| 1 |
<?php |
| 2 |
/** |
| 3 |
* WCPOS sync read surface. |
| 4 |
* |
| 5 |
* @package WCPOS\WooCommercePOS\API\V2 |
| 6 |
*/ |
| 7 |
|
| 8 |
namespace WCPOS\WooCommercePOS\API\V2; |
| 9 |
|
| 10 |
use WCPOS\WooCommercePOS\Logger; |
| 11 |
use WCPOS\WooCommercePOS\Sync\Api; |
| 12 |
use WCPOS\WooCommercePOS\Sync\Collections; |
| 13 |
use WCPOS\WooCommercePOS\Sync\Digest_Index; |
| 14 |
use WCPOS\WooCommercePOS\Sync\Endpoint_Permissions; |
| 15 |
use WCPOS\WooCommercePOS\Sync\Integrity_Digest; |
| 16 |
use WCPOS\WooCommercePOS\Sync\Request_Int_Param; |
| 17 |
use WP_Error; |
| 18 |
use WP_REST_Controller; |
| 19 |
use WP_REST_Request; |
| 20 |
use WP_REST_Response; |
| 21 |
use WP_REST_Server; |
| 22 |
|
| 23 |
// phpcs:disable Squiz.Commenting, Generic.Commenting -- Ported lab documentation is preserved verbatim. |
| 24 |
|
| 25 |
/** |
| 26 |
* Hash-backed range-checksum scan — the missing experiment between |
| 27 |
* range-checksum and revision-hash (see change-signal-matrix-2026-06-10.md): |
| 28 |
* sql-bypass detection at GROUP BY prices instead of full-hydration prices. |
| 29 |
* |
| 30 |
* GET /integrity/scan compares, per id-range bucket and entirely in SQL, |
| 31 |
* the BIT_XOR aggregate of CURRENT raw-row digests (class-integrity-digest |
| 32 |
* canonical expression) against the BIT_XOR aggregate of STORED hook-time |
| 33 |
* digests — one SQL pass per side. ?bucket=<n> drills into a single bucket |
| 34 |
* and returns the mismatching ids. |
| 35 |
* |
| 36 |
* Every one of those questions is asked through {@see Digest_Index}: this |
| 37 |
* controller validates the request, shapes the envelope and owns the |
| 38 |
* rebuild-scheduling policy — it knows no table, column or JOIN. |
| 39 |
*/ |
| 40 |
final class Integrity_Controller extends WP_REST_Controller { |
| 41 |
use Endpoint_Permissions; |
| 42 |
use Request_Int_Param; |
| 43 |
|
| 44 |
public const DEFAULT_BUCKET_SIZE = 1000; |
| 45 |
public const DEFAULT_LIMIT_BUCKETS = 50; |
| 46 |
|
| 47 |
/** |
| 48 |
* Consecutive drill-downs a bucket may report stale stored digests before |
| 49 |
* the rebuild is scheduled. Must stay ABOVE the client's own escalation |
| 50 |
* threshold (DEFAULT_HYBRID_POLICY.escalateToRevisionHashAfter = 2) so the |
| 51 |
* client has pulled the drifted ids before the stored side is re-baselined. |
| 52 |
*/ |
| 53 |
public const DRIFT_REBUILD_THRESHOLD = 3; |
| 54 |
|
| 55 |
/** Per-bucket drift streaks; never autoloaded, read only on the drill-down path. */ |
| 56 |
public const DRIFT_STREAK_OPTION = 'wcpos_integrity_drift_streaks'; |
| 57 |
|
| 58 |
/** Hard cap on tracked buckets so the option cannot grow unbounded. */ |
| 59 |
public const DRIFT_STREAK_MAX_BUCKETS = 256; |
| 60 |
|
| 61 |
private Integrity_Digest $digests; |
| 62 |
|
| 63 |
private Digest_Index $index; |
| 64 |
|
| 65 |
public function __construct( ?Integrity_Digest $digests = null, ?Digest_Index $index = null ) { |
| 66 |
$this->digests = $digests ?? new Integrity_Digest(); |
| 67 |
$this->index = $index ?? new Digest_Index(); |
| 68 |
} |
| 69 |
|
| 70 |
public function register_routes(): void { |
| 71 |
register_rest_route( |
| 72 |
Api::ROUTE_NAMESPACE, |
| 73 |
