Overview

Our XL e2e nightly job failed last night because Amazon did not have any g6e.48xlarge instances available at the time:

Recommended Solution

We should keep our existing workflow logic to launch a g6e.48xlarge instance in our desired subnet on us-east-2. However, if that instance fails to launch due to capacity constraints, we should then fall back and select another subnet.

We appear to have 3 subnets available to us within the us-east-2 region, and we can try the remaining two regions we have not touched.

Worst case scenario, we can create a follow-up issue to investigate the possibility of trying other regions too (as fallbacks). We just must be aware that different regions tend to have different pricing models.