HDDS-12010. Block ozone repair if service is running (apache#7758)

hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/RepairTool.java

@@ -17,6 +17,7 @@
  */
 package org.apache.hadoop.ozone.repair;
 
+import jakarta.annotation.Nullable;
 import org.apache.hadoop.hdds.cli.AbstractSubcommand;
 import picocli.CommandLine;
 
@@ -45,35 +46,57 @@ public abstract class RepairTool extends AbstractSubcommand implements Callable<
   /** Hook method for subclasses for performing actual repair task. */
   protected abstract void execute() throws Exception;
 
+  /** Which Ozone component should be verified to be offline. */
+  @Nullable
+  protected Component serviceToBeOffline() {
+    return null;
+  }
+
   @Override
   public final Void call() throws Exception {
     if (!dryRun) {
       confirmUser();
     }
-    execute();
+    if (isServiceStateOK()) {
+      execute();
+    }
     return null;
   }
 
-  protected boolean checkIfServiceIsRunning(String serviceName) {
-    String runningEnvVar = String.format("OZONE_%s_RUNNING", serviceName);
-    String pidEnvVar = String.format("OZONE_%s_PID", serviceName);
-    String isServiceRunning = System.getenv(runningEnvVar);
-    String servicePid = System.getenv(pidEnvVar);
-    if ("true".equals(isServiceRunning)) {
-      if (!force) {
-        error("Error: %s is currently running on this host with PID %s. " +
-            "Stop the service before running the repair tool.", serviceName, servicePid);
-        return true;
-      } else {
-        info("Warning: --force flag used. Proceeding despite %s being detected as running with PID %s.",
-            serviceName, servicePid);
-      }
-    } else {
-      info("No running %s service detected. Proceeding with repair.", serviceName);
+  private boolean isServiceStateOK() {
+    final Component service = serviceToBeOffline();
+
+    if (service == null) {
+      return true; // online tool
+    }
+
+    if (!isServiceRunning(service)) {
+      info("No running %s service detected. Proceeding with repair.", service);
+      return true;
+    }
+
+    String servicePid = getServicePid(service);
+
+    if (force) {
+      info("Warning: --force flag used. Proceeding despite %s being detected as running with PID %s.",
+          service, servicePid);
+      return true;
     }
+
+    error("Error: %s is currently running on this host with PID %s. " +
+        "Stop the service before running the repair tool.", service, servicePid);
+
     return false;
   }
 
+  private static String getServicePid(Component service) {
+    return System.getenv(String.format("OZONE_%s_PID", service));
+  }
+
+  private static boolean isServiceRunning(Component service) {
+    return "true".equals(System.getenv(String.format("OZONE_%s_RUNNING", service)));
+  }
+
   protected boolean isDryRun() {
     return dryRun;
   }
@@ -117,4 +140,11 @@ private String getConsoleReadLineWithFormat(String currentUser) {
         .nextLine()
         .trim();
   }
+
+  /** Ozone component for offline tools. */
+  protected enum Component {
+    DATANODE,
+    OM,
+    SCM,
+  }
 }

hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/TransactionInfoRepair.java

@@ -21,6 +21,7 @@
  */
 package org.apache.hadoop.ozone.repair;
 
+import jakarta.annotation.Nonnull;
 import org.apache.hadoop.hdds.cli.HddsVersionProvider;
 import org.apache.hadoop.hdds.scm.metadata.SCMDBDefinition;
 import org.apache.hadoop.hdds.utils.IOUtils;
@@ -69,16 +70,12 @@ public class TransactionInfoRepair extends RepairTool {
 
   @Override
   public void execute() throws Exception {
-    final Component component = getComponent();
-    if (checkIfServiceIsRunning(component.name())) {
-      return;
-    }
     List<ColumnFamilyHandle> cfHandleList = new ArrayList<>();
     List<ColumnFamilyDescriptor> cfDescList = RocksDBUtils.getColumnFamilyDescriptors(
         dbPath);
+    String columnFamilyName = getColumnFamily().getName();
 
     try (ManagedRocksDB db = ManagedRocksDB.open(dbPath, cfDescList, cfHandleList)) {
-      String columnFamilyName = component.columnFamilyDefinition.getName();
       ColumnFamilyHandle transactionInfoCfh = RocksDBUtils.getColumnFamilyHandle(columnFamilyName, cfHandleList);
       if (transactionInfoCfh == null) {
         throw new IllegalArgumentException(columnFamilyName +
@@ -111,7 +108,9 @@ public void execute() throws Exception {
     }
   }
 
