We should attempt to re-commit the changes from 5bd4de9 as part of SERVER-33142. Running infrastructure_teardown.py as part of the "post" phase of the Evergreen task should hopefully be more reliable than running the teardown script.

• We had issues where the Evergreen task is aborted and because the .tfstate file wasn't written to the /data/infrastructure_provisioning/terraform/ directory causing the teardown script to do nothing and leading to EC2 instances being leaked. Note that the infrastructure_provisioning.py script first writes the .tfstate file to the working directory and then copies it over.
• We've also had issues where the /data drive is unmounted before the teardown script has actually run due to multiple Evergreen host termination jobs running concurrently.

diff --git a/etc/system_perf.yml b/etc/system_perf.yml
index 2d6aa3021db..64037979606 100644
--- a/etc/system_perf.yml
+++ b/etc/system_perf.yml
@@ -42,9 +42,7 @@ post:
         working_dir: work
         script: |
           source ./dsienv.sh
-          if [ -e /data/infrastructure_provisioning/terraform/provisioned.${cluster} ]; then
-            mark_idle.sh
-          fi
+          infrastructure_teardown.py

     - command: shell.exec
       params: