feature #33939 [Cache] add TagAwareMarshaller to optimize data storag…

…e when using AbstractTagAwareAdapter (nicolas-grekas)

This PR was merged into the 4.4 branch.

Discussion
----------

[Cache] add TagAwareMarshaller to optimize data storage when using AbstractTagAwareAdapter

| Q             | A
| ------------- | ---
| Branch?       | 4.4
| Bug fix?      | no
| New feature?  | no
| Deprecations? | no
| Tickets       | #33924
| License       | MIT
| Doc PR        | -

This is the final touch in my series of PR that fixes the linked issue.

Remarkably, the solutions I implemented for this issue are completely different than the one I described there. Fortunately, the issues themselves were correctly identified.

Plannification of implementation is gambling :)

/cc @andrerom

Commits
-------

5a4a30c [Cache] add TagAwareMarshaller to optimize data storage when using AbstractTagAwareAdapter

src/Symfony/Component/Cache/Adapter/AbstractTagAwareAdapter.php

@@ -149,6 +149,16 @@ abstract protected function doDelete(array $ids, array $tagData = []): bool;
      */
     abstract protected function doInvalidate(array $tagIds): bool;
 
+    /**
+     * Returns the tags bound to the provided ids.
+     */
+    protected function doFetchTags(array $ids): iterable
+    {
+        foreach ($this->doFetch($ids) as $id => $value) {
+            yield $id => \is_array($value) && \is_array($value['tags'] ?? null) ? $value['tags'] : [];
+        }
+    }
+
     /**
      * {@inheritdoc}
      *
@@ -233,8 +243,8 @@ public function deleteItems(array $keys)
         }
 
         try {
-            foreach ($this->doFetch($ids) as $id => $value) {
-                foreach ($value['tags'] ?? [] as $tag) {
+            foreach ($this->doFetchTags($ids) as $id => $tags) {
+                foreach ($tags as $tag) {
                     $tagData[$this->getId(self::TAGS_PREFIX.$tag)][] = $id;
                 }
             }

src/Symfony/Component/Cache/Adapter/FilesystemTagAwareAdapter.php

@@ -11,8 +11,8 @@
 
 namespace Symfony\Component\Cache\Adapter;
 
-use Symfony\Component\Cache\Marshaller\DefaultMarshaller;
 use Symfony\Component\Cache\Marshaller\MarshallerInterface;
+use Symfony\Component\Cache\Marshaller\TagAwareMarshaller;
 use Symfony\Component\Cache\PruneableInterface;
 use Symfony\Component\Cache\Traits\FilesystemTrait;
 
@@ -37,7 +37,7 @@ class FilesystemTagAwareAdapter extends AbstractTagAwareAdapter implements Prune
 
     public function __construct(string $namespace = '', int $defaultLifetime = 0, string $directory = null, MarshallerInterface $marshaller = null)
     {
-        $this->marshaller = $marshaller ?? new DefaultMarshaller();
+        $this->marshaller = new TagAwareMarshaller($marshaller);
         parent::__construct('', $defaultLifetime);
         $this->init($namespace, $directory);
     }
@@ -130,6 +130,40 @@ protected function doSave(array $values, ?int $lifetime, array $addTagData = [],
         return $failed;
     }
 
+    /**
+     * {@inheritdoc}
+     */
+    protected function doFetchTags(array $ids): iterable
+    {
+        foreach ($ids as $id) {
+            $file = $this->getFile($id);
+            if (!file_exists($file) || !$h = @fopen($file, 'rb')) {
+                continue;
+            }
+
+            $meta = explode("\n", fread($h, 4096), 3)[2] ?? '';
+
+            // detect the compact format used in marshall() using magic numbers in the form 9D-..-..-..-..-00-..-..-..-5F
+            if (13 < \strlen($meta) && "\x9D" === $meta[0] && "\0" === $meta[5] && "\x5F" === $meta[9]) {
+                $meta[9] = "\0";
+                $tagLen = unpack('Nlen', $meta, 9)['len'];
+                $meta = substr($meta, 13, $tagLen);
+
+                if (0 < $tagLen -= \strlen($meta)) {
+                    $meta .= fread($h, $tagLen);
+                }
+
+                try {
+                    yield $id => '' === $meta ? [] : $this->marshaller->unmarshall($meta);
+                } catch (\Exception $e) {
+                    yield $id => [];
+                }
+            }
+
+            fclose($h);
+        }
+    }
+
     /**
      * {@inheritdoc}
      */

src/Symfony/Component/Cache/Adapter/RedisTagAwareAdapter.php

@@ -16,6 +16,7 @@
 use Predis\Response\Status;
 use Symfony\Component\Cache\Exception\InvalidArgumentException;
 use Symfony\Component\Cache\Marshaller\MarshallerInterface;
+use Symfony\Component\Cache\Marshaller\TagAwareMarshaller;
 use Symfony\Component\Cache\Traits\RedisTrait;
 
