GEODE-9060: Remove the member from a copy of replicates as GII candid… (

#6246)

* GEODE-9060: Remove the member from a copy of replicates as GII candidate if
it's not part of the same distributed system, but leave original replicates
unchanged.

(cherry picked from commit 76a5afd)

geode-core/src/distributedTest/java/org/apache/geode/internal/cache/persistence/PersistentRecoveryOrderDUnitTest.java

@@ -877,11 +877,43 @@ public void testSplitBrain() {
         Throwable thrown = catchThrowable(() -> {
           createReplicateRegion(regionName, getDiskDirs(getVMId()));
         });
-        assertThat(thrown).isInstanceOf(ConflictingPersistentDataException.class);
+        assertThat(thrown)
+            .isInstanceOf(ConflictingPersistentDataException.class)
+            .hasMessageContaining("was not part of the same distributed system as the local data");
       }
     });
   }
 
+  @Test
+  public void testRecoverableSplitBrain() {
+    vm2.invoke(() -> {
+      createReplicateRegion(regionName, getDiskDirs(getVMId()));
+    });
+    vm0.invoke(() -> {
+      createReplicateRegion(regionName, getDiskDirs(getVMId()));
+      putEntry("A", "B");
+      getCache().getRegion(regionName).close();
+    });
+
+    vm1.invoke(() -> {
+      createReplicateRegion(regionName, getDiskDirs(getVMId()));
+      validateEntry("A", "B");
+      updateEntry("A", "C");
+      getCache().getRegion(regionName).close();
+    });
+
+    // VM0 doesn't know that VM1 ever existed so it will start up.
+    vm0.invoke(() -> {
+      createReplicateRegion(regionName, getDiskDirs(getVMId()));
+      validateEntry("A", "C");
+    });
+
+    vm1.invoke(() -> {
+      createReplicateRegion(regionName, getDiskDirs(getVMId()));
+      validateEntry("A", "C");
+    });
+  }
+
   /**
    * Test to make sure that if if a member crashes while a GII is in progress, we wait for the
    * member to come back for starting.

geode-core/src/main/java/org/apache/geode/internal/cache/persistence/PersistenceAdvisorImpl.java

@@ -510,18 +510,18 @@ public Set<PersistentMemberID> getPersistedMembers() {
   public boolean checkMyStateOnMembers(Set<InternalDistributedMember> replicates)
       throws ReplyException {
     PersistentStateQueryResults remoteStates = getMyStateOnMembers(replicates);
+    Set<InternalDistributedMember> copyOfReplicates = null;
 
     persistenceAdvisorObserver.observe(regionPath);
 
     boolean equal = false;
+    PersistentMemberID myId = getPersistentID();
     for (Map.Entry<InternalDistributedMember, PersistentMemberState> entry : remoteStates
         .getStateOnPeers().entrySet()) {
       InternalDistributedMember member = entry.getKey();
       PersistentMemberID remoteId = remoteStates.getPersistentIds().get(member);
 
-      PersistentMemberID myId = getPersistentID();
       PersistentMemberState stateOnPeer = entry.getValue();
-
       if (PersistentMemberState.REVOKED.equals(stateOnPeer)) {
         throw new RevokedPersistentDataException(
             String.format(
@@ -533,7 +533,19 @@ public boolean checkMyStateOnMembers(Set<InternalDistributedMember> replicates)
         String message = String.format(
             "Region %s remote member %s with persistent data %s was not part of the same distributed system as the local data from %s",
             regionPath, member, remoteId, myId);
-        throw new ConflictingPersistentDataException(message);
+        // Conceptually the removed member due to not knowing current member, should be equal to
+        // existing replicates.
+        // It can still be used as GII provider candidate. Use a copyOfReplicates to avoid modifying
+        // the replicates.
+        if (copyOfReplicates == null) {
+          copyOfReplicates = new HashSet<>(replicates);
+        }
+        copyOfReplicates.remove(member);
+        if (copyOfReplicates.isEmpty()) {
+          throw new ConflictingPersistentDataException(message);
+        } else {
+          logger.info(message);
+        }
       }
 
       if (myId != null && stateOnPeer == PersistentMemberState.EQUAL) {