PHP cache slam defense using a semaphore to prevent dogpile effect (aka clobbering updates, stampeding herd or Slashdot effect).
Problem: too many requests hit your website at the same time while it tries to regenerate same content slamming your database, eg. when cache expired.
Solution: first request generates new content while all the subsequent requests get (stale) content from cache until it's refreshed by the first request.
Read http://www.sobstel.org/blog/preventing-dogpile-effect/ for more details.
In composer.json file:
"require": {
"sobstel/metaphore": "2.0.*"
}
or just composer require sobstel/metaphore
use Metaphore\Cache;
use Metaphore\Store\MemcachedStore;
// initialize $memcached object (new Memcached())
$cache = new Cache(new MemcachedStore($memcached));
$cache->cache('key', function() {
// generate content
}, 30);
-
__construct(ValueStoreInterface $valueStore, LockManager $lockManager = null)
-
cache($key, callable $callable, [$ttl, [$onNoStaleCacheCallable]])
- returns result -
delete($key)
-
getValue($key)
- returns Value object -
setResult($key, $result, Ttl $ttl)
- sets result (without anti-dogpile-effect mechanism) -
onNoStaleCache($callable)
-
getValueStore()
-
getLockManager()
Cache values and locks can be handled by different stores.
$valueStore = new Metaphore\MemcachedStore($memcached);
$lockStore = new Your\Custom\MySQLLockStore($connection);
$lockManager = new Metaphore\LockManager($lockStore);
$cache = new Metaphore\Cache($valueStore, $lockManager);
By default - if no 2nd argument passed to Cache constructor - value store is used as a lock store.
Sample use case might be to have custom MySQL GET_LOCK/RELEASE_LOCK for locks and still use in-built Memcached store for storing values.
You can pass simple integer value...
$cache->cache('key', callback, 30); // cache for 30 secs
.. or use more advanced Metaphore\TTl
object, which gives you control over grace period and lock ttl.
// $ttl, $grace_ttl, $lock_ttl
$ttl = new Ttl(30, 60, 15);
$cache->cache('key', callback, $ttl);
$ttl
- regular cache time (in seconds)$grace_ttl
- grace period, how long to allow to serve stale content while new one is being generated (in seconds), similar to HTTP's stale-while-revalidate, default is 60s$lock_ttl
- lock time, how long to prevent other request(s) from generating same content, default is 5s
Ttl value is added to current timestamp (time() + $ttl
).
In rare situations, when cache gets expired and there's no stale (generated earlier) content available, all requests will start generating new content.
You can add listener to catch this:
$cache->onNoStaleCache(function (NoStaleCacheEvent $event) {
Logger::log(sprintf('no stale cache detected for key %s', $event->getKey()));
});
You can also affect value that is returned:
$cache->onNoStaleCache(function (NoStaleCacheEvent $event) {
$event->setResult('new custom result');
});
Run all tests: phpunit
.
If no memcached or/and redis installed: phpunit --exclude-group=notisolated
or phpunit --exclude-group=memcached,redis
.