Merge pull request #1084 from msvinaykumar/deletepartitions

delete Partitions RM to MVP

design/KruizeConfiguration.md

@@ -0,0 +1,102 @@
+# Kruize Configuration
+
+The following environment variables are set using the `kubectl apply` command with the provided YAML configuration.
+
+## Logging Configuration
+
+- **LOGGING_LEVEL**
+    - Description: Sets the logging level for application logs.
+    - Value: "info"
+
+- **ROOT_LOGGING_LEVEL**
+    - Description: Sets the root logging level for the application.
+    - Value: "error"
+
+## Database Configuration
+
+- **dbdriver**
+    - Description: Database driver configuration for JDBC.
+    - Value: "jdbc:postgresql://"
+
+## Cluster Configuration
+
+- **clustertype**
+    - Description: Defines the cluster type.
+    - Value: "kubernetes"
+
+- **k8stype**
+    - Description: Kubernetes type configuration.
+    - Value: "openshift"
+
+- **authtype**
+    - Description: Authentication type for the cluster.
+    - Value: "openshift"
+
+## Monitoring Configuration
+
+- **monitoringagent**
+    - Description: Specifies the monitoring agent used.
+    - Value: "prometheus"
+
+- **monitoringservice**
+    - Description: Monitoring service configuration.
+    - Value: "prometheus-k8s"
+
+- **monitoringendpoint**
+    - Description: Monitoring endpoint configuration.
+    - Value: "prometheus-k8s"
+
+## Database Interaction Configuration
+
+- **savetodb**
+    - Description: Indicates whether to save to the database.
+    - Value: "true"
+
+## Hibernate Configuration
+
+- **hibernate_dialect**
+    - Description: Hibernate dialect configuration.
+    - Value: "org.hibernate.dialect.PostgreSQLDialect"
+
+- **hibernate_driver**
+    - Description: Hibernate database driver configuration.
+    - Value: "org.postgresql.Driver"
+
+- **hibernate_c3p0minsize**
+    - Description: Minimum size for the C3P0 connection pool.
+    - Value: "5"
+
+- **hibernate_c3p0maxsize**
+    - Description: Maximum size for the C3P0 connection pool.
+    - Value: "10"
+
+
+- **hibernate_c3p0timeout**
+    - Description: Timeout configuration for the C3P0 connection pool.
+    - Value: "300"
+
+- **hibernate_c3p0maxstatements**
+    - Description: Maximum statements configuration for the C3P0 connection pool.
+    - Value: "100"
+
+- **hibernate_hbm2ddlauto**
+    - Description: Hibernate DDL auto configuration.
+    - Value: "none"
+
+- **hibernate_showsql**
+    - Description: Enable or disable showing SQL statements in the logs.
+    - Value: "false"
+
+- **hibernate_timezone**
+    - Description: Timezone configuration for Hibernate.
+    - Value: "UTC"
+
+## Other Configuration
+
+- **deletepartitionsthreshold**
+    - Description: Threshold for deleting partitions.
+    - Value: "16"
+    - Details: The value represents the number of days, indicating the duration for which partitions belonging to Kruize
+      that are older than the specified number of days from today's date will be deleted. For example, if the value is
+      set to "16," Kruize will automatically delete partitions older than 16 days, helping manage and optimize storage
+      resources.

