Stop Scaling Vertically: Debugging Redis Memory Leaks in WooCommerce 10.2+

#woocommerce #wordpress #redis #performance

Building a high-traffic WooCommerce store is easy. Keeping it alive during a flash sale when your Redis instance is hemorrhaging memory? That’s where the real engineering begins.

I’ve seen 64GB bare-metal servers choke on 100 concurrent users not because of traffic, but because of a Silent Killer: In-memory datastore exhaustion.

The Anatomy of a Leak

A Redis memory leak in WooCommerce usually isn't a Redis bug. It’s an application-layer failure. Poorly coded plugins generate unique transients (session data, dynamic pricing, filter fragments) for every single user but fail to assign a TTL (Time-To-Live).

The Chain Reaction:

The Bloat: Unique keys accumulate infinitely.
The OOM: Redis hits its maxmemory limit.
The Fallback: Under a default noeviction policy, Redis rejects new writes.
The Crash: PHP-FPM workers stall waiting for MySQL, leading to a 502 Bad Gateway.

Step 1: Profiling via CLI (The No-Plugin Approach)

Don't trust WordPress dashboard graphs when your server is redlining. Get into the terminal.

# Check real-time stats
redis-cli --stat

# Ask the "Doctor" for a diagnostic
redis-cli MEMORY DOCTOR

If your keys count only goes up and never fluctuates, you have orphaned transients.

Step 2: Fixing the Eviction Policy

Most "off-the-shelf" Redis configs are set to noeviction or volatile-lru. In a chaotic WordPress ecosystem, volatile-lru is a trap because it only deletes keys with an expiration date.

The Fix: Edit your redis.conf and force a ruthless survival policy.

maxmemory 4gb
maxmemory-policy allkeys-lru

allkeys-lru doesn't care if a key has a TTL or not. If RAM is full, the least recently used data is evicted. Period. This keeps the checkout flow alive.

Step 3: Bypassing the Persistent Cache

For hyper-dynamic B2B data (like unique wholesale pricing per user), stop saving it to Redis. It’s a waste of resources.

Add this to your wp-config.php or a custom MU-plugin:

wp_cache_add_non_persistent_groups( array( 'wc_session_id', 'woocommerce_cart_hash' ) );

The Front-End Connection: DOM Bloat

Here’s the "Senior Engineer" secret: Your Elementor DOM tree is killing your backend. When you have 4,500+ DOM nodes, PHP takes longer to render the page. This keeps the Redis connection open longer, exhausting the connection pool and accelerating memory fragmentation.