Merge pull request #13089 from vespa-engine/mpolden/take-application-lock

Take both legacy lock and application lock

1 files changed, 22 insertions, 5 deletions

diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java
index 347c66d5426..2b64037a8b9 100644
--- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java
+++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/persistence/CuratorDatabaseClient.java
@@ -10,8 +10,8 @@ import com.yahoo.config.provision.HostName;
 import com.yahoo.config.provision.NodeFlavors;
 import com.yahoo.config.provision.NodeType;
 import com.yahoo.config.provision.Zone;
-import java.util.logging.Level;
 import com.yahoo.path.Path;
+import com.yahoo.transaction.Mutex;
 import com.yahoo.transaction.NestedTransaction;
 import com.yahoo.transaction.Transaction;
 import com.yahoo.vespa.curator.Curator;
@@ -364,19 +364,36 @@ public class CuratorDatabaseClient {
 
     /** Acquires the single cluster-global, reentrant lock for active nodes of this application */
     // TODO(mpolden): Remove when all config servers take the new lock
-    public Lock legacyLock(ApplicationId application) {
+    public Mutex legacyLock(ApplicationId application) {
         return legacyLock(application, defaultLockTimeout);
     }
 
     /** Acquires the single cluster-global, reentrant lock with the specified timeout for active nodes of this application */
     // TODO(mpolden): Remove when all config servers take the new lock
-    public Lock legacyLock(ApplicationId application, Duration timeout) {
+    public Mutex legacyLock(ApplicationId application, Duration timeout) {
+        Mutex legacyLock;
+        Mutex lock;
+        // Take the legacy node-repository lock
         try {
-            return db.lock(legacyLockPath(application), timeout);
+            legacyLock = db.lock(legacyLockPath(application), timeout);
+        } catch (UncheckedTimeoutException e) {
+            throw new ApplicationLockException(e);
         }
-        catch (UncheckedTimeoutException e) {
+        // Take the application lock (same as config server). This is likey already held at this point, but is
+        // re-entrant.
+        try {
+            lock = db.lock(lockPath(application), timeout);
+        } catch (UncheckedTimeoutException e) {
+            legacyLock.close();
             throw new ApplicationLockException(e);
         }
+        return () -> {
+            try {
+                lock.close();
+            } finally {
+                legacyLock.close();
+            }
+        };
     }
 
     /**