'/integrity/scan', |
| 74 |
array( |
| 75 |
'methods' => WP_REST_Server::READABLE, |
| 76 |
'callback' => array( $this, 'scan' ), |
| 77 |
'permission_callback' => array( $this, 'permissions_check' ), |
| 78 |
'args' => array( |
| 79 |
'bucket_size' => array( |
| 80 |
'default' => self::DEFAULT_BUCKET_SIZE, |
| 81 |
'sanitize_callback' => 'absint', |
| 82 |
), |
| 83 |
'after_id' => array( |
| 84 |
'default' => 0, |
| 85 |
'sanitize_callback' => 'absint', |
| 86 |
), |
| 87 |
'limit_buckets' => array( |
| 88 |
'default' => self::DEFAULT_LIMIT_BUCKETS, |
| 89 |
'sanitize_callback' => 'absint', |
| 90 |
), |
| 91 |
'bucket' => array( 'sanitize_callback' => 'absint' ), |
| 92 |
'collection' => array( |
| 93 |
'default' => 'products', |
| 94 |
'sanitize_callback' => 'sanitize_key', |
| 95 |
), |
| 96 |
'status' => array( |
| 97 |
'sanitize_callback' => static fn ( $status ) => 'publish' === $status ? 'publish' : '', |
| 98 |
), |
| 99 |
), |
| 100 |
) |
| 101 |
); |
| 102 |
|
| 103 |
register_rest_route( |
| 104 |
Api::ROUTE_NAMESPACE, |
| 105 |
'/integrity/rebuild', |
| 106 |
array( |
| 107 |
'methods' => WP_REST_Server::CREATABLE, |
| 108 |
'callback' => array( $this, 'rebuild' ), |
| 109 |
'permission_callback' => array( $this, 'admin_permissions_check' ), |
| 110 |
) |
| 111 |
); |
| 112 |
|
| 113 |
// Leg-3 existence reconcile (ADR 0014 increment 5b): the authoritative LIVE {id, digest, |
| 114 |
// object_type} for one bucket's id-range — the FULL current set, not the drift subset the scan |
| 115 |
// drill-down returns. The client set-differences its manifest against this to prune stale records. |
| 116 |
register_rest_route( |
| 117 |
Api::ROUTE_NAMESPACE, |
| 118 |
'/integrity/bucket', |
| 119 |
array( |
| 120 |
'methods' => WP_REST_Server::READABLE, |
| 121 |
'callback' => array( $this, 'bucket_list' ), |
| 122 |
'permission_callback' => array( $this, 'permissions_check' ), |
| 123 |
'args' => array( |
| 124 |
'bucket' => array( |
| 125 |
'default' => 0, |
| 126 |
'sanitize_callback' => 'absint', |
| 127 |
), |
| 128 |
'bucket_size' => array( |
| 129 |
'default' => self::DEFAULT_BUCKET_SIZE, |
| 130 |
'sanitize_callback' => 'absint', |
| 131 |
), |
| 132 |
// Which id-space to walk: 'products' (default — products+variations over wp_posts) or |
| 133 |
// 'customers' (wp_users). Each collection has its own digest source + numeric id-space. |
| 134 |
'collection' => array( |
| 135 |
'default' => 'products', |
| 136 |
'sanitize_callback' => 'sanitize_key', |
| 137 |
), |
| 138 |
'status' => array( |
| 139 |
'sanitize_callback' => static function ( $status ) { |
| 140 |
return 'publish' === $status ? 'publish' : ''; |
| 141 |
}, |
| 142 |
), |
| 143 |
), |
| 144 |
) |
| 145 |
); |
| 146 |
} |
| 147 |
|
| 148 |
/** |
| 149 |
* Backfill pre-existing rows and repair stored digest drift. |
| 150 |
*/ |
| 151 |
public function rebuild( WP_REST_Request $request ): WP_REST_Response { |
| 152 |
$started = microtime( true ); |
| 153 |
$result = $this->digests->rebuild(); |
| 154 |
|
| 155 |
return rest_ensure_response( |
| 156 |
array( |
| 157 |
'collection' => 'products', |
| 158 |
'writes' => $result['writes'], |
| 159 |
'orphans_deleted' => $result['orphans_deleted'], |
