From c31d7e0e80dedaac2a273e1656c8cb92baf642c8 Mon Sep 17 00:00:00 2001
From: Giacomo Marciani <mgiacomo@amazon.com>
Date: Fri, 19 Dec 2025 11:50:42 -0500
Subject: [PATCH] [Test] Add retry helper for flaky InSpec controls. Use such
 retry logic to make the yum repo check more resilient to transient failures.

---
 .../test/controls/lustre_spec.rb              | 20 +++++++++++--------
 .../test/libraries/retry_helper.rb            | 15 ++++++++++++++
 2 files changed, 27 insertions(+), 8 deletions(-)
 create mode 100644 cookbooks/aws-parallelcluster-shared/test/libraries/retry_helper.rb

diff --git a/cookbooks/aws-parallelcluster-environment/test/controls/lustre_spec.rb b/cookbooks/aws-parallelcluster-environment/test/controls/lustre_spec.rb
index 0d1f5b707d..616ed9fc5a 100644
--- a/cookbooks/aws-parallelcluster-environment/test/controls/lustre_spec.rb
+++ b/cookbooks/aws-parallelcluster-environment/test/controls/lustre_spec.rb
@@ -19,10 +19,12 @@
         its('version') { should cmp >= minimal_lustre_client_version }
       end
 
-      describe yum.repo('aws-fsx') do
-        it { should exist }
-        it { should be_enabled }
-        its('baseurl') { should include 'fsx-lustre-client-repo.s3.amazonaws.com' }
+      with_retry(retries: 3, delay: 5) do
+        describe yum.repo('aws-fsx') do
+          it { should exist }
+          it { should be_enabled }
+          its('baseurl') { should include 'fsx-lustre-client-repo.s3.amazonaws.com' }
+        end
       end
     end
   end
@@ -47,10 +49,12 @@
         its('version') { should cmp >= minimal_lustre_client_version }
       end
 
-      describe yum.repo('aws-fsx') do
-        it { should exist }
-        it { should be_enabled }
-        its('baseurl') { should include 'fsx-lustre-client-repo.s3.amazonaws.com' }
+      with_retry(retries: 3, delay: 5) do
+        describe yum.repo('aws-fsx') do
+          it { should exist }
+          it { should be_enabled }
+          its('baseurl') { should include 'fsx-lustre-client-repo.s3.amazonaws.com' }
+        end
       end
     end
   end
diff --git a/cookbooks/aws-parallelcluster-shared/test/libraries/retry_helper.rb b/cookbooks/aws-parallelcluster-shared/test/libraries/retry_helper.rb
new file mode 100644
index 0000000000..618c32a550
--- /dev/null
+++ b/cookbooks/aws-parallelcluster-shared/test/libraries/retry_helper.rb
@@ -0,0 +1,15 @@
+# Helper method to retry flaky InSpec checks
+def with_retry(retries: 3, delay: 5)
+  attempts = 0
+  begin
+    attempts += 1
+    yield
+  rescue => e
+    if attempts < retries
+      sleep delay
+      retry
+    else
+      raise e
+    end
+  end
+end