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 Sat, 29 Jul 2017 00:53:01 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.

When the colocated child bucket failed to initialize, remove the leader
bucket since all these buckets should be created atomically.


Project: http://git-wip-us.apache.org/repos/asf/geode/repo
Commit: http://git-wip-us.apache.org/repos/asf/geode/commit/1504e0f2
Tree: http://git-wip-us.apache.org/repos/asf/geode/tree/1504e0f2
Diff: http://git-wip-us.apache.org/repos/asf/geode/diff/1504e0f2

Branch: refs/heads/feature/GEM-1483
Commit: 1504e0f28e0f36a87eb8bd81a2026c1c7163a18d
Parents: 36665ae
Author: zhouxh <gzhou@pivotal.io>
Authored: Thu Jun 8 14:52:56 2017 -0700
Committer: zhouxh <gzhou@pivotal.io>
Committed: Fri Jul 28 17:51:47 2017 -0700

----------------------------------------------------------------------
 .../internal/cache/PRHARedundancyProvider.java  |  4 +-
 .../cache/PartitionedRegionDataStore.java       | 73 +++++++++++++++++++-
 2 files changed, 73 insertions(+), 4 deletions(-)
----------------------------------------------------------------------


http://git-wip-us.apache.org/repos/asf/geode/blob/1504e0f2/geode-core/src/main/java/org/apache/geode/internal/cache/PRHARedundancyProvider.java
----------------------------------------------------------------------
diff --git a/geode-core/src/main/java/org/apache/geode/internal/cache/PRHARedundancyProvider.java
b/geode-core/src/main/java/org/apache/geode/internal/cache/PRHARedundancyProvider.java
index f8e2108..c584b90 100644
--- a/geode-core/src/main/java/org/apache/geode/internal/cache/PRHARedundancyProvider.java
+++ b/geode-core/src/main/java/org/apache/geode/internal/cache/PRHARedundancyProvider.java
@@ -1798,9 +1798,9 @@ public class PRHARedundancyProvider {
   public void scheduleCreateMissingBuckets() {
     if (this.prRegion.getColocatedWith() != null
         && ColocationHelper.isColocationComplete(this.prRegion)) {
-      Runnable task = new CreateMissingBucketsTask(this);
+      Runnable task = new CreateMissingBucketsTask(PRHARedundancyProvider.this);
       final InternalResourceManager resourceManager =
-          this.prRegion.getGemFireCache().getInternalResourceManager();
+          prRegion.getGemFireCache().getInternalResourceManager();
       resourceManager.getExecutor().execute(task);
     }
   }

http://git-wip-us.apache.org/repos/asf/geode/blob/1504e0f2/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..3050054 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
@@ -758,6 +758,17 @@ public class PartitionedRegionDataStore implements HasCachePerfStats
{
       }
       this.getPartitionedRegion().checkReadiness();
       throw e;
+    } catch (RuntimeException validationException) {
+      // GEODE-3055
+      PartitionedRegion leader = ColocationHelper.getLeaderRegion(this.partitionedRegion);
+      boolean isLeader = leader.equals(this.partitionedRegion);
+      if (!isLeader) {
+        leader.getDataStore().removeBucket(bucketId, true);
+        logger.info("For bucket " + bucketId + ", failed to create cololcated child bucket
for "
+            + this.partitionedRegion.getFullPath() + ", removed leader region "
+            + leader.getFullPath() + " bucket.");
+      }
+      throw validationException;
     }
     // Determine the size of the bucket (the Region in this case is mirrored,
     // get initial image has populated the bucket, compute the size of the
@@ -1417,6 +1428,39 @@ 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 +1515,11 @@ 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
@@ -1480,10 +1528,33 @@ public class PartitionedRegionDataStore implements HasCachePerfStats
{
       // member is no longer hosting the bucket.
       writeLock.lock();
       try {
+        // forceRemovePrimary==true will enable remove the bucket even when:
+        // 1) it's primary
+        // 2) no other primary ready yet
+        // 3) colocated bucket and its child is not completely ready
         if (!forceRemovePrimary && bucketAdvisor.isPrimary()) {
           return false;
         }
 
+        if (isLeader) {
+          if (!forceRemovePrimary && !isRemotePrimaryReadyForColocatedChildren(bucketId))
{
+            return false;
+          }
+
+          InternalDistributedMember primary = bucketAdvisor.getPrimary();
+          if (!forceRemovePrimary && (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 +1584,6 @@ 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