| 160 |
'stored_total' => $result['stored_total'], |
| 161 |
'meta' => array( |
| 162 |
'duration_ms' => round( ( microtime( true ) - $started ) * 1000, 3 ), |
| 163 |
'rebuild_duration_ms' => $result['duration_ms'], |
| 164 |
'supported' => true, |
| 165 |
), |
| 166 |
) |
| 167 |
); |
| 168 |
} |
| 169 |
|
| 170 |
/** |
| 171 |
* The authoritative current {id, digest, object_type} for every live servable record whose id falls |
| 172 |
* in bucket [bucket*size, (bucket+1)*size). Digests come from the SAME 64-bit formula the manifest |
| 173 |
* stores ({@see Digest_Index}), so client and server compare apples-to-apples; object_type lets the |
| 174 |
* client route each pull/prune to the right lane (products and variations share the wp_posts id space). |
| 175 |
* |
| 176 |
* @return WP_Error|WP_REST_Response WP_Error (400) when the requested collection is not a supported |
| 177 |
* id-space bucket; a bucket listing otherwise. |
| 178 |
*/ |
| 179 |
public function bucket_list( WP_REST_Request $request ) { |
| 180 |
$bucket = $this->int_param( $request, 'bucket', 0, 0, PHP_INT_MAX ); |
| 181 |
$bucket_size = $this->int_param( $request, 'bucket_size', self::DEFAULT_BUCKET_SIZE, 1, 10000 ); |
| 182 |
$collection = $request->get_param( 'collection' ); |
| 183 |
$collection = \is_string( $collection ) ? $collection : 'products'; |
| 184 |
|
| 185 |
// Fail closed (review finding 7): the bucket walk only understands the id-space OWNER collections |
| 186 |
// (products — which also folds variations — customers, orders). An unsupported collection must NOT |
| 187 |
// silently fall into the products id-space via the else branch below and mis-report another |
| 188 |
// id-space as products. Reject with a 400 that names the offending collection. |
| 189 |
$supported = array_keys( Collections::with( 'digest' ) ); |
| 190 |
if ( ! \in_array( $collection, $supported, true ) ) { |
| 191 |
return new WP_Error( |
| 192 |
'woocommerce_pos_sync_unsupported_bucket_collection', |
| 193 |
\sprintf( 'integrity/bucket does not support the "%s" collection', $collection ), |
| 194 |
array( 'status' => 400 ) |
| 195 |
); |
| 196 |
} |
| 197 |
|
| 198 |
$range_start = $bucket * $bucket_size; |
| 199 |
$range_end = $range_start + $bucket_size; |
| 200 |
|
| 201 |
// Each collection has its OWN digest source + id-space (ADR 0015): products/variations over |
| 202 |
// wp_posts, customers over wp_users, orders over HPOS/CPT — the index picks the source from the |
| 203 |
// collection name. Products carry the servable scoping (readable-catalog `status` + the |
| 204 |
// POS-hidden `online_only` ids) so this authoritative list AGREES with the pull filter and the |
| 205 |
// reconcile prunes anything toggled hidden after it was pulled (ADR 0014 WP-M5). |
| 206 |
$ids = $this->index->bucket_listing( |
| 207 |
$collection, |
| 208 |
array( |
| 209 |
'start' => $range_start, |
| 210 |
'end' => $range_end, |
| 211 |
), |
| 212 |
array( 'status' => (string) $request->get_param( 'status' ) ) |
| 213 |
); |
| 214 |
|
| 215 |
return new WP_REST_Response( |
| 216 |
array( |
| 217 |
'collection' => $collection, |
| 218 |
'bucket' => $bucket, |
| 219 |
