Code Coverage |
||||||||||
Lines |
Functions and Methods |
Classes and Traits |
||||||||
| Total | |
91.69% |
706 / 770 |
|
67.35% |
33 / 49 |
CRAP | |
0.00% |
0 / 1 |
| WANObjectCache | |
91.81% |
706 / 769 |
|
67.35% |
33 / 49 |
241.65 | |
0.00% |
0 / 1 |
| __construct | |
100.00% |
11 / 11 |
|
100.00% |
1 / 1 |
2 | |||
| setLogger | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| newEmpty | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| get | |
100.00% |
19 / 19 |
|
100.00% |
1 / 1 |
4 | |||
| getMulti | |
95.65% |
22 / 23 |
|
0.00% |
0 / 1 |
5 | |||
| fetchKeys | |
96.92% |
63 / 65 |
|
0.00% |
0 / 1 |
17 | |||
| processCheckKeys | |
100.00% |
15 / 15 |
|
100.00% |
1 / 1 |
4 | |||
| set | |
100.00% |
21 / 21 |
|
100.00% |
1 / 1 |
2 | |||
| setMainValue | |
97.44% |
76 / 78 |
|
0.00% |
0 / 1 |
22 | |||
| delete | |
100.00% |
18 / 18 |
|
100.00% |
1 / 1 |
3 | |||
| getCheckKeyTime | |
100.00% |
2 / 2 |
|
100.00% |
1 / 1 |
1 | |||
| getMultiCheckKeyTime | |
100.00% |
20 / 20 |
|
100.00% |
1 / 1 |
4 | |||
| touchCheckKey | |
100.00% |
16 / 16 |
|
100.00% |
1 / 1 |
2 | |||
| resetCheckKey | |
100.00% |
9 / 9 |
|
100.00% |
1 / 1 |
2 | |||
| getWithSetCallback | |
100.00% |
24 / 24 |
|
100.00% |
1 / 1 |
8 | |||
| fetchOrRegenerate | |
100.00% |
141 / 141 |
|
100.00% |
1 / 1 |
30 | |||
| makeSisterKey | |
100.00% |
4 / 4 |
|
100.00% |
1 / 1 |
2 | |||
| getInterimValue | |
66.67% |
8 / 12 |
|
0.00% |
0 / 1 |
5.93 | |||
| setInterimValue | |
91.67% |
11 / 12 |
|
0.00% |
0 / 1 |
2.00 | |||
| getMultiWithSetCallback | |
85.19% |
23 / 27 |
|
0.00% |
0 / 1 |
3.03 | |||
| getMultiWithUnionSetCallback | |
84.21% |
48 / 57 |
|
0.00% |
0 / 1 |
10.39 | |||
| makeGlobalKey | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| makeKey | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| makeMultiKeys | |
100.00% |
10 / 10 |
|
100.00% |
1 / 1 |
4 | |||
| multiRemap | |
40.00% |
2 / 5 |
|
0.00% |
0 / 1 |
4.94 | |||
| watchErrors | |
0.00% |
0 / 1 |
|
0.00% |
0 / 1 |
2 | |||
| getLastError | |
0.00% |
0 / 5 |
|
0.00% |
0 / 1 |
30 | |||
| clearProcessCache | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| useInterimHoldOffCaching | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| getQoS | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| adaptiveTTL | |
100.00% |
5 / 5 |
|
100.00% |
1 / 1 |
2 | |||
| getWarmupKeyMisses | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
1 | |||
| getRouteKey | |
80.00% |
4 / 5 |
|
0.00% |
0 / 1 |
3.07 | |||
| scheduleAsyncRefresh | |
53.33% |
8 / 15 |
|
0.00% |
0 / 1 |
3.91 | |||
| isAcceptablyFreshValue | |
100.00% |
11 / 11 |
|
100.00% |
1 / 1 |
4 | |||
| isLotteryRefreshDue | |
100.00% |
7 / 7 |
|
100.00% |
1 / 1 |
2 | |||
| worthRefreshPopular | |
54.55% |
6 / 11 |
|
0.00% |
0 / 1 |
7.35 | |||
| worthRefreshExpiring | |
100.00% |
9 / 9 |
|
100.00% |
1 / 1 |
5 | |||
| isValid | |
100.00% |
1 / 1 |
|
100.00% |
1 / 1 |
2 | |||
| wrap | |
100.00% |
9 / 9 |
|
100.00% |
1 / 1 |
2 | |||
| unwrap | |
100.00% |
28 / 28 |
|
100.00% |
1 / 1 |
6 | |||
| determineKeyGroupForStats | |
100.00% |
2 / 2 |
|
100.00% |
1 / 1 |
1 | |||
| parsePurgeValue | |
90.91% |
10 / 11 |
|
0.00% |
0 / 1 |
5.02 | |||
| makeCheckPurgeValue | |
100.00% |
3 / 3 |
|
100.00% |
1 / 1 |
1 | |||
| getProcessCache | |
100.00% |
6 / 6 |
|
100.00% |
1 / 1 |
3 | |||
| getNonProcessCachedMultiKeys | |
100.00% |
8 / 8 |
|
100.00% |
1 / 1 |
5 | |||
| fetchWrappedValuesForWarmupCache | |
69.23% |
9 / 13 |
|
0.00% |
0 / 1 |
7.05 | |||
| timeSinceLoggedMiss | |
100.00% |
4 / 4 |
|
100.00% |
1 / 1 |
3 | |||
| getCurrentTime | n/a |
0 / 0 |
n/a |
0 / 0 |
2 | |||||
| setMockTime | n/a |
0 / 0 |
n/a |
0 / 0 |
2 | |||||
| startOperationSpan | |
27.78% |
5 / 18 |
|
0.00% |
0 / 1 |
14.42 | |||
| 1 | <?php |
| 2 | /** |
| 3 | * @license GPL-2.0-or-later |
| 4 | * @file |
| 5 | */ |
| 6 | |
| 7 | namespace Wikimedia\ObjectCache; |
| 8 | |
| 9 | use ArrayIterator; |
| 10 | use Closure; |
| 11 | use Exception; |
| 12 | use Psr\Log\LoggerAwareInterface; |
| 13 | use Psr\Log\LoggerInterface; |
| 14 | use Psr\Log\NullLogger; |
| 15 | use RuntimeException; |
| 16 | use UnexpectedValueException; |
| 17 | use Wikimedia\LightweightObjectStore\ExpirationAwareness; |
| 18 | use Wikimedia\Stats\StatsFactory; |
| 19 | use Wikimedia\Telemetry\NoopTracer; |
| 20 | use Wikimedia\Telemetry\SpanInterface; |
| 21 | use Wikimedia\Telemetry\TracerInterface; |
| 22 | |
| 23 | /** |
| 24 | * Multi-datacenter aware caching interface |
| 25 | * |
| 26 | * ### Using WANObjectCache |
| 27 | * |
| 28 | * %WANObjectCache (known as **WANCache**, pronounced whan-cache) improves performance |
| 29 | * by reducing database load, increasing web server capacity (fewer repeated computations) and |
| 30 | * providing faster access to data. The data cached here follows a "cache-aside" strategy, with |
| 31 | * data potentially derived from database rows. Generally speaking, cache data should be treated |
| 32 | * as equally up-to-date to data from a replica database, and is thus essentially subject to the |
| 33 | * same replication lag. |
| 34 | * |
| 35 | * The primary way to interact with this class is via the getWithSetCallback() method. |
| 36 | * |
| 37 | * Each data center has its own cache cluster, with web servers in a given datacenter |
| 38 | * populating and reading from the local datacenter only. The exceptions are methods delete(), |
| 39 | * touchCheckKey(), and resetCheckKey(), which also asynchronously broadcast the equivalent |
| 40 | * purge to other datacenters. |
| 41 | * |
| 42 | * To learn how this is used and configured at Wikimedia Foundation, |
| 43 | * refer to <https://wikitech.wikimedia.org/wiki/Memcached_for_MediaWiki>. |
| 44 | * |
| 45 | * For broader guidance on how to approach caching in MediaWiki at scale, |
| 46 | * refer to <https://wikitech.wikimedia.org/wiki/MediaWiki_Engineering/Guides/Backend_performance_practices>. |
| 47 | * |
| 48 | * For your code to "see" new values in a timely manner, you need to follow either the |
| 49 | * validation strategy, or the purge strategy. |
| 50 | * |
| 51 | * #### Strategy 1: Validation |
| 52 | * |
| 53 | * The validation strategy refers to the natural avoidance of stale data |
| 54 | * by one of the following means: |
| 55 | * |
| 56 | * - A) The cached value is immutable. |
| 57 | * |
| 58 | * If you can obtain all the information needed to uniquely describe the value, |
| 59 | * then the value never has to change or be purged. Instead, the key changes, |
| 60 | * which naturally creates a miss where you can compute the right value. |
| 61 | * For example, a transformation like parsing or transforming some input, |
| 62 | * could have a cache key like `example-myparser, option-xyz, v2, hash1234` |
| 63 | * which would describe the transformation, the version/parameters, and a hash |
| 64 | * of the exact input. |
| 65 | * |
| 66 | * This also naturally avoids oscillation or corruption in the context of multiple |
| 67 | * servers and data centers, where your code may not always be running the same version |
| 68 | * everywhere at the same time. Newer code would have its own set of cache keys, |
| 69 | * ensuring a deterministic outcome. |
| 70 | * - B) The value is cached with a low TTL. |
| 71 | * |
| 72 | * If you can tolerate a few seconds or minutes of delay before changes are reflected |
| 73 | * in the way your data is used, and if re-computation is quick, you can consider |
| 74 | * caching it with a "blind" TTL – using the value's age as your method of validation. |
| 75 | * - C) Validity is checked against an external source. |
| 76 | * |
| 77 | * Perhaps you prefer to utilize the old data as fallback or to help compute the new |
| 78 | * value, or for other reasons you need to have a stable key across input changes |
| 79 | * (e.g. cache by page title instead of revision ID). If you put the variable identifier |
| 80 | * (e.g. input hash, or revision ID) in the cache value, and validate this on retrieval |
| 81 | * then you don't need purging or expiration. |
| 82 | * |
| 83 | * After calling get() you can validate the ID inside the cached value against what |
| 84 | * you know. When needed, recompute the value and call set(). |
| 85 | * |
| 86 | * #### Strategy 2: Purge |
| 87 | * |
| 88 | * The purge strategy refers to the approach whereby your application knows that source |
| 89 | * data has changed and can react by purging the relevant cache keys. |
| 90 | * The simplest purge method is delete(). |
| 91 | * |
| 92 | * Note that cache updates and purges are not immediately visible to all application servers in |
| 93 | * all data centers. The cache should be treated like a replica database in this regard. |
| 94 | * If immediate synchronization is required, then solutions must be sought outside WANCache. |
| 95 | * |
| 96 | * Write operations like delete() and the "set" part of getWithSetCallback(), may return true as |
| 97 | * soon as the command has been sent or buffered to an open connection to the cache cluster. |
| 98 | * It will be processed and/or broadcasted asynchronously. |
| 99 | * |
| 100 | * @anchor wanobjectcache-deployment |
| 101 | * ### Deploying WANObjectCache |
| 102 | * |
| 103 | * There are two supported ways for sysadmins to set up multi-DC cache purging: |
| 104 | * |
| 105 | * - A) Set up mcrouter as the cache backend, with a memcached BagOStuff class for the 'cache' |
| 106 | * parameter, and a wildcard routing prefix for the 'broadcastRoutingPrefix' parameter. |
| 107 | * Configure mcrouter as follows: |
| 108 | * - Define a "<datacenter>" pool of memcached servers for each datacenter. |
| 109 | * - Define a "<datacenter>/wan" route to each datacenter, using "AllSyncRoute" for the |
| 110 | * routes that go to the local datacenter pool and "AllAsyncRoute" for the routes that |
| 111 | * go to remote datacenter pools. The child routes should use "HashRoute|<datacenter>". |
| 112 | * This allows for the use of a wildcard route for 'broadcastRoutingPrefix'. See |
| 113 | * https://github.com/facebook/mcrouter/wiki/Routing-Prefix and |
| 114 | * https://github.com/facebook/mcrouter/wiki/Multi-cluster-broadcast-setup. |
| 115 | * - In order to reroute operations from "down" servers to spare ("gutter") servers, use |
| 116 | * "FailoverWithExptimeRoute" (failover_exptime=60) instead of "HashRoute|<datacenter>" |
| 117 | * in the "AllSyncRoute"/"AllAsyncRoute" child routes. |
| 118 | * The "gutter" pool is a set of memcached servers that only handle failover traffic. |
| 119 | * Such servers should be carefully spread over different rows and racks. See |
| 120 | * https://github.com/facebook/mcrouter/wiki/List-of-Route-Handles#failoverroute |
| 121 | * - B) Set up dynomite as the cache backend, using a memcached BagOStuff class for the 'cache' |
| 122 | * parameter. Note that with this setup, all key setting operations will be broadcasted, |
| 123 | * rather than just purges. Writes will be eventually consistent via the Dynamo replication |
| 124 | * model. See https://github.com/Netflix/dynomite. |
| 125 | * |
| 126 | * Broadcasted operations like delete() and touchCheckKey() are intended to run |
| 127 | * immediately in the local datacenter and asynchronously in remote datacenters. |
| 128 | * |
| 129 | * This means that callers in all datacenters may see older values for however many |
| 130 | * milliseconds that the purge took to reach that datacenter. As with any cache, this |
| 131 | * should not be relied on for cases where reads are used to determine writes to source |
| 132 | * (e.g. non-cache) data stores, except when reading immutable data. |
| 133 | * |
| 134 | * Internally, access to a given key actually involves the use of one or more "sister" keys. |
| 135 | * A sister key is constructed by prefixing the base key with "WANCache:" (used to distinguish |
| 136 | * WANObjectCache formatted keys) and suffixing a colon followed by a single-character sister |
| 137 | * key type. The sister key types include the following: |
| 138 | * |
| 139 | * - `v`: used to store "regular" values (metadata-wrapped) and temporary purge "tombstones". |
| 140 | * - `t`: used to store "last purge" timestamps for "check" keys. |
| 141 | * - `m`: used to store temporary mutex locks to avoid cache stampedes. |
| 142 | * - `i`: used to store temporary interim values (metadata-wrapped) for tombstoned keys. |
| 143 | * |
| 144 | * @ingroup Cache |
| 145 | * @newable |
| 146 | * @since 1.26 |
| 147 | */ |
| 148 | class WANObjectCache implements |
| 149 | ExpirationAwareness, |
| 150 | IStoreKeyEncoder, |
| 151 | LoggerAwareInterface |
| 152 | { |
| 153 | /** @var BagOStuff The local datacenter cache */ |
| 154 | protected $cache; |
| 155 | /** @var MapCacheLRU[] Map of group PHP instance caches */ |
| 156 | protected $processCaches = []; |
| 157 | /** @var LoggerInterface */ |
| 158 | protected $logger; |
| 159 | /** @var StatsFactory */ |
| 160 | protected $stats; |
| 161 | /** @var callable|null Function that takes a WAN cache callback and runs it later */ |
| 162 | protected $asyncHandler; |
| 163 | |
| 164 | /** |
| 165 | * Routing prefix for operations that should be broadcasted to all data centers. |
| 166 | * |
| 167 | * If null, the there is only one datacenter or a backend proxy broadcasts everything. |
| 168 | * |
| 169 | * @var string|null |
| 170 | */ |
| 171 | protected $broadcastRoute; |
| 172 | /** @var bool Whether to use "interim" caching while keys are tombstoned */ |
| 173 | protected $useInterimHoldOffCaching = true; |
| 174 | /** @var float Unix timestamp of the oldest possible valid values */ |
| 175 | protected $epoch; |
| 176 | /** @var int Scheme to use for key coalescing (Hash Tags or Hash Stops) */ |
| 177 | protected $coalesceScheme; |
| 178 | |
| 179 | /** @var TracerInterface */ |
| 180 | private $tracer; |
| 181 | |
| 182 | /** @var array<int,array> List of (key, UNIX timestamp) tuples for get() cache misses */ |
| 183 | private $missLog; |
| 184 | |
| 185 | /** @var int Callback stack depth for getWithSetCallback() */ |
| 186 | private $callbackDepth = 0; |
| 187 | /** @var mixed[] Temporary warm-up cache */ |
| 188 | private $warmupCache = []; |
| 189 | /** @var int Key fetched */ |
| 190 | private $warmupKeyMisses = 0; |
| 191 | |
| 192 | /** @var float|null */ |
| 193 | private $wallClockOverride; |
| 194 | |
| 195 | /** Max expected seconds to pass between delete() and DB commit finishing */ |
| 196 | private const MAX_COMMIT_DELAY = 3; |
| 197 | /** Max expected seconds of combined lag from replication and "view snapshots" */ |
| 198 | private const MAX_READ_LAG = 7; |
| 199 | /** Seconds to tombstone keys on delete() and to treat keys as volatile after purges */ |
| 200 | public const HOLDOFF_TTL = self::MAX_COMMIT_DELAY + self::MAX_READ_LAG + 1; |
| 201 | |
| 202 | /** Consider regeneration if the key will expire within this many seconds */ |
| 203 | private const LOW_TTL = 60; |
| 204 | /** Max TTL, in seconds, to store keys when a data source has high replication lag */ |
| 205 | public const TTL_LAGGED = 30; |
| 206 | |
| 207 | /** Expected time-till-refresh, in seconds, if the key is accessed once per second */ |
| 208 | private const HOT_TTR = 900; |
| 209 | /** Minimum key age, in seconds, for expected time-till-refresh to be considered */ |
| 210 | private const AGE_NEW = 60; |
| 211 | |
| 212 | /** Idiom for getWithSetCallback() meaning "no cache stampede mutex" */ |
| 213 | private const TSE_NONE = -1; |
| 214 | |
| 215 | /** Idiom for set()/getWithSetCallback() meaning "no post-expiration persistence" */ |
| 216 | public const STALE_TTL_NONE = 0; |
| 217 | /** Idiom for set()/getWithSetCallback() meaning "no post-expiration grace period" */ |
| 218 | public const GRACE_TTL_NONE = 0; |
| 219 | /** Idiom for delete()/touchCheckKey() meaning "no hold-off period" */ |
| 220 | public const HOLDOFF_TTL_NONE = 0; |
| 221 | |
| 222 | /** @var float Idiom for getWithSetCallback() meaning "no minimum required as-of timestamp" */ |
| 223 | public const MIN_TIMESTAMP_NONE = 0.0; |
| 224 | |
| 225 | /** Default process cache name and max key count */ |
| 226 | private const PC_PRIMARY = 'primary:1000'; |
| 227 | |
| 228 | /** Idiom for get()/getMulti() to return extra information by reference */ |
| 229 | public const PASS_BY_REF = []; |
| 230 | |
| 231 | /** Use twemproxy-style Hash Tag key scheme (e.g. "{...}") */ |
| 232 | private const SCHEME_HASH_TAG = 1; |
| 233 | /** Use mcrouter-style Hash Stop key scheme (e.g. "...|#|") */ |
| 234 | private const SCHEME_HASH_STOP = 2; |
| 235 | |
| 236 | /** Seconds to keep dependency purge keys around */ |
| 237 | private const CHECK_KEY_TTL = self::TTL_YEAR; |
| 238 | /** Seconds to keep interim value keys for tombstoned keys around */ |
| 239 | private const INTERIM_KEY_TTL = 2; |
| 240 | |
| 241 | /** Seconds to keep lock keys around */ |
| 242 | private const LOCK_TTL = 10; |
| 243 | /** Seconds to ramp up the chance of regeneration due to expected time-till-refresh */ |
| 244 | private const RAMPUP_TTL = 30; |
| 245 | |
| 246 | /** @var float Tiny negative float to use when CTL comes up >= 0 due to clock skew */ |
| 247 | private const TINY_NEGATIVE = -0.000001; |
| 248 | /** @var float Tiny positive float to use when using "minTime" to assert an inequality */ |
| 249 | private const TINY_POSITIVE = 0.000001; |
| 250 | |
| 251 | /** Min millisecond set() backoff during hold-off (far less than INTERIM_KEY_TTL) */ |
| 252 | private const RECENT_SET_LOW_MS = 50; |
| 253 | /** Max millisecond set() backoff during hold-off (far less than INTERIM_KEY_TTL) */ |
| 254 | private const RECENT_SET_HIGH_MS = 100; |
| 255 | |
| 256 | /** Consider value generation somewhat high if it takes this many seconds or more */ |
| 257 | private const GENERATION_HIGH_SEC = 0.2; |
| 258 | |
| 259 | /** Key to the tombstone entry timestamp */ |
| 260 | private const PURGE_TIME = 0; |
| 261 | /** Key to the tombstone entry hold-off TTL */ |
| 262 | private const PURGE_HOLDOFF = 1; |
| 263 | |
| 264 | /** Cache format version number */ |
| 265 | private const VERSION = 1; |
| 266 | |
| 267 | /** Version number attribute for a key; keep value for b/c (< 1.36) */ |
| 268 | public const KEY_VERSION = 'version'; |
| 269 | /** Generation completion timestamp attribute for a key; keep value for b/c (< 1.36) */ |
| 270 | public const KEY_AS_OF = 'asOf'; |
| 271 | /** Logical TTL attribute for a key */ |
| 272 | public const KEY_TTL = 'ttl'; |
| 273 | /** Remaining TTL attribute for a key; keep value for b/c (< 1.36) */ |
| 274 | public const KEY_CUR_TTL = 'curTTL'; |
| 275 | /** Tombstone timestamp attribute for a key; keep value for b/c (< 1.36) */ |
| 276 | public const KEY_TOMB_AS_OF = 'tombAsOf'; |
| 277 | /** Highest "check" key timestamp for a key; keep value for b/c (< 1.36) */ |
| 278 | public const KEY_CHECK_AS_OF = 'lastCKPurge'; |
| 279 | |
| 280 | /** Value for a key */ |
| 281 | private const RES_VALUE = 0; |
| 282 | /** Version number attribute for a key */ |
| 283 | private const RES_VERSION = 1; |
| 284 | /** Generation completion timestamp attribute for a key */ |
| 285 | private const RES_AS_OF = 2; |
| 286 | /** Logical TTL attribute for a key */ |
| 287 | private const RES_TTL = 3; |
| 288 | /** Tombstone timestamp attribute for a key */ |
| 289 | private const RES_TOMB_AS_OF = 4; |
| 290 | /** Highest "check" key timestamp for a key */ |
| 291 | private const RES_CHECK_AS_OF = 5; |
| 292 | /** Highest "touched" timestamp for a key */ |
| 293 | private const RES_TOUCH_AS_OF = 6; |
