budget is |
| 436 |
// spent must not start another batch inside this same request — |
| 437 |
// strict hosts kill long requests at limits the plugin cannot see. |
| 438 |
// Position and totals were persisted with the last listing, so this |
| 439 |
// worker queues a follow-up worker for the same run, keeps the |
| 440 |
// site-wide slot, and exits. Only a worker that reaches the end of |
| 441 |
// the plan finishes the run below. Automatic runs chunk too (issue |
| 442 |
// #330): an hourly delta the size of a whole task used to run in |
| 443 |
// one cron request, die, and restart from zero every hour. |
| 444 |
if ( $handled < $expected |
| 445 |
&& ( $this->environment->now() - $chunk_started_at ) >= self::CHUNK_BUDGET_SECONDS ) { |
| 446 |
// Count the hand-off on the run record (issue #216): the finished |
| 447 |
// run's telemetry snapshot reports 1 + handoffs + revivals as its |
| 448 |
// worker total. Each worker hands off at most once, so the value |
| 449 |
// read at execute() start is still current here. |
| 450 |
$this->environment->update_run( |
| 451 |
$run_id, |
| 452 |
array( 'handoffs' => (int) ( $run['handoffs'] ?? 0 ) + 1 ) |
| 453 |
); |
| 454 |
$this->environment->enqueue_worker( $run_id ); |
| 455 |
return array( |
| 456 |
'state' => 'running', |
| 457 |
'found' => $found, |
| 458 |
'saved' => $saved, |
| 459 |
'failed' => $failed, |
| 460 |
'error' => $error, |
| 461 |
); |
| 462 |
} |
| 463 |
} |
| 464 |
|
| 465 |
$state = $stopped ? 'stopped' : ( '' === $error && 0 === $failed ? 'completed' : 'failed' ); |
| 466 |
$result = array( |
| 467 |
'state' => $state, |
| 468 |
'found' => $found, |
| 469 |
'saved' => $saved, |
| 470 |
'failed' => $failed, |
| 471 |
'error' => $error, |
| 472 |
); |
| 473 |
// Every completed run advances the task's last-sync watermark — not just |
| 474 |
// automatic ones (GitHub issue #202 follow-up). A completed manual run |
| 475 |
// has just written everything the task matches, so "changed since |
| 476 |
// completion" is exactly the right next window; and it is the manual run |
| 477 |
// that makes a task cron-eligible, so it must seed the watermark the |
| 478 |
// hourly sync requires (an automatic run refuses to start without one). |
| 479 |
if ( 'completed' === $state ) { |
| 480 |
$this->environment->advance_last_successful_sync_time( |
| 481 |
(int) $run['task_id'], |
| 482 |
$this->environment->now() |
| 483 |
); |
| 484 |
} |
| 485 |
$this->environment->finish_run( $run_id, $result ); |
| 486 |
|
| 487 |
return $result; |
| 488 |
} |
| 489 |
} |
| 490 |
|
