chore(ci): more stable spot request (#6212)

Coincidentally doubles the amount of memory we have available, but costs
seemed alright

.github/workflows/ci.yml

@@ -25,9 +25,8 @@ jobs:
       ebs_cache_size_gb: 256
       runner_concurrency: 20
       subaction: ${{ inputs.runner_action || 'start' }}
-      # This instance list has a spot placement score of 9/10. Note we used to just use m6a.32xlarge, which had a score of 1!
-      # https://us-east-2.console.aws.amazon.com/ec2/home?region=us-east-2#SpotPlacementScore:
-      ec2_instance_type: i4i.32xlarge m6a.32xlarge m6i.32xlarge m6id.32xlarge m6idn.32xlarge m6in.32xlarge m7a.32xlarge r6a.32xlarge r6i.32xlarge r6id.32xlarge
+      # Use instance types with low interruption rate in us-east-2 https://aws.amazon.com/ec2/spot/instance-advisor/
+      ec2_instance_type: r6in.32xlarge r6a.32xlarge i4i.32xlarge r7iz.32xlarge
       ec2_ami_id: ami-04d8422a9ba4de80f
       ec2_spot_instance_strategy: BestEffort
       ec2_instance_ttl: 40 # refreshed by jobs

.github/workflows/start-spot.yml

@@ -18,7 +18,8 @@ jobs:
       ebs_cache_size_gb: 256
       runner_concurrency: 20
       subaction: ${{ inputs.action }}
-      ec2_instance_type: m6a.32xlarge
+      # Use instance types with low interruption rate in us-east-2 https://aws.amazon.com/ec2/spot/instance-advisor/
+      ec2_instance_type: r6in.32xlarge r6a.32xlarge i4i.32xlarge r7iz.32xlarge
       ec2_ami_id: ami-04d8422a9ba4de80f
       ec2_instance_ttl: 40 # refreshed by jobs
     secrets: inherit