'bucket_size' => $bucket_size, |
| 220 |
'range' => array( |
| 221 |
'start' => $range_start, |
| 222 |
'end' => $range_end, |
| 223 |
), |
| 224 |
'ids' => $ids, |
| 225 |
), |
| 226 |
200 |
| 227 |
); |
| 228 |
} |
| 229 |
|
| 230 |
/** |
| 231 |
* GET /integrity/scan?bucket_size=&after_id=&limit_buckets=[&bucket=]. |
| 232 |
* |
| 233 |
* Bucket aggregate: BIT_XOR over per-row CRC32 digests, deliberately |
| 234 |
* instead of MD5(GROUP_CONCAT(... ORDER BY id)). XOR is commutative and |
| 235 |
* associative, so the aggregate needs no ORDER BY and carries no |
| 236 |
* group_concat_max_len truncation hazard (which the plain range-checksum |
| 237 |
* endpoint must patch per session); its state is a constant-size integer |
| 238 |
* regardless of bucket population. Collision properties are |
| 239 |
* detection-grade, not cryptographic: a drifted row escapes only on a |
| 240 |
* 2^-32 per-row CRC collision, and two simultaneous drifts in one bucket |
| 241 |
* cancel only when their XOR deltas are exactly equal — and the |
| 242 |
* drill-down re-verifies per id before anything is acted on. Bucket |
| 243 |
* record counts are compared alongside, so add/delete imbalances that |
| 244 |
* could cancel in XOR still flag. |
| 245 |
*/ |
| 246 |
public function scan( WP_REST_Request $request ) { |
| 247 |
$started = microtime( true ); |
| 248 |
$bucket_size = $this->int_param( $request, 'bucket_size', self::DEFAULT_BUCKET_SIZE, 1, 10000 ); |
| 249 |
$bucket_raw = $request->get_param( 'bucket' ); |
| 250 |
$collection = $request->get_param( 'collection' ); |
| 251 |
$collection = \is_string( $collection ) ? $collection : 'products'; |
| 252 |
|
| 253 |
if ( ! \in_array( $collection, array_keys( Collections::with( 'digest' ) ), true ) ) { |
| 254 |
return new WP_Error( |
| 255 |
'woocommerce_pos_sync_unsupported_scan_collection', |
| 256 |
\sprintf( 'integrity/scan does not support the "%s" collection', $collection ), |
| 257 |
array( 'status' => 400 ) |
| 258 |
); |
| 259 |
} |
| 260 |
|
| 261 |
if ( null !== $bucket_raw && '' !== $bucket_raw ) { |
| 262 |
if ( 'products' !== $collection ) { |
| 263 |
return new WP_Error( |
| 264 |
'woocommerce_pos_sync_unsupported_scan_drill_down_collection', |
| 265 |
\sprintf( 'integrity/scan drill-down does not support the "%s" collection', $collection ), |
| 266 |
array( 'status' => 400 ) |
| 267 |
); |
| 268 |
} |
| 269 |
$bucket = max( 0, (int) $bucket_raw ); |
| 270 |
if ( $this->maybe_schedule_digest_rebuild() ) { |
| 271 |
return rest_ensure_response( |
| 272 |
$this->envelope( |
| 273 |
array( |
| 274 |
'bucket_size' => $bucket_size, |
| 275 |
'bucket' => $bucket, |
| 276 |
), |
| 277 |
array(), |
| 278 |
true, |
| 279 |
$started, |
| 280 |
'drill-down: per-id stored-vs-current digest mismatches in one bucket.', |
| 281 |
true |
| 282 |
) |
| 283 |
); |
| 284 |
} |
| 285 |
|
| 286 |
return $this->drill_down( $bucket, $bucket_size, $started ); |
| 287 |
} |
| 288 |
|
| 289 |
$after_id = max( 0, (int) ( $request->get_param( 'after_id' ) ?? 0 ) ); |
| 290 |
$limit_buckets = $this->int_param( $request, 'limit_buckets', self::DEFAULT_LIMIT_BUCKETS, 1, 1000 ); |
| 291 |
|
| 292 |
// after_id pagination: the checkpoint is the last id covered by the |
