geode-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From zho...@apache.org
Subject [50/50] [abbrv] geode git commit: GEODE-3055: The old primary's the shadow bucket is not initialized when rebalance remove it. Thus the new primary candidate can never initialize from it. The fix is to wait until new primary exists before remove the old
Date Fri, 21 Jul 2017 06:42:02 GMT
GEODE-3055: The old primary's the shadow bucket is not
initialized when rebalance remove it. Thus the new primary candidate can
never initialize from it. The fix is to wait until new primary exists before
remove the old primary's bucket in rebalance.


Project: http://git-wip-us.apache.org/repos/asf/geode/repo
Commit: http://git-wip-us.apache.org/repos/asf/geode/commit/40fb5fdf
Tree: http://git-wip-us.apache.org/repos/asf/geode/tree/40fb5fdf
Diff: http://git-wip-us.apache.org/repos/asf/geode/diff/40fb5fdf

Branch: refs/heads/feature/GEM-1483
Commit: 40fb5fdf9e7053de7b1f1df0fc1adfd3fe78546b
Parents: 66cd31a
Author: zhouxh <gzhou@pivotal.io>
Authored: Thu Jun 8 14:52:56 2017 -0700
Committer: zhouxh <gzhou@pivotal.io>
Committed: Thu Jul 20 23:40:43 2017 -0700

----------------------------------------------------------------------
 .../cache/PartitionedRegionDataStore.java       | 58 +++++++++++++++++++-
 1 file changed, 56 insertions(+), 2 deletions(-)
----------------------------------------------------------------------


http://git-wip-us.apache.org/repos/asf/geode/blob/40fb5fdf/geode-core/src/main/java/org/apache/geode/internal/cache/PartitionedRegionDataStore.java
----------------------------------------------------------------------
diff --git a/geode-core/src/main/java/org/apache/geode/internal/cache/PartitionedRegionDataStore.java
b/geode-core/src/main/java/org/apache/geode/internal/cache/PartitionedRegionDataStore.java
index 0318c75..bf708a9 100644
--- a/geode-core/src/main/java/org/apache/geode/internal/cache/PartitionedRegionDataStore.java
+++ b/geode-core/src/main/java/org/apache/geode/internal/cache/PartitionedRegionDataStore.java
@@ -1417,6 +1417,39 @@ public class PartitionedRegionDataStore implements HasCachePerfStats
{
     }
   }
 
+  public boolean isRemotePrimaryReadyForColocatedChildren(int bucketId) {
+    boolean isRemotePrimaryReady = true;
+    InternalDistributedMember myId =
+        this.partitionedRegion.getDistributionManager().getDistributionManagerId();
+
+    List<PartitionedRegion> colocatedChildPRs =
+        ColocationHelper.getColocatedChildRegions(this.partitionedRegion);
+    if (colocatedChildPRs != null) {
+      for (PartitionedRegion pr : colocatedChildPRs) {
+        InternalDistributedMember primaryChild = pr.getBucketPrimary(bucketId);
+        if (logger.isDebugEnabled()) {
+          logger.debug("Checking colocated child bucket " + pr + ", bucketId=" + bucketId
+              + ", primary is " + primaryChild);
+        }
+        if (primaryChild == null || myId.equals(primaryChild)) {
+          if (logger.isDebugEnabled()) {
+            logger.debug("Colocated bucket region " + pr + " " + bucketId
+                + " does not have a remote primary yet. Not to remove.");
+          }
+          return false;
+        } else {
+          if (logger.isDebugEnabled()) {
+            logger
+                .debug(pr + " bucketId=" + bucketId + " has remote primary, checking its
children");
+          }
+          isRemotePrimaryReady = isRemotePrimaryReady
+              && pr.getDataStore().isRemotePrimaryReadyForColocatedChildren(bucketId);
+        }
+      }
+    }
+    return isRemotePrimaryReady;
+  }
+
   /**
    * Removes a redundant bucket hosted by this data store. The rebalancer invokes this method
    * directly or sends this member a message to invoke it.
@@ -1471,7 +1504,11 @@ public class PartitionedRegionDataStore implements HasCachePerfStats
{
 
       }
 
+      PartitionedRegion leader = ColocationHelper.getLeaderRegion(this.partitionedRegion);
+      boolean isLeader = leader.equals(this.partitionedRegion);
       BucketAdvisor bucketAdvisor = bucketRegion.getBucketAdvisor();
+      InternalDistributedMember myId =
+          this.partitionedRegion.getDistributionManager().getDistributionManagerId();
       Lock writeLock = bucketAdvisor.getActiveWriteLock();
 
       // Fix for 43613 - don't remove the bucket
@@ -1484,6 +1521,25 @@ public class PartitionedRegionDataStore implements HasCachePerfStats
{
           return false;
         }
 
+        if (isLeader) {
+          if (!isRemotePrimaryReadyForColocatedChildren(bucketId)) {
+            return false;
+          }
+
+          InternalDistributedMember primary = bucketAdvisor.getPrimary();
+          if (primary == null || myId.equals(primary)) {
+            if (logger.isDebugEnabled()) {
+              logger.debug("Bucket region " + bucketRegion
+                  + " does not have a remote primary yet. Not to remove.");
+            }
+            return false;
+          }
+
+          if (logger.isDebugEnabled()) {
+            logger.debug("Bucket region " + bucketRegion + " has primary at " + primary);
+          }
+        }
+
         // recurse down to each tier of children to remove first
         removeBucketForColocatedChildren(bucketId, forceRemovePrimary);
 
@@ -1513,8 +1569,6 @@ public class PartitionedRegionDataStore implements HasCachePerfStats
{
       // because it won't block write operations while we're trying to acquire
       // the activePrimaryMoveLock
       InternalDistributedMember primary = bucketAdvisor.getPrimary();
-      InternalDistributedMember myId =
-          this.partitionedRegion.getDistributionManager().getDistributionManagerId();
       if (!myId.equals(primary)) {
         StateFlushOperation flush = new StateFlushOperation(bucketRegion);
         int executor = DistributionManager.WAITING_POOL_EXECUTOR;


Mime
View raw message