Index truncation is no longer in transaction with primary records to avoid deadlock.

1 files changed, 13 insertions, 13 deletions

diff --git a/src/main/java/com/amazon/carbonado/repo/indexed/IndexedStorage.java b/src/main/java/com/amazon/carbonado/repo/indexed/IndexedStorage.java
index 1abb064..6b2fed9 100644
--- a/src/main/java/com/amazon/carbonado/repo/indexed/IndexedStorage.java
+++ b/src/main/java/com/amazon/carbonado/repo/indexed/IndexedStorage.java
@@ -142,20 +142,20 @@ class IndexedStorage<S extends Storable> implements Storage<S>, StorageAccess<S>
             return;

         }

 

-        Transaction txn = mRepository.enterTransaction();

-        try {

-            mMasterStorage.truncate();

-

-            // Now truncate the indexes.

-            for (IndexInfo info : mAllIndexInfoMap.values()) {

-                if (info instanceof ManagedIndex) {

-                    ((ManagedIndex) info).getIndexEntryStorage().truncate();

-                }

+        // Although the master storage and indexes should be truncated

+        // atomically, this cannot be guaranteed due to freedom of truncate

+        // implementation. Also, a transaction can be deadlock prone with

+        // concurrent queries and updates. Instead, truncate the master first,

+        // which may cause warnings to be logged if index entries refer to

+        // missing records. This is harmless.

+

+        mMasterStorage.truncate();

+

+        // Now truncate the indexes.

+        for (IndexInfo info : mAllIndexInfoMap.values()) {

+            if (info instanceof ManagedIndex) {

+                ((ManagedIndex) info).getIndexEntryStorage().truncate();

             }

-

-            txn.commit();

-        } finally {

-            txn.exit();

         }

     }