| 293 |
// previous window, so the next window starts at the following |
| 294 |
// bucket boundary. Both SQL sides share the same [start, end) id |
| 295 |
// window, so they always agree on which buckets are in scope. |
| 296 |
$first_bucket = $after_id > 0 ? ( (int) floor( $after_id / $bucket_size ) ) + 1 : 0; |
| 297 |
$window_start = $first_bucket * $bucket_size; |
| 298 |
$window_end = ( $first_bucket + $limit_buckets ) * $bucket_size; |
| 299 |
|
| 300 |
if ( 'products' === $collection && $this->maybe_schedule_digest_rebuild() ) { |
| 301 |
return rest_ensure_response( |
| 302 |
$this->envelope( |
| 303 |
array( |
| 304 |
'bucket_size' => $bucket_size, |
| 305 |
'after_id' => $window_end - 1, |
| 306 |
), |
| 307 |
array(), |
| 308 |
true, |
| 309 |
$started, |
| 310 |
'stored hook-time digests vs current raw-row digests, BIT_XOR(64-bit MD5-derived) per bucket; mismatch = content changed without hooks (or stored side not yet backfilled).', |
| 311 |
true |
| 312 |
) |
| 313 |
); |
| 314 |
} |
| 315 |
|
| 316 |
$filters = array( 'status' => (string) $request->get_param( 'status' ) ); |
| 317 |
$ttl = (int) apply_filters( 'woocommerce_pos_integrity_scan_cache_ttl', 120 ); |
| 318 |
$key_parts = array( $collection, $bucket_size, $filters, $window_start, $window_end ); |
| 319 |
$cache_key = 'wcpos_integrity_scan_' . md5( (string) wp_json_encode( $key_parts ) ); |
| 320 |
$cached = 0 < $ttl ? get_transient( $cache_key ) : false; |
| 321 |
|
| 322 |
// Both sides in one call: the stored hook-time aggregate and the current |
| 323 |
// raw-row aggregate over the SAME id window, plus the max id completion is |
| 324 |
// judged against (the larger of both sides, so orphaned stored digests past |
| 325 |
// the last live post still get scanned). |
| 326 |
$aggregates = false !== $cached ? $cached : $this->index->bucket_aggregates( |
| 327 |
array( |
| 328 |
'bucket_size' => $bucket_size, |
| 329 |
'start' => $window_start, |
| 330 |
'end' => $window_end, |
| 331 |
), |
| 332 |
$collection, |
| 333 |
$filters |
| 334 |
); |
| 335 |
if ( 0 < $ttl && false === $cached ) { |
| 336 |
set_transient( $cache_key, $aggregates, $ttl ); |
| 337 |
} |
| 338 |
|
| 339 |
return rest_ensure_response( |
| 340 |
$this->envelope( |
| 341 |
array( |
| 342 |
'bucket_size' => $bucket_size, |
| 343 |
'after_id' => $window_end - 1, |
| 344 |
), |
| 345 |
$aggregates['buckets'], |
| 346 |
$window_end > $aggregates['max_id'], |
| 347 |
$started, |
| 348 |
'stored hook-time digests vs current raw-row digests, BIT_XOR(64-bit MD5-derived) per bucket; mismatch = content changed without hooks (or stored side not yet backfilled).', |
| 349 |
false, |
| 350 |
$collection |
| 351 |
) |
| 352 |
); |
| 353 |
} |
| 354 |
|
| 355 |
/** |
| 356 |
* ?bucket=<n>: per-id comparison inside one bucket. Three mismatch |
| 357 |
* shapes: changed (both sides present, digests differ), missing_stored |
| 358 |
* (live row never digested — created without hooks or pre-backfill), |
| 359 |
* deleted (stored digest whose row is gone — hook-bypassing delete). |
| 360 |
*/ |
| 361 |
private function drill_down( int $bucket, int $bucket_size, float $started ) { |
