chore(ci): spot capacity and reaping (#6561)

AztecProtocol · May 21, 2024 · 8c639b5 · 8c639b5
1 parent 797115b
commit 8c639b5
Show file tree

Hide file tree

Showing 6 changed files with 6 additions and 6 deletions.
diff --git a/.github/ensure-builder/action.yml b/.github/ensure-builder/action.yml
@@ -24,7 +24,7 @@ runs:
         echo "spot_strategy=BestEffort" >> $GITHUB_OUTPUT
         echo "runner_label=$USERNAME-$runner_type" >> $GITHUB_OUTPUT
         if [[ $TYPE == builder-x86 ]]; then
-          # 128-core x86 instance types with least evictions
+          # 128-core x86 instance types, aws chooses for us based on capacity
           echo "instance_type=i4i.32xlarge m6a.32xlarge m6i.32xlarge m6id.32xlarge m6idn.32xlarge m6in.32xlarge m7a.32xlarge r6a.32xlarge r6i.32xlarge r6id.32xlarge r6in.32xlarge r7iz.32xlarge" >> $GITHUB_OUTPUT
           echo "ami_id=ami-04d8422a9ba4de80f" >> $GITHUB_OUTPUT
           echo "ebs_cache_size=256" >> $GITHUB_OUTPUT

diff --git a/.github/ensure-tester/action.yml b/.github/ensure-tester/action.yml
@@ -61,7 +61,7 @@ runs:
         ec2_subnet_id: subnet-4cfabd25
         ec2_security_group_id: sg-0ccd4e5df0dcca0c9
         ec2_key_name: "build-instance"
-        ec2_instance_tags: '[{"Key": "Keep-Alive", "Value": "true"}]'
+        ec2_instance_tags: '[]'
 
     # Set up a context for this run
     - name: Copy Repo to Tester

diff --git a/.github/spot-runner-action/dist/index.js b/.github/spot-runner-action/dist/index.js
@@ -327,7 +327,7 @@ class Ec2Instance {
                 LaunchTemplateConfigs: [fleetLaunchConfig],
                 ClientToken: clientToken,
                 SpotOptions: {
-                    AllocationStrategy: "capacity-optimized",
+                    AllocationStrategy: "price-capacity-optimized",
                 },
                 TargetCapacitySpecification: {
                     TotalTargetCapacity: 1,

diff --git a/.github/spot-runner-action/src/ec2.ts b/.github/spot-runner-action/src/ec2.ts
@@ -246,7 +246,7 @@ export class Ec2Instance {
       LaunchTemplateConfigs: [fleetLaunchConfig],
       ClientToken: clientToken,
       SpotOptions: {
-        AllocationStrategy: "capacity-optimized",
+        AllocationStrategy: "price-capacity-optimized",
       },
       TargetCapacitySpecification: {
         TotalTargetCapacity: 1,

diff --git a/scripts/run_on_builder b/scripts/run_on_builder
@@ -4,4 +4,4 @@ set -eu
 # Enter the repo root
 cd "$(dirname "$0")/.."
 
-ssh -o ControlMaster=auto -o ControlPath=~/.ssh_mux_%h_%p_%r -o ControlPersist=10m -o TCPKeepAlive=no -o ServerAliveCountMax=20 -o ServerAliveInterval=30 -o StrictHostKeyChecking=no -i "$BUILDER_SPOT_KEY" ubuntu@"$BUILDER_SPOT_IP" "$@"
+ssh -o ControlMaster=auto -o ControlPath=~/.ssh_mux_%h_%p_%r -o ControlPersist=30s -o TCPKeepAlive=no -o ServerAliveCountMax=5 -o ServerAliveInterval=30 -o StrictHostKeyChecking=no -i "$BUILDER_SPOT_KEY" ubuntu@"$BUILDER_SPOT_IP" "$@"
diff --git a/scripts/run_on_tester b/scripts/run_on_tester
@@ -4,4 +4,4 @@ set -eu
 # Enter the repo root
 cd "$(dirname "$0")/.."
 
-ssh -o ControlMaster=auto -o ControlPath=~/.ssh_mux_%h_%p_%r -o ControlPersist=10m -o TCPKeepAlive=no -o ServerAliveCountMax=20 -o ServerAliveInterval=30 -o StrictHostKeyChecking=no -i "$SPOT_KEY" ubuntu@"$SPOT_IP" "$@"
+ssh -o ControlMaster=auto -o ControlPath=~/.ssh_mux_%h_%p_%r -o ControlPersist=30s -o TCPKeepAlive=no -o ServerAliveCountMax=5 -o ServerAliveInterval=30 -o StrictHostKeyChecking=no -i "$SPOT_KEY" ubuntu@"$SPOT_IP" "$@"