[CASSANDRA-14952] NPE when using allocate_tokens_for_keyspace and add new DC

• Type: Bug
• Status: Resolved
• Priority: Low
• Resolution: Fixed
• Fix Version/s: 3.0.19, 3.11.5, 4.0-alpha1, 4.0
• Component/s: Cluster/Gossip
• Severity: Low
• Since Version: 3.0 alpha 1

Source Control Link

https://github.com/apache/cassandra/commit/2374a74eba6a4df84f9bda3fd311916c820e9cd6

Test and Documentation Plan

.

• Assignee: Jaydeepkumar Chovatia
• Reporter: Jaydeepkumar Chovatia

Authors

Jaydeepkumar Chovatia, Michael Semb Wever

Reviewers

Michael Semb Wever

• Created: 04/Jan/19 22:38
• Updated: 15/May/20 08:04
• Resolved: 09/Aug/19 17:54

Description

Received following NPE while bootstrapping very first node in the new datacenter with allocate_tokens_for_keyspace yaml option

INFO  21:44:13 JOINING: getting bootstrap token
Exception (java.lang.NullPointerException) encountered during startup: null
java.lang.NullPointerException
	at org.apache.cassandra.dht.tokenallocator.TokenAllocation.getStrategy(TokenAllocation.java:208)
	at org.apache.cassandra.dht.tokenallocator.TokenAllocation.getStrategy(TokenAllocation.java:170)
	at org.apache.cassandra.dht.tokenallocator.TokenAllocation.allocateTokens(TokenAllocation.java:55)
	at org.apache.cassandra.dht.BootStrapper.allocateTokens(BootStrapper.java:206)
	at org.apache.cassandra.dht.BootStrapper.getBootstrapTokens(BootStrapper.java:173)
	at org.apache.cassandra.service.StorageService.joinTokenRing(StorageService.java:854)
	at org.apache.cassandra.service.StorageService.initServer(StorageService.java:666)
	at org.apache.cassandra.service.StorageService.initServer(StorageService.java:579)
	at org.apache.cassandra.service.CassandraDaemon.setup(CassandraDaemon.java:351)
	at org.apache.cassandra.service.CassandraDaemon.activate(CassandraDaemon.java:586)
	at org.apache.cassandra.service.CassandraDaemon.main(CassandraDaemon.java:714)

Please find reproducible steps here:
1. Set allocate_tokens_for_keyspace property with Networktopologystrategy say {{Networktopologystrategy, 'dc1' : 1, 'dc2' : 1}}
2. Start first node in dc1
3. Now bootstrap second node in dc2, it will throw above exception.

RCA:
doAddEndpoint is invoked from the bootstrap and at this time local node's rack information is available

However with have allocate_tokens_for_keyspace option, daemon tries to access rack information even before calling bootstrap function, at this place which results in NPE

Fix:
Since this is applicable to only very first node for new dc, we can check for null as:

diff --git a/src/java/org/apache/cassandra/dht/tokenallocator/TokenAllocation.java b/src/java/org/apache/cassandra/dht/tokenallocator/TokenAllocation.java
index 8d8a6ffeca..e162757d95 100644
--- a/src/java/org/apache/cassandra/dht/tokenallocator/TokenAllocation.java
+++ b/src/java/org/apache/cassandra/dht/tokenallocator/TokenAllocation.java
@@ -205,7 +205,11 @@ public class TokenAllocation
         final int replicas = rs.getReplicationFactor(dc);
 
         Topology topology = tokenMetadata.getTopology();
-        int racks = topology.getDatacenterRacks().get(dc).asMap().size();
+        int racks = 1;
+        if (topology.getDatacenterRacks().get(dc) != null)
+        {
+            racks = topology.getDatacenterRacks().get(dc).asMap().size();
+        }
 
         if (racks >= replicas)
         {

Let me know your comments.

Issue Links

• relates to: CASSANDRA-12681 Reject empty options and invalid DC names in replication configuration while creating or altering a keyspace.