Solving Complex Scaling Problems by Prashant Kumar and Abhishek Jain of Myntra - Redis Day Bangalore 2020
Download as PPTX, PDF0 likes145 views
Redis was used by Myntra to solve several complex scaling problems. It was used to build a scalable user segment service to support high read throughput of up to 5 million requests per minute with low latency. Redis allowed the service to scale beyond a single instance and included features like automatic backups and memory management. Redis also helped build a scalable mobile verification platform to reliably handle 100,000 requests per minute and scale to support higher future volumes. It was used as both a transient store and persistent backend. Finally, Redis locks helped build a scalable A/B testing platform by allowing experiments to be created and updated in an orderly concurrent fashion.
Solving Complex Scaling Problems by Prashant Kumar and Abhishek Jain of Myntra - Redis Day Bangalore 2020
- 1. Redis to Solve Complex Scale Problems Abhishek Jain, Prashant Kumar Engineering, Platforms @Myntra
- 2. PRESENTED BY 1 Why Redis? 2 Scalable Segment Service 3 Scalable Mobile Verification Platform Agenda: 4 5 Scalable A/B Test Platform Redis as a Managed Service
- 3. PRESENTED BY • Turn underutilised resources on a server (CPU/RAM) into valuable resources. • Scale beyond a single instance of cache, if more headroom is required, while some other caching services such as memache cannot. • Regular backups at disk to recover and get boot-up with minimal loss of data. • Minimal memory fragmentation, to allow better reads. • Ability to setup a cluster which helps with data sharding. • Setting up user defined max-memory usage. • User defined policies to either stop or evict data based on recent usages or expiry. Why Redis?
- 4. PRESENTED BY Any B2C product will have user segmentation to perform A/B Testing, Personalization or any behavioral analysis, certain analysis leads to create new segments as well • Support very high read throughput, RPM: ~5mil • latency should not be more than 5ms • Service should be scalable to support for next few years Redis for Scalable Segment Service
- 5. PRESENTED BY Redis seemed to fit the bill perfectly here, as it fulfills all our requirements here. • Challenge: During writes almost entire redis database (keys) updated. Read throughput affected during updates. • Solution: ⎼ Before any write could happen at redis, we put up a queue (kafka), which throttles writes ⎼ Read and Write pools of redis connections were made separate. Redis for Scalable Segment Service
- 6. PRESENTED BY Mobile verification via OTP is a popular mechanism to perform authentication for different purposes, like Sign-In, 2 Factor Auth etc. • OTP generation, Large scale Read/Writes, Time Bound deliveries were top level challenges • Store OTP in RDBMS for high consistency. What if RDBMS goes down? • RPM to serve: ~100k • What if number of request doubles, or 10 times in next 5 years? • How do we make it 99.999% available service? Should be highly consistent and reliable. • We also wanted to persist those OTP sent against mobile number for other purposes, but not within the transactional flow Redis for Scalable Mobile Verification Platform
- 7. PRESENTED BY We introduced Augmented Data Store system • Used Redis as a transient store & NoSQL DB for persistence and fallback option • OTP data must expires in 15-20 minutes, so using ephemeral memory • Redis works in cluster mode with Master-Slave, RF=1 • Redis helps in scale due to cluster mode, much better than any schema based DB Redis for Scalable Mobile Verification Platform
- 8. PRESENTED BY In any B2C product, decision making is a tricky job, So A/B Testing experimentation helps, and performing large scale A/B is complex because of concurrency, so distributed lock comes in picture. • Using redis locks to serve concurrent users creating A/B experiments • Lock prevents multiple concurrent processes stepping on each other • Many times locking over a request is required to make it transactional Redis for Scalable A/B Test Platform
- 9. PRESENTED BY For creation of experiments • Slotting concept, each experiment is associated with certain slots with each slot containing a part of the population set • If multiple concurrent users try to create/update, execution happens in an orderly fashion while acquiring read and write locks over the slots. Redis for Scalable A/B Test Platform
- 10. PRESENTED BY • 1-click-deploy redis cluster ready to boot up and use within a few minutes. • Auto-Monitoring setup, with cluster health checks, commands monitoring, done via Graphana. • Auto-backups with replication can also be configured with the setup before deploying if required. Redis as a Managed Service
- 11. Thank You!