In a distributed database, data is split into multiple tablets which reside on different nodes. So it is not just tables, but indexes, that are distributed across nodes. So let’s examine how to optimally design distributed indexes to get the best query performance.
In storage systems based on LSM/MVCC, when a record is deleted, it’s not immediately removed. A delete marker (typically called a Tombstone) is placed on the record. As the number of tombstones increases, it could adversely affect the performance of a scan. In this blog, we are going to understand this problem and come up with solutions to address it.
In a previous blog, we developed an application-level sharding layout to avoid hotspots. With that layout in mind, where order is maintained within each shard, let’s discuss how to design a query to return data with pagination while maintaining the global ordering.
Some data model choices in distributed databases cause data to grow in one node before it moves to another node. This will cause one node to become a hotspot for reads and writes. This article explains how to avoid that.