New Relic and NVIDIA released the first observability integration making it easy for companies to monitor the health and performance of their AI appli

ByteByteGo Newsletter

submited by

Style Pass

2024-07-06 20:30:04

New Relic and NVIDIA released the first observability integration making it easy for companies to monitor the health and performance of their AI applications built with NVIDIA NIM.

Async Processing Move time-consuming and resource-intensive tasks to background workers using async processing to scale out new requests.

Vertical Scaling and Replication Figma used a single, large Amazon RDS database. As a first step, they upgraded to the largest instance available (from r5.12xlarge to r5.24xlarge). They also created multiple read replicas to scale read traffic and added PgBouncer as a connection pooler to limit the impact of a growing number of connections.

Vertical Partitioning The next step was vertical partitioning. They migrated high-traffic tables like “Figma Files” and “Organizations” into their separate databases. Multiple PgBouncer instances were used to manage the connections for these separate databases.

Horizontal Partitioning Over time, some tables crossed several terabytes of data and billions of rows. Postgres Vacuum became an issue and max IOPS exceeded the limits of Amazon RDS at the time. To solve this, Figma implemented horizontal partitioning by splitting large tables across multiple physical databases. A new DBProxy service was built to handle routing and query execution.