HDDS-11727. Block ozone repair om fso-tree if OM is running (#7589)

apache · Jan 3, 2025 · d013188 · d013188
1 parent 69206e9
commit d013188
Show file tree

Hide file tree

Showing 3 changed files with 45 additions and 0 deletions.
diff --git a/hadoop-ozone/dist/src/shell/ozone/ozone b/hadoop-ozone/dist/src/shell/ozone/ozone
@@ -224,6 +224,7 @@ function ozonecmd_case
       OZONE_RUN_ARTIFACT_NAME="ozone-tools"
     ;;
     repair)
+      check_running_ozone_services
       OZONE_CLASSNAME=org.apache.hadoop.ozone.repair.OzoneRepair
       OZONE_DEBUG_OPTS="${OZONE_DEBUG_OPTS} ${RATIS_OPTS} ${OZONE_MODULE_ACCESS_ARGS}"
       OZONE_RUN_ARTIFACT_NAME="ozone-tools"
@@ -245,6 +246,25 @@ function ozonecmd_case
   esac
 }
 
+## @description  Check for running Ozone services using PID files.
+## @audience     public
+function check_running_ozone_services
+{
+  OZONE_PID_DIR="/tmp"
+
+  local services=("om" "scm" "datanode")
+
+  for service in "${services[@]}"; do
+    for pid_file in ${OZONE_PID_DIR}/ozone-*-${service}.pid; do
+      if [[ -f "${pid_file}" ]]; then
+        if kill -0 "$(cat "${pid_file}")" 2>/dev/null; then
+          export "OZONE_${service^^}_RUNNING=true"
+        fi
+      fi
+    done
+  done
+}
+
 ## @description  turn off logging for CLI by default
 ## @audience     private
 function ozone_suppress_shell_log

diff --git a/hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/RepairTool.java b/hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/RepairTool.java
@@ -18,13 +18,18 @@
 package org.apache.hadoop.ozone.repair;
 
 import org.apache.hadoop.hdds.cli.AbstractSubcommand;
+import picocli.CommandLine;
 
 import java.io.PrintWriter;
 import java.util.concurrent.Callable;
 
 /** Parent class for all actionable repair commands. */
 public abstract class RepairTool extends AbstractSubcommand implements Callable<Void> {
 
+  @CommandLine.Option(names = {"--force"},
+      description = "Use this flag if you want to bypass the check in false-positive cases.")
+  private boolean force;
+
   /** Hook method for subclasses for performing actual repair task. */
   protected abstract void execute() throws Exception;
 
@@ -34,6 +39,23 @@ public final Void call() throws Exception {
     return null;
   }
 
+  protected boolean checkIfServiceIsRunning(String serviceName) {
+    String envVariable = String.format("OZONE_%s_RUNNING", serviceName);
+    String runningServices = System.getenv(envVariable);
+    if ("true".equals(runningServices)) {
+      if (!force) {
+        error("Error: %s is currently running on this host. " +
+              "Stop the service before running the repair tool.", serviceName);
+        return true;
+      } else {
+        info("Warning: --force flag used. Proceeding despite %s being detected as running.", serviceName);
+      }
+    } else {
+      info("No running %s service detected. Proceeding with repair.", serviceName);
+    }
+    return false;
+  }
+
   protected void info(String msg, Object... args) {
     out().println(formatMessage(msg, args));
   }

diff --git a/hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/om/FSORepairCLI.java b/hadoop-ozone/tools/src/main/java/org/apache/hadoop/ozone/repair/om/FSORepairCLI.java
@@ -55,6 +55,9 @@ public class FSORepairCLI extends RepairTool {
 
   @Override
   public void execute() throws Exception {
+    if (checkIfServiceIsRunning("OM")) {
+      return;
+    }
     if (repair) {
       info("FSO Repair Tool is running in repair mode");
     } else {