 /**
@@ -67,7 +68,7 @@ public function __construct($redisClient, string $namespace = '', int $defaultLi
             throw new InvalidArgumentException(sprintf('Unsupported Predis cluster connection: only "%s" is, "%s" given.', PredisCluster::class, \get_class($redisClient->getConnection())));
         }
 
-        $this->init($redisClient, $namespace, $defaultLifetime, $marshaller);
+        $this->init($redisClient, $namespace, $defaultLifetime, new TagAwareMarshaller($marshaller));
     }
 
     /**
@@ -119,6 +120,42 @@ protected function doSave(array $values, ?int $lifetime, array $addTagData = [],
         return $failed;
     }
 
+    /**
+     * {@inheritdoc}
+     */
+    protected function doFetchTags(array $ids): iterable
+    {
+        $lua = <<<'EOLUA'
+            local v = redis.call('GET', KEYS[1])
+
+            if not v or v:len() <= 13 or v:byte(1) ~= 0x9D or v:byte(6) ~= 0 or v:byte(10) ~= 0x5F then
+                return ''
+            end
+
+            return v:sub(14, 13 + v:byte(13) + v:byte(12) * 256 + v:byte(11) * 65536)
+EOLUA;
+
+        if ($this->redis instanceof \Predis\ClientInterface) {
+            $evalArgs = [$lua, 1, &$id];
+        } else {
+            $evalArgs = [$lua, [&$id], 1];
+        }
+
+        $results = $this->pipeline(function () use ($ids, &$id, $evalArgs) {
+            foreach ($ids as $id) {
+                yield 'eval' => $evalArgs;
+            }
+        });
+
+        foreach ($results as $id => $result) {
+            try {
+                yield $id => !\is_string($result) || '' === $result ? [] : $this->marshaller->unmarshall($result);
+            } catch (\Exception $e) {
+                yield $id => [];
+            }
+        }
+    }
+
     /**
      * {@inheritdoc}
      */

src/Symfony/Component/Cache/CHANGELOG.md

@@ -7,6 +7,7 @@ CHANGELOG
  * added support for connecting to Redis Sentinel clusters
  * added argument `$prefix` to `AdapterInterface::clear()`
  * improved `RedisTagAwareAdapter` to support Redis server >= 2.8 and up to 4B items per tag
+ * added `TagAwareMarshaller` for optimized data storage when using `AbstractTagAwareAdapter`
  * [BC BREAK] `RedisTagAwareAdapter` is not compatible with `RedisCluster` from `Predis` anymore, use `phpredis` instead
 