-  private Component getComponent() {
+  @Override
+  @Nonnull
+  protected Component serviceToBeOffline() {
     final String parent = spec().parent().name();
     switch (parent) {
     case "om":
@@ -123,14 +122,15 @@ private Component getComponent() {
     }
   }
 
-  private enum Component {
-    OM(OMDBDefinition.TRANSACTION_INFO_TABLE),
-    SCM(SCMDBDefinition.TRANSACTIONINFO);
-
-    private final DBColumnFamilyDefinition<String, TransactionInfo> columnFamilyDefinition;
-
-    Component(DBColumnFamilyDefinition<String, TransactionInfo> columnFamilyDefinition) {
-      this.columnFamilyDefinition = columnFamilyDefinition;
+  private DBColumnFamilyDefinition<String, TransactionInfo> getColumnFamily() {
+    Component component = serviceToBeOffline();
+    switch (component) {
+    case OM:
+      return OMDBDefinition.TRANSACTION_INFO_TABLE;
+    case SCM:
+      return SCMDBDefinition.TRANSACTIONINFO;
+    default:
+      throw new IllegalStateException("This tool does not support component: " + component);
     }
   }
 }

hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/om/FSORepairTool.java

@@ -17,6 +17,7 @@
  */
 package org.apache.hadoop.ozone.repair.om;
 
+import jakarta.annotation.Nonnull;
 import org.apache.commons.io.FileUtils;
 import org.apache.hadoop.hdds.conf.ConfigurationSource;
 import org.apache.hadoop.hdds.conf.OzoneConfiguration;
@@ -97,11 +98,14 @@ public class FSORepairTool extends RepairTool {
       description = "Verbose output. Show all intermediate steps.")
   private boolean verbose;
 
+  @Nonnull
+  @Override
+  protected Component serviceToBeOffline() {
+    return Component.OM;
+  }
+
   @Override
   public void execute() throws Exception {
-    if (checkIfServiceIsRunning("OM")) {
-      return;
-    }
     try {
       Impl repairTool = new Impl();
       repairTool.run();

hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/om/SnapshotChainRepair.java

@@ -18,6 +18,7 @@
 
 package org.apache.hadoop.ozone.repair.om;
 
+import jakarta.annotation.Nonnull;
 import org.apache.hadoop.hdds.utils.IOUtils;
 import org.apache.hadoop.hdds.utils.db.StringCodec;
 import org.apache.hadoop.hdds.utils.db.managed.ManagedRocksDB;
@@ -76,11 +77,14 @@ public class SnapshotChainRepair extends RepairTool {
       description = "Path previous snapshotId to set for the given snapshot")
   private UUID pathPreviousSnapshotId;
 
+  @Nonnull
+  @Override
+  protected Component serviceToBeOffline() {
+    return Component.OM;
+  }
+
   @Override
   public void execute() throws Exception {
-    if (checkIfServiceIsRunning("OM")) {
-      return;
-    }
     List<ColumnFamilyHandle> cfHandleList = new ArrayList<>();
     List<ColumnFamilyDescriptor> cfDescList = RocksDBUtils.getColumnFamilyDescriptors(dbPath);
 

hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/scm/cert/RecoverSCMCertificate.java

@@ -17,6 +17,7 @@
  */
 package org.apache.hadoop.ozone.repair.scm.cert;
 
+import jakarta.annotation.Nonnull;
 import org.apache.hadoop.hdds.conf.OzoneConfiguration;
 import org.apache.hadoop.hdds.security.SecurityConfig;
 import org.apache.hadoop.hdds.security.x509.certificate.authority.CAType;
@@ -71,11 +72,14 @@ public class RecoverSCMCertificate extends RepairTool {
       description = "SCM DB Path")
   private String dbPath;
 
+  @Nonnull
+  @Override
+  protected Component serviceToBeOffline() {
+    return Component.SCM;
+  }
+
   @Override
   public void execute() throws Exception {
-    if (checkIfServiceIsRunning("SCM")) {
-      return;
-    }
     dbPath = removeTrailingSlashIfNeeded(dbPath);
     String tableName = VALID_SCM_CERTS.getName();
     DBDefinition dbDefinition =