| 294 | /** Remaining TTL attribute for a key */ |
| 295 | private const RES_CUR_TTL = 7; |
| 296 | |
| 297 | /** Key to WAN cache version number; stored in blobs */ |
| 298 | private const FLD_FORMAT_VERSION = 0; |
| 299 | /** Key to the cached value; stored in blobs */ |
| 300 | private const FLD_VALUE = 1; |
| 301 | /** Key to the original TTL; stored in blobs */ |
| 302 | private const FLD_TTL = 2; |
| 303 | /** Key to the cache timestamp; stored in blobs */ |
| 304 | private const FLD_TIME = 3; |
| 305 | /** Key to the flags bit field (reserved number) */ |
| 306 | private const /** @noinspection PhpUnusedPrivateFieldInspection */ FLD_FLAGS = 4; |
| 307 | /** Key to collection cache version number; stored in blobs */ |
| 308 | private const FLD_VALUE_VERSION = 5; |
| 309 | private const /** @noinspection PhpUnusedPrivateFieldInspection */ FLD_GENERATION_TIME = 6; |
| 310 | |
| 311 | /** Single character component for value keys */ |
| 312 | private const TYPE_VALUE = 'v'; |
| 313 | /** Single character component for timestamp check keys */ |
| 314 | private const TYPE_TIMESTAMP = 't'; |
| 315 | /** Single character component for mutex lock keys */ |
| 316 | private const TYPE_MUTEX = 'm'; |
| 317 | /** Single character component for interim value keys */ |
| 318 | private const TYPE_INTERIM = 'i'; |
| 319 | |
| 320 | /** Value prefix of purge values */ |
| 321 | private const PURGE_VAL_PREFIX = 'PURGED'; |
| 322 | |
| 323 | /** |
| 324 | * @stable to call |
| 325 | * @param array $params |
| 326 | * - cache : BagOStuff object for a persistent cache |
| 327 | * - logger : LoggerInterface object |
| 328 | * - stats : StatsFactory object. Since 1.43, constructing a WANObjectCache object |
| 329 | * with an IBufferingStatsdDataFactory stats collector will emit a warning. |
| 330 | * - asyncHandler : A function that takes a callback and runs it later. If supplied, |
| 331 | * whenever a preemptive refresh would be triggered in getWithSetCallback(), the |
| 332 | * current cache value is still used instead. However, the async-handler function |
| 333 | * receives a WAN cache callback that, when run, will execute the value generation |
| 334 | * callback supplied by the getWithSetCallback() caller. The result will be saved |
| 335 | * as normal. The handler is expected to call the WAN cache callback at an opportune |
| 336 | * time (e.g. HTTP post-send), though generally within a few 100ms. [optional] |
| 337 | * - broadcastRoutingPrefix: a routing prefix used to broadcast certain operations to all |
| 338 | * datacenters; See also <https://github.com/facebook/mcrouter/wiki/Config-Files>. |
| 339 | * This prefix takes the form `/<datacenter>/<name of wan route>/`, where `datacenter` |
| 340 | * is usually a wildcard to select all matching routes (e.g. the WAN cluster in all DCs). |
| 341 | * See also <https://github.com/facebook/mcrouter/wiki/Multi-cluster-broadcast-setup>. |
| 342 | * This is required when using mcrouter as a multi-region backing store proxy. [optional] |
| 343 | * - epoch: lowest UNIX timestamp a value/tombstone must have to be valid. [optional] |
| 344 | * - coalesceScheme: which key scheme to use in order to encourage the backend to place any |
| 345 | * "helper" keys for a "value" key within the same cache server. This reduces network |
| 346 | * overhead and reduces the chance the single downed cache server causes disruption. |
| 347 | * Use "hash_stop" with mcrouter and "hash_tag" with dynomite. [default: "hash_stop"] |
| 348 | * - tracer: TracerInterface instance where per-operation spans will be recorded |
| 349 | */ |
| 350 | public function __construct( array $params ) { |
| 351 | $this->cache = $params['cache']; |
| 352 | $this->broadcastRoute = $params['broadcastRoutingPrefix'] ?? null; |
| 353 | $this->epoch = $params['epoch'] ?? 0; |
| 354 | if ( ( $params['coalesceScheme'] ?? '' ) === 'hash_tag' ) { |
| 355 | // https://redis.io/topics/cluster-spec |
| 356 | // https://github.com/twitter/twemproxy/blob/v0.4.1/notes/recommendation.md#hash-tags |
| 357 | // https://github.com/Netflix/dynomite/blob/v0.7.0/notes/recommendation.md#hash-tags |
| 358 | $this->coalesceScheme = self::SCHEME_HASH_TAG; |
| 359 | } else { |
| 360 | // https://github.com/facebook/mcrouter/wiki/Key-syntax |
| 361 | $this->coalesceScheme = self::SCHEME_HASH_STOP; |
| 362 | } |
| 363 | |
| 364 | $this->setLogger( $params['logger'] ?? new NullLogger() ); |
| 365 | $this->tracer = $params['tracer'] ?? new NoopTracer(); |
| 366 | $this->stats = $params['stats'] ?? StatsFactory::newNull(); |
| 367 | |
| 368 | $this->asyncHandler = $params['asyncHandler'] ?? null; |
| 369 | $this->missLog = array_fill( 0, 10, [ '', 0.0 ] ); |
| 370 | } |
| 371 | |
| 372 | public function setLogger( LoggerInterface $logger ): void { |
| 373 | $this->logger = $logger; |
| 374 | } |
| 375 | |
| 376 | /** |
| 377 | * Get an instance that wraps EmptyBagOStuff |
| 378 | */ |
| 379 | public static function newEmpty(): static { |
| 380 | return new static( [ 'cache' => new EmptyBagOStuff() ] ); |
| 381 | } |
| 382 | |
| 383 | /** |
| 384 | * Fetch the value of a key from cache |
| 385 | * |
| 386 | * If supplied, $curTTL is set to the remaining TTL (current time left): |
| 387 | * - a) INF; if $key exists, has no TTL, and is not purged by $checkKeys |
| 388 | * - b) float (>=0); if $key exists, has a TTL, and is not purged by $checkKeys |
| 389 | * - c) float (<0); if $key is tombstoned, stale, or existing but purged by $checkKeys |
| 390 | * - d) null; if $key does not exist and is not tombstoned |
| 391 | * |
| 392 | * If a key is tombstoned, $curTTL will reflect the time since delete(). |
| 393 | * |
| 394 | * The timestamp of $key will be checked against the last-purge timestamp |
| 395 | * of each of $checkKeys. Those $checkKeys not in cache will have the last-purge |
| 396 | * initialized to the current timestamp. If any of $checkKeys have a timestamp |
| 397 | * greater than that of $key, then $curTTL will reflect how long ago $key |
| 398 | * became invalid. Callers can use $curTTL to know when the value is stale. |
| 399 | * The $checkKeys parameter allow mass key purges by updating a single key: |
| 400 | * - a) Each "check" key represents "last purged" of some source data |
| 401 | * - b) Callers pass in relevant "check" keys as $checkKeys in get() |
| 402 | * - c) When the source data that "check" keys represent changes, |
| 403 | * the touchCheckKey() method is called on them |
| 404 | * |
| 405 | * Source data entities might exist in a DB that uses snapshot isolation |
| 406 | * (e.g. the default REPEATABLE-READ in innoDB). Even for mutable data, that |
| 407 | * isolation can largely be maintained by doing the following: |
| 408 | * - a) Calling delete() on entity change *and* creation, before DB commit |
| 409 | * - b) Keeping transaction duration shorter than the delete() hold-off TTL |
| 410 | * - c) Disabling interim key caching via useInterimHoldOffCaching() before get() calls |
| 411 | * |
| 412 | * However, pre-snapshot values might still be seen if an update was made |
| 413 | * in a remote datacenter but the purge from delete() didn't relay yet. |
| 414 | * |
| 415 | * Consider using getWithSetCallback(), which has cache slam avoidance and key |
| 416 | * versioning features, instead of bare get()/set() calls. |
| 417 | * |
| 418 | * Do not use this method on versioned keys accessed via getWithSetCallback(). |
| 419 | * |
| 420 | * When using the $info parameter, it should be passed in as WANObjectCache::PASS_BY_REF. |
| 421 | * In that case, it becomes a key metadata map. Otherwise, for backwards compatibility, |
| 422 | * $info becomes the value generation timestamp (null if the key is nonexistant/tombstoned). |
| 423 | * Key metadata map fields include: |
| 424 | * - WANObjectCache::KEY_VERSION: value version number; null if key is nonexistant |
| 425 | * - WANObjectCache::KEY_AS_OF: value generation timestamp (UNIX); null if key is nonexistant |
| 426 | * - WANObjectCache::KEY_TTL: assigned TTL (seconds); null if key is nonexistant/tombstoned |
| 427 | * - WANObjectCache::KEY_CUR_TTL: remaining TTL (seconds); null if key is nonexistant |
| 428 | * - WANObjectCache::KEY_TOMB_AS_OF: tombstone timestamp (UNIX); null if key is not tombstoned |
| 429 | * - WANObjectCache::KEY_CHECK_AS_OF: highest "check" key timestamp (UNIX); null if none |
| 430 | * |
| 431 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 432 | * @param float|null &$curTTL Seconds of TTL left [returned] |
| 433 | * @param string[] $checkKeys Map of (integer or cache key => "check" key(s)); |
| 434 | * "check" keys must also be made with makeKey()/makeGlobalKey() |
| 435 | * @param array &$info Metadata map [returned] |
| 436 | * @return mixed Value of cache key; false on failure |
| 437 | */ |
| 438 | final public function get( $key, &$curTTL = null, array $checkKeys = [], &$info = [] ) { |
| 439 | // Note that an undeclared variable passed as $info starts as null (not the default). |
| 440 | // Also, if no $info parameter is provided, then it doesn't matter how it changes here. |
| 441 | $legacyInfo = ( $info !== self::PASS_BY_REF ); |
| 442 | |
| 443 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 444 | $span = $this->startOperationSpan( __FUNCTION__, $key, $checkKeys ); |
| 445 | |
| 446 | $now = $this->getCurrentTime(); |
| 447 | $res = $this->fetchKeys( [ $key ], $checkKeys, $now )[$key]; |
| 448 | |
| 449 | $curTTL = $res[self::RES_CUR_TTL]; |
| 450 | $info = $legacyInfo |
| 451 | ? $res[self::RES_AS_OF] |
| 452 | : [ |
| 453 | self::KEY_VERSION => $res[self::RES_VERSION], |
| 454 | self::KEY_AS_OF => $res[self::RES_AS_OF], |
| 455 | self::KEY_TTL => $res[self::RES_TTL], |
| 456 | self::KEY_CUR_TTL => $res[self::RES_CUR_TTL], |
| 457 | self::KEY_TOMB_AS_OF => $res[self::RES_TOMB_AS_OF], |
| 458 | self::KEY_CHECK_AS_OF => $res[self::RES_CHECK_AS_OF] |
| 459 | ]; |
| 460 | |
| 461 | if ( $curTTL === null || $curTTL <= 0 ) { |
| 462 | // Log the timestamp in case a corresponding set() call does not provide "walltime" |
| 463 | unset( $this->missLog[array_key_first( $this->missLog )] ); |
| 464 | $this->missLog[] = [ $key, $this->getCurrentTime() ]; |
| 465 | } |
| 466 | |
| 467 | return $res[self::RES_VALUE]; |
| 468 | } |
| 469 | |
| 470 | /** |
| 471 | * Fetch the value of several keys from cache |
| 472 | * |
| 473 | * $curTTLs becomes a map of only present/tombstoned $keys to their current time-to-live. |
| 474 | * |
| 475 | * $checkKeys holds the "check" keys used to validate values of applicable keys. The |
| 476 | * integer indexes hold "check" keys that apply to all of $keys while the string indexes |
| 477 | * hold "check" keys that only apply to the cache key with that name. The logic of "check" |
| 478 | * keys otherwise works the same as in WANObjectCache::get(). |
| 479 | * |
| 480 | * When using the $info parameter, it should be passed in as WANObjectCache::PASS_BY_REF. |
| 481 | * In that case, it becomes a mapping of all the $keys to their metadata maps, each in the |
| 482 | * style of WANObjectCache::get(). Otherwise, for backwards compatibility, $info becomes a |
| 483 | * map of only present/tombstoned $keys to their value generation timestamps. |
| 484 | * |
| 485 | * @see WANObjectCache::get() |
| 486 | * |
| 487 | * @param string[] $keys List/map with makeKey()/makeGlobalKey() cache keys as values |
| 488 | * @param array<string,float> &$curTTLs Map of (key => seconds of TTL left) [returned] |
| 489 | * @param string[]|string[][] $checkKeys Map of (integer or cache key => "check" key(s)); |
| 490 | * "check" keys must also be made with makeKey()/makeGlobalKey() |
| 491 | * @param array<string,array> &$info Map of (key => metadata map) [returned] |
| 492 | * @return array<string,mixed> Map of (key => value) for existing values in order of $keys |
| 493 | */ |
| 494 | final public function getMulti( |
| 495 | array $keys, |
| 496 | &$curTTLs = [], |
| 497 | array $checkKeys = [], |
| 498 | &$info = [] |
| 499 | ) { |
| 500 | // Note that an undeclared variable passed as $info starts as null (not the default). |
| 501 | // Also, if no $info parameter is provided, then it doesn't matter how it changes here. |
| 502 | $legacyInfo = ( $info !== self::PASS_BY_REF ); |
| 503 | |
| 504 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 505 | $span = $this->startOperationSpan( __FUNCTION__, $keys, $checkKeys ); |
| 506 | |
| 507 | $curTTLs = []; |
| 508 | $info = []; |
| 509 | $valuesByKey = []; |
| 510 | |
| 511 | $now = $this->getCurrentTime(); |
| 512 | $resByKey = $this->fetchKeys( $keys, $checkKeys, $now ); |
| 513 | foreach ( $resByKey as $key => $res ) { |
| 514 | if ( $res[self::RES_VALUE] !== false ) { |
| 515 | $valuesByKey[$key] = $res[self::RES_VALUE]; |
| 516 | } |
| 517 | |
| 518 | if ( $res[self::RES_CUR_TTL] !== null ) { |
| 519 | $curTTLs[$key] = $res[self::RES_CUR_TTL]; |
| 520 | } |
| 521 | $info[$key] = $legacyInfo |
| 522 | ? $res[self::RES_AS_OF] |
| 523 | : [ |
| 524 | self::KEY_VERSION => $res[self::RES_VERSION], |
| 525 | self::KEY_AS_OF => $res[self::RES_AS_OF], |
| 526 | self::KEY_TTL => $res[self::RES_TTL], |
| 527 | self::KEY_CUR_TTL => $res[self::RES_CUR_TTL], |
| 528 | self::KEY_TOMB_AS_OF => $res[self::RES_TOMB_AS_OF], |
| 529 | self::KEY_CHECK_AS_OF => $res[self::RES_CHECK_AS_OF] |
| 530 | ]; |
| 531 | } |
| 532 | |
| 533 | return $valuesByKey; |
| 534 | } |
| 535 | |
| 536 | /** |
| 537 | * Fetch the value and key metadata of several keys from cache |
| 538 | * |
| 539 | * $checkKeys holds the "check" keys used to validate values of applicable keys. |
| 540 | * The integer indexes hold "check" keys that apply to all of $keys while the string |
| 541 | * indexes hold "check" keys that only apply to the cache key with that name. |
| 542 | * |
| 543 | * @param string[] $keys List/map with makeKey()/makeGlobalKey() cache keys as values |
| 544 | * @param string[]|string[][] $checkKeys Map of (integer or cache key => "check" key(s)); |
| 545 | * "check" keys must also be made with makeKey()/makeGlobalKey() |
| 546 | * @param float $now The current UNIX timestamp |
| 547 | * @param callable|null $touchedCb Callback yielding a UNIX timestamp from a value, or null |
| 548 | * @return array<string,array> Map of (key => WANObjectCache::RESULT_* map) in order of $keys |
| 549 | * @note Callable type hints are not used to avoid class-autoloading |
| 550 | */ |
| 551 | protected function fetchKeys( array $keys, array $checkKeys, float $now, $touchedCb = null ) { |
| 552 | $resByKey = []; |
| 553 | |
| 554 | // List of all sister keys that need to be fetched from cache |
| 555 | $allSisterKeys = []; |
| 556 | // Order-corresponding value sister key list for the base key list ($keys) |
| 557 | $valueSisterKeys = []; |
| 558 | // List of "check" sister keys to compare all value sister keys against |
| 559 | $checkSisterKeysForAll = []; |
| 560 | // Map of (base key => additional "check" sister key(s) to compare against) |
| 561 | $checkSisterKeysByKey = []; |
| 562 | |
| 563 | foreach ( $keys as $key ) { |
| 564 | $sisterKey = $this->makeSisterKey( $key, self::TYPE_VALUE ); |
| 565 | $allSisterKeys[] = $sisterKey; |
| 566 | $valueSisterKeys[] = $sisterKey; |
| 567 | } |
| 568 | |
| 569 | foreach ( $checkKeys as $i => $checkKeyOrKeyGroup ) { |
| 570 | // Note: avoid array_merge() inside loop in case there are many keys |
| 571 | if ( is_int( $i ) ) { |
| 572 | // Single "check" key that applies to all base keys |
| 573 | $sisterKey = $this->makeSisterKey( $checkKeyOrKeyGroup, self::TYPE_TIMESTAMP ); |
| 574 | $allSisterKeys[] = $sisterKey; |
| 575 | $checkSisterKeysForAll[] = $sisterKey; |
| 576 | } else { |
| 577 | // List of "check" keys that apply to a specific base key |
| 578 | foreach ( (array)$checkKeyOrKeyGroup as $checkKey ) { |
| 579 | $sisterKey = $this->makeSisterKey( $checkKey, self::TYPE_TIMESTAMP ); |
| 580 | $allSisterKeys[] = $sisterKey; |
| 581 | $checkSisterKeysByKey[$i][] = $sisterKey; |
| 582 | } |
| 583 | } |
| 584 | } |
| 585 | |
| 586 | if ( $this->warmupCache ) { |
| 587 | // Get the wrapped values of the sister keys from the warmup cache |
| 588 | $wrappedBySisterKey = $this->warmupCache; |
| 589 | $sisterKeysMissing = array_diff( $allSisterKeys, array_keys( $wrappedBySisterKey ) ); |
| 590 | if ( $sisterKeysMissing ) { |
| 591 | $this->warmupKeyMisses += count( $sisterKeysMissing ); |
| 592 | $wrappedBySisterKey += $this->cache->getMulti( $sisterKeysMissing ); |
| 593 | } |
| 594 | } else { |
| 595 | // Fetch the wrapped values of the sister keys from the backend |
| 596 | $wrappedBySisterKey = $this->cache->getMulti( $allSisterKeys ); |
| 597 | } |
| 598 | |
| 599 | // List of "check" sister key purge timestamps to compare all value sister keys against |
| 600 | $ckPurgesForAll = $this->processCheckKeys( |
| 601 | $checkSisterKeysForAll, |
| 602 | $wrappedBySisterKey, |
| 603 | $now |
| 604 | ); |
| 605 | // Map of (base key => extra "check" sister key purge timestamp(s) to compare against) |
| 606 | $ckPurgesByKey = []; |
| 607 | foreach ( $checkSisterKeysByKey as $keyWithCheckKeys => $checkKeysForKey ) { |
| 608 | $ckPurgesByKey[$keyWithCheckKeys] = $this->processCheckKeys( |
| 609 | $checkKeysForKey, |
| 610 | $wrappedBySisterKey, |
| 611 | $now |
| 612 | ); |
| 613 | } |
| 614 | |
| 615 | // Unwrap and validate any value found for each base key (under the value sister key) |
| 616 | foreach ( |
| 617 | array_map( null, $valueSisterKeys, $keys ) |
| 618 | as [ $valueSisterKey, $key ] |
| 619 | ) { |
| 620 | if ( array_key_exists( $valueSisterKey, $wrappedBySisterKey ) ) { |
| 621 | // Key exists as either a live value or tombstone value |
| 622 | $wrapped = $wrappedBySisterKey[$valueSisterKey]; |
| 623 | } else { |
| 624 | // Key does not exist |
| 625 | $wrapped = false; |
| 626 | } |
| 627 | |
| 628 | $res = $this->unwrap( $wrapped, $now ); |
| 629 | $value = $res[self::RES_VALUE]; |
| 630 | |
| 631 | foreach ( array_merge( $ckPurgesForAll, $ckPurgesByKey[$key] ?? [] ) as $ckPurge ) { |
| 632 | $res[self::RES_CHECK_AS_OF] = max( |
| 633 | $ckPurge[self::PURGE_TIME], |
| 634 | $res[self::RES_CHECK_AS_OF] |
| 635 | ); |
| 636 | // Timestamp marking the end of the hold-off period for this purge |
| 637 | $holdoffDeadline = $ckPurge[self::PURGE_TIME] + $ckPurge[self::PURGE_HOLDOFF]; |
| 638 | // Check if the value was generated during the hold-off period |
| 639 | if ( $value !== false && $holdoffDeadline >= $res[self::RES_AS_OF] ) { |
| 640 | // How long ago this value was purged by *this* "check" key |
| 641 | $ago = min( $ckPurge[self::PURGE_TIME] - $now, self::TINY_NEGATIVE ); |
| 642 | // How long ago this value was purged by *any* known "check" key |
| 643 | $res[self::RES_CUR_TTL] = min( $res[self::RES_CUR_TTL], $ago ); |
| 644 | } |
| 645 | } |
| 646 | |
| 647 | if ( $touchedCb !== null && $value !== false ) { |
| 648 | $touched = $touchedCb( $value ); |
| 649 | if ( $touched !== null && $touched >= $res[self::RES_AS_OF] ) { |
| 650 | $res[self::RES_CUR_TTL] = min( |
| 651 | $res[self::RES_CUR_TTL], |
| 652 | $res[self::RES_AS_OF] - $touched, |
| 653 | self::TINY_NEGATIVE |
| 654 | ); |
| 655 | } |
| 656 | } else { |
| 657 | $touched = null; |
| 658 | } |
| 659 | |
| 660 | $res[self::RES_TOUCH_AS_OF] = max( $res[self::RES_TOUCH_AS_OF], $touched ); |
| 661 | |
| 662 | $resByKey[$key] = $res; |
| 663 | } |
| 664 | |
| 665 | return $resByKey; |
| 666 | } |
| 667 | |
| 668 | /** |
| 669 | * @param string[] $checkSisterKeys List of "check" sister keys |
