puma integration not aggregating metrics as expected for ECS/Fargate cluster

[stephenwiebe]

I've created a Datadog agent image with the puma integration, and have been trying it out in our Fargate cluster of auto-scaling Puma instances. In our cluster, each Puma instance runs 5 workers, and we scale between 4 and 14 Puma instances. Each instance has an attached Datadog agent with the Puma integration installed. The Datadog puma.* metrics show up, but they don't seem to be aggregated correctly at all. For puma.running for example, which should vary between 20 (5*4) and 70 (5*14), the sum (almost) always seems to be 15. If I show a "weighted" sum, then things appear a little more correct as long as I fine-tune the metrics to set the update interval to 15 seconds (the default setting on the agent). See this screenshot for what I mean:

The other puma metrics also seem to be off in a similar way, which leads me to believe something is not working correctly in aggregating these metrics together. All Puma instances and Datadog agents are configured identically.