manifests/crc/default-db-included-installation/minikube/kruize-crc-minikube.yaml

@@ -320,4 +320,47 @@ spec:
     - name: nginx-config-volume
       configMap:
         name: nginx-config
-
+---
+apiVersion: batch/v1
+kind: CronJob
+metadata:
+  name: kruize-delete-partition-cronjob
+  namespace: monitoring
+spec:
+  schedule: "0 0 25 * *" # Run on 25th of every month at midnight
+  jobTemplate:
+    spec:
+      template:
+        spec:
+          containers:
+            - name: kruizedeletejob
+              image: kruize/autotune_operator:0.0.20.1_rm
+              imagePullPolicy: Always
+              volumeMounts:
+                - name: config-volume
+                  mountPath: /etc/config
+              command:
+                - sh
+                - -c
+                - |
+                  /home/autotune/app/target/bin/RetentionPartition
+              args: [ "" ]
+              env:
+                - name: START_AUTOTUNE
+                  value: "false"
+                - name: LOGGING_LEVEL
+                  value: "info"
+                - name: ROOT_LOGGING_LEVEL
+                  value: "error"
+                - name: DB_CONFIG_FILE
+                  value: "/etc/config/dbconfigjson"
+                - name: KRUIZE_CONFIG_FILE
+                  value: "/etc/config/kruizeconfigjson"
+                - name: deletepartitionsthreshold
+                  value: "15"
+          volumes:
+            - name: config-volume
+              configMap:
+                name: kruizeconfig
+          restartPolicy: OnFailure
+---

manifests/crc/default-db-included-installation/openshift/kruize-crc-openshift.yaml

@@ -289,6 +289,49 @@ spec:
                 name: kruizeconfig
           restartPolicy: OnFailure
 ---
+apiVersion: batch/v1
+kind: CronJob
+metadata:
+  name: kruize-delete-partition-cronjob
+  namespace: openshift-tuning
+spec:
+  schedule: "0 0 25 * *" # Run on 25th of every month at midnight
+  jobTemplate:
+    spec:
+      template:
+        spec:
+          containers:
+            - name: kruizedeletejob
+              image: kruize/autotune_operator:0.0.20.1_rm
+              imagePullPolicy: Always
+              volumeMounts:
+                - name: config-volume
+                  mountPath: /etc/config
+              command:
+                - sh
+                - -c
+                - |
+                  /home/autotune/app/target/bin/RetentionPartition
+              args: [ "" ]
+              env:
+                - name: START_AUTOTUNE
+                  value: "false"
+                - name: LOGGING_LEVEL
+                  value: "info"
+                - name: ROOT_LOGGING_LEVEL
+                  value: "error"
+                - name: DB_CONFIG_FILE
+                  value: "/etc/config/dbconfigjson"
+                - name: KRUIZE_CONFIG_FILE
+                  value: "/etc/config/kruizeconfigjson"
+                - name: deletepartitionsthreshold
+                  value: "15"
+          volumes:
+            - name: config-volume
+              configMap:
+                name: kruizeconfig
+          restartPolicy: OnFailure
+---
 apiVersion: monitoring.coreos.com/v1
 kind: ServiceMonitor
 metadata:

pom.xml

@@ -191,6 +191,10 @@
                             <mainClass>com.autotune.jobs.CreatePartition</mainClass>
                             <name>CreatePartition</name>
                         </program>
+                        <program>
+                            <mainClass>com.autotune.jobs.RetentionPartition</mainClass>
+                            <name>RetentionPartition</name>
+                        </program>
                     </programs>
                 </configuration>
                 <executions>

src/main/java/com/autotune/database/dao/ExperimentDAO.java

@@ -66,6 +66,8 @@ public interface ExperimentDAO {
     // Get KruizeResult Record
     List<KruizeResultsEntry> getKruizeResultsEntry(String experiment_name, String cluster_name, Timestamp interval_start_time, Timestamp interval_end_time) throws Exception;
 
+    void deletePartitions(int thresholdDaysCount);
+
     public void addPartitions(String tableName, String month, String year, int dayOfTheMonth, String partitionType) throws Exception;
 
 }

src/main/java/com/autotune/database/dao/ExperimentDAOImpl.java

@@ -24,10 +24,13 @@
 import org.slf4j.LoggerFactory;
 
 import java.sql.Timestamp;
+import java.text.SimpleDateFormat;
 import java.time.LocalDateTime;
 import java.time.YearMonth;
 import java.time.temporal.ChronoUnit;
 import java.util.ArrayList;
+import java.util.Calendar;
+import java.util.Date;
 import java.util.List;
 import java.util.stream.IntStream;
 
