Bengaluru Systems Meetup #2

At Gojek, we have a love-hate kinda relationship with document stores. In one such use case, the design of our entity model made sense that a document store (ahem, mongoDb) should be used. Now serving a write traffic of a whopping ~25k RPM, we were bound to hit a wall, and we ended up hitting a cache eviction wall.

This talk explores our debugging process of the issue and all the insights regarding the storage engine which I gathered while deep-diving mongo’s source code at v6.0.

By the end of the talk, the participants would appreciate the nuances regarding:

• How does the storage engine store data?
• Internals of cache loading and evictions within mongo’s storage engine.
• How to tune your write-heavy database to solve this problem (unfortunately not covered in attached blog post).