From d321e42d9e7720ba9449e41d7cf50d13c1f170ac Mon Sep 17 00:00:00 2001
From: Max Gautier <mg@max.gautier.name>
Date: Mon, 13 May 2024 10:20:48 +0000
Subject: [PATCH] CI: Factorize ansible-playbook flags (#11173)

We have inconsistent sets of options passed to the playbooks during our
CI runs.

Don't run ansible-playbook directly, instead factorize the execution in
a bash function using all the common flags.
Also remove various ENABLE_* variables and instead directly test for the
relevant conditions at execution time, as this makes it more obvious and
does not force one to go back and forth in the script.
---
 tests/scripts/testcases_run.sh | 101 +++++++++++++++++----------------
 1 file changed, 52 insertions(+), 49 deletions(-)

diff --git a/tests/scripts/testcases_run.sh b/tests/scripts/testcases_run.sh
index 7cd4671a7..ec22a4dcb 100755
--- a/tests/scripts/testcases_run.sh
+++ b/tests/scripts/testcases_run.sh
@@ -47,21 +47,6 @@ if [[ "$CI_JOB_NAME" =~ "ubuntu" ]]; then
   CI_TEST_ADDITIONAL_VARS="-e ansible_python_interpreter=/usr/bin/python3"
 fi
 
-ENABLE_020_TEST="true"
-ENABLE_030_TEST="true"
-ENABLE_040_TEST="true"
-if [[ "$CI_JOB_NAME" =~ "macvlan" ]]; then
-  ENABLE_020_TEST="false"
-  ENABLE_030_TEST="false"
-  ENABLE_040_TEST="false"
-fi
-
-if [[ "$CI_JOB_NAME" =~ "hardening" ]]; then
-  # TODO: We need to remove this condition by finding alternative container
-  # image instead of netchecker which doesn't work at hardening environments.
-  ENABLE_040_TEST="false"
-fi
-
 # Check out latest tag if testing upgrade
 test "${UPGRADE_TEST}" != "false" && git fetch --all && git checkout "$KUBESPRAY_VERSION"
 # Checkout the CI vars file so it is available
@@ -69,21 +54,41 @@ test "${UPGRADE_TEST}" != "false" && git checkout "${CI_COMMIT_SHA}" tests/files
 test "${UPGRADE_TEST}" != "false" && git checkout "${CI_COMMIT_SHA}" ${CI_TEST_REGISTRY_MIRROR}
 test "${UPGRADE_TEST}" != "false" && git checkout "${CI_COMMIT_SHA}" ${CI_TEST_SETTING}
 
+
+run_playbook () {
+playbook=$1
+shift
+# We can set --limit here and still pass it as supplemental args because `--limit`  is a 'last one wins' option
+ansible-playbook --limit "all:!fake_hosts" \
+     $ANSIBLE_LOG_LEVEL \
+    -e @${CI_TEST_SETTING} \
+    -e @${CI_TEST_REGISTRY_MIRROR} \
+    -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} \
+    -e local_release_dir=${PWD}/downloads \
+    "$@" \
+    ${playbook}
+}
+
 # Create cluster
-ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} -e @${CI_TEST_VARS} -e local_release_dir=${PWD}/downloads --limit "all:!fake_hosts" cluster.yml
+run_playbook cluster.yml
 
 # Repeat deployment if testing upgrade
-if [ "${UPGRADE_TEST}" != "false" ]; then
-  test "${UPGRADE_TEST}" == "basic" && PLAYBOOK="cluster.yml"
-  test "${UPGRADE_TEST}" == "graceful" && PLAYBOOK="upgrade-cluster.yml"
-  git checkout "${CI_COMMIT_SHA}"
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} -e @${CI_TEST_VARS} -e local_release_dir=${PWD}/downloads --limit "all:!fake_hosts" $PLAYBOOK
-fi
+case "${UPGRADE_TEST}" in
+
+    "basic")
+        run_playbook cluster.yml
+        ;;
+    "graceful")
+        run_playbook upgrade-cluster.yml
+        ;;
+    *)
+        ;;
+esac
 
 # Test control plane recovery
 if [ "${RECOVER_CONTROL_PLANE_TEST}" != "false" ]; then
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} -e @${CI_TEST_VARS} -e local_release_dir=${PWD}/downloads --limit "${RECOVER_CONTROL_PLANE_TEST_GROUPS}:!fake_hosts" -e reset_confirmation=yes reset.yml
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} -e @${CI_TEST_VARS} -e local_release_dir=${PWD}/downloads -e etcd_retries=10 --limit "etcd:kube_control_plane:!fake_hosts" recover-control-plane.yml
+    run_playbook reset.yml --limit "${RECOVER_CONTROL_PLANE_TEST_GROUPS}:!fake_hosts" -e reset_confirmation=yes
+    run_playbook recover-control-plane.yml -e etcd_retries=10 --limit "etcd:kube_control_plane:!fake_hosts"
 fi
 