| 362 |
$range_start = $bucket * $bucket_size; |
| 363 |
$range_end = $range_start + $bucket_size; |
| 364 |
|
| 365 |
$rows = $this->index->bucket_drift( |
| 366 |
array( |
| 367 |
'start' => $range_start, |
| 368 |
'end' => $range_end, |
| 369 |
) |
| 370 |
); |
| 371 |
|
| 372 |
// Each drifted id carries its collection (the hash-checksum id-space holds |
| 373 |
// BOTH products and variations) so the host pulls the right path instead of |
| 374 |
// assuming 'products' (ADR 0005 — engine reads DriftedId.collection). |
| 375 |
$changes = array(); |
| 376 |
foreach ( $rows as $row ) { |
| 377 |
$collection = Collections::collection_for_object_type( (string) ( $row['object_type'] ?? '' ) ); |
| 378 |
if ( null === $collection ) { |
| 379 |
// Fail closed (#421 increment 5): the digest SQL constrains |
| 380 |
// object_type to this id-space's own types, so this cannot |
| 381 |
// happen live — but an unknown value must never masquerade as |
| 382 |
// 'products' and mis-route a pull. |
| 383 |
Logger::log( \sprintf( 'WCPOS sync: dropped drifted-id row with unknown object_type "%s" (id %d)', (string) ( $row['object_type'] ?? '' ), (int) $row['id'] ) ); |
| 384 |
|
| 385 |
continue; |
| 386 |
} |
| 387 |
$changes[] = array( |
| 388 |
'id' => (int) $row['id'], |
| 389 |
'status' => (string) $row['status'], |
| 390 |
'collection' => $collection, |
| 391 |
// Unsigned 64-bit (ADR 0014 M1): keep as strings — a (int) cast (and JS Number) can't |
| 392 |
// hold values above PHP_INT_MAX / 2^53 without precision loss. |
| 393 |
'stored_digest' => null === $row['stored_digest'] ? null : (string) $row['stored_digest'], |
| 394 |
'current_digest' => null === $row['current_digest'] ? null : (string) $row['current_digest'], |
| 395 |
); |
| 396 |
} |
| 397 |
|
| 398 |
$this->maybe_schedule_stale_digest_rebuild( $bucket, $bucket_size, $changes ); |
| 399 |
|
| 400 |
return rest_ensure_response( |
| 401 |
$this->envelope( |
| 402 |
array( |
| 403 |
'bucket_size' => $bucket_size, |
| 404 |
'bucket' => $bucket, |
| 405 |
), |
| 406 |
$changes, |
| 407 |
true, |
| 408 |
$started, |
| 409 |
'drill-down: per-id stored-vs-current digest mismatches in one bucket.' |
| 410 |
) |
| 411 |
); |
| 412 |
} |
| 413 |
|
| 414 |
/** |
| 415 |
* Schedule one guarded rebuild when product digests are empty or use an old formula. |
| 416 |
*/ |
| 417 |
private function maybe_schedule_digest_rebuild(): bool { |
| 418 |
$needs_rebuild = $this->index->needs_product_rebuild(); |
| 419 |
$stored_fingerprint = (string) get_option( Digest_Index::FORMULA_FP_OPTION, '' ); |
| 420 |
if ( '' === $stored_fingerprint ) { |
| 421 |
$stored_fingerprint = Digest_Index::legacy_formula_fingerprint(); |
| 422 |
update_option( Digest_Index::FORMULA_FP_OPTION, $stored_fingerprint, false ); |
| 423 |
} |
| 424 |
|
| 425 |
if ( ! $needs_rebuild && Digest_Index::digest_formula_fingerprint() === $stored_fingerprint ) { |
| 426 |
return false; |
| 427 |
} |
| 428 |
|
| 429 |
$this->save_drift_streaks( array() ); |
| 430 |
$this->schedule_guarded_rebuild(); |
| 431 |
|
| 432 |
return true; |
| 433 |
} |
| 434 |
|
| 435 |
/** |
| 436 |
* Schedule the rebuild when a bucket keeps reporting the SAME stale stored |