| 670 | * @param mixed[] $wrappedBySisterKey Preloaded map of (sister key => wrapped value) |
| 671 | * @param float $now UNIX timestamp |
| 672 | * @return array[] List of purge value arrays |
| 673 | */ |
| 674 | private function processCheckKeys( |
| 675 | array $checkSisterKeys, |
| 676 | array $wrappedBySisterKey, |
| 677 | float $now |
| 678 | ) { |
| 679 | $purges = []; |
| 680 | |
| 681 | foreach ( $checkSisterKeys as $timeKey ) { |
| 682 | $purge = isset( $wrappedBySisterKey[$timeKey] ) |
| 683 | ? $this->parsePurgeValue( $wrappedBySisterKey[$timeKey] ) |
| 684 | : null; |
| 685 | |
| 686 | if ( $purge === null ) { |
| 687 | // No holdoff when lazy creating a check key, use cache right away (T344191) |
| 688 | $wrapped = $this->makeCheckPurgeValue( $now, self::HOLDOFF_TTL_NONE, $purge ); |
| 689 | $this->cache->add( |
| 690 | $timeKey, |
| 691 | $wrapped, |
| 692 | self::CHECK_KEY_TTL, |
| 693 | $this->cache::WRITE_BACKGROUND |
| 694 | ); |
| 695 | } |
| 696 | |
| 697 | $purges[] = $purge; |
| 698 | } |
| 699 | |
| 700 | return $purges; |
| 701 | } |
| 702 | |
| 703 | /** |
| 704 | * Set the value of a key in cache |
| 705 | * |
| 706 | * Simply calling this method when source data changes is not valid because |
| 707 | * the changes do not replicate to the other WAN sites. In that case, delete() |
| 708 | * should be used instead. This method is intended for use on cache misses. |
| 709 | * |
| 710 | * If data was read using "view snapshots" (e.g. innodb REPEATABLE-READ), |
| 711 | * use 'since' to avoid the following race condition: |
| 712 | * - a) T1 starts |
| 713 | * - b) T2 updates a row, calls delete(), and commits |
| 714 | * - c) The HOLDOFF_TTL passes, expiring the delete() tombstone |
| 715 | * - d) T1 reads the row and calls set() due to a cache miss |
| 716 | * - e) Stale value is stuck in cache |
| 717 | * |
| 718 | * Setting 'lag' and 'since' help avoids keys getting stuck in stale states. |
| 719 | * |
| 720 | * Be aware that this does not update the process cache for getWithSetCallback() |
| 721 | * callers. Keys accessed via that method are not generally meant to also be set |
| 722 | * using this primitive method. |
| 723 | * |
| 724 | * Consider using getWithSetCallback(), which has cache slam avoidance and key |
| 725 | * versioning features, instead of bare get()/set() calls. |
| 726 | * |
| 727 | * Do not use this method on versioned keys accessed via getWithSetCallback(). |
| 728 | * |
| 729 | * Example usage: |
| 730 | * @code |
| 731 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 732 | * $setOpts = Database::getCacheSetOptions( $dbr ); |
| 733 | * // Fetch the row from the DB |
| 734 | * $row = $dbr->selectRow( ... ); |
| 735 | * $key = $cache->makeKey( 'building', $buildingId ); |
| 736 | * $cache->set( $key, $row, $cache::TTL_DAY, $setOpts ); |
| 737 | * @endcode |
| 738 | * |
| 739 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 740 | * @param mixed $value Value to set for the cache key |
| 741 | * @param int $ttl Seconds to live. Special values are: |
| 742 | * - WANObjectCache::TTL_INDEFINITE: Cache forever (default) |
| 743 | * - WANObjectCache::TTL_UNCACHEABLE: Do not cache (if the key exists, it is not deleted) |
| 744 | * @param array $opts Options map: |
| 745 | * - lag: Highest seconds of replication lag potentially affecting reads used to generate |
| 746 | * the value. This should not be affected by the duration of transaction "view snapshots" |
| 747 | * (e.g. innodb REPEATABLE-READ) nor the time elapsed since the first read (though both |
| 748 | * increase staleness). For reads using view snapshots, only the replication lag during |
| 749 | * snapshot initialization matters. Use false if replication is stopped/broken on a |
| 750 | * replica server involved in the reads. |
| 751 | * Default: 0 seconds |
| 752 | * - since: UNIX timestamp indicative of the highest possible staleness caused by the |
| 753 | * duration of transaction "view snapshots" (e.g. innodb REPEATABLE-READ) and the time |
| 754 | * elapsed since the first read. This should not be affected by replication lag. |
| 755 | * Default: 0 seconds |
| 756 | * - pending: Whether this data is possibly from an uncommitted write transaction. |
| 757 | * Generally, other threads should not see values from the future and |
| 758 | * they certainly should not see ones that ended up getting rolled back. |
| 759 | * Default: false |
| 760 | * - lockTSE: If excessive replication/snapshot lag is detected, then store the value |
| 761 | * with this TTL and flag it as stale. This is only useful if the reads for this key |
| 762 | * use getWithSetCallback() with "lockTSE" set. Note that if "staleTTL" is set |
| 763 | * then it will still add on to this TTL in the excessive lag scenario. |
| 764 | * Default: WANObjectCache::TSE_NONE |
| 765 | * - staleTTL: Seconds to keep the key around if it is stale. The get()/getMulti() |
| 766 | * methods return such stale values with a $curTTL of 0, and getWithSetCallback() |
| 767 | * will call the generation callback in such cases, passing in the old value |
| 768 | * and its as-of time to the callback. This is useful if adaptiveTTL() is used |
| 769 | * on the old value's as-of time when it is verified as still being correct. |
| 770 | * Default: WANObjectCache::STALE_TTL_NONE |
| 771 | * - segmentable: Allow partitioning of the value if it is a large string. |
| 772 | * Default: false. |
| 773 | * - creating: Optimize for the case where the key does not already exist. |
| 774 | * Default: false |
| 775 | * - version: Integer version number signifying the format of the value. |
| 776 | * Default: null |
| 777 | * - walltime: How long the value took to generate in seconds. Default: null |
| 778 | * @phpcs:ignore Generic.Files.LineLength |
| 779 | * @phan-param array{lag?:float|int,since?:float|int,pending?:bool,lockTSE?:int,staleTTL?:int,creating?:bool,version?:int,walltime?:int|float,segmentable?:bool} $opts |
| 780 | * @note Options added in 1.28: staleTTL |
| 781 | * @note Options added in 1.33: creating |
| 782 | * @note Options added in 1.34: version, walltime |
| 783 | * @note Options added in 1.40: segmentable |
| 784 | * @return bool Success |
| 785 | */ |
| 786 | final public function set( $key, $value, $ttl = self::TTL_INDEFINITE, array $opts = [] ) { |
| 787 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 788 | $span = $this->startOperationSpan( __FUNCTION__, $key ); |
| 789 | |
| 790 | $keygroup = $this->determineKeyGroupForStats( $key ); |
| 791 | |
| 792 | $ok = $this->setMainValue( |
| 793 | $key, |
| 794 | $value, |
| 795 | $ttl, |
| 796 | $opts['version'] ?? null, |
| 797 | $opts['walltime'] ?? null, |
| 798 | $opts['lag'] ?? 0, |
| 799 | $opts['since'] ?? null, |
| 800 | $opts['pending'] ?? false, |
| 801 | $opts['lockTSE'] ?? self::TSE_NONE, |
| 802 | $opts['staleTTL'] ?? self::STALE_TTL_NONE, |
| 803 | $opts['segmentable'] ?? false, |
| 804 | $opts['creating'] ?? false |
| 805 | ); |
| 806 | |
| 807 | $this->stats->getCounter( 'wanobjectcache_set_total' ) |
| 808 | ->setLabel( 'keygroup', $keygroup ) |
| 809 | ->setLabel( 'result', ( $ok ? 'ok' : 'error' ) ) |
| 810 | ->increment(); |
| 811 | |
| 812 | return $ok; |
| 813 | } |
| 814 | |
| 815 | /** |
| 816 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 817 | * @param mixed $value |
| 818 | * @param int|float $ttl |
| 819 | * @param int|null $version |
| 820 | * @param float|null $walltime |
| 821 | * @param float|int|bool $dataReplicaLag |
| 822 | * @param float|int|null $dataReadSince |
| 823 | * @param bool $dataPendingCommit |
| 824 | * @param int $lockTSE |
| 825 | * @param int $staleTTL |
| 826 | * @param bool $segmentable |
| 827 | * @param bool $creating |
| 828 | * @return bool Success |
| 829 | */ |
| 830 | private function setMainValue( |
| 831 | $key, |
| 832 | $value, |
| 833 | $ttl, |
| 834 | ?int $version, |
| 835 | ?float $walltime, |
| 836 | $dataReplicaLag, |
| 837 | $dataReadSince, |
| 838 | bool $dataPendingCommit, |
| 839 | int $lockTSE, |
| 840 | int $staleTTL, |
| 841 | bool $segmentable, |
| 842 | bool $creating |
| 843 | ) { |
| 844 | if ( $ttl < 0 ) { |
| 845 | // not cacheable |
| 846 | return true; |
| 847 | } |
| 848 | |
| 849 | $now = $this->getCurrentTime(); |
| 850 | |
| 851 | // T413673: Handle PHP8.5 case where TTL is infinite. |
| 852 | if ( is_finite( $ttl ) ) { |
| 853 | $ttl = (int)$ttl; |
| 854 | } else { |
| 855 | $ttl = self::TTL_INDEFINITE; |
| 856 | } |
| 857 | $walltime ??= $this->timeSinceLoggedMiss( $key, $now ); |
| 858 | $dataSnapshotLag = ( $dataReadSince !== null ) ? max( 0, $now - $dataReadSince ) : 0; |
| 859 | $dataCombinedLag = $dataReplicaLag + $dataSnapshotLag; |
| 860 | |
| 861 | // Forbid caching data that only exists within an uncommitted transaction. Also, lower |
| 862 | // the TTL when the data has a "since" time so far in the past that a delete() tombstone, |
| 863 | // made after that time, could have already expired (the key is no longer write-holed). |
| 864 | // The mitigation TTL depends on whether this data lag is assumed to systemically effect |
| 865 | // regeneration attempts in the near future. The TTL also reflects regeneration wall time. |
| 866 | if ( $dataPendingCommit ) { |
| 867 | // Case A: data comes from an uncommitted write transaction |
| 868 | $mitigated = 'pending writes'; |
| 869 | // Data might never be committed; rely on a less problematic regeneration attempt |
| 870 | $mitigationTTL = self::TTL_UNCACHEABLE; |
| 871 | } elseif ( $dataSnapshotLag > self::MAX_READ_LAG ) { |
| 872 | // Case B: high snapshot lag |
| 873 | $pregenSnapshotLag = ( $walltime !== null ) ? ( $dataSnapshotLag - $walltime ) : 0; |
| 874 | if ( ( $pregenSnapshotLag + self::GENERATION_HIGH_SEC ) > self::MAX_READ_LAG ) { |
| 875 | // Case B1: generation started when transaction duration was already long |
| 876 | $mitigated = 'snapshot lag (late generation)'; |
| 877 | // Probably non-systemic; rely on a less problematic regeneration attempt |
| 878 | $mitigationTTL = self::TTL_UNCACHEABLE; |
| 879 | } else { |
| 880 | // Case B2: slow generation made transaction duration long |
| 881 | $mitigated = 'snapshot lag (high generation time)'; |
| 882 | // Probably systemic; use a low TTL to avoid stampedes/uncacheability |
| 883 | $mitigationTTL = self::TTL_LAGGED; |
| 884 | } |
| 885 | } elseif ( $dataReplicaLag === false || $dataReplicaLag > self::MAX_READ_LAG ) { |
| 886 | // Case C: low/medium snapshot lag with high replication lag |
| 887 | $mitigated = 'replication lag'; |
| 888 | // Probably systemic; use a low TTL to avoid stampedes/uncacheability |
| 889 | $mitigationTTL = self::TTL_LAGGED; |
| 890 | } elseif ( $dataCombinedLag > self::MAX_READ_LAG ) { |
| 891 | $pregenCombinedLag = ( $walltime !== null ) ? ( $dataCombinedLag - $walltime ) : 0; |
| 892 | // Case D: medium snapshot lag with medium replication lag |
| 893 | if ( ( $pregenCombinedLag + self::GENERATION_HIGH_SEC ) > self::MAX_READ_LAG ) { |
| 894 | // Case D1: generation started when read lag was too high |
| 895 | $mitigated = 'read lag (late generation)'; |
| 896 | // Probably non-systemic; rely on a less problematic regeneration attempt |
| 897 | $mitigationTTL = self::TTL_UNCACHEABLE; |
| 898 | } else { |
| 899 | // Case D2: slow generation made read lag too high |
| 900 | $mitigated = 'read lag (high generation time)'; |
| 901 | // Probably systemic; use a low TTL to avoid stampedes/uncacheability |
| 902 | $mitigationTTL = self::TTL_LAGGED; |
| 903 | } |
| 904 | } else { |
| 905 | // Case E: new value generated with recent data |
| 906 | $mitigated = null; |
| 907 | // Nothing to mitigate |
| 908 | $mitigationTTL = null; |
| 909 | } |
| 910 | |
| 911 | if ( $mitigationTTL === self::TTL_UNCACHEABLE ) { |
| 912 | $this->logger->warning( |
| 913 | "Rejected set() for {cachekey} due to $mitigated.", |
| 914 | [ |
| 915 | 'cachekey' => $key, |
| 916 | 'lag' => $dataReplicaLag, |
| 917 | 'age' => $dataSnapshotLag, |
| 918 | 'walltime' => $walltime |
| 919 | ] |
| 920 | ); |
| 921 | |
| 922 | // no-op the write for being unsafe |
| 923 | return true; |
| 924 | } |
| 925 | |
| 926 | // TTL to use in staleness checks (does not effect persistence layer TTL) |
| 927 | $logicalTTL = null; |
| 928 | |
| 929 | if ( $mitigationTTL !== null ) { |
| 930 | // New value was generated from data that is old enough to be risky |
| 931 | if ( $lockTSE >= 0 ) { |
| 932 | // Persist the value as long as normal, but make it count as stale sooner |
| 933 | $logicalTTL = min( $ttl ?: INF, $mitigationTTL ); |
| 934 | } else { |
| 935 | // Persist the value for a shorter duration |
| 936 | $ttl = min( $ttl ?: INF, $mitigationTTL ); |
| 937 | } |
| 938 | |
| 939 | $this->logger->warning( |
| 940 | "Lowered set() TTL for {cachekey} due to $mitigated.", |
| 941 | [ |
| 942 | 'cachekey' => $key, |
| 943 | 'lag' => $dataReplicaLag, |
| 944 | 'age' => $dataSnapshotLag, |
| 945 | 'walltime' => $walltime |
| 946 | ] |
| 947 | ); |
| 948 | } |
| 949 | |
| 950 | // Wrap that value with time/TTL/version metadata |
| 951 | $wrapped = $this->wrap( $value, $logicalTTL ?: $ttl, $version, $now ); |
| 952 | $storeTTL = $ttl + $staleTTL; |
| 953 | |
| 954 | $flags = $this->cache::WRITE_BACKGROUND; |
| 955 | if ( $segmentable ) { |
| 956 | $flags |= $this->cache::WRITE_ALLOW_SEGMENTS; |
| 957 | } |
| 958 | |
| 959 | if ( $creating ) { |
| 960 | $ok = $this->cache->add( |
| 961 | $this->makeSisterKey( $key, self::TYPE_VALUE ), |
| 962 | $wrapped, |
| 963 | $storeTTL, |
| 964 | $flags |
| 965 | ); |
| 966 | } else { |
| 967 | $ok = $this->cache->merge( |
| 968 | $this->makeSisterKey( $key, self::TYPE_VALUE ), |
| 969 | static function ( $cache, $key, $cWrapped ) use ( $wrapped ) { |
| 970 | // A string value means that it is a tombstone; do nothing in that case |
| 971 | return ( is_string( $cWrapped ) ) ? false : $wrapped; |
| 972 | }, |
| 973 | $storeTTL, |
| 974 | $this->cache::MAX_CONFLICTS_ONE, |
| 975 | $flags |
| 976 | ); |
| 977 | } |
| 978 | |
| 979 | return $ok; |
| 980 | } |
| 981 | |
| 982 | /** |
| 983 | * Purge a key from all datacenters |
| 984 | * |
| 985 | * This should only be called when the underlying data (being cached) |
| 986 | * changes in a significant way. This deletes the key and starts a hold-off |
| 987 | * period where the key cannot be written to for a few seconds (HOLDOFF_TTL). |
| 988 | * This is done to avoid the following race condition: |
| 989 | * - a) Some DB data changes and delete() is called on a corresponding key |
| 990 | * - b) A request refills the key with a stale value from a lagged DB |
| 991 | * - c) The stale value is stuck there until the key is expired/evicted |
| 992 | * |
| 993 | * This is implemented by storing a special "tombstone" value at the cache |
| 994 | * key that this class recognizes; get() calls will return false for the key |
| 995 | * and any set() calls will refuse to replace tombstone values at the key. |
| 996 | * For this to always avoid stale value writes, the following must hold: |
| 997 | * - a) Replication lag is bounded to being less than HOLDOFF_TTL; or |
| 998 | * - b) If lag is higher, the DB will have gone into read-only mode already |
| 999 | * |
| 1000 | * Note that set() can also be lag-aware and lower the TTL if it's high. |
| 1001 | * |
| 1002 | * Be aware that this does not clear the process cache. Even if it did, callbacks |
| 1003 | * used by getWithSetCallback() might still return stale data in the case of either |
| 1004 | * uncommitted or not-yet-replicated changes (callback generally use replica DBs). |
| 1005 | * |
| 1006 | * When using potentially long-running ACID transactions, a good pattern is |
| 1007 | * to use a pre-commit hook to issue the delete(). This means that immediately |
| 1008 | * after commit, callers will see the tombstone in cache upon purge relay. |
| 1009 | * It also avoids the following race condition: |
| 1010 | * - a) T1 begins, changes a row, and calls delete() |
| 1011 | * - b) The HOLDOFF_TTL passes, expiring the delete() tombstone |
| 1012 | * - c) T2 starts, reads the row and calls set() due to a cache miss |
| 1013 | * - d) T1 finally commits |
| 1014 | * - e) Stale value is stuck in cache |
| 1015 | * |
| 1016 | * Example usage: |
| 1017 | * @code |
| 1018 | * $dbw->startAtomic( __METHOD__ ); // start of request |
| 1019 | * ... <execute some stuff> ... |
| 1020 | * // Update the row in the DB |
| 1021 | * $dbw->update( ... ); |
| 1022 | * $key = $cache->makeKey( 'homes', $homeId ); |
| 1023 | * // Purge the corresponding cache entry just before committing |
| 1024 | * $dbw->onTransactionPreCommitOrIdle( function() use ( $cache, $key ) { |
| 1025 | * $cache->delete( $key ); |
| 1026 | * } ); |
| 1027 | * ... <execute some stuff> ... |
| 1028 | * $dbw->endAtomic( __METHOD__ ); // end of request |
| 1029 | * @endcode |
| 1030 | * |
| 1031 | * The $ttl parameter can be used when purging values that have not actually changed |
| 1032 | * recently. For example, user-requested purges or cache cleanup scripts might not need |
| 1033 | * to invoke a hold-off period on cache backfills, so they can use HOLDOFF_TTL_NONE. |
| 1034 | * |
| 1035 | * Note that $ttl limits the effective range of 'lockTSE' for getWithSetCallback(). |
| 1036 | * |
| 1037 | * If called twice on the same key, then the last hold-off TTL takes precedence. For |
| 1038 | * idempotence, the $ttl should not vary for different delete() calls on the same key. |
| 1039 | * |
| 1040 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1041 | * @param int $ttl Tombstone TTL; Default: WANObjectCache::HOLDOFF_TTL |
| 1042 | * @return bool True if the item was purged or not found, false on failure |
| 1043 | */ |
| 1044 | final public function delete( $key, $ttl = self::HOLDOFF_TTL ) { |
| 1045 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1046 | $span = $this->startOperationSpan( __FUNCTION__, $key ); |
| 1047 | |
| 1048 | // Purge values must be stored under the value key so that WANObjectCache::set() |
| 1049 | // can atomically merge values without accidentally undoing a recent purge and thus |
| 1050 | // violating the holdoff TTL restriction. |
| 1051 | $valueSisterKey = $this->makeSisterKey( $key, self::TYPE_VALUE ); |
| 1052 | |
| 1053 | if ( $ttl <= 0 ) { |
| 1054 | // A client or cache cleanup script is requesting a cache purge, so there is no |
| 1055 | // volatility period due to replica DB lag. Any recent change to an entity cached |
| 1056 | // in this key should have triggered an appropriate purge event. |
| 1057 | $ok = $this->cache->delete( $this->getRouteKey( $valueSisterKey ), $this->cache::WRITE_BACKGROUND ); |
| 1058 | } else { |
| 1059 | // A cacheable entity recently changed, so there might be a volatility period due |
| 1060 | // to replica DB lag. Clients usually expect their actions to be reflected in any |
| 1061 | // of their subsequent web request. This is attainable if (a) purge relay lag is |
| 1062 | // lower than the time it takes for subsequent request by the client to arrive, |
| 1063 | // and, (b) DB replica queries have "read-your-writes" consistency due to DB lag |
| 1064 | // mitigation systems. |
| 1065 | $now = $this->getCurrentTime(); |
| 1066 | // Set the key to the purge value in all datacenters |
| 1067 | $purge = self::PURGE_VAL_PREFIX . ':' . (int)$now; |
| 1068 | $ok = $this->cache->set( |
| 1069 | $this->getRouteKey( $valueSisterKey ), |
| 1070 | $purge, |
| 1071 | $ttl, |
| 1072 | $this->cache::WRITE_BACKGROUND |
| 1073 | ); |
| 1074 | } |
| 1075 | |
| 1076 | $keygroup = $this->determineKeyGroupForStats( $key ); |
