fix(chart): node preStop - refresh node status in loop

Signed-off-by: Viet Nguyen Duc <[email protected]>

Author: VietND96

README.md

@@ -1,6 +1,10 @@
-![Build & test](https://github.com/SeleniumHQ/docker-selenium/workflows/Build%20&%20test/badge.svg?branch=trunk)
-![Deployments](https://github.com/SeleniumHQ/docker-selenium/workflows/Deploys/badge.svg)
-![Helm Charts](https://github.com/SeleniumHQ/docker-selenium/workflows/Lint%20and%20Test%20Helm%20Charts/badge.svg)
+[![Build & test](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/build-test.yml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/build-test.yml)
+[![Test Docker Selenium](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/test-video.yml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/test-video.yml)
+[![Test Helm Charts](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/helm-chart-test.yml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/helm-chart-test.yml)
+[![Deploys](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/deploy.yml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/deploy.yml)
+[![Release Charts](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/helm-chart-release.yml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/helm-chart-release.yml)
+[![Nightly](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/nightly.yaml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/nightly.yaml)
+[![Update Dev/Beta Browser Images](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/update-dev-beta-browser-images.yml/badge.svg)](https://github.com/SeleniumHQ/docker-selenium/actions/workflows/update-dev-beta-browser-images.yml)
 
 # Docker images for the Selenium Grid Server
 

charts/selenium-grid/configs/node/nodePreStop.sh

@@ -3,12 +3,14 @@
 probe_name="lifecycle.${1:-"preStop"}"
 
 max_time=3
+retry_time=5
 
 ID=$(echo $RANDOM)
 tmp_node_file="/tmp/nodeProbe${ID}"
 
 function on_exit() {
   rm -rf ${tmp_node_file}
+  echo "$(date +%FT%T%Z) [${probe_name}] - Exiting Node preStop..."
 }
 trap on_exit EXIT
 
@@ -89,8 +91,12 @@ if curl -m ${max_time} -sfk ${SE_SERVER_PROTOCOL}://127.0.0.1:${SE_NODE_PORT}/st
     fi
     signal_node_to_drain
     # Wait for the current session to be finished if any
-    while curl -m ${max_time} -sfk ${SE_SERVER_PROTOCOL}://127.0.0.1:${SE_NODE_PORT}/status -o ${tmp_node_file};
-    do
+    while true; do
+      # Attempt the cURL request and capture the exit status
+      endpoint_http_code=$(curl --retry ${retry_time} -m ${max_time} -sfk ${SE_SERVER_PROTOCOL}://127.0.0.1:${SE_NODE_PORT}/status -o ${tmp_node_file} -w "%{http_code}")
+      endpoint_status=$?
+      echo "$(date +%FT%T%Z) [${probe_name}] - Fetch the Node status via cURL with exit status: ${endpoint_status}, HTTP code: ${endpoint_http_code}"
+
       SLOT_HAS_SESSION=$(jq -e ".value.node.slots[]|select(.session != null).id.id" ${tmp_node_file} | tr -d '"' || "")
       if [ -z "${SLOT_HAS_SESSION}" ]; then
         echo "$(date +%FT%T%Z) [${probe_name}] - There is no session running. Node is ready to be terminated."
@@ -99,7 +105,13 @@ if curl -m ${max_time} -sfk ${SE_SERVER_PROTOCOL}://127.0.0.1:${SE_NODE_PORT}/st
         exit 0
       else
         echo "$(date +%FT%T%Z) [${probe_name}] - Node preStop is waiting for current session on slot ${SLOT_HAS_SESSION} to be finished. Node details: message: $(jq -r '.value.message' ${tmp_node_file} || "unknown"), availability: $(jq -r '.value.node.availability' ${tmp_node_file} || "unknown")"
-        sleep 1;
+        sleep 2;
+      fi
+
+      # If the cURL command failed, break the loop
+      if [ ${endpoint_status} -ne 0 ] || [ "${endpoint_http_code}" != "200" ]; then
+        echo "$(date +%FT%T%Z) [${probe_name}] - Node endpoint returned status ${endpoint_http_code:-"exit ${endpoint_status}"}, probably Node draining complete!"
+        break
       fi
     done
 else

tests/bootstrap.sh

@@ -9,6 +9,7 @@ fi
 
 python -m pip install selenium==4.19.0 \
                       docker===7.0.0 \
+                      chardet \
                       | grep -v 'Requirement already satisfied'
 
 if [ "${SELENIUM_GRID_PROTOCOL}" = "https" ]; then

tests/charts/ci/base-auth-ingress-values.yaml

@@ -1,6 +1,7 @@
 global:
   seleniumGrid:
     logLevel: INFO
+    stdoutProbeLog: true
 
 ingress:
   className: nginx