Your Workers Memory Is Now Right There in the Dashboard!
Hi there, Shiichan here! Today I've got a nice little update for everyone running Workers and Durable Objects: you can now see your memory usage right in the dashboard!
Cloudflare ChangelogWhat was announced?
This one comes from Cloudflare's Changelog. Memory usage for Workers and Durable Objects is now visible in the dashboard! A new Memory Usage chart shows up in the Workers Metrics tab, displaying how much memory your V8 isolate is using across the P50, P90, P99, and P999 percentiles. Memory is measured at every invocation.
The story so far
Until now, it was hard to see from the outside how much memory your Worker was actually using. Was memory creeping up because of a leak, or did you still have plenty of headroom? You often wouldn't notice until you hit the limit.
What changes
Now you can watch memory gradually climb, which helps you catch leaks early. Line it up with deployment markers and you can pin down which release changed your memory profile. You can also learn your baseline memory footprint before you ever get close to the limit.
Dive Deep
Workers isolates have a 128 MB per-isolate limit, and the Memory Usage chart shows how close you are to that ceiling.
For Durable Objects, the chart reflects in-memory state like class properties, caches, and WebSocket connections. That state sticks around until the DO is hibernated or evicted, so anything you want to keep for the long haul should live in storage. In the dashboard you can also filter by DO ID or name to zoom in on a specific object.
To grab this data programmatically, use the GraphQL Analytics API with the workersInvocationsAdaptive dataset. The fields quantiles.memoryUsageBytesP50 through quantiles.memoryUsageBytesP999 return percentile values in bytes. And if you want to dig deeper into the cause, you can also profile memory with DevTools.
Wrap-up
- A new Memory Usage chart in the Workers Metrics tab shows V8 isolate memory at P50, P90, P99, and P999
- Measured per invocation, great for catching leaks early and correlating with deployments
- Isolates cap at 128 MB; for Durable Objects it tracks in-memory state (kept until hibernation or eviction)
- You can pull the same numbers programmatically via the GraphQL Analytics API's workersInvocationsAdaptive dataset
- A small but handy update for anyone who wants fewer memory mysteries when running Workers and Durable Objects!