| 1077 | |
| 1078 | $this->stats->getCounter( 'wanobjectcache_delete_total' ) |
| 1079 | ->setLabel( 'keygroup', $keygroup ) |
| 1080 | ->setLabel( 'result', ( $ok ? 'ok' : 'error' ) ) |
| 1081 | ->increment(); |
| 1082 | |
| 1083 | return $ok; |
| 1084 | } |
| 1085 | |
| 1086 | /** |
| 1087 | * Fetch the value of a timestamp "check" key |
| 1088 | * |
| 1089 | * The key will be *initialized* to the current time if not set, |
| 1090 | * so only call this method if this behavior is actually desired |
| 1091 | * |
| 1092 | * The timestamp can be used to check whether a cached value is valid. |
| 1093 | * Callers should not assume that this returns the same timestamp in |
| 1094 | * all datacenters due to relay delays. |
| 1095 | * |
| 1096 | * The level of staleness can roughly be estimated from this key, but |
| 1097 | * if the key was evicted from cache, such calculations may show the |
| 1098 | * time since expiry as ~0 seconds. |
| 1099 | * |
| 1100 | * Note that "check" keys won't collide with other regular keys. |
| 1101 | * |
| 1102 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1103 | * @return float UNIX timestamp |
| 1104 | */ |
| 1105 | final public function getCheckKeyTime( $key ) { |
| 1106 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1107 | $span = $this->startOperationSpan( __FUNCTION__, $key ); |
| 1108 | |
| 1109 | return $this->getMultiCheckKeyTime( [ $key ] )[$key]; |
| 1110 | } |
| 1111 | |
| 1112 | /** |
| 1113 | * Fetch the values of each timestamp "check" key |
| 1114 | * |
| 1115 | * This works like getCheckKeyTime() except it takes a list of keys |
| 1116 | * and returns a map of timestamps instead of just that of one key |
| 1117 | * |
| 1118 | * This might be useful if both: |
| 1119 | * - a) a class of entities each depend on hundreds of other entities |
| 1120 | * - b) these other entities are depended upon by millions of entities |
| 1121 | * |
| 1122 | * The later entities can each use a "check" key to purge their dependee entities. |
| 1123 | * However, it is expensive for the former entities to verify against all of the relevant |
| 1124 | * "check" keys during each getWithSetCallback() call. A less expensive approach is to do |
| 1125 | * these verifications only after a "time-till-verify" (TTV) has passed. This is a middle |
| 1126 | * ground between using blind TTLs and using constant verification. The adaptiveTTL() method |
| 1127 | * can be used to dynamically adjust the TTV. Also, the initial TTV can make use of the |
| 1128 | * last-modified times of the dependent entities (either from the DB or the "check" keys). |
| 1129 | * |
| 1130 | * Example usage: |
| 1131 | * @code |
| 1132 | * $value = $cache->getWithSetCallback( |
| 1133 | * $cache->makeGlobalKey( 'wikibase-item', $id ), |
| 1134 | * self::INITIAL_TTV, // initial time-till-verify |
| 1135 | * function ( $oldValue, &$ttv, &$setOpts, $oldAsOf ) use ( $checkKeys, $cache ) { |
| 1136 | * $now = time(); |
| 1137 | * // Use $oldValue if it passes max ultimate age and "check" key comparisons |
| 1138 | * if ( $oldValue && |
| 1139 | * $oldAsOf > max( $cache->getMultiCheckKeyTime( $checkKeys ) ) && |
| 1140 | * ( $now - $oldValue['ctime'] ) <= self::MAX_CACHE_AGE |
| 1141 | * ) { |
| 1142 | * // Increase time-till-verify by 50% of last time to reduce overhead |
| 1143 | * $ttv = $cache->adaptiveTTL( $oldAsOf, self::MAX_TTV, self::MIN_TTV, 1.5 ); |
| 1144 | * // Unlike $oldAsOf, "ctime" is the ultimate age of the cached data |
| 1145 | * return $oldValue; |
| 1146 | * } |
| 1147 | * |
| 1148 | * $mtimes = []; // dependency last-modified times; passed by reference |
| 1149 | * $value = [ 'data' => $this->fetchEntityData( $mtimes ), 'ctime' => $now ]; |
| 1150 | * // Guess time-till-change among the dependencies, e.g. 1/(total change rate) |
| 1151 | * $ttc = 1 / array_sum( array_map( |
| 1152 | * function ( $mtime ) use ( $now ) { |
| 1153 | * return 1 / ( $mtime ? ( $now - $mtime ) : 900 ); |
| 1154 | * }, |
| 1155 | * $mtimes |
| 1156 | * ) ); |
| 1157 | * // The time-to-verify should not be overly pessimistic nor optimistic |
| 1158 | * $ttv = min( max( $ttc, self::MIN_TTV ), self::MAX_TTV ); |
| 1159 | * |
| 1160 | * return $value; |
| 1161 | * }, |
| 1162 | * [ 'staleTTL' => $cache::TTL_DAY ] // keep around to verify and re-save |
| 1163 | * ); |
| 1164 | * @endcode |
| 1165 | * |
| 1166 | * @see WANObjectCache::getCheckKeyTime() |
| 1167 | * @see WANObjectCache::getWithSetCallback() |
| 1168 | * |
| 1169 | * @param string[] $keys Cache keys made with makeKey()/makeGlobalKey() |
| 1170 | * @return float[] Map of (key => UNIX timestamp) |
| 1171 | * @since 1.31 |
| 1172 | */ |
| 1173 | final public function getMultiCheckKeyTime( array $keys ) { |
| 1174 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1175 | $span = $this->startOperationSpan( __FUNCTION__, $keys ); |
| 1176 | |
| 1177 | $checkSisterKeysByKey = []; |
| 1178 | foreach ( $keys as $key ) { |
| 1179 | $checkSisterKeysByKey[$key] = $this->makeSisterKey( $key, self::TYPE_TIMESTAMP ); |
| 1180 | } |
| 1181 | |
| 1182 | $wrappedBySisterKey = $this->cache->getMulti( $checkSisterKeysByKey ); |
| 1183 | $wrappedBySisterKey += array_fill_keys( $checkSisterKeysByKey, false ); |
| 1184 | |
| 1185 | $now = $this->getCurrentTime(); |
| 1186 | $times = []; |
| 1187 | foreach ( $checkSisterKeysByKey as $key => $checkSisterKey ) { |
| 1188 | $purge = $this->parsePurgeValue( $wrappedBySisterKey[$checkSisterKey] ); |
| 1189 | if ( $purge === null ) { |
| 1190 | $wrapped = $this->makeCheckPurgeValue( $now, self::HOLDOFF_TTL_NONE, $purge ); |
| 1191 | $this->cache->add( |
| 1192 | $checkSisterKey, |
| 1193 | $wrapped, |
| 1194 | self::CHECK_KEY_TTL, |
| 1195 | $this->cache::WRITE_BACKGROUND |
| 1196 | ); |
| 1197 | } |
| 1198 | |
| 1199 | $times[$key] = $purge[self::PURGE_TIME]; |
| 1200 | } |
| 1201 | |
| 1202 | return $times; |
| 1203 | } |
| 1204 | |
| 1205 | /** |
| 1206 | * Increase the last-purge timestamp of a "check" key in all datacenters |
| 1207 | * |
| 1208 | * This method should only be called when some heavily referenced data changes in |
| 1209 | * a significant way, such that it is impractical to call delete() on all the cache |
| 1210 | * keys that should be purged. The get*() method calls used to fetch these keys must |
| 1211 | * include the given "check" key in the relevant "check" keys argument/option. |
| 1212 | * |
| 1213 | * A "check" key essentially represents a last-modified time of an entity. When the |
| 1214 | * key is touched, the timestamp will be updated to the current time. Keys fetched |
| 1215 | * using get*() calls, that include the "check" key, will be seen as purged. |
| 1216 | * |
| 1217 | * The timestamp of the "check" key is treated as being HOLDOFF_TTL seconds in the |
| 1218 | * future by get*() methods in order to avoid race conditions where keys are updated |
| 1219 | * with stale values (e.g. from a lagged replica DB). A high TTL is set on the "check" |
| 1220 | * key, making it possible to know the timestamp of the last change to the corresponding |
| 1221 | * entities in most cases. This might use more cache space than resetCheckKey(). |
| 1222 | * |
| 1223 | * When a few important keys get a large number of hits, a high cache time is usually |
| 1224 | * desired as well as "lockTSE" logic. The resetCheckKey() method is less appropriate |
| 1225 | * in such cases since the "time since expiry" cannot be inferred, causing any get() |
| 1226 | * after the reset to treat the key as being "hot", resulting in more stale value usage. |
| 1227 | * |
| 1228 | * Note that "check" keys won't collide with other regular keys. |
| 1229 | * |
| 1230 | * @see WANObjectCache::get() |
| 1231 | * @see WANObjectCache::getWithSetCallback() |
| 1232 | * @see WANObjectCache::resetCheckKey() |
| 1233 | * |
| 1234 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1235 | * @param int $holdoff HOLDOFF_TTL or HOLDOFF_TTL_NONE constant |
| 1236 | * @return bool True if the item was purged or not found, false on failure |
| 1237 | */ |
| 1238 | public function touchCheckKey( $key, $holdoff = self::HOLDOFF_TTL ) { |
| 1239 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1240 | $span = $this->startOperationSpan( __FUNCTION__, $key ); |
| 1241 | |
| 1242 | $checkSisterKey = $this->makeSisterKey( $key, self::TYPE_TIMESTAMP ); |
| 1243 | |
| 1244 | $now = $this->getCurrentTime(); |
| 1245 | $purge = $this->makeCheckPurgeValue( $now, $holdoff ); |
| 1246 | $ok = $this->cache->set( |
| 1247 | $this->getRouteKey( $checkSisterKey ), |
| 1248 | $purge, |
| 1249 | self::CHECK_KEY_TTL, |
| 1250 | $this->cache::WRITE_BACKGROUND |
| 1251 | ); |
| 1252 | |
| 1253 | $keygroup = $this->determineKeyGroupForStats( $key ); |
| 1254 | |
| 1255 | $this->stats->getCounter( 'wanobjectcache_check_total' ) |
| 1256 | ->setLabel( 'keygroup', $keygroup ) |
| 1257 | ->setLabel( 'result', ( $ok ? 'ok' : 'error' ) ) |
| 1258 | ->increment(); |
| 1259 | |
| 1260 | return $ok; |
| 1261 | } |
| 1262 | |
| 1263 | /** |
| 1264 | * Clear the last-purge timestamp of a "check" key in all datacenters |
| 1265 | * |
| 1266 | * Similar to touchCheckKey(), in that keys fetched using get*() calls, that include |
| 1267 | * the given "check" key, will be seen as purged. However, there are some differences: |
| 1268 | * - a) The "check" key will be deleted from all caches and lazily |
| 1269 | * re-initialized when accessed (rather than set everywhere) |
| 1270 | * - b) Thus, dependent keys will be known to be stale, but not |
| 1271 | * for how long (they are treated as "just" purged), which |
| 1272 | * effects any lockTSE logic in getWithSetCallback() |
| 1273 | * - c) Since "check" keys are initialized only on the server the key hashes |
| 1274 | * to, any temporary ejection of that server will cause the value to be |
| 1275 | * seen as purged as a new server will initialize the "check" key. |
| 1276 | * |
| 1277 | * The advantage over touchCheckKey() is that the "check" keys, which have high TTLs, |
| 1278 | * will only be created when a get*() method actually uses those keys. This is better |
| 1279 | * when a large number of "check" keys must be changed in a short period of time. |
| 1280 | * |
| 1281 | * Note that "check" keys won't collide with other regular keys. |
| 1282 | * |
| 1283 | * @see WANObjectCache::get() |
| 1284 | * @see WANObjectCache::getWithSetCallback() |
| 1285 | * @see WANObjectCache::touchCheckKey() |
| 1286 | * |
| 1287 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1288 | * @return bool True if the item was purged or not found, false on failure |
| 1289 | */ |
| 1290 | public function resetCheckKey( $key ) { |
| 1291 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1292 | $span = $this->startOperationSpan( __FUNCTION__, $key ); |
| 1293 | |
| 1294 | $checkSisterKey = $this->makeSisterKey( $key, self::TYPE_TIMESTAMP ); |
| 1295 | $ok = $this->cache->delete( $this->getRouteKey( $checkSisterKey ), $this->cache::WRITE_BACKGROUND ); |
| 1296 | |
| 1297 | $keygroup = $this->determineKeyGroupForStats( $key ); |
| 1298 | |
| 1299 | $this->stats->getCounter( 'wanobjectcache_reset_total' ) |
| 1300 | ->setLabel( 'keygroup', $keygroup ) |
| 1301 | ->setLabel( 'result', ( $ok ? 'ok' : 'error' ) ) |
| 1302 | ->increment(); |
| 1303 | |
| 1304 | return $ok; |
| 1305 | } |
| 1306 | |
| 1307 | /** |
| 1308 | * Method to fetch/regenerate a cache key |
| 1309 | * |
| 1310 | * On cache miss, the key will be set to the callback result via set() |
| 1311 | * (unless the callback returns false) and that result will be returned. |
| 1312 | * The arguments supplied to the callback are: |
| 1313 | * - $oldValue: prior cache value or false if none was present |
| 1314 | * - &$ttl: alterable reference to the TTL to be assigned to the new value |
| 1315 | * - &$setOpts: alterable reference to the set() options to be used with the new value |
| 1316 | * - $oldAsOf: generation UNIX timestamp of $oldValue or null if not present (since 1.28) |
| 1317 | * - $params: custom field/value map as defined by $cbParams (since 1.35) |
| 1318 | * |
| 1319 | * It is strongly recommended to set the 'lag' and 'since' fields to avoid race conditions |
| 1320 | * that can cause stale values to get stuck at keys. Usually, callbacks ignore the current |
| 1321 | * value, but it can be used to maintain "most recent X" values that come from time or |
| 1322 | * sequence based source data, provided that the "as of" id/time is tracked. Note that |
| 1323 | * preemptive regeneration and $checkKeys can result in a non-false current value. |
| 1324 | * |
| 1325 | * Usage of $checkKeys is similar to get() and getMulti(). However, rather than the caller |
| 1326 | * having to inspect a "current time left" variable (e.g. $curTTL, $curTTLs), a cache |
| 1327 | * regeneration will automatically be triggered using the callback. |
| 1328 | * |
| 1329 | * The $ttl argument and "hotTTR" option (in $opts) use time-dependent randomization |
| 1330 | * to avoid stampedes. Keys that are slow to regenerate and either heavily used |
| 1331 | * or subject to explicit (unpredictable) purges, may need additional mechanisms. |
| 1332 | * The simplest way to avoid stampedes for such keys is to use 'lockTSE' (in $opts). |
| 1333 | * If explicit purges are needed, also: |
| 1334 | * - a) Pass $key into $checkKeys |
| 1335 | * - b) Use touchCheckKey( $key ) instead of delete( $key ) |
| 1336 | * |
| 1337 | * This applies cache server I/O stampede protection against duplicate cache sets. |
| 1338 | * This is important when the callback is slow and/or yields large values for a key. |
| 1339 | * |
| 1340 | * Example usage (typical key): |
| 1341 | * @code |
| 1342 | * $catInfo = $cache->getWithSetCallback( |
| 1343 | * // Key to store the cached value under |
| 1344 | * $cache->makeKey( 'cat-attributes', $catId ), |
| 1345 | * // Time-to-live (in seconds) |
| 1346 | * $cache::TTL_MINUTE, |
| 1347 | * // Function that derives the new key value |
| 1348 | * function ( $oldValue, &$ttl, array &$setOpts ) { |
| 1349 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 1350 | * // Account for any snapshot/replica DB lag |
| 1351 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 1352 | * |
| 1353 | * return $dbr->selectRow( ... ); |
| 1354 | * } |
| 1355 | * ); |
| 1356 | * @endcode |
| 1357 | * |
| 1358 | * Example usage (key that is expensive and hot): |
| 1359 | * @code |
| 1360 | * $catConfig = $cache->getWithSetCallback( |
| 1361 | * // Key to store the cached value under |
| 1362 | * $cache->makeKey( 'site-cat-config' ), |
| 1363 | * // Time-to-live (in seconds) |
| 1364 | * $cache::TTL_DAY, |
| 1365 | * // Function that derives the new key value |
| 1366 | * function ( $oldValue, &$ttl, array &$setOpts ) { |
| 1367 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 1368 | * // Account for any snapshot/replica DB lag |
| 1369 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 1370 | * |
| 1371 | * return CatConfig::newFromRow( $dbr->selectRow( ... ) ); |
| 1372 | * }, |
| 1373 | * [ |
| 1374 | * // Calling touchCheckKey() on this key purges the cache |
| 1375 | * 'checkKeys' => [ $cache->makeKey( 'site-cat-config' ) ], |
| 1376 | * // Try to only let one datacenter thread manage cache updates at a time |
| 1377 | * 'lockTSE' => 30, |
| 1378 | * // Avoid querying cache servers multiple times in a web request |
| 1379 | * 'pcTTL' => $cache::TTL_PROC_LONG |
| 1380 | * ] |
| 1381 | * ); |
| 1382 | * @endcode |
| 1383 | * |
| 1384 | * Example usage (key with dynamic dependencies): |
| 1385 | * @code |
| 1386 | * $catState = $cache->getWithSetCallback( |
| 1387 | * // Key to store the cached value under |
| 1388 | * $cache->makeKey( 'cat-state', $cat->getId() ), |
| 1389 | * // Time-to-live (seconds) |
| 1390 | * $cache::TTL_HOUR, |
| 1391 | * // Function that derives the new key value |
| 1392 | * function ( $oldValue, &$ttl, array &$setOpts ) { |
| 1393 | * // Determine new value from the DB |
| 1394 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 1395 | * // Account for any snapshot/replica DB lag |
| 1396 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 1397 | * |
| 1398 | * return CatState::newFromResults( $dbr->select( ... ) ); |
| 1399 | * }, |
| 1400 | * [ |
| 1401 | * // The "check" keys that represent things the value depends on; |
| 1402 | * // Calling touchCheckKey() on any of them purges the cache |
| 1403 | * 'checkKeys' => [ |
| 1404 | * $cache->makeKey( 'sustenance-bowls', $cat->getRoomId() ), |
| 1405 | * $cache->makeKey( 'people-present', $cat->getHouseId() ), |
| 1406 | * $cache->makeKey( 'cat-laws', $cat->getCityId() ), |
| 1407 | * ] |
| 1408 | * ] |
| 1409 | * ); |
| 1410 | * @endcode |
| 1411 | * |
| 1412 | * Example usage (key that is expensive with too many DB dependencies for "check" keys): |
| 1413 | * @code |
| 1414 | * $catToys = $cache->getWithSetCallback( |
| 1415 | * // Key to store the cached value under |
| 1416 | * $cache->makeKey( 'cat-toys', $catId ), |
| 1417 | * // Time-to-live (seconds) |
| 1418 | * $cache::TTL_HOUR, |
| 1419 | * // Function that derives the new key value |
| 1420 | * function ( $oldValue, &$ttl, array &$setOpts ) { |
| 1421 | * // Determine new value from the DB |
| 1422 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 1423 | * // Account for any snapshot/replica DB lag |
| 1424 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 1425 | * |
| 1426 | * return CatToys::newFromResults( $dbr->select( ... ) ); |
| 1427 | * }, |
| 1428 | * [ |
| 1429 | * // Get the highest timestamp of any of the cat's toys |
| 1430 | * 'touchedCallback' => function ( $value ) use ( $catId ) { |
| 1431 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 1432 | * $ts = $dbr->selectField( 'cat_toys', 'MAX(ct_touched)', ... ); |
| 1433 | * |
| 1434 | * return wfTimestampOrNull( TS::UNIX, $ts ); |
| 1435 | * }, |
| 1436 | * // Avoid DB queries for repeated access |
| 1437 | * 'pcTTL' => $cache::TTL_PROC_SHORT |
| 1438 | * ] |
| 1439 | * ); |
| 1440 | * @endcode |
| 1441 | * |
| 1442 | * Example usage (hot key holding most recent 100 events): |
| 1443 | * @code |
| 1444 | * $lastCatActions = $cache->getWithSetCallback( |
| 1445 | * // Key to store the cached value under |
| 1446 | * $cache->makeKey( 'cat-last-actions', 100 ), |
| 1447 | * // Time-to-live (in seconds) |
| 1448 | * 10, |
| 1449 | * // Function that derives the new key value |
| 1450 | * function ( $oldValue, &$ttl, array &$setOpts ) { |
| 1451 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 1452 | * // Account for any snapshot/replica DB lag |
| 1453 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 1454 | * |
| 1455 | * // Start off with the last cached list |
| 1456 | * $list = $oldValue ?: []; |
| 1457 | * // Fetch the last 100 relevant rows in descending order; |
| 1458 | * // only fetch rows newer than $list[0] to reduce scanning |
| 1459 | * $rows = iterator_to_array( $dbr->select( ... ) ); |
| 1460 | * // Merge them and get the new "last 100" rows |
| 1461 | * return array_slice( array_merge( $new, $list ), 0, 100 ); |
| 1462 | * }, |
| 1463 | * [ |
| 1464 | * // Try to only let one datacenter thread manage cache updates at a time |
| 1465 | * 'lockTSE' => 30, |
| 1466 | * // Use a magic value when no cache value is ready rather than stampeding |
| 1467 | * 'busyValue' => 'computing' |
| 1468 | * ] |
| 1469 | * ); |
| 1470 | * @endcode |
| 1471 | * |
| 1472 | * Example usage (key holding an LRU subkey:value map; this can avoid flooding cache with |
