This document discusses using ElastiCache for Redis in production. It begins with an introduction to the author and their role at gumi Inc. It then provides an overview of Redis operations and data persistence methods. The document outlines gumi's migration from memcached to Redis, including their use of master-slave replication and multiple databases. It notes potential issues with ElastiCache for Redis like lack of auto failover and inability to change instance types. In summary, while some features are still missing, ElastiCache for Redis allows quick node provisioning and has proven reliable for gumi despite initial function gaps.
result management system report for college project
Use ElastiCache for Redis in Production Guide
1. Use case for using
the ElastiCache for Redis
in Production.
2. Table of contents
• Self-introduction
• The minimum knowledge needed Redis
operation
• gumi’s Redis Operation
• Warning for operating ElastiCache for Redis
• Summary
3. Table of contents
• Self-introduction
• The minimum knowledge of Redis operation
• gumi’s Redis Operation
• Warning for operating ElastiCache for Redis
• Summary
4. Self-introduction
• Tomonori Honma
• gumi Inc.
• System Operation Engineer
• All Game Server’s operarion in Japan
• I don’t develop Game Events
• My favorite AWS Service:Amazon S3
@CkReal
5. Table of contents
• Self-introduction
• The minimum knowledge needed Redis
operation
• gumi’s Redis Operation
• Warning for operating ElastiCache for Redis
• Summary
6. The minimum knowledge
of Redis operation
• Key Value Store
• In-memory is very fast
• When is storing data over memory in Server, is
controlled by maxmemory-policy(Parameters
Group) settings. (default:volatile-lru)
• Two methods for Data Persistence
• Append-only file
• RDB file
• Can use multiple databases
7. Table of contents
• Self-introduction
• The minimum knowledge needed Redis
operation
• gumi’s Redis Operation
• Warning for operating ElastiCache for Redis
• Summary
10. The Use case
of Application
• User’s actions in game events can be immediately
reflected in Redis.
• example
• Realtime Ranking
• Matching process of each player
• Card Deck Information
11. The Use case
of Operation
• Error Tracking Tool(Sentry) use as a Buffer
• Recent Operation tools sometimes uses Redis
• For quick checks, EC2 is very troublesome.
12. gumi’s Redis Operation
• Using almost master-slave
• A few games use master only
• master : AOF and no RDB, slave : AOF and RDB
• One GET operation takes sometimes 2.47sec???
• Fork time in different systems
• Multiple backups of RDB are saved
• This is to avoid problem where master DB is
flushed and that wipes out slave backup too.
• The use multiple databases in one Redis node
• Instead of memcached, use redis
• Developer is difficult to understand the timing of
memcached's eviction.
13. Table of contents
• Self-introduction
• The minimum knowledge needed Redis
operation
• gumi’s Redis Operation
• Warning for operating ElastiCache for Redis
• Summary
14. Possible and
Not (yet) possible
Possible Not (yet) possible
No EC2 instance!!
Import data file from S3
Configure AOF setting
Use awscli
Promotion of replica node
Auto failover like RDS
Backup RDB file to S3
Change instance type
Point-In-Time-Recovery
15. Possible and
Not (yet) possible
Possible Not (yet) possible
No EC2 instance!!
Import data file from S3
Configure AOF setting
Use awscli
Promotion of replica node
Auto failover like RDS
Backup RDB file to S3
Change instance type
Point-In-Time-Recovery
2014/04/24 Release
Backup time includes time
taken to upload RDB file to S3?
EC2 : few min
ElastiCache : Tens of minutes
16. Possible and
Not (yet) possible
Possible Not (yet) possible
No EC2 instance!!
Import data file from S3
Configure AOF setting
Use awscli
Promotion of replica node
Auto failover like RDS
Backup RDB file to S3
Change instance type
Point-In-Time-Recovery
17. Possible and
Not (yet) possible
Possible Not (yet) possible
No EC2 instance!!
Import data file from S3
Configure AOF setting
Use awscli
Promotion of replica node
Auto failover like RDS
Backup RDB file to S3
Change instance type
Point-In-Time-Recovery
ElastiCache for Redis is still not
high-performance like RDS
18. ElastiCache for Redis
impression
• There is nothing wrong in performance
• When you create a snapshot, it is recommended to run
from slave
• Backup and Restore ElastiCache Redis Clusters
• appendonly settings is not available in cache.t1.micro
• We have not experienced a big problem with
ElastiCache for Redis
• When AOF file is corrupted, I have restored from
RDB file of slave server.
19. Table of contents
• Self-introduction
• The minimum knowledge needed Redis
operation
• gumi’s Redis Operation
• Warning for operating ElastiCache for Redis
• Summary
20. Summary
• Can prepare redis nodes in a short period of time
• Since we have very few operation enginners in
our company, this is very helpful :)
• Initially, functions we assumed to exist actually
were not yet implemented
• But feature set has grown
• Currently, there are no big problems
• we are increasing the number of production nodes