| 437 |
* digests drill-down after drill-down. |
| 438 |
* |
| 439 |
* A 'changed' row means the stored side has a digest that no longer matches |
| 440 |
* the row — i.e. a write that bypassed the hooks (bulk import, WP-CLI, |
| 441 |
* direct SQL, a migration plugin). Pulling cannot fix that: the client |
| 442 |
* refreshes its own copy, but nothing rewrites the STORED digest, so the |
| 443 |
* bucket mismatches forever and the till shows a permanent "records need |
| 444 |
* attention" for data that is already correct. Observed on dev-pro |
| 445 |
* 2026-08-19: 138 products drifted by a hookless bulk edit, re-escalated |
| 446 |
* every sweep, local copies byte-identical to the server. |
| 447 |
* |
| 448 |
* Why a STREAK and not the first sight of drift: for a hookless write the |
| 449 |
* integrity scan is the ONLY signal — such a write bypasses the sequence |
| 450 |
* log too — so re-baselining on first detection would erase the one thing |
| 451 |
* telling clients to re-pull. The client drills a mismatched bucket every |
| 452 |
* sweep and issues targeted pulls each time; it escalates at 2 consecutive |
| 453 |
* post-pull mismatches (DEFAULT_HYBRID_POLICY.escalateToRevisionHashAfter). |
| 454 |
* Rebuilding at 3 therefore only ever fires AFTER the client has pulled and |
| 455 |
* still sees the mismatch — the point at which the drift is provably the |
| 456 |
* stored side's problem, not a delivery problem. |
| 457 |
* |
| 458 |
* @param int $bucket Bucket just drilled down. |
| 459 |
* @param int $bucket_size Number of ids covered by the bucket. |
| 460 |
* @param array $changes Rows the drill-down is returning. |
| 461 |
*/ |
| 462 |
private function maybe_schedule_stale_digest_rebuild( int $bucket, int $bucket_size, array $changes ): void { |
| 463 |
$stale = 0; |
| 464 |
foreach ( $changes as $change ) { |
| 465 |
if ( 'changed' === ( $change['status'] ?? '' ) ) { |
| 466 |
++$stale; |
| 467 |
} |
| 468 |
} |
| 469 |
|
| 470 |
$streaks = get_option( self::DRIFT_STREAK_OPTION, array() ); |
| 471 |
if ( ! \is_array( $streaks ) ) { |
| 472 |
$streaks = array(); |
| 473 |
} |
| 474 |
|
| 475 |
$streak_key = $bucket_size . ':' . $bucket; |
| 476 |
|
| 477 |
if ( 0 === $stale ) { |
| 478 |
// Reconciled (or only deletions/missing_stored left) — forget it. |
| 479 |
if ( isset( $streaks[ $streak_key ] ) ) { |
| 480 |
unset( $streaks[ $streak_key ] ); |
| 481 |
$this->save_drift_streaks( $streaks ); |
| 482 |
} |
| 483 |
|
| 484 |
return; |
| 485 |
} |
| 486 |
|
| 487 |
$streak = ( (int) ( $streaks[ $streak_key ] ?? 0 ) ) + 1; |
| 488 |
|
| 489 |
if ( $streak < self::DRIFT_REBUILD_THRESHOLD ) { |
| 490 |
$streaks[ $streak_key ] = $streak; |
| 491 |
$this->save_drift_streaks( $streaks ); |
| 492 |
|
| 493 |
return; |
| 494 |
} |
| 495 |
|
| 496 |
Logger::log( |
| 497 |
\sprintf( |
| 498 |
'WCPOS sync: bucket %d (size %d) reported %d stale stored digest(s) on %d consecutive drill-downs; scheduling an integrity digest rebuild.', |
| 499 |
$bucket, |
| 500 |
$bucket_size, |
| 501 |
$stale, |
| 502 |
$streak |
| 503 |
) |
| 504 |
); |
| 505 |
|
| 506 |
// The rebuild re-digests the whole product space, so every bucket's |
| 507 |
