clarax commented on a change in pull request #3723:
URL: https://github.com/apache/hbase/pull/3723#discussion_r728645078



##########
File path: 
hbase-balancer/src/main/java/org/apache/hadoop/hbase/master/balancer/LoadCandidateGenerator.java
##########
@@ -34,27 +35,51 @@ BalanceAction generate(BalancerClusterState cluster) {
   private int pickLeastLoadedServer(final BalancerClusterState cluster, int 
thisServer) {
     Integer[] servers = cluster.serverIndicesSortedByRegionCount;
 
-    int index = 0;
-    while (servers[index] == null || servers[index] == thisServer) {
-      index++;
-      if (index == servers.length) {
-        return -1;
+    int selectedIndex = -1;
+    double currentLargestRandom = -1;
+    for (int i = 0; i < servers.length; i++) {
+      if (servers[i] == null || servers[i] == thisServer) {
+        continue;
+      }
+      if (selectedIndex != -1
+        && cluster.getNumRegionsComparator().compare(servers[i], 
servers[selectedIndex]) != 0) {
+        // Exhausted servers of the same region count
+        break;
+      }
+      // we don't know how many servers have the same region count, we will 
randomly select one

Review comment:
       The server are sorted by region counts. And on a large cluster, we often 
have many servers with same count.  instead of always picking the first one 
which may not work, we want to pick a random one from the candidate to unstuck 
balancer and generate more random candidates.. 




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to