When testing cloudwatch metric collection for EC2 instances in our preprod environment, which has of the order of 10s of instances, we see metric collections taking longer than the default 5m collection period. This leads to fragmented collections and dashboards which look like they are missing data.

Investigations into the reasons why are ongoing within the cloud monitoring team. But the expected outcome is some optimizations to improve the performance of the metric collection code for cloudwatch. This is inextricably linked to elastic/integrations#2913, but we are approaching this from the performance side only, not cost. If issues end up improving both issues, that's great.