test: Improvements to wait_for_pods function (kubeflow#11162)

Signed-off-by: hbelmiro <helber.belmiro@gmail.com>

.github/workflows/backend.yml

@@ -34,17 +34,17 @@ jobs:
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
-      - name: Create KFP cluster
-        uses: ./.github/actions/kfp-tekton-cluster
-      - name: Set up Python 3.10
+      - name: Set up Python 3.9
         uses: actions/setup-python@v4
         with:
-          python-version: '3.10'
+          python-version: '3.9'
       - name: Install sdk
         run: |
           python3 -m venv .venv
           . .venv/bin/activate
           pip install -e sdk/python
+      - name: Create KFP cluster
+        uses: ./.github/actions/kfp-tekton-cluster
       - name: "flip coin test"
         run: |
           . .venv/bin/activate

.github/workflows/e2e-test.yml

@@ -23,6 +23,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
@@ -46,6 +51,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
@@ -69,6 +79,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
@@ -92,6 +107,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
@@ -115,6 +135,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
@@ -144,6 +169,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 

.github/workflows/kfp-kubernetes-execution-tests.yml

@@ -7,6 +7,7 @@ on:
   pull_request:
     paths:
       - '.github/workflows/kfp-kubernetes-execution-tests.yml'
+      - 'scripts/deploy/github/**'
       - 'sdk/python/**'
       - 'api/v2alpha1/**'
       - 'kubernetes_platform/**'
@@ -18,17 +19,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: '3.9'
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
       - name: Forward API port
         run: ./scripts/deploy/github/forward-port.sh "kubeflow" "ml-pipeline" 8888 8888
 
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: '3.9'
-
       - name: apt-get update
         run: sudo apt-get update
 

.github/workflows/kfp-samples.yml

@@ -6,6 +6,7 @@ on:
       - master
   pull_request:
     paths:
+      - 'scripts/deploy/github/**'
       - 'samples/**'
       - 'backend/src/v2/**'
       - '.github/workflows/kfp-samples.yml'
@@ -21,7 +22,7 @@ jobs:
     - name: Set up Python
       uses: actions/setup-python@v2
       with:
-        python-version: 3.8
+        python-version: 3.9
 
     - name: Create KFP cluster
       uses: ./.github/actions/kfp-cluster

.github/workflows/kubeflow-pipelines-integration-v2.yml

@@ -7,6 +7,7 @@ on:
   pull_request:
     paths:
       - '.github/workflows/kubeflow-pipelines-integration-v2.yml'
+      - 'scripts/deploy/github/**'
       - 'samples'
       - 'core'
       - 'backend'

.github/workflows/periodic.yml

@@ -10,6 +10,10 @@ jobs:
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
       - name: Port forward kfp apiserver

.github/workflows/sdk-execution.yml

@@ -7,6 +7,7 @@ on:
   pull_request:
     paths:
       - '.github/workflows/sdk-execution.yml'
+      - 'scripts/deploy/github/**'
       - 'sdk/python/**'
       - 'api/v2alpha1/**'
 
@@ -17,17 +18,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 
       - name: Forward API port
         run: ./scripts/deploy/github/forward-port.sh "kubeflow" "ml-pipeline" 8888 8888
 
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: 3.8
-
       - name: apt-get update
         run: sudo apt-get update
 

.github/workflows/upgrade-test.yml

@@ -7,6 +7,7 @@ on:
   pull_request:
     paths:
       - '.github/workflows/upgrade-test.yml'
+      - 'scripts/deploy/github/**'
       - 'backend/**'
       - 'manifests/kustomize/**'
 
@@ -17,6 +18,11 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+
       - name: Create KFP cluster
         uses: ./.github/actions/kfp-cluster
 

scripts/deploy/github/deploy-kfp-tekton.sh

@@ -40,16 +40,14 @@ then
   exit 1
 fi
 
-# Check if all pods are running - allow 20 retries (10 minutes)
-wait_for_pods kubeflow 40 30 || EXIT_CODE=$?
+# Check if all pods are running - (10 minutes)
+wait_for_pods || EXIT_CODE=$?
 if [[ $EXIT_CODE -ne 0 ]]
 then
   echo "Deploy unsuccessful. Not all pods running."
   exit 1
 fi
 
-echo "List Kubeflow: "
-kubectl get pod -n kubeflow
 collect_artifacts kubeflow
 
 echo "List Tekton control plane: "

scripts/deploy/github/deploy-kfp.sh

@@ -41,16 +41,14 @@ then
   exit 1
 fi
 
-# Check if all pods are running - allow 20 retries (10 minutes)
-wait_for_pods kubeflow 40 30 || EXIT_CODE=$?
+# Check if all pods are running - (10 minutes)
+wait_for_pods || EXIT_CODE=$?
 if [[ $EXIT_CODE -ne 0 ]]
 then
   echo "Deploy unsuccessful. Not all pods running."
   exit 1
 fi
 