| 1473 | * keys for an unlimited set of (constraint,situation) pairs, thereby avoiding elevated |
| 1474 | * cache evictions and wasted memory): |
| 1475 | * @code |
| 1476 | * $catSituationTolerabilityCache = $this->cache->getWithSetCallback( |
| 1477 | * // Group by constraint ID/hash, cat family ID/hash, or something else useful |
| 1478 | * $this->cache->makeKey( 'cat-situation-tolerability-checks', $groupKey ), |
| 1479 | * WANObjectCache::TTL_DAY, // rarely used groups should fade away |
| 1480 | * // The $scenarioKey format is $constraintId:<ID/hash of $situation> |
| 1481 | * function ( $cacheMap ) use ( $scenarioKey, $constraintId, $situation ) { |
| 1482 | * $lruCache = MapCacheLRU::newFromArray( $cacheMap ?: [], self::CACHE_SIZE ); |
| 1483 | * $result = $lruCache->get( $scenarioKey ); // triggers LRU bump if present |
| 1484 | * if ( $result === null || $this->isScenarioResultExpired( $result ) ) { |
| 1485 | * $result = $this->checkScenarioTolerability( $constraintId, $situation ); |
| 1486 | * $lruCache->set( $scenarioKey, $result, 3 / 8 ); |
| 1487 | * } |
| 1488 | * // Save the new LRU cache map and reset the map's TTL |
| 1489 | * return $lruCache->toArray(); |
| 1490 | * }, |
| 1491 | * [ |
| 1492 | * // Once map is > 1 sec old, consider refreshing |
| 1493 | * 'ageNew' => 1, |
| 1494 | * // Update within 5 seconds after "ageNew" given a 1hz cache check rate |
| 1495 | * 'hotTTR' => 5, |
| 1496 | * // Avoid querying cache servers multiple times in a request; this also means |
| 1497 | * // that a request can only alter the value of any given constraint key once |
| 1498 | * 'pcTTL' => WANObjectCache::TTL_PROC_LONG |
| 1499 | * ] |
| 1500 | * ); |
| 1501 | * $tolerability = isset( $catSituationTolerabilityCache[$scenarioKey] ) |
| 1502 | * ? $catSituationTolerabilityCache[$scenarioKey] |
| 1503 | * : $this->checkScenarioTolerability( $constraintId, $situation ); |
| 1504 | * @endcode |
| 1505 | * |
| 1506 | * @see WANObjectCache::get() |
| 1507 | * @see WANObjectCache::set() |
| 1508 | * |
| 1509 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1510 | * @param int $ttl Nominal seconds-to-live for newly computed values. Special values are: |
| 1511 | * - WANObjectCache::TTL_INDEFINITE: Cache forever (subject to LRU-style evictions) |
| 1512 | * - WANObjectCache::TTL_UNCACHEABLE: Do not cache (if the key exists, it is not deleted) |
| 1513 | * @param callable $callback Value generation function |
| 1514 | * @param array $opts Options map: |
| 1515 | * - checkKeys: List of "check" keys. The key at $key will be seen as stale when either |
| 1516 | * touchCheckKey() or resetCheckKey() is called on any of the keys in this list. This |
| 1517 | * is useful if thousands or millions of keys depend on the same entity. The entity can |
| 1518 | * simply have its "check" key updated whenever the entity is modified. |
| 1519 | * Default: []. |
| 1520 | * - graceTTL: If the key is stale due to a purge (by "checkKeys" or "touchedCallback") |
| 1521 | * less than this many seconds ago, consider reusing the stale value. The odds of a |
| 1522 | * refresh become more likely over time, becoming certain once the grace period is |
| 1523 | * reached. This can reduce traffic spikes when millions of keys are compared to the |
| 1524 | * same "check" key and touchCheckKey() or resetCheckKey() is called on that "check" key. |
| 1525 | * This option is not useful for avoiding traffic spikes in the case of the key simply |
| 1526 | * expiring on account of its TTL (use "lowTTL" instead). |
| 1527 | * Default: WANObjectCache::GRACE_TTL_NONE. |
| 1528 | * - lockTSE: If the value is stale and the "time since expiry" (TSE) is less than the given |
| 1529 | * number of seconds ago, then reuse the stale value if another such thread is already |
| 1530 | * regenerating the value. The TSE of the key is influenced by purges (e.g. via delete(), |
| 1531 | * "checkKeys", "touchedCallback"), and various other options (e.g. "staleTTL"). A low |
| 1532 | * enough TSE is assumed to indicate a high enough key access rate to justify stampede |
| 1533 | * avoidance. Note that no cache value exists after deletion, expiration, or eviction |
| 1534 | * at the storage-layer; to prevent stampedes during these cases, use "busyValue". |
| 1535 | * Default: WANObjectCache::TSE_NONE. |
| 1536 | * - busyValue: Specify a placeholder value to use when no value exists and another thread |
| 1537 | * is currently regenerating it. This assures that cache stampedes cannot happen if the |
| 1538 | * value falls out of cache. This also mitigates stampedes when value regeneration |
| 1539 | * becomes very slow (greater than $ttl/"lowTTL"). If this is a closure, then it will |
| 1540 | * be invoked to get the placeholder when needed. |
| 1541 | * Default: null. |
| 1542 | * - pcTTL: Process cache the value in this PHP instance for this many seconds. This avoids |
| 1543 | * network I/O when a key is read several times. This will not cache when the callback |
| 1544 | * returns false, however. Note that any purges will not be seen while process cached; |
| 1545 | * since the callback should use replica DBs and they may be lagged or have snapshot |
| 1546 | * isolation anyway, this should not typically matter. |
| 1547 | * Default: WANObjectCache::TTL_UNCACHEABLE. |
| 1548 | * - pcGroup: Process cache group to use instead of the primary one. If set, this must be |
| 1549 | * of the format ALPHANUMERIC_NAME:MAX_KEY_SIZE, e.g. "mydata:10". Use this for storing |
| 1550 | * large values, small yet numerous values, or some values with a high cost of eviction. |
| 1551 | * It is generally preferable to use a class constant when setting this value. |
| 1552 | * This has no effect unless pcTTL is used. |
| 1553 | * Default: WANObjectCache::PC_PRIMARY. |
| 1554 | * - version: Integer version number. This lets callers make breaking changes to the format |
| 1555 | * of cached values without causing problems for sites that use non-instantaneous code |
| 1556 | * deployments. Old and new code will recognize incompatible versions and purges from |
| 1557 | * both old and new code will been seen by each other. When this method encounters an |
| 1558 | * incompatibly versioned value at the provided key, a "variant key" will be used for |
| 1559 | * reading from and saving to cache. The variant key is specific to the key and version |
| 1560 | * number provided to this method. If the variant key value is older than that of the |
| 1561 | * provided key, or the provided key is non-existant, then the variant key will be seen |
| 1562 | * as non-existant. Therefore, delete() calls purge the provided key's variant keys. |
| 1563 | * The "checkKeys" and "touchedCallback" options still apply to variant keys as usual. |
| 1564 | * Avoid storing class objects, as this reduces compatibility (due to serialization). |
| 1565 | * Default: null. |
| 1566 | * - minAsOf: Reject values if they were generated before this UNIX timestamp. |
| 1567 | * This is useful if the source of a key is suspected of having possibly changed |
| 1568 | * recently, and the caller wants any such changes to be reflected. |
| 1569 | * Default: WANObjectCache::MIN_TIMESTAMP_NONE. |
| 1570 | * - lowTTL: Consider pre-emptive updates once the current TTL (seconds) of the key is less |
| 1571 | * than this. It becomes more likely over time, and is more likely when a key is popular, |
| 1572 | * becoming certain as the key approaches expiry. This can avoid cache stampedes by |
| 1573 | * regenerating your value on a random request before the key expires, instead of letting |
| 1574 | * it expire and all requests having to regen together. |
| 1575 | * This feature is enabled by default (LOW_TTL=60s) but should have little to no impact |
| 1576 | * on most keys, while automatically avoiding stampedes on hot keys. Set to 0 to disable. |
| 1577 | * See also: WANObjectCache::worthRefreshExpiring. |
| 1578 | * Default: WANObjectCache::LOW_TTL. |
| 1579 | * - hotTTR: Schedule pre-emptive updates on popular keys once every $hotTTR seconds. |
| 1580 | * If one of your keys becomes popular (more than 1 req/s), we schedule an async refresh on |
| 1581 | * a random request roughly once every $hotTTR seconds. For keys with an even higher rate, |
| 1582 | * this will happen sooner. During the first few seconds after a value is generated, |
| 1583 | * this option is ignored as controlled by the "ageNew" option. This feature is enabled by |
| 1584 | * default to avoid stale data on heavily referenced keys (e.g. due to lost purges), |
| 1585 | * and should only impact keys that are hot. Set to 0 to disable this feature. |
| 1586 | * See also: WANObjectCache::worthRefreshPopular. |
| 1587 | * Default: WANObjectCache::HOT_TTR. |
| 1588 | * - ageNew: Start pre-emptive "hotTTR" updates after a key reaches this age in seconds. |
| 1589 | * Set `hotTTR: 0` to disable this feature. Setting ageNew to zero does not disable |
| 1590 | * the hotTTR feature. |
| 1591 | * Default: WANObjectCache::AGE_NEW. |
| 1592 | * - staleTTL: Seconds to keep the key around if it is stale. This means that on cache |
| 1593 | * miss the callback may get $oldValue/$oldAsOf values for keys that have already been |
| 1594 | * expired for this specified time. This is useful if adaptiveTTL() is used on the old |
| 1595 | * value's as-of time when it is verified as still being correct. |
| 1596 | * Default: WANObjectCache::STALE_TTL_NONE |
| 1597 | * - touchedCallback: A callback that takes the current value and returns a UNIX timestamp |
| 1598 | * indicating the last time a dynamic dependency changed. Null can be returned if there |
| 1599 | * are no relevant dependency changes to check. This can be used to check against things |
| 1600 | * like last-modified times of files or DB timestamp fields. This should generally not be |
| 1601 | * used for small and easily queried values in a DB if the callback itself ends up doing |
| 1602 | * a similarly expensive DB query to check a timestamp. Usages of this option makes the |
| 1603 | * most sense for values that are moderately to highly expensive to regenerate and easy |
| 1604 | * to query for dependency timestamps. The use of "pcTTL" reduces timestamp queries. |
| 1605 | * Default: null. |
| 1606 | * - segmentable: Allow partitioning of the value if it is a large string. Default: false. |
| 1607 | * |
| 1608 | * @param array $cbParams Custom field/value map to pass to the callback (since 1.35) |
| 1609 | * @phpcs:ignore Generic.Files.LineLength |
| 1610 | * @phan-param array{checkKeys?:string[],graceTTL?:int,lockTSE?:int,busyValue?:mixed,pcTTL?:int,pcGroup?:string,version?:int,minAsOf?:float|int,hotTTR?:int,lowTTL?:int,ageNew?:int,staleTTL?:int,touchedCallback?:callable,segmentable?:bool} $opts |
| 1611 | * @return mixed Value found or written to the key |
| 1612 | * @note Options added in 1.28: version, busyValue, hotTTR, ageNew, pcGroup, minAsOf |
| 1613 | * @note Options added in 1.31: staleTTL, graceTTL |
| 1614 | * @note Options added in 1.33: touchedCallback |
| 1615 | * @note Callable type hints are not used to avoid class-autoloading |
| 1616 | */ |
| 1617 | final public function getWithSetCallback( |
| 1618 | $key, $ttl, $callback, array $opts = [], array $cbParams = [] |
| 1619 | ) { |
| 1620 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1621 | $span = $this->startOperationSpan( __FUNCTION__, $key ); |
| 1622 | |
| 1623 | $version = $opts['version'] ?? null; |
| 1624 | $pcTTL = $opts['pcTTL'] ?? self::TTL_UNCACHEABLE; |
| 1625 | $pCache = ( $pcTTL >= 0 ) |
| 1626 | ? $this->getProcessCache( $opts['pcGroup'] ?? self::PC_PRIMARY ) |
| 1627 | : null; |
| 1628 | |
| 1629 | // Use the process cache if requested as long as no outer cache callback is running. |
| 1630 | // Nested callback process cache use is not lag-safe with regard to HOLDOFF_TTL since |
| 1631 | // process cached values are more lagged than persistent ones as they are not purged. |
| 1632 | if ( $pCache && $this->callbackDepth == 0 ) { |
| 1633 | $cached = $pCache->get( $key, $pcTTL, false ); |
| 1634 | if ( $cached !== false ) { |
| 1635 | $this->logger->debug( "getWithSetCallback($key): process cache hit" ); |
| 1636 | return $cached; |
| 1637 | } |
| 1638 | } |
| 1639 | |
| 1640 | [ $value, $valueVersion, $curAsOf ] = $this->fetchOrRegenerate( $key, $ttl, $callback, $opts, $cbParams ); |
| 1641 | if ( $valueVersion !== $version ) { |
| 1642 | // Current value has a different version; use the variant key for this version. |
| 1643 | // Regenerate the variant value if it is not newer than the main value at $key |
| 1644 | // so that purges to the main key propagate to the variant value. |
| 1645 | $this->logger->debug( "getWithSetCallback($key): using variant key" ); |
| 1646 | [ $value ] = $this->fetchOrRegenerate( |
| 1647 | $this->makeGlobalKey( 'WANCache-key-variant', md5( $key ), (string)$version ), |
| 1648 | $ttl, |
| 1649 | $callback, |
| 1650 | [ 'version' => null, 'minAsOf' => $curAsOf ] + $opts, |
| 1651 | $cbParams |
| 1652 | ); |
| 1653 | } |
| 1654 | |
| 1655 | // Update the process cache if enabled |
| 1656 | if ( $pCache && $value !== false ) { |
| 1657 | $pCache->set( $key, $value ); |
| 1658 | } |
| 1659 | |
| 1660 | return $value; |
| 1661 | } |
| 1662 | |
| 1663 | /** |
| 1664 | * Do the actual I/O for getWithSetCallback() when needed |
| 1665 | * |
| 1666 | * @see WANObjectCache::getWithSetCallback() |
| 1667 | * |
| 1668 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1669 | * @param int $ttl |
| 1670 | * @param callable $callback |
| 1671 | * @param array $opts |
| 1672 | * @param array $cbParams |
| 1673 | * @return array Ordered list of the following: |
| 1674 | * - Cached or regenerated value |
| 1675 | * - Cached or regenerated value version number or null if not versioned |
| 1676 | * - Timestamp of the current cached value at the key or null if there is no value |
| 1677 | * @note Callable type hints are not used to avoid class-autoloading |
| 1678 | */ |
| 1679 | private function fetchOrRegenerate( $key, $ttl, $callback, array $opts, array $cbParams ) { |
| 1680 | $checkKeys = $opts['checkKeys'] ?? []; |
| 1681 | $graceTTL = $opts['graceTTL'] ?? self::GRACE_TTL_NONE; |
| 1682 | $minAsOf = $opts['minAsOf'] ?? self::MIN_TIMESTAMP_NONE; |
| 1683 | $hotTTR = $opts['hotTTR'] ?? self::HOT_TTR; |
| 1684 | $lowTTL = $opts['lowTTL'] ?? min( self::LOW_TTL, $ttl ); |
| 1685 | $ageNew = $opts['ageNew'] ?? self::AGE_NEW; |
| 1686 | $touchedCb = $opts['touchedCallback'] ?? null; |
| 1687 | $startTime = $this->getCurrentTime(); |
| 1688 | |
| 1689 | $keygroup = $this->determineKeyGroupForStats( $key ); |
| 1690 | |
| 1691 | // Get the current key value and its metadata |
| 1692 | $curState = $this->fetchKeys( [ $key ], $checkKeys, $startTime, $touchedCb )[$key]; |
| 1693 | $curValue = $curState[self::RES_VALUE]; |
| 1694 | |
| 1695 | // Use the cached value if it exists and is not due for synchronous regeneration |
| 1696 | if ( $this->isAcceptablyFreshValue( $curState, $graceTTL, $minAsOf ) ) { |
| 1697 | if ( !$this->isLotteryRefreshDue( $curState, $lowTTL, $ageNew, $hotTTR, $startTime ) ) { |
| 1698 | $this->stats->getTiming( 'wanobjectcache_getwithset_seconds' ) |
| 1699 | ->setLabel( 'keygroup', $keygroup ) |
| 1700 | ->setLabel( 'result', 'hit' ) |
| 1701 | ->setLabel( 'reason', 'good' ) |
| 1702 | ->observe( 1e3 * ( $this->getCurrentTime() - $startTime ) ); |
| 1703 | |
| 1704 | return [ $curValue, $curState[self::RES_VERSION], $curState[self::RES_AS_OF] ]; |
| 1705 | } elseif ( $this->scheduleAsyncRefresh( $key, $ttl, $callback, $opts, $cbParams ) ) { |
| 1706 | $this->logger->debug( "fetchOrRegenerate($key): hit with async refresh" ); |
| 1707 | |
| 1708 | $this->stats->getTiming( 'wanobjectcache_getwithset_seconds' ) |
| 1709 | ->setLabel( 'keygroup', $keygroup ) |
| 1710 | ->setLabel( 'result', 'hit' ) |
| 1711 | ->setLabel( 'reason', 'refresh' ) |
| 1712 | ->observe( 1e3 * ( $this->getCurrentTime() - $startTime ) ); |
| 1713 | |
| 1714 | return [ $curValue, $curState[self::RES_VERSION], $curState[self::RES_AS_OF] ]; |
| 1715 | } else { |
| 1716 | $this->logger->debug( "fetchOrRegenerate($key): hit with sync refresh" ); |
| 1717 | } |
| 1718 | } |
| 1719 | |
| 1720 | $isKeyTombstoned = ( $curState[self::RES_TOMB_AS_OF] !== null ); |
| 1721 | // Use the interim key as a temporary alternative if the key is tombstoned |
| 1722 | if ( $isKeyTombstoned ) { |
| 1723 | $volState = $this->getInterimValue( $key, $minAsOf, $startTime, $touchedCb ); |
| 1724 | $volValue = $volState[self::RES_VALUE]; |
| 1725 | } else { |
| 1726 | $volState = $curState; |
| 1727 | $volValue = $curValue; |
| 1728 | } |
| 1729 | |
| 1730 | // During the volatile "hold-off" period that follows a purge of the key, the value |
| 1731 | // will be regenerated many times if frequently accessed. This is done to mitigate |
| 1732 | // the effects of backend replication lag as soon as possible. However, throttle the |
| 1733 | // overhead of locking and regeneration by reusing values recently written to cache |
| 1734 | // tens of milliseconds ago. Verify the "as of" time against the last purge event. |
| 1735 | $lastPurgeTime = max( |
| 1736 | // RES_TOUCH_AS_OF depends on the value (possibly from the interim key) |
| 1737 | $volState[self::RES_TOUCH_AS_OF], |
| 1738 | $curState[self::RES_TOMB_AS_OF], |
| 1739 | $curState[self::RES_CHECK_AS_OF] |
| 1740 | ); |
| 1741 | $safeMinAsOf = max( $minAsOf, $lastPurgeTime + self::TINY_POSITIVE ); |
| 1742 | |
| 1743 | if ( $volState[self::RES_VALUE] === false || $volState[self::RES_AS_OF] < $safeMinAsOf ) { |
| 1744 | $isExtremelyNewValue = false; |
| 1745 | } else { |
| 1746 | $age = $startTime - $volState[self::RES_AS_OF]; |
| 1747 | $isExtremelyNewValue = ( $age < mt_rand( self::RECENT_SET_LOW_MS, self::RECENT_SET_HIGH_MS ) / 1e3 ); |
| 1748 | } |
| 1749 | if ( $isExtremelyNewValue ) { |
| 1750 | $this->logger->debug( "fetchOrRegenerate($key): volatile hit" ); |
| 1751 | |
| 1752 | $this->stats->getTiming( 'wanobjectcache_getwithset_seconds' ) |
| 1753 | ->setLabel( 'keygroup', $keygroup ) |
| 1754 | ->setLabel( 'result', 'hit' ) |
| 1755 | ->setLabel( 'reason', 'volatile' ) |
| 1756 | ->observe( 1e3 * ( $this->getCurrentTime() - $startTime ) ); |
| 1757 | |
| 1758 | return [ $volValue, $volState[self::RES_VERSION], $curState[self::RES_AS_OF] ]; |
| 1759 | } |
| 1760 | |
| 1761 | $lockTSE = $opts['lockTSE'] ?? self::TSE_NONE; |
| 1762 | $busyValue = $opts['busyValue'] ?? null; |
| 1763 | $staleTTL = $opts['staleTTL'] ?? self::STALE_TTL_NONE; |
| 1764 | $segmentable = $opts['segmentable'] ?? false; |
| 1765 | $version = $opts['version'] ?? null; |
| 1766 | |
| 1767 | // Determine whether one thread per datacenter should handle regeneration at a time |
| 1768 | $useRegenerationLock = |
| 1769 | // Note that since tombstones no-op set(), $lockTSE and $curTTL cannot be used to |
| 1770 | // deduce the key hotness because |$curTTL| will always keep increasing until the |
| 1771 | // tombstone expires or is overwritten by a new tombstone. Also, even if $lockTSE |
| 1772 | // is not set, constant regeneration of a key for the tombstone lifetime might be |
| 1773 | // very expensive. Assume tombstoned keys are possibly hot in order to reduce |
| 1774 | // the risk of high regeneration load after the delete() method is called. |
| 1775 | $isKeyTombstoned || |
| 1776 | // Assume a key is hot if requested soon ($lockTSE seconds) after purge. |
| 1777 | // This avoids stampedes when timestamps from $checkKeys/$touchedCb bump. |
