Rootconf Mini 2024

Objective

At Flipkart, we have seen the huge adoption of the home grown managed platforms running over hybrid cloud by all the engineering teams working at massive scale, and DbaaS platforms are protagonists of this story. It becomes paramount that these platforms can maintain high resilience, high availability to deliver sustained performance and continuous optimisations to handle adoption at scale. In this talk, we delve into:

• What does it take to measure the availability in realtime as well as platform resiliency aspects for disaster recovery strategy for hybrid cloud topology.
• How does these platforms intend to deliver further values with continous optimisation and recommendations to improve productivity of platform maintainers.
• Our attempts to capitalize language models to extract intelligence for multiple platform level optimisations.

What will be the scope of this session? Key problems, challenges being addressed and lessons learned

Appropriate format for this session - 40 mins talk

• Building blocks [10 mins]

Any managed DbaaS platform is not mere multiple dbs running, rather a stitched and tuned ecosystem of various components working in harmony.

• Availability computation [5 mins]

  • Everyone projects multi-nine numbers, or else need HA setup, but how do you decide between Availability v/s Uptime?
  • DbaaS realtime scientific way of Availability Computation.

• Resiliency trade-off for DbaaS [10 mins]

DbaaS resiliency trade-offs over hybrid cloud setup during important sale events like Big Billion Days.

• Optimisation measures taken [5 mins]

Productivity improvements, Oncall reduction and Opex tightning

• Intelligence in-house (thank you open source LLMs) [10 mins]

  • Insights generation and recommendations for platform maintainers.
  • Self-serve troubleshooting and debugging capabities for platform users.
  • Platform operational readiness through agents.

Target audience and Takeaways for this session. Problem/pain to be solved for the audience.

• This talk with particularly pique interests of the engineers who are fed-up with maintaining various self-managed stacks and keen on evaluating platform first approach. [(1) Building blocks]
• Backend engineers evaluating systemic availability computation, DR usecases and productivity measures [(2) Availability computation]
• Architects looking to resilient multi cloud design and optimise opex working with datastores. [(3) Resiliency and 4) Optimisations]
• AI-for-DB or AI-for-Platforms connoisseurs [(5) Intelligence in-house]
• It’s also highly relevant for developers working on large-scale distributed systems requiring extreme backend scaling with high throughput low latency performance needs. [Across]