 4.3.0

src/Symfony/Component/Cache/Marshaller/TagAwareMarshaller.php

@@ -0,0 +1,89 @@
+<?php
+
+/*
+ * This file is part of the Symfony package.
+ *
+ * (c) Fabien Potencier <fabien@symfony.com>
+ *
+ * For the full copyright and license information, please view the LICENSE
+ * file that was distributed with this source code.
+ */
+
+namespace Symfony\Component\Cache\Marshaller;
+
+/**
+ * A marshaller optimized for data structures generated by AbstractTagAwareAdapter.
+ *
+ * @author Nicolas Grekas <p@tchwork.com>
+ */
+class TagAwareMarshaller implements MarshallerInterface
+{
+    private $marshaller;
+
+    public function __construct(MarshallerInterface $marshaller = null)
+    {
+        $this->marshaller = $marshaller ?? new DefaultMarshaller();
+    }
+
+    /**
+     * {@inheritdoc}
+     */
+    public function marshall(array $values, ?array &$failed): array
+    {
+        $failed = $notSerialized = $serialized = [];
+
+        foreach ($values as $id => $value) {
+            if (\is_array($value) && \is_array($value['tags'] ?? null) && \array_key_exists('value', $value) && \count($value) === 2 + (\is_string($value['meta'] ?? null) && 8 === \strlen($value['meta']))) {
+                // if the value is an array with keys "tags", "value" and "meta", use a compact serialization format
+                // magic numbers in the form 9D-..-..-..-..-00-..-..-..-5F allow detecting this format quickly in unmarshall()
+
+                $v = $this->marshaller->marshall($value, $f);
+
+                if ($f) {
+                    $f = [];
+                    $failed[] = $id;
+                } else {
+                    if ([] === $value['tags']) {
+                        $v['tags'] = '';
+                    }
+
+                    $serialized[$id] = "\x9D".($value['meta'] ?? "\0\0\0\0\0\0\0\0").pack('N', \strlen($v['tags'])).$v['tags'].$v['value'];
+                    $serialized[$id][9] = "\x5F";
+                }
+            } else {
+                // other arbitratry values are serialized using the decorated marshaller below
+                $notSerialized[$id] = $value;
+            }
+        }
+
+        if ($notSerialized) {
+            $serialized += $this->marshaller->marshall($notSerialized, $f);
+            $failed = array_merge($failed, $f);
+        }
+
+        return $serialized;
+    }
+
+    /**
+     * {@inheritdoc}
+     */
+    public function unmarshall(string $value)
+    {
+        // detect the compact format used in marshall() using magic numbers in the form 9D-..-..-..-..-00-..-..-..-5F
+        if (13 >= \strlen($value) || "\x9D" !== $value[0] || "\0" !== $value[5] || "\x5F" !== $value[9]) {
+            return $this->marshaller->unmarshall($value);
+        }
+
+        // data consists of value, tags and metadata which we need to unpack
+        $meta = substr($value, 1, 12);
+        $meta[8] = "\0";
+        $tagLen = unpack('Nlen', $meta, 8)['len'];
+        $meta = substr($meta, 0, 8);
+
+        return [
+            'value' => $this->marshaller->unmarshall(substr($value, 13 + $tagLen)),
+            'tags' => $tagLen ? $this->marshaller->unmarshall(substr($value, 13, $tagLen)) : [],
+            'meta' => "\0\0\0\0\0\0\0\0" === $meta ? null : $meta,
+        ];
+    }
+}

src/Symfony/Component/Cache/Traits/AbstractAdapterTrait.php

@@ -51,11 +51,13 @@ public function getItem($key)
             foreach ($this->doFetch([$id]) as $value) {
                 $isHit = true;
             }
+
+            return $f($key, $value, $isHit);
         } catch (\Exception $e) {
             CacheItem::log($this->logger, 'Failed to fetch key "{key}": '.$e->getMessage(), ['key' => $key, 'exception' => $e]);
         }
 
-        return $f($key, $value, $isHit);
+        return $f($key, null, false);
     }
 
     /**

src/Symfony/Component/Cache/Traits/RedisTrait.php

@@ -445,25 +445,26 @@ private function pipeline(\Closure $generator, $redis = null): \Generator
             $results = [];
             foreach ($generator() as $command => $args) {
                 $results[] = $redis->{$command}(...$args);
-                $ids[] = $args[0];
+                $ids[] = 'eval' === $command ? ($redis instanceof \Predis\ClientInterface ? $args[2] : $args[1][0]) : $args[0];
             }
         } elseif ($redis instanceof \Predis\ClientInterface) {
             $results = $redis->pipeline(static function ($redis) use ($generator, &$ids) {
                 foreach ($generator() as $command => $args) {
                     $redis->{$command}(...$args);
-                    $ids[] = $args[0];
+                    $ids[] = 'eval' === $command ? $args[2] : $args[0];
                 }
             });
         } elseif ($redis instanceof \RedisArray) {
             $connections = $results = $ids = [];
             foreach ($generator() as $command => $args) {
-                if (!isset($connections[$h = $redis->_target($args[0])])) {
+                $id = 'eval' === $command ? $args[1][0] : $args[0];
+                if (!isset($connections[$h = $redis->_target($id)])) {
                     $connections[$h] = [$redis->_instance($h), -1];
                     $connections[$h][0]->multi(\Redis::PIPELINE);
                 }
                 $connections[$h][0]->{$command}(...$args);
                 $results[] = [$h, ++$connections[$h][1]];
-                $ids[] = $args[0];
+                $ids[] = $id;
             }
             foreach ($connections as $h => $c) {
                 $connections[$h] = $c[0]->exec();
@@ -475,7 +476,7 @@ private function pipeline(\Closure $generator, $redis = null): \Generator
             $redis->multi(\Redis::PIPELINE);
             foreach ($generator() as $command => $args) {
                 $redis->{$command}(...$args);
-                $ids[] = $args[0];
+                $ids[] = 'eval' === $command ? $args[1][0] : $args[0];
             }
             $results = $redis->exec();
         }