Foundry Standard Agent Setup (BYO Cosmos DB), run-state-v1 container throttled at 1000 RU/s

Question

Foundry Standard Agent Setup (BYO Cosmos DB), run-state-v1 container throttled at 1000 RU/s

Nutpapon Limsupavanich 20

Hi. I'm running a Foundry Standard agent setup with Bring-Your-Own (BYO) thread storage backed by my own Azure Cosmos DB for NoSQL account.

Setup:

Standard agent setup (own Cosmos DB, Storage, AI Search)
Cosmos DB: no total throughput limited, provisioned throughput
Database: enterprise_memory
Per the docs, 5 containers are auto-provisioned per project, each with a documented minimum of 1000 RU/s:

Problem: The run-state-v1 container — which tracks agent run execution state plus internal/end-user messages tied to those runs — is consistently hitting RU limits and getting throttled (429s) at the documented 1000 RU/s minimum. I had to manually scale it up to 3000 RU/s to stop the throttling.

Any guidance from someone who has tuned this in production, or from the Foundry/Cosmos DB product team, would be appreciated. Happy to share metrics/diagnostics if useful.

Thank you,

Nutpapon Lim

Himaja Y 375 Reputation points Microsoft External Staff Moderator

2026-09-03T14:32:16.02+00:00

Hi @Nutpapon Limsupavanich ,

I hope you had a chance to review the information shared earlier, and I hope this information has been helpful! If you still have questions, please let us know what is needed in the comments so the question can be answered.

1 answer

Your answer

Himaja Y 375 Reputation points Microsoft External Staff Moderator

2026-09-03T14:32:16.02+00:00

Hi @Nutpapon Limsupavanich ,

I hope you had a chance to review the information shared earlier, and I hope this information has been helpful! If you still have questions, please let us know what is needed in the comments so the question can be answered.

Answer 1

Hi @Nutpapon Limsupavanich ,

Thank you for reaching out to the Microsoft Q&A forum.

What you're observing is likely expected behavior for workloads with moderate to high agent activity.

The 1000 RU/s configured for the run-state-v1 container is generally the minimum provisioned throughput, not necessarily a production sizing recommendation. Since this container stores agent execution state, run status updates, and conversation messages, it typically experiences the highest read/write volume among the auto-created containers.

A few things to check:

RU consumption and 429 metrics for the run-state-v1 container.
Partition key distribution to identify any hot partitions.
Concurrent agent runs/threads during peak usage.
Document size and indexing policy, as larger documents and excessive indexing can significantly increase RU consumption.

Since increasing throughput from 1000 RU/s to 3000 RU/s eliminated throttling, the behavior appears workload-driven rather than indicative of a platform issue.

If possible, please share:

Peak concurrent agent runs
429/throttling metrics
Partition key design
RU utilization graphs

This will help determine whether the throttling is due to workload scale, partition hot spots, or an underlying service-related concern.