Cache stampede prevention in Redis - Step-by-Step Execution

Concept Flow - Cache stampede prevention

Client requests data

↓

Check cache for data

↓

Return cached

↓

Lock acquired?

↓

Fetch from DB

↓

Update cache

↓

Return data to client

When a client requests data, the system checks the cache. If missing, it tries to acquire a lock to fetch from the database and update the cache. Others wait for the cache to be updated to prevent many DB hits.

Execution Sample

Redis

if cache.exists(key):
    return cache.get(key)
if cache.setnx(lock_key, 1):
    data = db.query()
    cache.set(key, data)
    cache.delete(lock_key)
else:
    wait_for_cache_update()
    return cache.get(key)

This code checks cache, acquires a lock to fetch DB data if missing, updates cache, and others wait to avoid multiple DB hits.

Execution Table

Step	Action	Cache State	Lock State	DB Query	Returned Data
1	Client requests data	empty	no lock	no query	none
2	Check cache for key	empty	no lock	no query	none
3	Cache miss, try to set lock	empty	lock set	no query	none
4	Lock acquired, query DB	empty	lock set	query executed	none
5	Update cache with data	data cached	lock set	query done	none
6	Delete lock	data cached	no lock	query done	none
7	Return cached data	data cached	no lock	query done	data returned
8	Another client requests data	data cached	no lock	no query	data returned
9	Check cache for key	data cached	no lock	no query	data returned
10	Cache hit, return data	data cached	no lock	no query	data returned
11	Client requests data, cache empty	empty	lock set by another	no query	none
12	Lock not acquired, wait	empty	lock set	no query	none
13	Cache updated by lock owner	data cached	no lock	query done	none
14	Return cached data after wait	data cached	no lock	query done	data returned

💡 Execution stops after data is returned from cache or DB and lock is released.

Variable Tracker

Variable	Start	After Step 3	After Step 5	After Step 6	After Step 10	After Step 14
cache	empty	empty	data cached	data cached	data cached	data cached
lock	no lock	lock set	lock set	no lock	no lock	no lock
db_query	no query	no query	query executed	query done	no query	query done
returned_data	none	none	none	none	data returned	data returned

Key Moments - 3 Insights

Why do we use a lock before querying the database?

The lock prevents multiple clients from querying the database simultaneously when the cache is empty, as shown in steps 3-6 in the execution_table.

What happens if a client cannot acquire the lock?

The client waits for the cache to be updated by the lock owner, then reads from the cache, as shown in steps 11-14.

Why do we delete the lock after updating the cache?

Deleting the lock signals other clients that the cache is ready to be read, preventing deadlocks and allowing them to get data from cache (step 6).

Visual Quiz - 3 Questions

Look at the execution_table at step 4. What is happening?

A Cache is being updated with data

B Client returns cached data

C Database query is executed after lock acquired

D Lock is deleted

💡 Hint

Check the 'DB Query' and 'Lock State' columns at step 4 in execution_table.

At which step does the lock get released?

A Step 3

B Step 6

C Step 5

D Step 10

💡 Hint

Look at the 'Lock State' column in execution_table to find when it changes from 'lock set' to 'no lock'.

If the cache is already filled, what happens when a client requests data?

A Client returns cached data immediately

B Client queries the database

C Client waits for lock to be released

D Client sets a new lock

💡 Hint

Refer to steps 8-10 in execution_table where cache is 'data cached'.

Concept Snapshot

Cache Stampede Prevention:
- Check cache first for data.
- If missing, try to set a lock (setnx).
- Lock owner queries DB and updates cache.
- Others wait for cache update to avoid multiple DB hits.
- Delete lock after cache update to release waiting clients.

Full Transcript

Cache stampede prevention is a technique to avoid many clients querying the database at once when cache data is missing. The process starts when a client requests data and checks the cache. If the data is not found, the client tries to acquire a lock. If successful, it queries the database, updates the cache, and releases the lock. Other clients that fail to get the lock wait until the cache is updated and then read from the cache. This prevents multiple expensive database queries and reduces load. The execution table shows each step, including cache state, lock state, database queries, and returned data. Key moments include why the lock is used, what happens when the lock is not acquired, and why the lock is deleted after updating the cache. The visual quiz tests understanding of these steps. The snapshot summarizes the main points for quick reference.