Browser Run Gets 4x Capacity, 50% Faster Quick Actions, Powered by Cloudflare Containers
Deploy the new Browser Run limits and migrate container state to D1 with Queues for batch writes.
Deploy the new Browser Run limits and migrate container state to D1 with Queues for batch writes.
Summary
Browser Run now supports 60 browsers per minute and 120 concurrent instances, a four‑fold increase over the previous limits, and Quick Action response times have dropped by more than 50 %. No changes are required from customers; the improvements are live today. The upgrade was enabled by migrating from the older Browser Isolation (BISO) platform to Cloudflare Containers with Durable Object‑enabled Containers, which solved startup latency and global distribution issues.
To reduce latency between the Durable Object and the container, regional pre‑warmed DO‑backed browser pools were created, ensuring the maximum distance between the two is minimized. Container state was moved from Workers KV to D1, leveraging SQLite transactions for atomic assignment and preventing race conditions. Batch writes of 100 rows were introduced via Cloudflare Queues with a 1 s timeout, achieving a P95 latency of 0.1 ms.
When the primary queue lags, a fallback to a backup region is used to keep state fresh. The new architecture allows rapid scaling to meet the growing demand from AI agents and other use cases, delivering higher capacity and faster performance without requiring customer redeployments.
Key changes
- Increased Browser Run limits to 60 browsers per minute and 120 concurrent instances.
- Quick Action response times improved by more than 50 %.
- Migrated to Cloudflare Containers with Durable Object‑enabled Containers, replacing BISO.
- Implemented regional pre‑warmed DO‑backed browser pools to reduce latency between DO and container.
- Switched container state storage from Workers KV to D1 with transactional updates.
- Added batch writes of 100 rows via Cloudflare Queues with a 1 s timeout.
- Achieved a P95 batch write latency of 0.1 ms.
- Configured fallback to a backup region when the primary queue lags.