QuestionQ70

VerifiedDevelop AI solutions by using Azure data management services

A recommendation system stores 3 million embeddings in Azure Database for PostgreSQL.

During peak hours, P95 similarity-query latency rises and the cache hit ratio drops significantly.

You suspect the vector-index working set no longer fits in memory, causing additional disk reads.

You must scale server resources and validate the result.

Which four actions should you take, in order?

Drag & Drop
Explanation

A low cache-hit ratio together with increased similarity-query latency indicates more reads are being served from storage rather than memory. Establishing a pre-change benchmark and plan provides a comparison point; increasing memory per vCore addresses the working-set shortfall; observing memory pressure and cache behavior during peak load confirms the suspected mechanism; and repeating the identical benchmark verifies the P95/P99 latency outcome. Higher embedding dimensionality increases index and memory requirements, and autovacuum settings do not correct a vector-index working-set memory deficit.

Community Discussion

No comments yet. Be the first to start the discussion!