Error is transient. Subsequent runs worked.

@Stack
We too got one NPE in hbck. Still not found the reason. Not sure if it is same as this one.

@Ram , Yes this is the same issue.. I got the reason.
The scenario is like this as in our test.
There is one table and there was a case of one region of that table was not assigned with any of the RS. HBCK tool fixing this issue. After that HBCK will run again.
At this time getHTableDescriptors () is not finding any table in the cluster and return null and so reportTablesInFlux() -> errors.print("Number of Tables: " + allTables.length); gives a NPE

Why at this time no tables getting out of getHTableDescriptors () [Even though one table is there in the cluster is] this table is modified recently. HBCK just changed the HRegionInfo of the region of the table by assigning it to one of the RS.

For fix
1. We need null check in reportTablesInFlux() I think
2. When HBCK rerun after the fix we can set timelag =0?

I started a run of the unit test suite testing this fix – for a method like this, I prefer returning empty arrays instead of null arrays.

diff --git src/main/java/org/apache/hadoop/hbase/client/HBaseAdmin.java src/main/java/org/apache/hadoop/hbase/cli
index ee16e72..44b7c11 100644
--- src/main/java/org/apache/hadoop/hbase/client/HBaseAdmin.java
+++ src/main/java/org/apache/hadoop/hbase/client/HBaseAdmin.java
@@ -1691,7 +1691,7 @@ public class HBaseAdmin implements Abortable, Closeable {
  /**
  * Get tableDescriptors
  * @param tableNames List of table names
- * @return HTD[] the tableDescriptor
+ * @return HTD[] the tableDescriptor (never null)
  * @throws IOException if a remote or network exception occurs
  */
   public HTableDescriptor[] getTableDescriptors(List<String> tableNames)
diff --git src/main/java/org/apache/hadoop/hbase/client/HConnectionManager.java src/main/java/org/apache/hadoop/h
index 820e2a9..f183b15 100644
--- src/main/java/org/apache/hadoop/hbase/client/HConnectionManager.java
+++ src/main/java/org/apache/hadoop/hbase/client/HConnectionManager.java
@@ -2195,7 +2195,7 @@ public class HConnectionManager {
 
     @Override
     public HTableDescriptor[] getHTableDescriptors(List<String> tableNames) throws IOException {
-      if (tableNames == null || tableNames.isEmpty()) return null;
+      if (tableNames == null || tableNames.isEmpty()) return new HTableDescriptor[0];
       MasterKeepAliveConnection master = getKeepAliveMaster();
       try {
         return master.getHTableDescriptors(tableNames);

@Jon
Yes null check I also dont like to put...
Also what about 2. When HBCK rerun after the fix we can set timelag =0?

I think #2 makes sense, but would need to be tested to verify (it is a legacy of the original hbck – I didn't change this).

Anoop – do you guys want to take this on or should I?

Jon, I can provide a patch tomorrow addressing both the points I have mentioned.[If it is ok with you]

Anoop – go for it.

Returning empty array is valid. I dug a little into the master side as well – it returns an empty array in the case where an invalid set of table names is passed.

Patch for trunk and 0.94

Got a doubt on my patch now
We should track the skipped regions? or the included regions in the 1st run...

The NPE problem as such getting fixed by the patch HBASE-5928.
We can close this issue.

Also there is another point ( issue ) with HBCK rerun which is mentioned as 2nd point in my above comment. Better to raise a new issue to handle that.
@Jon Ok with you?

Raised https://issues.apache.org/jira/browse/HBASE-6015 to track the second point

The issue with NPE is fixed as part of HBASE-5928.