| 1778 | ( |
| 1779 | $curState[self::RES_CUR_TTL] !== null && |
| 1780 | $curState[self::RES_CUR_TTL] <= 0 && |
| 1781 | abs( $curState[self::RES_CUR_TTL] ) <= $lockTSE |
| 1782 | ) || |
| 1783 | // Assume a key is hot if there is no value and a busy fallback is given. |
| 1784 | // This avoids stampedes on eviction or preemptive regeneration taking too long. |
| 1785 | ( $busyValue !== null && $volValue === false ); |
| 1786 | |
| 1787 | // If a regeneration lock is required, threads that do not get the lock will try to use |
| 1788 | // the stale value, the interim value, or the $busyValue placeholder, in that order. If |
| 1789 | // none of those are set then all threads will bypass the lock and regenerate the value. |
| 1790 | $mutexKey = $this->makeSisterKey( $key, self::TYPE_MUTEX ); |
| 1791 | // Note that locking is not bypassed due to I/O errors; this avoids stampedes |
| 1792 | $hasLock = $useRegenerationLock && $this->cache->add( $mutexKey, 1, self::LOCK_TTL ); |
| 1793 | if ( $useRegenerationLock && !$hasLock ) { |
| 1794 | // Determine if there is stale or volatile cached value that is still usable |
| 1795 | // @phan-suppress-next-line PhanTypeMismatchArgumentNullable False positive |
| 1796 | if ( $this->isValid( $volValue, $volState[self::RES_AS_OF], $minAsOf ) ) { |
| 1797 | $this->logger->debug( "fetchOrRegenerate($key): returning stale value" ); |
| 1798 | |
| 1799 | $this->stats->getTiming( 'wanobjectcache_getwithset_seconds' ) |
| 1800 | ->setLabel( 'keygroup', $keygroup ) |
| 1801 | ->setLabel( 'result', 'hit' ) |
| 1802 | ->setLabel( 'reason', 'stale' ) |
| 1803 | ->observe( 1e3 * ( $this->getCurrentTime() - $startTime ) ); |
| 1804 | |
| 1805 | return [ $volValue, $volState[self::RES_VERSION], $curState[self::RES_AS_OF] ]; |
| 1806 | } elseif ( $busyValue !== null ) { |
| 1807 | $miss = is_infinite( $minAsOf ) ? 'renew' : 'miss'; |
| 1808 | $this->logger->debug( "fetchOrRegenerate($key): busy $miss" ); |
| 1809 | |
| 1810 | $this->stats->getTiming( 'wanobjectcache_getwithset_seconds' ) |
| 1811 | ->setLabel( 'keygroup', $keygroup ) |
| 1812 | ->setLabel( 'result', $miss ) |
| 1813 | ->setLabel( 'reason', 'busy' ) |
| 1814 | ->observe( 1e3 * ( $this->getCurrentTime() - $startTime ) ); |
| 1815 | |
| 1816 | $placeholderValue = ( $busyValue instanceof Closure ) ? $busyValue() : $busyValue; |
| 1817 | |
| 1818 | return [ $placeholderValue, $version, $curState[self::RES_AS_OF] ]; |
| 1819 | } |
| 1820 | } |
| 1821 | |
| 1822 | // Generate the new value given any prior value with a matching version |
| 1823 | $setOpts = []; |
| 1824 | $preCallbackTime = $this->getCurrentTime(); |
| 1825 | ++$this->callbackDepth; |
| 1826 | // https://github.com/phan/phan/issues/4419 |
| 1827 | /** @noinspection PhpUnusedLocalVariableInspection */ |
| 1828 | $value = null; |
| 1829 | try { |
| 1830 | $value = $callback( |
| 1831 | ( $curState[self::RES_VERSION] === $version ) ? $curValue : false, |
| 1832 | $ttl, |
| 1833 | $setOpts, |
| 1834 | ( $curState[self::RES_VERSION] === $version ) ? $curState[self::RES_AS_OF] : null, |
| 1835 | $cbParams |
| 1836 | ); |
| 1837 | } finally { |
| 1838 | --$this->callbackDepth; |
| 1839 | } |
| 1840 | $postCallbackTime = $this->getCurrentTime(); |
| 1841 | |
| 1842 | // How long it took to generate the value |
| 1843 | $walltime = max( $postCallbackTime - $preCallbackTime, 0.0 ); |
| 1844 | |
| 1845 | $this->stats->getTiming( 'wanobjectcache_regen_seconds' ) |
| 1846 | ->setLabel( 'keygroup', $keygroup ) |
| 1847 | ->observe( 1e3 * $walltime ); |
| 1848 | |
| 1849 | // Attempt to save the newly generated value if applicable |
| 1850 | if ( |
| 1851 | // Callback yielded a cacheable value |
| 1852 | ( $value !== false && $ttl >= 0 ) && |
| 1853 | // Current thread was not raced out of a regeneration lock or key is tombstoned |
| 1854 | ( !$useRegenerationLock || $hasLock || $isKeyTombstoned ) |
| 1855 | ) { |
| 1856 | // If the key is write-holed then use the (volatile) interim key as an alternative |
| 1857 | if ( $isKeyTombstoned ) { |
| 1858 | $this->setInterimValue( |
| 1859 | $key, |
| 1860 | $value, |
| 1861 | $lockTSE, |
| 1862 | $version, |
| 1863 | $segmentable |
| 1864 | ); |
| 1865 | } else { |
| 1866 | $this->setMainValue( |
| 1867 | $key, |
| 1868 | $value, |
| 1869 | $ttl, |
| 1870 | $version, |
| 1871 | $walltime, |
| 1872 | // @phan-suppress-next-line PhanCoalescingAlwaysNull |
| 1873 | $setOpts['lag'] ?? 0, |
| 1874 | // @phan-suppress-next-line PhanCoalescingAlwaysNull |
| 1875 | $setOpts['since'] ?? $preCallbackTime, |
| 1876 | // @phan-suppress-next-line PhanCoalescingAlwaysNull |
| 1877 | $setOpts['pending'] ?? false, |
| 1878 | $lockTSE, |
| 1879 | $staleTTL, |
| 1880 | $segmentable, |
| 1881 | ( $curValue === false ) |
| 1882 | ); |
| 1883 | } |
| 1884 | } |
| 1885 | |
| 1886 | if ( $hasLock ) { |
| 1887 | $this->cache->delete( $mutexKey, $this->cache::WRITE_BACKGROUND ); |
| 1888 | } |
| 1889 | |
| 1890 | $miss = is_infinite( $minAsOf ) ? 'renew' : 'miss'; |
| 1891 | $this->logger->debug( "fetchOrRegenerate($key): $miss, new value computed" ); |
| 1892 | |
| 1893 | $this->stats->getTiming( 'wanobjectcache_getwithset_seconds' ) |
| 1894 | ->setLabel( 'keygroup', $keygroup ) |
| 1895 | ->setLabel( 'result', $miss ) |
| 1896 | ->setLabel( 'reason', 'compute' ) |
| 1897 | ->observe( 1e3 * ( $this->getCurrentTime() - $startTime ) ); |
| 1898 | |
| 1899 | return [ $value, $version, $curState[self::RES_AS_OF] ]; |
| 1900 | } |
| 1901 | |
| 1902 | /** |
| 1903 | * Get a sister key that should be collocated with a base cache key |
| 1904 | * |
| 1905 | * The keys will bear the WANCache prefix and use the configured coalescing scheme |
| 1906 | * |
| 1907 | * @param string $baseKey Cache key made with makeKey()/makeGlobalKey() |
| 1908 | * @param string $typeChar Consistent hashing agnostic suffix character matching [a-zA-Z] |
| 1909 | * @return string Sister key |
| 1910 | */ |
| 1911 | private function makeSisterKey( string $baseKey, string $typeChar ) { |
| 1912 | if ( $this->coalesceScheme === self::SCHEME_HASH_STOP ) { |
| 1913 | // Key style: "WANCache:<base key>|#|<character>" |
| 1914 | $sisterKey = 'WANCache:' . $baseKey . '|#|' . $typeChar; |
| 1915 | } else { |
| 1916 | // Key style: "WANCache:{<base key>}:<character>" |
| 1917 | $sisterKey = 'WANCache:{' . $baseKey . '}:' . $typeChar; |
| 1918 | } |
| 1919 | return $sisterKey; |
| 1920 | } |
| 1921 | |
| 1922 | /** |
| 1923 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1924 | * @param float $minAsOf Minimum acceptable value "as of" UNIX timestamp |
| 1925 | * @param float $now Fetch time to determine "age" metadata |
| 1926 | * @param callable|null $touchedCb Function to find the max "dependency touched" UNIX timestamp |
| 1927 | * @return array<int,mixed> Result map/n-tuple from unwrap() |
| 1928 | * @phan-return array{0:mixed,1:mixed,2:?float,3:?int,4:?float,5:?float,6:?float,7:?float} |
| 1929 | * @note Callable type hints are not used to avoid class-autoloading |
| 1930 | */ |
| 1931 | private function getInterimValue( $key, $minAsOf, $now, $touchedCb ) { |
| 1932 | if ( $this->useInterimHoldOffCaching ) { |
| 1933 | $interimSisterKey = $this->makeSisterKey( $key, self::TYPE_INTERIM ); |
| 1934 | $wrapped = $this->cache->get( $interimSisterKey ); |
| 1935 | $res = $this->unwrap( $wrapped, $now ); |
| 1936 | if ( $res[self::RES_VALUE] !== false && $res[self::RES_AS_OF] >= $minAsOf ) { |
| 1937 | if ( $touchedCb !== null ) { |
| 1938 | // Update "last purge time" since the $touchedCb timestamp depends on $value |
| 1939 | // Get the new "touched timestamp", accounting for callback-checked dependencies |
| 1940 | $res[self::RES_TOUCH_AS_OF] = max( |
| 1941 | $touchedCb( $res[self::RES_VALUE] ), |
| 1942 | $res[self::RES_TOUCH_AS_OF] |
| 1943 | ); |
| 1944 | } |
| 1945 | |
| 1946 | return $res; |
| 1947 | } |
| 1948 | } |
| 1949 | |
| 1950 | return $this->unwrap( false, $now ); |
| 1951 | } |
| 1952 | |
| 1953 | /** |
| 1954 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 1955 | * @param mixed $value |
| 1956 | * @param int|float $ttl |
| 1957 | * @param int|null $version Value version number |
| 1958 | * @param bool $segmentable |
| 1959 | * @return bool Success |
| 1960 | */ |
| 1961 | private function setInterimValue( |
| 1962 | $key, |
| 1963 | $value, |
| 1964 | $ttl, |
| 1965 | ?int $version, |
| 1966 | bool $segmentable |
| 1967 | ) { |
| 1968 | $now = $this->getCurrentTime(); |
| 1969 | $ttl = max( self::INTERIM_KEY_TTL, (int)$ttl ); |
| 1970 | |
| 1971 | // Wrap that value with time/TTL/version metadata |
| 1972 | $wrapped = $this->wrap( $value, $ttl, $version, $now ); |
| 1973 | |
| 1974 | $flags = $this->cache::WRITE_BACKGROUND; |
| 1975 | if ( $segmentable ) { |
| 1976 | $flags |= $this->cache::WRITE_ALLOW_SEGMENTS; |
| 1977 | } |
| 1978 | |
| 1979 | return $this->cache->set( |
| 1980 | $this->makeSisterKey( $key, self::TYPE_INTERIM ), |
| 1981 | $wrapped, |
| 1982 | $ttl, |
| 1983 | $flags |
| 1984 | ); |
| 1985 | } |
| 1986 | |
| 1987 | /** |
| 1988 | * Method to fetch multiple cache keys at once with regeneration |
| 1989 | * |
| 1990 | * This works the same as getWithSetCallback() except: |
| 1991 | * - a) The $keys argument must be the result of WANObjectCache::makeMultiKeys() |
| 1992 | * - b) The $callback argument expects a function that returns an entity value, using |
| 1993 | * boolean "false" if it does not exist. The callback takes the following arguments: |
| 1994 | * - $id: ID of the entity to query |
| 1995 | * - $oldValue: prior cache value or false if none was present |
| 1996 | * - &$ttl: reference to the TTL to be assigned to the new value (alterable) |
| 1997 | * - &$setOpts: reference to the new value set() options (alterable) |
| 1998 | * - $oldAsOf: generation UNIX timestamp of $oldValue or null if not present |
| 1999 | * - c) The return value is a map of (cache key => value) in the order of $keyedIds |
| 2000 | * |
| 2001 | * @see WANObjectCache::getWithSetCallback() |
| 2002 | * @see WANObjectCache::getMultiWithUnionSetCallback() |
| 2003 | * |
| 2004 | * Example usage: |
| 2005 | * @code |
| 2006 | * $rows = $cache->getMultiWithSetCallback( |
| 2007 | * // Map of cache keys to entity IDs |
| 2008 | * $cache->makeMultiKeys( |
| 2009 | * $this->fileVersionIds(), |
| 2010 | * function ( $id, $cache ) { |
| 2011 | * return $cache->makeKey( 'file-version', $id ); |
| 2012 | * } |
| 2013 | * ), |
| 2014 | * // Time-to-live (in seconds) |
| 2015 | * $cache::TTL_DAY, |
| 2016 | * // Function that derives the new key value |
| 2017 | * function ( $id, $oldValue, &$ttl, array &$setOpts ) { |
| 2018 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 2019 | * // Account for any snapshot/replica DB lag |
| 2020 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 2021 | * |
| 2022 | * // Load the row for this file |
| 2023 | * $queryInfo = File::getQueryInfo(); |
| 2024 | * $row = $dbr->selectRow( |
| 2025 | * $queryInfo['tables'], |
| 2026 | * $queryInfo['fields'], |
| 2027 | * [ 'id' => $id ], |
| 2028 | * __METHOD__, |
| 2029 | * [], |
| 2030 | * $queryInfo['joins'] |
| 2031 | * ); |
| 2032 | * |
| 2033 | * return $row ? (array)$row : false; |
| 2034 | * }, |
| 2035 | * [ |
| 2036 | * // Process cache for 30 seconds |
| 2037 | * 'pcTTL' => 30, |
| 2038 | * // Use a dedicated 500 item cache (initialized on-the-fly) |
| 2039 | * 'pcGroup' => 'file-versions:500' |
| 2040 | * ] |
| 2041 | * ); |
| 2042 | * $files = array_map( self::newFromRow( ... ), $rows ); |
| 2043 | * @endcode |
| 2044 | * |
| 2045 | * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys() |
| 2046 | * @param int $ttl Seconds to live for key updates |
| 2047 | * @param callable $callback Callback that yields entity generation callbacks |
| 2048 | * @param array $opts Options map similar to that of getWithSetCallback() |
| 2049 | * @return mixed[] Map of (cache key => value) in the same order as $keyedIds |
| 2050 | * @since 1.28 |
| 2051 | */ |
| 2052 | final public function getMultiWithSetCallback( |
| 2053 | ArrayIterator $keyedIds, $ttl, callable $callback, array $opts = [] |
| 2054 | ) { |
| 2055 | $span = $this->startOperationSpan( __FUNCTION__, '' ); |
| 2056 | if ( $span->getContext()->isSampled() ) { |
| 2057 | $span->setAttributes( [ |
| 2058 | 'org.wikimedia.wancache.multi_count' => $keyedIds->count(), |
| 2059 | 'org.wikimedia.wancache.ttl' => $ttl, |
| 2060 | ] ); |
| 2061 | } |
| 2062 | // Batch load required keys into the in-process warmup cache |
| 2063 | $this->warmupCache = $this->fetchWrappedValuesForWarmupCache( |
| 2064 | $this->getNonProcessCachedMultiKeys( $keyedIds, $opts ), |
| 2065 | $opts['checkKeys'] ?? [] |
| 2066 | ); |
| 2067 | $this->warmupKeyMisses = 0; |
| 2068 | |
| 2069 | // The required callback signature includes $id as the first argument for convenience |
| 2070 | // to distinguish different items. To reuse the code in getWithSetCallback(), wrap the |
| 2071 | // callback with a proxy callback that has the standard getWithSetCallback() signature. |
| 2072 | // This is defined only once per batch to avoid closure creation overhead. |
| 2073 | $proxyCb = static function ( $oldValue, &$ttl, &$setOpts, $oldAsOf, $params ) |
| 2074 | use ( $callback ) |
| 2075 | { |
| 2076 | return $callback( $params['id'], $oldValue, $ttl, $setOpts, $oldAsOf ); |
| 2077 | }; |
| 2078 | |
| 2079 | // Get the order-preserved result map using the warm-up cache |
| 2080 | $values = []; |
| 2081 | foreach ( $keyedIds as $key => $id ) { |
| 2082 | $values[$key] = $this->getWithSetCallback( |
| 2083 | $key, |
| 2084 | $ttl, |
| 2085 | $proxyCb, |
| 2086 | $opts, |
| 2087 | [ 'id' => $id ] |
| 2088 | ); |
| 2089 | } |
| 2090 | |
| 2091 | $this->warmupCache = []; |
| 2092 | |
| 2093 | return $values; |
| 2094 | } |
| 2095 | |
| 2096 | /** |
| 2097 | * Method to fetch/regenerate multiple cache keys at once |
| 2098 | * |
| 2099 | * This works the same as getWithSetCallback() except: |
| 2100 | * - a) The $keys argument expects the result of WANObjectCache::makeMultiKeys() |
| 2101 | * - b) The $callback argument expects a function that returns a map of (ID => new value), |
| 2102 | * using boolean "false" for entities that could not be found, for all entity IDs in |
| 2103 | * $ids. The callback takes the following arguments: |
| 2104 | * - $ids: list of entity IDs that require value generation |
| 2105 | * - &$ttls: reference to the (entity ID => new TTL) map (alterable) |
| 2106 | * - &$setOpts: reference to the new value set() options (alterable) |
| 2107 | * - c) The return value is a map of (cache key => value) in the order of $keyedIds |
| 2108 | * - d) The "lockTSE" and "busyValue" options are ignored |
| 2109 | * |
| 2110 | * @see WANObjectCache::getWithSetCallback() |
| 2111 | * @see WANObjectCache::getMultiWithSetCallback() |
| 2112 | * |
| 2113 | * Example usage: |
| 2114 | * @code |
| 2115 | * $rows = $cache->getMultiWithUnionSetCallback( |
| 2116 | * // Map of cache keys to entity IDs |
| 2117 | * $cache->makeMultiKeys( |
| 2118 | * $this->fileVersionIds(), |
| 2119 | * function ( $id ) use ( $cache ) { |
| 2120 | * return $cache->makeKey( 'file-version', $id ); |
| 2121 | * } |
| 2122 | * ), |
| 2123 | * // Time-to-live (in seconds) |
| 2124 | * $cache::TTL_DAY, |
| 2125 | * // Function that derives the new key value |
| 2126 | * function ( array $ids, array &$ttls, array &$setOpts ) { |
| 2127 | * $dbr = MediaWikiServices::getInstance()->getConnectionProvider()->getReplicaDatabase(); |
| 2128 | * // Account for any snapshot/replica DB lag |
| 2129 | * $setOpts += Database::getCacheSetOptions( $dbr ); |
| 2130 | * |
| 2131 | * // Load the rows for these files |
| 2132 | * $rows = array_fill_keys( $ids, false ); |
| 2133 | * $queryInfo = File::getQueryInfo(); |
| 2134 | * $res = $dbr->select( |
| 2135 | * $queryInfo['tables'], |
| 2136 | * $queryInfo['fields'], |
| 2137 | * [ 'id' => $ids ], |
| 2138 | * __METHOD__, |
| 2139 | * [], |
| 2140 | * $queryInfo['joins'] |
| 2141 | * ); |
| 2142 | * foreach ( $res as $row ) { |
| 2143 | * $rows[$row->id] = $row; |
| 2144 | * $mtime = wfTimestamp( TS::UNIX, $row->timestamp ); |
| 2145 | * $ttls[$row->id] = $this->adaptiveTTL( $mtime, $ttls[$row->id] ); |
| 2146 | * } |
| 2147 | * |
| 2148 | * return $rows; |
| 2149 | * }, |
| 2150 | * ] |
| 2151 | * ); |
| 2152 | * $files = array_map( self::newFromRow( ... ), $rows ); |
| 2153 | * @endcode |
| 2154 | * |
| 2155 | * @param ArrayIterator $keyedIds Result of WANObjectCache::makeMultiKeys() |
| 2156 | * @param int $ttl Seconds to live for key updates |
| 2157 | * @param callable $callback Callback that yields entity generation callbacks |
| 2158 | * @param array $opts Options map similar to that of getWithSetCallback() |
| 2159 | * @return mixed[] Map of (cache key => value) in the same order as $keyedIds |
| 2160 | * @since 1.30 |
| 2161 | */ |
| 2162 | final public function getMultiWithUnionSetCallback( |
| 2163 | ArrayIterator $keyedIds, $ttl, callable $callback, array $opts = [] |
| 2164 | ) { |
| 2165 | $span = $this->startOperationSpan( __FUNCTION__, '' ); |
| 2166 | if ( $span->getContext()->isSampled() ) { |
| 2167 | $span->setAttributes( [ |
| 2168 | 'org.wikimedia.wancache.multi_count' => $keyedIds->count(), |
| 2169 | 'org.wikimedia.wancache.ttl' => $ttl, |
| 2170 | ] ); |
| 2171 | } |
| 2172 | $checkKeys = $opts['checkKeys'] ?? []; // TODO: ??? |
| 2173 | $minAsOf = $opts['minAsOf'] ?? self::MIN_TIMESTAMP_NONE; |
| 2174 | |
| 2175 | // unset incompatible keys |
| 2176 | unset( $opts['lockTSE'] ); |
| 2177 | unset( $opts['busyValue'] ); |
| 2178 | |
| 2179 | // Batch load required keys into the in-process warmup cache |
| 2180 | $keysByIdGet = $this->getNonProcessCachedMultiKeys( $keyedIds, $opts ); |
| 2181 | $this->warmupCache = $this->fetchWrappedValuesForWarmupCache( $keysByIdGet, $checkKeys ); |
| 2182 | $this->warmupKeyMisses = 0; |
| 2183 | |
| 2184 | // IDs of entities known to be in need of generation |
| 2185 | $idsRegen = []; |
| 2186 | |
| 2187 | // Find out which keys are missing/deleted/stale |
| 2188 | $now = $this->getCurrentTime(); |
| 2189 | $resByKey = $this->fetchKeys( $keysByIdGet, $checkKeys, $now ); |
| 2190 | foreach ( $keysByIdGet as $id => $key ) { |
| 2191 | $res = $resByKey[$key]; |
| 2192 | if ( |
| 2193 | $res[self::RES_VALUE] === false || |
| 2194 | $res[self::RES_CUR_TTL] < 0 || |
| 2195 | $res[self::RES_AS_OF] < $minAsOf |
| 2196 | ) { |
| 2197 | $idsRegen[] = $id; |
| 2198 | } |
| 2199 | } |
| 2200 | |
| 2201 | // Run the callback to populate the generation value map for all required IDs |
| 2202 | $newSetOpts = []; |
| 2203 | $newTTLsById = array_fill_keys( $idsRegen, $ttl ); |
| 2204 | $newValsById = $idsRegen ? $callback( $idsRegen, $newTTLsById, $newSetOpts ) : []; |
| 2205 | |
| 2206 | $method = __METHOD__; |
| 2207 | // The required callback signature includes $id as the first argument for convenience |
| 2208 | // to distinguish different items. To reuse the code in getWithSetCallback(), wrap the |
| 2209 | // callback with a proxy callback that has the standard getWithSetCallback() signature. |
| 2210 | // This is defined only once per batch to avoid closure creation overhead. |
| 2211 | $proxyCb = function ( $oldValue, &$ttl, &$setOpts, $oldAsOf, $params ) |
| 2212 | use ( $callback, $newValsById, $newTTLsById, $newSetOpts, $method ) |
| 2213 | { |
| 2214 | $id = $params['id']; |
| 2215 | |
| 2216 | if ( array_key_exists( $id, $newValsById ) ) { |
| 2217 | // Value was already regenerated as expected, so use the value in $newValsById |
