[TEST1]

catalogs/catalog-hadoop/build.gradle.kts

@@ -102,7 +102,7 @@ tasks.test {
 
     doFirst {
       environment("GRAVITINO_CI_HIVE_DOCKER_IMAGE", "datastrato/gravitino-ci-hive:0.1.12")
-      environment("GRAVITINO_CI_KERBEROS_HIVE_DOCKER_IMAGE", "datastrato/gravitino-ci-kerberos-hive:0.1.0")
+      environment("GRAVITINO_CI_KERBEROS_HIVE_DOCKER_IMAGE", "datastrato/gravitino-ci-kerberos-hive:0.1.2")
     }
 
     val init = project.extra.get("initIntegrationTest") as (Test) -> Unit

catalogs/catalog-hive/build.gradle.kts

@@ -165,7 +165,7 @@ tasks.test {
 
     doFirst {
       environment("GRAVITINO_CI_HIVE_DOCKER_IMAGE", "datastrato/gravitino-ci-hive:0.1.12")
-      environment("GRAVITINO_CI_KERBEROS_HIVE_DOCKER_IMAGE", "datastrato/gravitino-ci-kerberos-hive:0.1.1")
+      environment("GRAVITINO_CI_KERBEROS_HIVE_DOCKER_IMAGE", "datastrato/gravitino-ci-kerberos-hive:0.1.2")
     }
 
     val init = project.extra.get("initIntegrationTest") as (Test) -> Unit

dev/docker/kerberos-hive/start.sh

@@ -7,8 +7,10 @@
 # start ssh
 HOSTNAME=`hostname`
 service ssh start
-ssh-keyscan localhost > /root/.ssh/known_hosts
+ssh-keyscan ${HOSTNAME} >> /root/.ssh/known_hosts
+ssh-keyscan localhost >> /root/.ssh/known_hosts
 ssh-keyscan 0.0.0.0 >> /root/.ssh/known_hosts
+ssh-keyscan 127.0.0.1 >> /root/.ssh/known_hosts
 
 # init the Kerberos database
 echo -e "${PASS}\n${PASS}" | kdb5_util create -s
@@ -65,6 +67,37 @@ ${HADOOP_HOME}/sbin/hadoop-daemon.sh start namenode
 
 echo "Starting DataNode..."
 ${HADOOP_HOME}/sbin/start-secure-dns.sh
+sleep 5
+
+# Check if the DataNode is running
+ps -ef | grep DataNode | grep -v "color=auto"
+if [[ $? -ne 0 ]]; then
+  echo "DataNode failed to start, please check the logs"
+  ehco "HDFS DataNode log start----------------------------"
+  cat ${HADOOP_HOME}/bin/logs/hadoop-root-datanode-*.log
+  exit 1
+fi
+
+retry_times=0
+ready=0
+while [[ ${retry_times} -lt 10 ]]; do
+  hdfs_ready=$(hdfs dfsadmin -report | grep "Live datanodes" | awk '{print $3}')
+  if [[ ${hdfs_ready} == "(1):" ]]; then
+    echo "HDFS is ready, retry_times = ${retry_times}"
+    let "ready=0"
+    break
+  fi
+  sleep 10
+  retry_times=$((retry_times+1))
+done
+
+if [[ ${ready} -ne 0 ]]; then
+  echo "HDFS is not ready"
+  ehco "HDFS DataNode log start---------------------------"
+  cat ${HADOOP_HOME}/bin/logs/hadoop-root-datanode-*.log
+  exit 1
+fi
+
 
 # start mysql and create databases/users for hive
 chown -R mysql:mysql /var/lib/mysql

docs/docker-image-details.md

@@ -89,6 +89,9 @@ You can use these kinds of Docker images to facilitate integration testing of al
 You can use this kind of image to test the catalog of Apache Hive with kerberos enable
 
 Changelog
+- gravitino-ci-kerberos-hive:0.1.2
+  - Add `${HOSTNAME} >> /root/.ssh/known_hosts` to the startup script.
+  - Add check for the status of DataNode, if the DataNode is not running or ready within 100s, the container will exit.
 
 - gravitino-ci-kerberos-hive:0.1.1
   - Add a principal for Gravitino web server named 'HTTP/localhost@HADOOPKRB'.