@@ -75,6 +78,57 @@ public ValidationOutputData addExperimentToDB(KruizeExperimentEntry kruizeExperi
         return validationOutputData;
     }
 
+    /**
+     * Deletes database partitions based on a specified threshold day count.
+     * <p>
+     * This method iterates through all Kruize tables in the database, extracts the date part
+     * from their names, and compares it with a cutoff date = currentDate - thresholdDaysCount. Tables with dates before the cutoff
+     * date are eligible for deletion, and the method executes SQL statements to drop these tables.
+     *
+     * @param thresholdDaysCount The number of days to be used as the threshold for partition deletion.
+     *                           Tables with dates older than this threshold will be deleted.
+     * @throws RuntimeException if any exception occurs during the deletion process. The exception
+     *                          details are logged, and the deletion process continues for other tables.
+     */
+    @Override
+    public void deletePartitions(int thresholdDaysCount) {
+        LOGGER.info("Threshold is set to {}", thresholdDaysCount);
+        // Calculate the date 'daysCount' days ago
+        Calendar calendar = Calendar.getInstance();
+        calendar.add(Calendar.DAY_OF_YEAR, -thresholdDaysCount);
+        Date cutoffDate = calendar.getTime();
+        String yyyyMMdd = "yyyyMMdd";
+        try (Session session = KruizeHibernateUtil.getSessionFactory().openSession()) {
+            List<String> tablenames = session.createNativeQuery(SELECT_ALL_KRUIZE_TABLES).getResultList();
+            for (String tableName : tablenames) {   // Since tableName cannot be null, there is no need to implement null handling; it can be skipped.
+                String datePart = null;
+                if (tableName.startsWith(DBConstants.TABLE_NAMES.KRUIZE_RESULTS + "_")) {
+                    datePart = tableName.substring((DBConstants.TABLE_NAMES.KRUIZE_RESULTS + "_").length());
+                } else if (tableName.startsWith(DBConstants.TABLE_NAMES.KRUIZE_RECOMMENDATIONS + "_")) {
+                    datePart = tableName.substring((DBConstants.TABLE_NAMES.KRUIZE_RECOMMENDATIONS + "_").length());
+                }
+                if (null != datePart) {
+                    Date tableDate = new SimpleDateFormat(yyyyMMdd).parse(datePart);
+                    // Compare the date part with the cutoffDate  (cutoffDate = todaysDate - thresholdDaysCount)
+                    if (tableDate.after(cutoffDate)) {
+                        LOGGER.debug("Table not eligible for deletion: " + tableName);
+                    } else {
+                        LOGGER.debug("Table found for deletion: " + tableName);
+                        try {
+                            Transaction tx = session.beginTransaction();
+                            session.createNativeQuery("DROP TABLE " + tableName).executeUpdate();
+                            tx.commit();
+                        } catch (Exception ignored) {
+                            LOGGER.error("Exception occurred while deleting the partition: {}", ignored.getMessage());
+                        }
+                    }
+                }
+            }
+        } catch (Exception e) {
+            LOGGER.error("Exception occurred while deleting the partition: {}", e.getMessage());
+        }
+    }
+
     @Override
     public void addPartitions(String tableName, String month, String year, int dayOfTheMonth, String partitionType) {
         Transaction tx;

src/main/java/com/autotune/database/helper/DBConstants.java

@@ -44,6 +44,8 @@ public static final class SQLQUERY {
         public static final String DELETE_FROM_RESULTS_BY_EXP_NAME = "DELETE FROM KruizeResultsEntry k WHERE k.experiment_name = :experimentName";
         public static final String DELETE_FROM_RECOMMENDATIONS_BY_EXP_NAME = "DELETE FROM KruizeRecommendationEntry k WHERE k.experiment_name = :experimentName";
         public static final String DB_PARTITION_DATERANGE = "CREATE TABLE IF NOT EXISTS %s_%s%s%s PARTITION OF %s FOR VALUES FROM ('%s-%s-%s 00:00:00.000') TO ('%s-%s-%s 23:59:59');";
+        public static final String SELECT_ALL_KRUIZE_TABLES = "SELECT table_name FROM information_schema.tables WHERE table_schema = 'public' " +
+                "and (table_name like 'kruize_results_%' or table_name like 'kruize_recommendations_%') ";
     }
 