| 2218 | $newValue = $newValsById[$id]; |
| 2219 | $ttl = $newTTLsById[$id]; |
| 2220 | $setOpts = $newSetOpts; |
| 2221 | } else { |
| 2222 | // Pre-emptive/popularity refresh and version mismatch cases are not detected |
| 2223 | // above and thus $newValsById has no entry. Run $callback on this single entity. |
| 2224 | $ttls = [ $id => $ttl ]; |
| 2225 | $result = $callback( [ $id ], $ttls, $setOpts ); |
| 2226 | if ( !isset( $result[$id] ) ) { |
| 2227 | // T303092 |
| 2228 | $this->logger->warning( |
| 2229 | $method . ' failed due to {id} not set in result {result}', [ |
| 2230 | 'id' => $id, |
| 2231 | 'result' => json_encode( $result ) |
| 2232 | ] ); |
| 2233 | } |
| 2234 | $newValue = $result[$id]; |
| 2235 | $ttl = $ttls[$id]; |
| 2236 | } |
| 2237 | |
| 2238 | return $newValue; |
| 2239 | }; |
| 2240 | |
| 2241 | // Get the order-preserved result map using the warm-up cache |
| 2242 | $values = []; |
| 2243 | foreach ( $keyedIds as $key => $id ) { |
| 2244 | $values[$key] = $this->getWithSetCallback( |
| 2245 | $key, |
| 2246 | $ttl, |
| 2247 | $proxyCb, |
| 2248 | $opts, |
| 2249 | [ 'id' => $id ] |
| 2250 | ); |
| 2251 | } |
| 2252 | |
| 2253 | $this->warmupCache = []; |
| 2254 | |
| 2255 | return $values; |
| 2256 | } |
| 2257 | |
| 2258 | /** |
| 2259 | * @see BagOStuff::makeGlobalKey() |
| 2260 | * @since 1.27 |
| 2261 | * @param string $keygroup Key group component, should be under 48 characters. |
| 2262 | * @param string|int ...$components Additional, ordered, key components for entity IDs |
| 2263 | * @return string Colon-separated, keyspace-prepended, ordered list of encoded components |
| 2264 | */ |
| 2265 | public function makeGlobalKey( $keygroup, ...$components ) { |
| 2266 | return $this->cache->makeGlobalKey( $keygroup, ...$components ); |
| 2267 | } |
| 2268 | |
| 2269 | /** |
| 2270 | * @see BagOStuff::makeKey() |
| 2271 | * @since 1.27 |
| 2272 | * @param string $keygroup Key group component, should be under 48 characters. |
| 2273 | * @param string|int ...$components Additional, ordered, key components for entity IDs |
| 2274 | * @return string Colon-separated, keyspace-prepended, ordered list of encoded components |
| 2275 | */ |
| 2276 | public function makeKey( $keygroup, ...$components ) { |
| 2277 | return $this->cache->makeKey( $keygroup, ...$components ); |
| 2278 | } |
| 2279 | |
| 2280 | /** |
| 2281 | * Get an iterator of (cache key => entity ID) for a list of entity IDs |
| 2282 | * |
| 2283 | * The $callback argument expects a function that returns the key for an entity ID via |
| 2284 | * makeKey()/makeGlobalKey(). There should be no network nor filesystem I/O used in the |
| 2285 | * callback. The entity ID/key mapping must be 1:1 or an exception will be thrown. |
| 2286 | * |
| 2287 | * The callback takes the following arguments: |
| 2288 | * - $id: An entity ID |
| 2289 | * - $cache: This WANObjectCache instance |
| 2290 | * |
| 2291 | * Example usage for the default keyspace: |
| 2292 | * @code |
| 2293 | * $keyedIds = $cache->makeMultiKeys( |
| 2294 | * $urls, |
| 2295 | * function ( $url, $cache ) { |
| 2296 | * return $cache->makeKey( 'example-url', $url ); |
| 2297 | * } |
| 2298 | * ); |
| 2299 | * @endcode |
| 2300 | * |
| 2301 | * Example usage for mixed default and global keyspace: |
| 2302 | * @code |
| 2303 | * $keyedIds = $cache->makeMultiKeys( |
| 2304 | * $filters, |
| 2305 | * function ( $filter, $cache ) { |
| 2306 | * return self::isCentral( $filter ) |
| 2307 | * ? $cache->makeGlobalKey( 'example-filter', $filter ) |
| 2308 | * : $cache->makeKey( 'example-filter', $filter ) |
| 2309 | * } |
| 2310 | * ); |
| 2311 | * @endcode |
| 2312 | * |
| 2313 | * @see WANObjectCache::makeKey() |
| 2314 | * @see WANObjectCache::makeGlobalKey() |
| 2315 | * |
| 2316 | * @param string[]|int[] $ids List of entity IDs |
| 2317 | * @param callable $keyCallback Function returning makeKey()/makeGlobalKey() on the input ID |
| 2318 | * @return ArrayIterator Iterator of (cache key => ID); order of $ids is preserved |
| 2319 | * @since 1.28 |
| 2320 | */ |
| 2321 | final public function makeMultiKeys( array $ids, $keyCallback ) { |
| 2322 | $idByKey = []; |
| 2323 | foreach ( $ids as $id ) { |
| 2324 | $key = $keyCallback( $id, $this ); |
| 2325 | // Edge case: ignore key collisions due to duplicate $ids like "42" and 42 |
| 2326 | if ( !isset( $idByKey[$key] ) ) { |
| 2327 | $idByKey[$key] = $id; |
| 2328 | } elseif ( (string)$id !== (string)$idByKey[$key] ) { |
| 2329 | throw new UnexpectedValueException( |
| 2330 | "Cache key collision; IDs ('$id','{$idByKey[$key]}') map to '$key'" |
| 2331 | ); |
| 2332 | } |
| 2333 | } |
| 2334 | |
| 2335 | return new ArrayIterator( $idByKey ); |
| 2336 | } |
| 2337 | |
| 2338 | /** |
| 2339 | * Get an (ID => value) map from (i) a non-unique list of entity IDs, and (ii) the list |
| 2340 | * of corresponding entity values by first appearance of each ID in the entity ID list |
| 2341 | * |
| 2342 | * For use with getMultiWithSetCallback() and getMultiWithUnionSetCallback(). |
| 2343 | * |
| 2344 | * *Only* use this method if the entity ID/key mapping is trivially 1:1 without exception. |
| 2345 | * Key generation method must utilize the *full* entity ID in the key (not a hash of it). |
| 2346 | * |
| 2347 | * Example usage: |
| 2348 | * @code |
| 2349 | * $poems = $cache->getMultiWithSetCallback( |
| 2350 | * $cache->makeMultiKeys( |
| 2351 | * $uuids, |
| 2352 | * function ( $uuid ) use ( $cache ) { |
| 2353 | * return $cache->makeKey( 'poem', $uuid ); |
| 2354 | * } |
| 2355 | * ), |
| 2356 | * $cache::TTL_DAY, |
| 2357 | * function ( $uuid ) use ( $url ) { |
| 2358 | * return $this->http->run( [ 'method' => 'GET', 'url' => "$url/$uuid" ] ); |
| 2359 | * } |
| 2360 | * ); |
| 2361 | * $poemsByUUID = $cache->multiRemap( $uuids, $poems ); |
| 2362 | * @endcode |
| 2363 | * |
| 2364 | * @see WANObjectCache::makeMultiKeys() |
| 2365 | * @see WANObjectCache::getMultiWithSetCallback() |
| 2366 | * @see WANObjectCache::getMultiWithUnionSetCallback() |
| 2367 | * |
| 2368 | * @param string[]|int[] $ids Entity ID list makeMultiKeys() |
| 2369 | * @param mixed[] $res Result of getMultiWithSetCallback()/getMultiWithUnionSetCallback() |
| 2370 | * @return mixed[] Map of (ID => value); order of $ids is preserved |
| 2371 | * @since 1.34 |
| 2372 | */ |
| 2373 | final public function multiRemap( array $ids, array $res ) { |
| 2374 | if ( count( $ids ) !== count( $res ) ) { |
| 2375 | // If makeMultiKeys() is called on a list of non-unique IDs, then the resulting |
| 2376 | // ArrayIterator will have less entries due to "first appearance" de-duplication |
| 2377 | $ids = array_keys( array_fill_keys( $ids, true ) ); |
| 2378 | if ( count( $ids ) !== count( $res ) ) { |
| 2379 | throw new UnexpectedValueException( "Multi-key result does not match ID list" ); |
| 2380 | } |
| 2381 | } |
| 2382 | |
| 2383 | return array_combine( $ids, $res ); |
| 2384 | } |
| 2385 | |
| 2386 | /** |
| 2387 | * Get a "watch point" token that can be used to get the "last error" to occur after now |
| 2388 | * |
| 2389 | * @return int A token that the current error event |
| 2390 | * @since 1.38 |
| 2391 | */ |
| 2392 | public function watchErrors() { |
| 2393 | return $this->cache->watchErrors(); |
| 2394 | } |
| 2395 | |
| 2396 | /** |
| 2397 | * Get the "last error" registry |
| 2398 | * |
| 2399 | * The method should be invoked by a caller as part of the following pattern: |
| 2400 | * - The caller invokes watchErrors() to get a "since token" |
| 2401 | * - The caller invokes a sequence of cache operation methods |
| 2402 | * - The caller invokes getLastError() with the "since token" |
| 2403 | * |
| 2404 | * External callers can also invoke this method as part of the following pattern: |
| 2405 | * - The caller invokes clearLastError() |
| 2406 | * - The caller invokes a sequence of cache operation methods |
| 2407 | * - The caller invokes getLastError() |
| 2408 | * |
| 2409 | * @param int $watchPoint Only consider errors from after this "watch point" [optional] |
| 2410 | * @return int BagOStuff:ERR_* constant for the "last error" registry |
| 2411 | * @note Parameters added in 1.38: $watchPoint |
| 2412 | */ |
| 2413 | final public function getLastError( $watchPoint = 0 ) { |
| 2414 | $code = $this->cache->getLastError( $watchPoint ); |
| 2415 | switch ( $code ) { |
| 2416 | case BagOStuff::ERR_NONE: |
| 2417 | return BagOStuff::ERR_NONE; |
| 2418 | case BagOStuff::ERR_NO_RESPONSE: |
| 2419 | return BagOStuff::ERR_NO_RESPONSE; |
| 2420 | case BagOStuff::ERR_UNREACHABLE: |
| 2421 | return BagOStuff::ERR_UNREACHABLE; |
| 2422 | default: |
| 2423 | return BagOStuff::ERR_UNEXPECTED; |
| 2424 | } |
| 2425 | } |
| 2426 | |
| 2427 | /** |
| 2428 | * Clear the in-process caches; useful for testing |
| 2429 | * |
| 2430 | * @since 1.27 |
| 2431 | */ |
| 2432 | public function clearProcessCache() { |
| 2433 | $this->processCaches = []; |
| 2434 | } |
| 2435 | |
| 2436 | /** |
| 2437 | * Enable or disable the use of brief caching for tombstoned keys |
| 2438 | * |
| 2439 | * When a key is purged via delete(), there normally is a period where caching |
| 2440 | * is hold-off limited to an extremely short time. This method will disable that |
| 2441 | * caching, forcing the callback to run for any of: |
| 2442 | * - WANObjectCache::getWithSetCallback() |
| 2443 | * - WANObjectCache::getMultiWithSetCallback() |
| 2444 | * - WANObjectCache::getMultiWithUnionSetCallback() |
| 2445 | * |
| 2446 | * This is useful when both: |
| 2447 | * - a) the database used by the callback is known to be up-to-date enough |
| 2448 | * for some particular purpose (e.g. replica DB has applied transaction X) |
| 2449 | * - b) the caller needs to exploit that fact, and therefore needs to avoid the |
| 2450 | * use of inherently volatile and possibly stale interim keys |
| 2451 | * |
| 2452 | * @see WANObjectCache::delete() |
| 2453 | * @param bool $enabled Whether to enable interim caching |
| 2454 | * @since 1.31 |
| 2455 | */ |
| 2456 | final public function useInterimHoldOffCaching( $enabled ) { |
| 2457 | $this->useInterimHoldOffCaching = $enabled; |
| 2458 | } |
| 2459 | |
| 2460 | /** |
| 2461 | * @param int $flag BagOStuff::ATTR_* class constant |
| 2462 | * @return int BagOStuff::QOS_* class constant |
| 2463 | * @since 1.28 |
| 2464 | */ |
| 2465 | public function getQoS( $flag ) { |
| 2466 | return $this->cache->getQoS( $flag ); |
| 2467 | } |
| 2468 | |
| 2469 | /** |
| 2470 | * Get a TTL that is higher for objects that have not changed recently |
| 2471 | * |
| 2472 | * This is useful for keys that get explicit purges and DB or purge relay |
| 2473 | * lag is a potential concern (especially how it interacts with CDN cache) |
| 2474 | * |
| 2475 | * Example usage: |
| 2476 | * @code |
| 2477 | * // Last-modified time of page |
| 2478 | * $mtime = wfTimestamp( TS::UNIX, $page->getTimestamp() ); |
| 2479 | * // Get adjusted TTL. If $mtime is 3600 seconds ago and $minTTL/$factor left at |
| 2480 | * // defaults, then $ttl is 3600 * .2 = 720. If $minTTL was greater than 720, then |
| 2481 | * // $ttl would be $minTTL. If $maxTTL was smaller than 720, $ttl would be $maxTTL. |
| 2482 | * $ttl = $cache->adaptiveTTL( $mtime, $cache::TTL_DAY ); |
| 2483 | * @endcode |
| 2484 | * |
| 2485 | * Another use case is when there are no applicable "last modified" fields in the DB, |
| 2486 | * and there are too many dependencies for explicit purges to be viable, and the rate of |
| 2487 | * change to relevant content is unstable, and it is highly valued to have the cached value |
| 2488 | * be as up-to-date as possible. |
| 2489 | * |
| 2490 | * Example usage: |
| 2491 | * @code |
| 2492 | * $query = "<some complex query>"; |
| 2493 | * $idListFromComplexQuery = $cache->getWithSetCallback( |
| 2494 | * $cache->makeKey( 'complex-graph-query', $hashOfQuery ), |
| 2495 | * GraphQueryClass::STARTING_TTL, |
| 2496 | * function ( $oldValue, &$ttl, array &$setOpts, $oldAsOf ) use ( $query, $cache ) { |
| 2497 | * $gdb = $this->getReplicaGraphDbConnection(); |
| 2498 | * // Account for any snapshot/replica DB lag |
| 2499 | * $setOpts += GraphDatabase::getCacheSetOptions( $gdb ); |
| 2500 | * |
| 2501 | * $newList = iterator_to_array( $gdb->query( $query ) ); |
| 2502 | * sort( $newList, SORT_NUMERIC ); // normalize |
| 2503 | * |
| 2504 | * $minTTL = GraphQueryClass::MIN_TTL; |
| 2505 | * $maxTTL = GraphQueryClass::MAX_TTL; |
| 2506 | * if ( $oldValue !== false ) { |
| 2507 | * // Note that $oldAsOf is the last time this callback ran |
| 2508 | * $ttl = ( $newList === $oldValue ) |
| 2509 | * // No change: cache for 150% of the age of $oldValue |
| 2510 | * ? $cache->adaptiveTTL( $oldAsOf, $maxTTL, $minTTL, 1.5 ) |
| 2511 | * // Changed: cache for 50% of the age of $oldValue |
| 2512 | * : $cache->adaptiveTTL( $oldAsOf, $maxTTL, $minTTL, .5 ); |
| 2513 | * } |
| 2514 | * |
| 2515 | * return $newList; |
| 2516 | * }, |
| 2517 | * [ |
| 2518 | * // Keep stale values around for doing comparisons for TTL calculations. |
| 2519 | * // High values improve long-tail keys hit-rates, though might waste space. |
| 2520 | * 'staleTTL' => GraphQueryClass::GRACE_TTL |
| 2521 | * ] |
| 2522 | * ); |
| 2523 | * @endcode |
| 2524 | * |
| 2525 | * @param int|float|string|null $mtime UNIX timestamp; null if none |
| 2526 | * @param int $maxTTL Maximum TTL (seconds) |
| 2527 | * @param int $minTTL Minimum TTL (seconds); Default: 30 |
| 2528 | * @param float $factor Value in the range (0,1); Default: .2 |
| 2529 | * @return int Adaptive TTL |
| 2530 | * @since 1.28 |
| 2531 | */ |
| 2532 | public function adaptiveTTL( $mtime, $maxTTL, $minTTL = 30, $factor = 0.2 ) { |
| 2533 | // handle fractional seconds and string integers |
| 2534 | $mtime = (int)$mtime; |
| 2535 | if ( $mtime <= 0 ) { |
| 2536 | // no last-modified time provided |
| 2537 | return $minTTL; |
| 2538 | } |
| 2539 | |
| 2540 | $age = (int)$this->getCurrentTime() - $mtime; |
| 2541 | |
| 2542 | return (int)min( $maxTTL, max( $minTTL, $factor * $age ) ); |
| 2543 | } |
| 2544 | |
| 2545 | /** |
| 2546 | * @internal For use by unit tests only |
| 2547 | * @return int |
| 2548 | * @since 1.30 |
| 2549 | */ |
| 2550 | final public function getWarmupKeyMisses() { |
| 2551 | // Number of misses in $this->warmupCache during the last call to certain methods |
| 2552 | return $this->warmupKeyMisses; |
| 2553 | } |
| 2554 | |
| 2555 | /** |
| 2556 | * @param string $sisterKey |
| 2557 | * @return string |
| 2558 | */ |
| 2559 | protected function getRouteKey( string $sisterKey ) { |
| 2560 | if ( $this->broadcastRoute !== null ) { |
| 2561 | if ( $sisterKey[0] === '/' ) { |
| 2562 | throw new RuntimeException( "Sister key '$sisterKey' already contains a route." ); |
| 2563 | } |
| 2564 | return $this->broadcastRoute . $sisterKey; |
| 2565 | } |
| 2566 | return $sisterKey; |
| 2567 | } |
| 2568 | |
| 2569 | /** |
| 2570 | * Schedule a deferred cache regeneration if possible |
| 2571 | * |
| 2572 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 2573 | * @param int $ttl Seconds to live |
| 2574 | * @param callable $callback |
| 2575 | * @param array $opts |
| 2576 | * @param array $cbParams |
| 2577 | * @return bool Success |
| 2578 | * @note Callable type hints are not used to avoid class-autoloading |
| 2579 | */ |
| 2580 | private function scheduleAsyncRefresh( $key, $ttl, $callback, array $opts, array $cbParams ) { |
| 2581 | if ( !$this->asyncHandler ) { |
| 2582 | return false; |
| 2583 | } |
| 2584 | // Update the cache value later, such during post-send of an HTTP request. This forces |
| 2585 | // cache regeneration by setting "minAsOf" to infinity, meaning that no existing value |
| 2586 | // is considered valid. Furthermore, note that preemptive regeneration is not applicable |
| 2587 | // to invalid values, so there is no risk of infinite preemptive regeneration loops. |
| 2588 | $func = $this->asyncHandler; |
| 2589 | $func( function () use ( $key, $ttl, $callback, $opts, $cbParams ) { |
| 2590 | $opts['minAsOf'] = INF; |
| 2591 | try { |
| 2592 | $this->fetchOrRegenerate( $key, $ttl, $callback, $opts, $cbParams ); |
| 2593 | } catch ( Exception $e ) { |
| 2594 | // Log some context for easier debugging |
| 2595 | $this->logger->error( 'Async refresh failed for {key}', [ |
| 2596 | 'key' => $key, |
| 2597 | 'ttl' => $ttl, |
| 2598 | 'exception' => $e |
| 2599 | ] ); |
| 2600 | throw $e; |
| 2601 | } |
| 2602 | } ); |
| 2603 | |
| 2604 | return true; |
| 2605 | } |
| 2606 | |
| 2607 | /** |
| 2608 | * Check if a key value is non-false, new enough, and either fresh or "gracefully" stale |
| 2609 | * |
| 2610 | * @param array $res Current value WANObjectCache::RES_* data map |
| 2611 | * @param int $graceTTL Consider using stale values if $curTTL is greater than this |
| 2612 | * @param float $minAsOf Minimum acceptable value "as of" UNIX timestamp |
| 2613 | * @return bool |
| 2614 | */ |
| 2615 | private function isAcceptablyFreshValue( $res, $graceTTL, $minAsOf ) { |
| 2616 | if ( !$this->isValid( $res[self::RES_VALUE], $res[self::RES_AS_OF], $minAsOf ) ) { |
| 2617 | // Value does not exists or is too old |
| 2618 | return false; |
| 2619 | } |
| 2620 | |
| 2621 | $curTTL = $res[self::RES_CUR_TTL]; |
| 2622 | if ( $curTTL > 0 ) { |
| 2623 | // Value is definitely still fresh |
| 2624 | return true; |
| 2625 | } |
| 2626 | |
| 2627 | // Remaining seconds during which this stale value can be used |
| 2628 | $curGraceTTL = $graceTTL + $curTTL; |
| 2629 | |
| 2630 | return ( $curGraceTTL > 0 ) |
| 2631 | // Chance of using the value decreases as $curTTL goes from 0 to -$graceTTL |
| 2632 | ? !$this->worthRefreshExpiring( $curGraceTTL, $graceTTL, $graceTTL ) |
| 2633 | // Value is too stale to fall in the grace period |
| 2634 | : false; |
| 2635 | } |
| 2636 | |
| 2637 | /** |
| 2638 | * Check if a key is due for randomized regeneration due to near-expiration/popularity |
| 2639 | * |
| 2640 | * @param array $res Current value WANObjectCache::RES_* data map |
| 2641 | * @param float $lowTTL Consider a refresh when $curTTL is less than this; the "low" threshold |
| 2642 | * @param int $ageNew Age of key when this might recommend refreshing (seconds) |
| 2643 | * @param int $hotTTR Age of key when it should be refreshed if popular (seconds) |
| 2644 | * @param float $now The current UNIX timestamp |
| 2645 | * @return bool |
| 2646 | */ |
| 2647 | protected function isLotteryRefreshDue( $res, $lowTTL, $ageNew, $hotTTR, $now ) { |
| 2648 | $curTTL = $res[self::RES_CUR_TTL]; |
| 2649 | $logicalTTL = $res[self::RES_TTL]; |
| 2650 | $asOf = $res[self::RES_AS_OF]; |
| 2651 | |
| 2652 | return ( |
| 2653 | $this->worthRefreshExpiring( $curTTL, $logicalTTL, $lowTTL ) || |
| 2654 | $this->worthRefreshPopular( $asOf, $ageNew, $hotTTR, $now ) |
| 2655 | ); |
| 2656 | } |
| 2657 | |
| 2658 | /** |
| 2659 | * Check if a key is due for randomized regeneration due to its popularity |
| 2660 | * |
| 2661 | * Suppose we cache a value with a TTL of 50min (3000 seconds) and use the default |
| 2662 | * ageNew=60s and hotTTR=900s options. During the first 60s your key is safe from refreshes. |
| 2663 | * Then we ramp up the chance for the 30s from age 60s to 90s. After that, we apply a constant |
| 2664 | * chance until the key expires. The chance is calculated such that if your key is very hot |
| 2665 | * (>1 req/s), we automatically refresh it roughly once every 900s. |
| 2666 | * |
| 2667 | * See also WANObjectCache::worthRefreshExpiring which plays a role closer to the expiry. |