// streak is moot once it is queued. |
| 508 |
$this->save_drift_streaks( array() ); |
| 509 |
$this->schedule_guarded_rebuild(); |
| 510 |
} |
| 511 |
|
| 512 |
/** |
| 513 |
* Persist the per-bucket streak map, bounded so a pathological catalogue |
| 514 |
* cannot grow an unbounded option. Never autoloaded — it is read only on |
| 515 |
* the drill-down path. |
| 516 |
* |
| 517 |
* @param array $streaks Bucket-size:bucket => consecutive drifted drill-downs. |
| 518 |
*/ |
| 519 |
private function save_drift_streaks( array $streaks ): void { |
| 520 |
if ( \count( $streaks ) > self::DRIFT_STREAK_MAX_BUCKETS ) { |
| 521 |
arsort( $streaks ); |
| 522 |
$streaks = \array_slice( $streaks, 0, self::DRIFT_STREAK_MAX_BUCKETS, true ); |
| 523 |
} |
| 524 |
|
| 525 |
if ( array() === $streaks ) { |
| 526 |
delete_option( self::DRIFT_STREAK_OPTION ); |
| 527 |
|
| 528 |
return; |
| 529 |
} |
| 530 |
|
| 531 |
update_option( self::DRIFT_STREAK_OPTION, $streaks, false ); |
| 532 |
} |
| 533 |
|
| 534 |
/** |
| 535 |
* Queue one rebuild behind the owner-token lease. Shared by both triggers. |
| 536 |
*/ |
| 537 |
private function schedule_guarded_rebuild(): void { |
| 538 |
if ( false === get_transient( Integrity_Digest::REBUILD_LOCK ) ) { |
| 539 |
// Owner-token lease: a rebuild outliving the TTL must not delete a |
| 540 |
// SUCCESSOR's lock in its finally (the callback captures the token |
| 541 |
// at start and only releases a matching lease). |
| 542 |
$token = uniqid( 'wcpos_rebuild_', true ); |
| 543 |
set_transient( Integrity_Digest::REBUILD_LOCK, $token, Integrity_Digest::REBUILD_LOCK_TTL ); |
| 544 |
if ( wp_next_scheduled( Integrity_Digest::REBUILD_HOOK ) ) { |
| 545 |
// An identical event is already queued (e.g. a concurrent scan |
| 546 |
// won the race, or a prior lock expired before cron fired): |
| 547 |
// keep the fresh lease and do not stack another event. |
| 548 |
return; |
| 549 |
} |
| 550 |
if ( false === wp_schedule_single_event( time(), Integrity_Digest::REBUILD_HOOK ) ) { |
| 551 |
if ( get_transient( Integrity_Digest::REBUILD_LOCK ) === $token ) { |
| 552 |
delete_transient( Integrity_Digest::REBUILD_LOCK ); |
| 553 |
} |
| 554 |
Logger::error( 'WCPOS sync: failed to schedule integrity digest rebuild.' ); |
| 555 |
} |
| 556 |
} |
| 557 |
} |
| 558 |
|
| 559 |
/** |
| 560 |
* Same envelope shape as class-changes-controller.php so the matrix client plumbing stays uniform. |
| 561 |
*/ |
| 562 |
private function envelope( array $checkpoint, array $changes, bool $complete, float $started, ?string $note = null, bool $rebuilding = false, string $collection = 'products' ): array { |
| 563 |
$meta = array( |
| 564 |
'duration_ms' => round( ( microtime( true ) - $started ) * 1000, 3 ), |
| 565 |
'supported' => true, |
| 566 |
); |
| 567 |
if ( null !== $note ) { |
| 568 |
$meta['note'] = $note; |
| 569 |
} |
| 570 |
if ( $rebuilding ) { |
| 571 |
$meta['rebuilding'] = true; |
| 572 |
} |
| 573 |
|
| 574 |
return array( |
| 575 |
'collection' => $collection, |
| 576 |
'checkpoint' => $checkpoint, |
| 577 |
'changes' => $changes, |
| 578 |
'complete' => $complete, |
| 579 |
'meta' => $meta, |
| 580 |
); |
| 581 |
} |
| 582 |
} |
| 583 |
|