 # Test collection build and install by installing our collection, emptying our repository, adding
@@ -114,61 +119,59 @@ EOF
   # Write remove-node.yml
 cat > remove-node.yml <<EOF
 - name: Remove node from Kubernetes
-  ansible.builtin.import_playbook: kubernetes_sigs.kubespray.remote-node
+  ansible.builtin.import_playbook: kubernetes_sigs.kubespray.remove_node
 EOF
 
 fi
-
 # Tests Cases
 ## Test Master API
-ansible-playbook --limit "all:!fake_hosts" -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} tests/testcases/010_check-apiserver.yml $ANSIBLE_LOG_LEVEL
+run_playbook tests/testcases/010_check-apiserver.yml
+run_playbook tests/testcases/015_check-nodes-ready.yml
 
 ## Test that all nodes are Ready
-ansible-playbook --limit "all:!fake_hosts" -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} tests/testcases/015_check-nodes-ready.yml $ANSIBLE_LOG_LEVEL
 
-## Test that all pods are Running
-if [ "${ENABLE_020_TEST}" = "true" ]; then
-ansible-playbook --limit "all:!fake_hosts" -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} tests/testcases/020_check-pods-running.yml $ANSIBLE_LOG_LEVEL
-fi
-
-## Test pod creation and ping between them
-if [ "${ENABLE_030_TEST}" = "true" ]; then
-ansible-playbook --limit "all:!fake_hosts" -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} tests/testcases/030_check-network.yml $ANSIBLE_LOG_LEVEL
-fi
-
-## Advanced DNS checks
-if [ "${ENABLE_040_TEST}" = "true" ]; then
-  ansible-playbook --limit "all:!fake_hosts" -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} tests/testcases/040_check-network-adv.yml $ANSIBLE_LOG_LEVEL
+if [[ ! ( "$CI_JOB_NAME" =~ "macvlan" ) ]]; then
+    run_playbook tests/testcases/020_check-pods-running.yml
+    run_playbook tests/testcases/030_check-network.yml
+    if [[ ! ( "$CI_JOB_NAME" =~ "hardening" ) ]]; then
+      # TODO: We need to remove this condition by finding alternative container
+      # image instead of netchecker which doesn't work at hardening environments.
+      run_playbook tests/testcases/040_check-network-adv.yml
+    fi
 fi
 
 ## Kubernetes conformance tests
-ansible-playbook -i ${ANSIBLE_INVENTORY} -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} --limit "all:!fake_hosts" tests/testcases/100_check-k8s-conformance.yml $ANSIBLE_LOG_LEVEL
+run_playbook tests/testcases/100_check-k8s-conformance.yml
 
 if [ "${IDEMPOT_CHECK}" = "true" ]; then
   ## Idempotency checks 1/5 (repeat deployment)
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} ${CI_TEST_ADDITIONAL_VARS} -e @${CI_TEST_VARS} -e local_release_dir=${PWD}/downloads --limit "all:!fake_hosts" cluster.yml
+  run_playbook cluster.yml
 
   ## Idempotency checks 2/5 (Advanced DNS checks)
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} --limit "all:!fake_hosts" tests/testcases/040_check-network-adv.yml
+  if [[ ! ( "$CI_JOB_NAME" =~ "hardening" ) ]]; then
+      run_playbook tests/testcases/040_check-network-adv.yml
+  fi
 
   if [ "${RESET_CHECK}" = "true" ]; then
     ## Idempotency checks 3/5 (reset deployment)
-    ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR}  -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} -e reset_confirmation=yes --limit "all:!fake_hosts" reset.yml
+    run_playbook reset.yml -e reset_confirmation=yes
 
     ## Idempotency checks 4/5 (redeploy after reset)
-    ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} -e local_release_dir=${PWD}/downloads --limit "all:!fake_hosts" cluster.yml
+    run_playbook cluster.yml
 
     ## Idempotency checks 5/5 (Advanced DNS checks)
-    ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR} -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} --limit "all:!fake_hosts" tests/testcases/040_check-network-adv.yml
+    if [[ ! ( "$CI_JOB_NAME" =~ "hardening" ) ]]; then
+        run_playbook tests/testcases/040_check-network-adv.yml
+    fi
   fi
 fi
 
 # Test node removal procedure
 if [ "${REMOVE_NODE_CHECK}" = "true" ]; then
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR}  -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} -e skip_confirmation=yes -e node=${REMOVE_NODE_NAME} --limit "all:!fake_hosts" remove-node.yml
+  run_playbook remove-node.yml -e skip_confirmation=yes -e node=${REMOVE_NODE_NAME}
 fi
 
 # Clean up at the end, this is to allow stage1 tests to include cleanup test
 if [ "${RESET_CHECK}" = "true" ]; then
-  ansible-playbook ${ANSIBLE_LOG_LEVEL} -e @${CI_TEST_SETTING} -e @${CI_TEST_REGISTRY_MIRROR}  -e @${CI_TEST_VARS} ${CI_TEST_ADDITIONAL_VARS} -e reset_confirmation=yes --limit "all:!fake_hosts" reset.yml
+  run_playbook reset.yml -e reset_confirmation=yes
 fi