| 2668 | * |
| 2669 | * The chance is calculated as follows: At 1 req/s, a 1:900 chance of refresh should trigger |
| 2670 | * every 900s. We adjust for 60s being safe (ageNew) and 30s lower chance (RAMPUP_TTL), which |
| 2671 | * works out to 1:825 requests, or a 0.1212% chance to truly trigger every 900s. |
| 2672 | * |
| 2673 | * | Age | Description |
| 2674 | * | ---- | ----------- |
| 2675 | * | 0s | miss, regen, set |
| 2676 | * | 1s | get, hit (0% chance of async regen) |
| 2677 | * | 60s | get, hit (0% chance of async regen) |
| 2678 | * | 63s | get, hit (0.01212% chance of async regen, or 1:8250 requests) |
| 2679 | * | 70s | get, hit (0.04040% chance of async regen, or 1:2475 requests) |
| 2680 | * | 90s | get, hit (0.12% chance of async regen, or 1:825 requests) |
| 2681 | * | >90s | get, hit (0.12% chance of async regen, or 1:825 requests) |
| 2682 | * |
| 2683 | * This feature exists to reduce negative impact of lost or delayed cache purges. The impact |
| 2684 | * of a heavily referenced key being stale is worse than that of a rarely referenced key. |
| 2685 | * Unlike simply lowering $ttl, rare keys are largely unaffected to allow for a high cache-hit |
| 2686 | * ratio for the "long-tail" of less-used keys. Similar to worthRefreshExpiring(), this uses |
| 2687 | * randomization to avoid causing cache stampedes. |
| 2688 | * |
| 2689 | * @param float $asOf UNIX timestamp of the value |
| 2690 | * @param int $ageNew Age of key when this may recommend a refresh (seconds) |
| 2691 | * @param int $hotTTR Age of key by which this should recommend a refresh if popular (seconds) |
| 2692 | * @param float $now The current UNIX timestamp |
| 2693 | * @return bool |
| 2694 | */ |
| 2695 | protected function worthRefreshPopular( $asOf, $ageNew, $hotTTR, $now ) { |
| 2696 | if ( $ageNew < 0 || $hotTTR <= 0 ) { |
| 2697 | return false; |
| 2698 | } |
| 2699 | |
| 2700 | $age = $now - $asOf; |
| 2701 | $timeOld = $age - $ageNew; |
| 2702 | if ( $timeOld <= 0 ) { |
| 2703 | return false; |
| 2704 | } |
| 2705 | |
| 2706 | $popularHitsPerSec = 1; |
| 2707 | // Lifecycle is: new, ramp-up refresh chance, full refresh chance. |
| 2708 | // Note that the "expected # of refreshes" for the ramp-up time range is half |
| 2709 | // of what it would be if P(refresh) was at its full value during that time range. |
| 2710 | $refreshWindowSec = max( $hotTTR - $ageNew - self::RAMPUP_TTL / 2, 1 ); |
| 2711 | // P(refresh) * (# hits in $refreshWindowSec) = (expected # of refreshes) |
| 2712 | // P(refresh) * ($refreshWindowSec * $popularHitsPerSec) = 1 (by definition) |
| 2713 | // P(refresh) = 1/($refreshWindowSec * $popularHitsPerSec) |
| 2714 | $chance = 1 / ( $popularHitsPerSec * $refreshWindowSec ); |
| 2715 | // Ramp up $chance from 0 to its nominal value over RAMPUP_TTL seconds to avoid stampedes |
| 2716 | $chance *= ( $timeOld <= self::RAMPUP_TTL ) ? $timeOld / self::RAMPUP_TTL : 1; |
| 2717 | |
| 2718 | return ( mt_rand( 1, 1_000_000_000 ) <= 1_000_000_000 * $chance ); |
| 2719 | } |
| 2720 | |
| 2721 | /** |
| 2722 | * Check if a key is nearing expiration and thus due for randomized regeneration. |
| 2723 | * |
| 2724 | * Suppose we cache a value with a TTL of 50min (3000 seconds), and a lowTTL of 100s. |
| 2725 | * |
| 2726 | * | Time passed | Remaining TTL | Description |
| 2727 | * | ----------- | ------------- | ----------- |
| 2728 | * | t=0 | 3000s | miss, regen, set |
| 2729 | * | t=100 | 2900s | get, hit |
| 2730 | * | t=2899 | 101s | get, hit |
| 2731 | * | t=2900 | 100s | get, hit (0% chance of async regen) |
| 2732 | * | t=2950 | 50s | get, hit (6% chance of async regen) |
| 2733 | * | t=2975 | 25s | get, hit (31% chance of async regen) |
| 2734 | * | t=2995 | 5s | get, hit (81% chance of async regen) |
| 2735 | * | t=2999 | 1s | get, hit (96% chance of async regen) |
| 2736 | * | t=3000 | 0s | miss, regen, set |
| 2737 | * |
| 2738 | * We only consider randomized regeneration once $curTTL is less than $lowTTL (i.e. near expiry). |
| 2739 | * We do not consider it once $curTTL is <= 0 (i.e. value expired). |
| 2740 | * |
| 2741 | * The chance of returning true increases steadily from 0% to 100% as |
| 2742 | * $curTTL counts down from the "low" threshold to 0 seconds. |
| 2743 | * |
| 2744 | * If the original TTL passed to WANObjectCache::getWithSetCallback is below $lowTTL (the author |
| 2745 | * choose a TTL under 60s, and did not set a custom 'lowTTL' option), then we use the original |
| 2746 | * TTL as the low threshold. This ensures the ramp up always starts slowly at 0%, |
| 2747 | * instead of breaking caching for intentionally short TTLs (T264787). |
| 2748 | * |
| 2749 | * This method uses deadline-aware randomization in order to handle wide ranges |
| 2750 | * of request rates without the need for complex options or state keeping. |
| 2751 | * |
| 2752 | * @param float $curTTL Approximate TTL left on the key |
| 2753 | * @param float $logicalTTL Full logical TTL assigned to the key; 0 for "infinite" |
| 2754 | * @param float $lowTTL Consider a refresh once $curTTL is less than this; the "low" threshold |
| 2755 | * @return bool |
| 2756 | */ |
| 2757 | protected function worthRefreshExpiring( $curTTL, $logicalTTL, $lowTTL ) { |
| 2758 | if ( $lowTTL <= 0 ) { |
| 2759 | return false; |
| 2760 | } |
| 2761 | // T264787: avoid having keys start off with a high chance of being refreshed; |
| 2762 | // the point where refreshing becomes possible cannot precede the key lifetime. |
| 2763 | $effectiveLowTTL = min( $lowTTL, $logicalTTL ?: INF ); |
| 2764 | |
| 2765 | // How long the value was in the "low TTL" phase |
| 2766 | $timeOld = $effectiveLowTTL - $curTTL; |
| 2767 | if ( $timeOld <= 0 || $timeOld >= $effectiveLowTTL ) { |
| 2768 | return false; |
| 2769 | } |
| 2770 | |
| 2771 | // Ratio of the low TTL phase that has elapsed (r) |
| 2772 | $ttrRatio = $timeOld / $effectiveLowTTL; |
| 2773 | // Use p(r) as the monotonically increasing "chance of refresh" function, |
| 2774 | // having p(0)=0 and p(1)=1. The value expires at the nominal expiry. |
| 2775 | $chance = $ttrRatio ** 4; |
| 2776 | |
| 2777 | return ( mt_rand( 1, 1_000_000_000 ) <= 1_000_000_000 * $chance ); |
| 2778 | } |
| 2779 | |
| 2780 | /** |
| 2781 | * Check that a wrapper value exists and has an acceptable age |
| 2782 | * |
| 2783 | * @param array|false $value Value wrapper or false |
| 2784 | * @param float $asOf Value generation "as of" timestamp |
| 2785 | * @param float $minAsOf Minimum acceptable value "as of" UNIX timestamp |
| 2786 | * @return bool |
| 2787 | */ |
| 2788 | protected function isValid( $value, $asOf, $minAsOf ) { |
| 2789 | return ( $value !== false && $asOf >= $minAsOf ); |
| 2790 | } |
| 2791 | |
| 2792 | /** |
| 2793 | * @param mixed $value |
| 2794 | * @param int $ttl Seconds to live or zero for "indefinite" |
| 2795 | * @param int|null $version Value version number or null if not versioned |
| 2796 | * @param float $now Unix Current timestamp just before calling set() |
| 2797 | * @return array |
| 2798 | */ |
| 2799 | private function wrap( $value, $ttl, $version, $now ) { |
| 2800 | // Returns keys in ascending integer order for PHP7 array packing: |
| 2801 | // https://nikic.github.io/2014/12/22/PHPs-new-hashtable-implementation.html |
| 2802 | $wrapped = [ |
| 2803 | self::FLD_FORMAT_VERSION => self::VERSION, |
| 2804 | self::FLD_VALUE => $value, |
| 2805 | self::FLD_TTL => $ttl, |
| 2806 | self::FLD_TIME => $now |
| 2807 | ]; |
| 2808 | if ( $version !== null ) { |
| 2809 | $wrapped[self::FLD_VALUE_VERSION] = $version; |
| 2810 | } |
| 2811 | |
| 2812 | return $wrapped; |
| 2813 | } |
| 2814 | |
| 2815 | /** |
| 2816 | * @param array|string|false $wrapped The entry at a cache key (false if key is nonexistant) |
| 2817 | * @param float $now Unix Current timestamp (preferably pre-query) |
| 2818 | * @return array<int,mixed> Result map/n-tuple that includes the following: |
| 2819 | * - WANObjectCache::RES_VALUE: value or false if absent/tombstoned/malformed |
| 2820 | * - WANObjectCache::KEY_VERSION: value version number; null if there is no value |
| 2821 | * - WANObjectCache::KEY_AS_OF: value generation timestamp (UNIX); null if there is no value |
| 2822 | * - WANObjectCache::KEY_TTL: assigned logical TTL (seconds); null if there is no value |
| 2823 | * - WANObjectCache::KEY_TOMB_AS_OF: tombstone timestamp (UNIX); null if not tombstoned |
| 2824 | * - WANObjectCache::RES_CHECK_AS_OF: null placeholder for highest "check" key timestamp |
| 2825 | * - WANObjectCache::RES_TOUCH_AS_OF: null placeholder for highest "touched" timestamp |
| 2826 | * - WANObjectCache::KEY_CUR_TTL: remaining logical TTL (seconds) (negative if tombstoned) |
| 2827 | * @phan-return array{0:mixed,1:mixed,2:?float,3:?int,4:?float,5:?float,6:?float,7:?float} |
| 2828 | */ |
| 2829 | private function unwrap( $wrapped, $now ) { |
| 2830 | // https://nikic.github.io/2014/12/22/PHPs-new-hashtable-implementation.html |
| 2831 | $res = [ |
| 2832 | // Attributes that only depend on the fetched key value |
| 2833 | self::RES_VALUE => false, |
| 2834 | self::RES_VERSION => null, |
| 2835 | self::RES_AS_OF => null, |
| 2836 | self::RES_TTL => null, |
| 2837 | self::RES_TOMB_AS_OF => null, |
| 2838 | // Attributes that depend on caller-specific "check" keys or "touched callbacks" |
| 2839 | self::RES_CHECK_AS_OF => null, |
| 2840 | self::RES_TOUCH_AS_OF => null, |
| 2841 | self::RES_CUR_TTL => null |
| 2842 | ]; |
| 2843 | |
| 2844 | if ( is_array( $wrapped ) ) { |
| 2845 | // Entry expected to be a cached value; validate it |
| 2846 | if ( |
| 2847 | ( $wrapped[self::FLD_FORMAT_VERSION] ?? null ) === self::VERSION && |
| 2848 | $wrapped[self::FLD_TIME] >= $this->epoch |
| 2849 | ) { |
| 2850 | if ( $wrapped[self::FLD_TTL] > 0 ) { |
| 2851 | // Get the approximate time left on the key |
| 2852 | $age = $now - $wrapped[self::FLD_TIME]; |
| 2853 | $curTTL = max( $wrapped[self::FLD_TTL] - $age, 0.0 ); |
| 2854 | } else { |
| 2855 | // Key had no TTL, so the time left is unbounded |
| 2856 | $curTTL = INF; |
| 2857 | } |
| 2858 | $res[self::RES_VALUE] = $wrapped[self::FLD_VALUE]; |
| 2859 | $res[self::RES_VERSION] = $wrapped[self::FLD_VALUE_VERSION] ?? null; |
| 2860 | $res[self::RES_AS_OF] = $wrapped[self::FLD_TIME]; |
| 2861 | $res[self::RES_CUR_TTL] = $curTTL; |
| 2862 | $res[self::RES_TTL] = $wrapped[self::FLD_TTL]; |
| 2863 | } |
| 2864 | } else { |
| 2865 | // Entry expected to be a tombstone; parse it |
| 2866 | $purge = $this->parsePurgeValue( $wrapped ); |
| 2867 | if ( $purge !== null ) { |
| 2868 | // Tombstoned keys should always have a negative "current TTL" |
| 2869 | $curTTL = min( $purge[self::PURGE_TIME] - $now, self::TINY_NEGATIVE ); |
| 2870 | $res[self::RES_CUR_TTL] = $curTTL; |
| 2871 | $res[self::RES_TOMB_AS_OF] = $purge[self::PURGE_TIME]; |
| 2872 | } |
| 2873 | } |
| 2874 | |
| 2875 | return $res; |
| 2876 | } |
| 2877 | |
| 2878 | /** |
| 2879 | * @param string $key Cache key in the format `<keyspace>:<keygroup>[:<other components>]...` |
| 2880 | * as formatted by WANObjectCache::makeKey() or ::makeKeyGlobal. |
| 2881 | * @return string The key group of this cache key |
| 2882 | */ |
| 2883 | private function determineKeyGroupForStats( $key ) { |
| 2884 | $parts = explode( ':', $key, 3 ); |
| 2885 | // Fallback in case the key was not made by makeKey. |
| 2886 | // Replace dots because they are special in StatsD (T232907) |
| 2887 | return strtr( $parts[1] ?? $parts[0], '.', '_' ); |
| 2888 | } |
| 2889 | |
| 2890 | /** |
| 2891 | * Extract purge metadata from cached value if it is a valid purge value |
| 2892 | * |
| 2893 | * Valid purge values come from makeTombstonePurgeValue()/makeCheckKeyPurgeValue() |
| 2894 | * |
| 2895 | * @param mixed $value Cached value |
| 2896 | * @return array|null Tuple of (UNIX timestamp, hold-off seconds); null if value is invalid |
| 2897 | */ |
| 2898 | private function parsePurgeValue( $value ) { |
| 2899 | if ( !is_string( $value ) ) { |
| 2900 | return null; |
| 2901 | } |
| 2902 | |
| 2903 | $segments = explode( ':', $value, 3 ); |
| 2904 | $prefix = $segments[0]; |
| 2905 | if ( $prefix !== self::PURGE_VAL_PREFIX ) { |
| 2906 | // Not a purge value |
| 2907 | return null; |
| 2908 | } |
| 2909 | |
| 2910 | $timestamp = (float)$segments[1]; |
| 2911 | // makeTombstonePurgeValue() doesn't store hold-off TTLs |
| 2912 | $holdoff = isset( $segments[2] ) ? (int)$segments[2] : self::HOLDOFF_TTL; |
| 2913 | |
| 2914 | if ( $timestamp < $this->epoch ) { |
| 2915 | // Purge value is too old |
| 2916 | return null; |
| 2917 | } |
| 2918 | |
| 2919 | return [ self::PURGE_TIME => $timestamp, self::PURGE_HOLDOFF => $holdoff ]; |
| 2920 | } |
| 2921 | |
| 2922 | /** |
| 2923 | * @param float $timestamp UNIX timestamp |
| 2924 | * @param int $holdoff In seconds |
| 2925 | * @param array|null &$purge Unwrapped purge value array [returned] |
| 2926 | * @return string Wrapped purge value; format is "PURGED:<timestamp>:<holdoff>" |
| 2927 | */ |
| 2928 | private function makeCheckPurgeValue( float $timestamp, int $holdoff, ?array &$purge = null ) { |
| 2929 | $normalizedTime = (int)$timestamp; |
| 2930 | // Purge array that matches what parsePurgeValue() would have returned |
| 2931 | $purge = [ self::PURGE_TIME => (float)$normalizedTime, self::PURGE_HOLDOFF => $holdoff ]; |
| 2932 | |
| 2933 | return self::PURGE_VAL_PREFIX . ":$normalizedTime:$holdoff"; |
| 2934 | } |
| 2935 | |
| 2936 | /** |
| 2937 | * @param string $group |
| 2938 | * @return MapCacheLRU |
| 2939 | */ |
| 2940 | private function getProcessCache( $group ) { |
| 2941 | if ( !isset( $this->processCaches[$group] ) ) { |
| 2942 | [ , $size ] = explode( ':', $group ); |
| 2943 | $this->processCaches[$group] = new MapCacheLRU( (int)$size ); |
| 2944 | if ( $this->wallClockOverride !== null ) { |
| 2945 | $this->processCaches[$group]->setMockTime( $this->wallClockOverride ); |
| 2946 | } |
| 2947 | } |
| 2948 | |
| 2949 | return $this->processCaches[$group]; |
| 2950 | } |
| 2951 | |
| 2952 | /** |
| 2953 | * @param ArrayIterator $keys |
| 2954 | * @param array $opts |
| 2955 | * @return string[] Map of (ID => cache key) |
| 2956 | */ |
| 2957 | private function getNonProcessCachedMultiKeys( ArrayIterator $keys, array $opts ) { |
| 2958 | $pcTTL = $opts['pcTTL'] ?? self::TTL_UNCACHEABLE; |
| 2959 | |
| 2960 | $keysMissing = []; |
| 2961 | if ( $pcTTL > 0 && $this->callbackDepth == 0 ) { |
| 2962 | $pCache = $this->getProcessCache( $opts['pcGroup'] ?? self::PC_PRIMARY ); |
| 2963 | foreach ( $keys as $key => $id ) { |
| 2964 | if ( !$pCache->has( $key, $pcTTL ) ) { |
| 2965 | $keysMissing[$id] = $key; |
| 2966 | } |
| 2967 | } |
| 2968 | } |
| 2969 | |
| 2970 | return $keysMissing; |
| 2971 | } |
| 2972 | |
| 2973 | /** |
| 2974 | * @param string[] $keys Cache keys made with makeKey()/makeGlobalKey() |
| 2975 | * @param string[]|string[][] $checkKeys Map of (integer or cache key => "check" key(s)); |
| 2976 | * "check" keys must also be made with makeKey()/makeGlobalKey() |
| 2977 | * @return array<string,mixed> Map of (sister key => value, or, false if not found) |
| 2978 | */ |
| 2979 | private function fetchWrappedValuesForWarmupCache( array $keys, array $checkKeys ) { |
| 2980 | if ( !$keys ) { |
| 2981 | return []; |
| 2982 | } |
| 2983 | |
| 2984 | // Get all the value keys to fetch... |
| 2985 | $sisterKeys = []; |
| 2986 | foreach ( $keys as $baseKey ) { |
| 2987 | $sisterKeys[] = $this->makeSisterKey( $baseKey, self::TYPE_VALUE ); |
| 2988 | } |
| 2989 | // Get all the "check" keys to fetch... |
| 2990 | foreach ( $checkKeys as $i => $checkKeyOrKeyGroup ) { |
| 2991 | // Note: avoid array_merge() inside loop in case there are many keys |
| 2992 | if ( is_int( $i ) ) { |
| 2993 | // Single "check" key that applies to all value keys |
| 2994 | $sisterKeys[] = $this->makeSisterKey( $checkKeyOrKeyGroup, self::TYPE_TIMESTAMP ); |
| 2995 | } else { |
| 2996 | // List of "check" keys that apply to a specific value key |
| 2997 | foreach ( (array)$checkKeyOrKeyGroup as $checkKey ) { |
| 2998 | $sisterKeys[] = $this->makeSisterKey( $checkKey, self::TYPE_TIMESTAMP ); |
| 2999 | } |
| 3000 | } |
| 3001 | } |
| 3002 | |
| 3003 | $wrappedBySisterKey = $this->cache->getMulti( $sisterKeys ); |
| 3004 | $wrappedBySisterKey += array_fill_keys( $sisterKeys, false ); |
| 3005 | |
| 3006 | return $wrappedBySisterKey; |
| 3007 | } |
| 3008 | |
| 3009 | /** |
| 3010 | * @param string $key Cache key made with makeKey()/makeGlobalKey() |
| 3011 | * @param float $now Current UNIX timestamp |
| 3012 | * @return float|null Seconds since the last logged get() miss for this key, or, null |
| 3013 | */ |
| 3014 | private function timeSinceLoggedMiss( $key, $now ) { |
| 3015 | // phpcs:ignore Generic.CodeAnalysis.AssignmentInCondition.Found |
| 3016 | for ( end( $this->missLog ); $miss = current( $this->missLog ); prev( $this->missLog ) ) { |
| 3017 | if ( $miss[0] === $key ) { |
| 3018 | return ( $now - $miss[1] ); |
| 3019 | } |
| 3020 | } |
| 3021 | |
| 3022 | return null; |
| 3023 | } |
| 3024 | |
| 3025 | /** |
| 3026 | * @return float UNIX timestamp |
| 3027 | * @codeCoverageIgnore |
| 3028 | */ |
| 3029 | protected function getCurrentTime() { |
| 3030 | return $this->wallClockOverride ?: microtime( true ); |
| 3031 | } |
| 3032 | |
| 3033 | /** |
| 3034 | * @param float|null &$time Mock UNIX timestamp for testing |
| 3035 | * @codeCoverageIgnore |
| 3036 | */ |
| 3037 | public function setMockTime( &$time ) { |
| 3038 | $this->wallClockOverride =& $time; |
| 3039 | $this->cache->setMockTime( $time ); |
| 3040 | foreach ( $this->processCaches as $pCache ) { |
| 3041 | $pCache->setMockTime( $time ); |
| 3042 | } |
| 3043 | } |
| 3044 | |
| 3045 | /** |
| 3046 | * Convenience function to start an OpenTelemetry span for the given operation. |
| 3047 | * The span is deactivated and ended once the returned object goes out of scope, |
| 3048 | * but to be sure of timings, callers should call finish() on the returned object. |
| 3049 | * |
| 3050 | * @param string $opName Name of the operation to instrument (e.g. GET) |
| 3051 | * @param string|string[] $keys Cache keys related to the operation |
| 3052 | * @param string[]|string[][] $checkKeys 1/2D array of check keys related to the operation |
| 3053 | * |
| 3054 | * @return SpanInterface |
| 3055 | */ |
| 3056 | private function startOperationSpan( $opName, $keys, $checkKeys = [] ) { |
| 3057 | $span = $this->tracer->createSpan( "WANObjectCache::$opName" ) |
| 3058 | ->setSpanKind( SpanInterface::SPAN_KIND_CLIENT ) |
| 3059 | ->start(); |
| 3060 | |
| 3061 | if ( !$span->getContext()->isSampled() ) { |
| 3062 | return $span; |
| 3063 | } |
| 3064 | |
| 3065 | $keys = is_array( $keys ) ? implode( ' ', $keys ) : $keys; |
| 3066 | |
| 3067 | if ( count( $checkKeys ) > 0 ) { |
| 3068 | $checkKeys = array_map( |
| 3069 | static fn ( $checkKeyOrKeyGroup ) => |
| 3070 | is_array( $checkKeyOrKeyGroup ) |
| 3071 | ? implode( ' ', $checkKeyOrKeyGroup ) |
| 3072 | : $checkKeyOrKeyGroup, |
| 3073 | $checkKeys ); |
| 3074 | |
| 3075 | $checkKeys = implode( ' ', $checkKeys ); |
| 3076 | $span->setAttributes( [ 'org.wikimedia.wancache.check_keys' => $checkKeys ] ); |
| 3077 | } |
| 3078 | |
| 3079 | $span->setAttributes( [ 'org.wikimedia.wancache.keys' => $keys ] ); |
| 3080 | |
| 3081 | $span->activate(); |
| 3082 | return $span; |
| 3083 | } |
| 3084 | } |
| 3085 | |
| 3086 | /** @deprecated class alias since 1.43 */ |
| 3087 | class_alias( WANObjectCache::class, 'WANObjectCache' ); |