In this talk we will cover: 1. What is Memcache? 2. Why do we need Memcache? 3. How to Use Memcache on Google App Engine? 4. Caveats And Solutions

1. App Engine Memcache Basics

2. App Engine: Memcache Basics
Who? Why?
Ido Green
Solutions Architect
plus.google.com/greenido
greenido.wordpress.com

3. App Engine: Memcache Basics
Topics We Cover in This Lesson
● What is Memcache?
● Why do we need Memcache?
● How to Use Memcache?
● Caveats And Solutions

4. App Engine: Memcache Basics
What Is Memcache?

5. What is Memcache?
What is Memcache?
● Memcache is an in-memory Key-Value Pairs data store
○ Put a value with a key
○ Get a value with a key
value
key
"user001" : "John Doe"
"user002" : "Larry Page"
● key or value can be anything that is serializable
● Memcache is a shared service accessed via App Engine APIs.

6. What Is Memcache
What Do We Use Memcache For?
●
●
●
Caching In Front of Datastore
○ Cache entities for low-latency reads
○ Integrated into most ORM frameworks (ndb, Obectify ..)
Caching for Read heavy operations
○ User authentication token and session data
○ APIs call or other computation results
Semi-durable Shared state Cross App Instances
○ Sessions
○ Counters / Metrics
○ Application Configurations

7. App Engine: Memcache Basics
Why Do We Need Memcache?

8. Why do we need Memcache?
Why Do We Use Memcache?
● Improve Application Performance
● Reduce Application Cost

9. Why do we need Memcache?
How Fast Can Memcache be? And How Much?
Datastore Query Latency
Memcache Read Latency

10. How to Use Memcache

11. How to Use Memcache
Memcache APIs
● Java
○ JCache APIs
○ GAE Low-Level Memcache APIs
○ Objectify for Datastore
● Python
○ google.appengine.api.memcache module
○ ndb for Datastore

12. How to Use Memcache
General Memcache Usage Pattern
Coordinate data read with Datastore:
● Check if Memcache value exists
○ if it does, display/use cached value directly; otherwise
○ fetch the value from Datastore and write the value to
Memcache
Coordinate data write with Datastore:
● Update Memcache value
○ to handle race condition, leverage put if
untouched/compare and set to detect race conditions
● Write the value to Datastore
○ optionally, leverage the task queue for background writes

13. How to Use Memcache
Using GAE Client Library (Java)
import com.google.appengine.api.memcache;
...
MemcacheService syncCache = MemcacheServiceFactory.
getMemcacheService();
// Act on error as cache miss
syncCache.setErrorHandler(ErrorHandlers.
getConsistentLogAndContinue(Level.INFO));
value = (byte[]) syncCache.get(key); // read from cache
if (value == null) {
value = getDataFromDb(key); // fetch value from datastore
syncCache.put(key, value); // write to cache
// (key and value must be
serializable)
}
...

14. How to Use Memcache
Using GAE Client Library (Python)
from google.appengine.api import memcache
...
value = memcache.get(key)
if value is None:
value = get_value_from_db(key)
if not memcache.add(key, value):
logging.error('Memcache add failed.')
...

15. How to Use Memcache
Batch Operations
● getAll(), putAll(), deleteAll()
A single read or write operation for multiple memcache
entries
Note
● Further improve Memcache performance by reducing network traffics
● Batch size < 32 MB

16. How to Use Memcache
Atomic Operations
● increment(key, delta), incrementAll(...),
Provide atomic increment of numeric value(s)
● getIdentifiable(), putIfUntouched() in Java, or cas() in Python
A mechanism to update a value consistently by concurrent requests
Note
●
Helps managing memcache data consistency in
multi-instances/concurrent environment

17. How to Use Memcache
Some Other Features
● Asynchronous calls
Provides a mechanism to make a non-blocking call for memcache
operations
● Namespace
Logically separates data layers for different application purposes (like
multi tenancy) across many GAE services such as Datastore,
Memcache, Task Queue etc...

18. Caveats And Solutions

19. Caveats And Solutions
Caveat: Memcache Is Volatile
Entries can be evicted anytime by various reasons:
●
●
●
entry reaches expiration
entry is evicted because memcache memory is full
memcache server fails
Tip
●
It's important to handle cache-miss gracefully!
●
Implements write-through logic by backing
memcache with datastore in your application!
○ Psst... Objectify and ndb do it for you.

20. Caveats And Solutions
Caveat: Memcache Is Not Transactional
Instance 1 reads $100
$100
$100
Instance 2 reads $100
$100
$100
Instance 2 deducts $20
$80
$80
Instance 1 deducts $30
$70
$70
Tip
Using getIdentifiable() and putIfUntouched(...)
for optimistic locking

21. Caveats And Solutions
Caveat: Memcache Is A Limited Resource
My Application Does NOT Have Enough Memcache!
Tips
●
Only need to cache what is useful and necessary!
●
Your application should function without memcache!
●
Dedicated Memcache

22. Caveats And Solutions
Dedicated Memcache
Enabling dedicated memcache from the admin console
1. Provides fixed cache capacity that is exclusive to your application
2. Billed by the GB per hour of cache size. Charged in 15 minute
increments
3. Gives you additional control over cache size. More predictable
performance
Note: Whether shared or dedicated, Memcache is not a durable storage.
Plan for your application to function without Memcache

23. Discussion And Questions
Key Takeaways
1. Memcache is supported natively in GAE. Take advantage of it to
improve your GAE application performance.
2. Memcache supports open standard JCache API. Many advanced
features are available by GAE Memcache APIs to suit your
application's need i.e. Batch, Atomic, Asynchronous operations
3. Seamless integration with GAE Datastore in a few libraries like
Python ndb and Java Objectify.
4. Read-frequently and write-rarely data is most suitable in combining
with Memcache
5. Handle Memcache's volatility in your application
6. Use Memcache wisely, it is not an unlimited resource

24. App Engine: Memcache Basics
Thank you!
Questions?