-echo "List Kubeflow: "
-kubectl get pod -n kubeflow
 collect_artifacts kubeflow
 
 echo "Finished KFP deployment."

scripts/deploy/github/helper-functions.sh

@@ -56,57 +56,9 @@ wait_for_namespace () {
 }
 
 wait_for_pods () {
-    if [[ $# -ne 3 ]]
-    then
-        echo "Usage: wait_for_pods namespace max_retries sleep_time"
-        return 1
-    fi 
-
-    local namespace=$1
-    local max_retries=$2
-    local sleep_time=$3
-
-    local i=0
-
-    while [[ $i -lt $max_retries ]]
-    do
-        local pods
-        local statuses
-        local num_pods
-        local num_running
-        pods=$(kubectl get pod -n "$namespace")
-        # echo "$pods"
-        # kubectl get pvc -n "$namespace"
-
-        if [[ -z $pods ]]
-        then
-            echo "no pod is up yet"
-        else
-            # Using quotations around variables to keep column format in echo
-            # Remove 1st line (header line) -> trim whitespace -> cut statuses column (3rd column)
-            # Might be overkill to parse down to specific columns :).
-            statuses=$(echo "$pods" | tail -n +2 | tr -s ' '  | cut -d ' ' -f 3)
-            num_pods=$(echo "$statuses" | wc -l | xargs)
-            num_running=$(echo "$statuses" | grep -ow "Running\|Completed" | wc -l | xargs)
-
-            local msg="${num_running}/${num_pods} pods running in \"${namespace}\"."
-
-            if [[ $num_running -ne $num_pods ]]
-            then
-                # for debugging
-                # kubectl get pod -n "$namespace" | grep '0/1' | awk '{print $1}' | xargs kubectl describe pod -n "$namespace"
-                echo "$msg Checking again in ${sleep_time}s."
-            else
-                echo "$msg"
-                return 0
-            fi
-        fi
-
-        sleep "$sleep_time"
-        i=$((i+1))
-    done
-
-    return 1
+    C_DIR="${BASH_SOURCE%/*}"
+    pip install -r "${C_DIR}"/../../../sdk/python/requirements.txt
+    python "${C_DIR}"/kfp-readiness/wait_for_pods.py
 }
 
 deploy_with_retries () {

scripts/deploy/github/kfp-readiness/wait_for_pods.py

@@ -0,0 +1,76 @@
+import logging
+import time
+import urllib3
+import sys
+from kubernetes import client, config
+
+urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
+
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
+
+namespace = 'kubeflow'
+
+config.load_kube_config()
+v1 = client.CoreV1Api()
+
+
+def get_pod_statuses():
+    pods = v1.list_namespaced_pod(namespace=namespace)
+    statuses = {}
+    for pod in pods.items:
+        pod_name = pod.metadata.name
+        pod_status = pod.status.phase
+        container_statuses = pod.status.container_statuses or []
+        ready_containers = sum(1 for status in container_statuses if status.ready)
+        total_containers = len(container_statuses)
+        statuses[pod_name] = (pod_status, ready_containers, total_containers)
+    return statuses
+
+
+def all_pods_ready(statuses):
+    return all(pod_status == 'Running' and ready == total
+               for pod_status, ready, total in statuses.values())
+
+
+def check_pods(calm_time=10, timeout=600, retries_after_ready=5):
+    start_time = time.time()
+    stable_count = 0
+    previous_statuses = {}
+
+    while time.time() - start_time < timeout:
+        current_statuses = get_pod_statuses()
+
+        logging.info("Checking pod statuses...")
+        for pod_name, (pod_status, ready, total) in current_statuses.items():
+            logging.info(f"Pod {pod_name} - Status: {pod_status}, Ready: {ready}/{total}")
+
+        if current_statuses == previous_statuses:
+            if all_pods_ready(current_statuses):
+                stable_count += 1
+                if stable_count >= retries_after_ready:
+                    logging.info("All pods are calm and fully ready.")
+                    break
+                else:
+                    logging.info(
+                        f"Pods are calm but have only been stable for {stable_count}/{retries_after_ready} retries.")
+            else:
+                stable_count = 0
+        else:
+            stable_count = 0
+
+        previous_statuses = current_statuses
+        logging.info(f"Pods are still stabilizing. Retrying in {calm_time} seconds...")
+        time.sleep(calm_time)
+    else:
+        raise Exception("Pods did not stabilize within the timeout period.")
+
+    logging.info("Final pod statuses:")
+    for pod_name, (pod_status, ready, total) in previous_statuses.items():
+        if pod_status == 'Running' and ready == total:
+            logging.info(f"Pod {pod_name} is fully ready ({ready}/{total})")
+        else:
+            logging.info(f"Pod {pod_name} is not ready (Status: {pod_status}, Ready: {ready}/{total})")
+
+
+if __name__ == "__main__":
+    check_pods()