     public static final class TABLE_NAMES {

src/main/java/com/autotune/jobs/RetentionPartition.java

@@ -3,10 +3,9 @@
 import com.autotune.analyzer.exceptions.K8sTypeNotSupportedException;
 import com.autotune.analyzer.exceptions.MonitoringAgentNotFoundException;
 import com.autotune.analyzer.exceptions.MonitoringAgentNotSupportedException;
-import com.autotune.database.init.KruizeHibernateUtil;
+import com.autotune.database.dao.ExperimentDAOImpl;
 import com.autotune.operator.InitializeDeployment;
-import org.hibernate.Session;
-import org.hibernate.SessionFactory;
+import com.autotune.operator.KruizeDeploymentInfo;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
@@ -15,19 +14,14 @@ public class RetentionPartition {
     private static final Logger LOGGER = LoggerFactory.getLogger(RetentionPartition.class);
 
     public static void main(String[] args) {
-        LOGGER.info("Checking Liveliness probe DB connection...");
+        LOGGER.info("RetentionPartition");
         try {
             InitializeDeployment.setup_deployment_info();
+            new ExperimentDAOImpl().deletePartitions(KruizeDeploymentInfo.delete_partition_threshold_in_days);
         } catch (Exception | K8sTypeNotSupportedException | MonitoringAgentNotSupportedException |
                  MonitoringAgentNotFoundException e) {
             e.printStackTrace();
             System.exit(1);
         }
-        SessionFactory factory = KruizeHibernateUtil.getSessionFactory();
-
-        Session session = factory.openSession();
-
-        session.close();
-        LOGGER.info("DB Liveliness probe connection successful!");
     }
 }

src/main/java/com/autotune/operator/KruizeDeploymentInfo.java

@@ -31,6 +31,7 @@
 import java.util.Hashtable;
 
 import static com.autotune.analyzer.utils.AnalyzerConstants.AutotuneConfigConstants.*;
+import static com.autotune.utils.KruizeConstants.KRUIZE_CONFIG_DEFAULT_VALUE.DELETE_PARTITION_THRESHOLD_IN_DAYS;
 
 /**
  * Contains information about the current deployment by parsing the autotune config map
@@ -68,7 +69,7 @@ public class KruizeDeploymentInfo {
     public static Integer bulk_update_results_limit = 100;
 
     public static int generate_recommendations_date_range_limit_in_days = 15;
-
+    public static Integer delete_partition_threshold_in_days = DELETE_PARTITION_THRESHOLD_IN_DAYS;
     private static Hashtable<String, Class> tunableLayerPair;
     //private static KubernetesClient kubernetesClient;
     private static KubeEventLogger kubeEventLogger;

src/main/java/com/autotune/utils/KruizeConstants.java

@@ -418,6 +418,7 @@ public static final class KRUIZE_CONFIG_ENV_NAME {
         public static final String AUTOTUNE_MODE = "autotunemode";
         public static final String EM_ONLY_MODE = "emonly";
         public static final String BULK_UPDATE_RESULTS_LIMIT = "bulkresultslimit";
+        public static final String DELETE_PARTITION_THRESHOLD_IN_DAYS = "deletepartitionsthreshold";
         public static final String SETTINGS_SAVE_TO_DB = "savetodb";
         public static final String SETTINGS_DB_DRIVER = "dbdriver";
         public static final String SETTINGS_HIBERNATE_DIALECT = "hibernate_dialect";
@@ -478,4 +479,8 @@ private RecommendationDurationRanges() {
 
 
     }
+
+    public static final class KRUIZE_CONFIG_DEFAULT_VALUE {
+        public static final int DELETE_PARTITION_THRESHOLD_IN_DAYS = 16;
+    }
 }