From Rising RU Costs to Stable Performance (Azure Cosmos DB Conf 2026)
Anurag Dutt walks through a realistic Azure Cosmos DB scenario where a high-volume ingestion and querying workload starts to suffer from rising RU costs, unpredictable latency, and uneven throughput as it scales.
Full summary based on transcript
Problem pattern: cost growth and unstable performance at scale
As Cosmos DB workloads scale, early design decisions can lead to:
- Rising RU consumption
- Unpredictable latency
- Uneven throughput
Diagnostic approach using Azure platform signals
Anurag Dutt uses Azure platform metrics, diagnostics, and query insights to identify common root causes, including:
- Hot partitions
- Cross-partition queries
- Over-indexing
- Inefficient throughput provisioning
Targeted improvements to restore stability and reduce cost
The session focuses on applying targeted changes based on what the diagnostics reveal, including:
- Addressing hot partitions (partitioning strategy adjustments)
- Reducing cross-partition query impact (query and data access pattern changes)
- Fixing over-indexing (indexing policy tuning)
- Improving throughput provisioning (right-sizing and provisioning strategy)
Practical takeaway
The video aims to provide a repeatable framework for diagnosing and resolving common Azure Cosmos DB performance and cost issues using built-in Azure observability and query tooling.