Yongle Zhang created CASSANDRA-16266:
----------------------------------------
Summary: Stress testing a mixed cluster with C* 2.1.0 (seed) and
2.0.0 causes NPE
Key: CASSANDRA-16266
URL: https://issues.apache.org/jira/browse/CASSANDRA-16266
Project: Cassandra
Issue Type: Bug
Reporter: Yongle Zhang
Steps to reproduce:
# setup a mixed cluster with C* 2.1.0 (seed node) and C* 2.0.0
# run the stress testing tool, e.g.,
{code:java}
/cassandra/tools/bin/cassandra-stress write n=1000 -rate threads=50 -node
250.16.238.1,250.16.238.2{code}
NPE:
{code:java}
ERROR [InternalResponseStage:2] 2020-07-22 08:29:36,170 CassandraDaemon.java
(line 186) Exception in thread Thread[InternalResponseStage:2,5,main]
java.lang.NullPointerException
at
org.apache.cassandra.serializers.BooleanSerializer.deserialize(BooleanSerializer.java:33)
at
org.apache.cassandra.serializers.BooleanSerializer.deserialize(BooleanSerializer.java:24)
at org.apache.cassandra.db.marshal.AbstractType.compose(AbstractType.java:142)
at
org.apache.cassandra.cql3.UntypedResultSet$Row.getBoolean(UntypedResultSet.java:106)
at
org.apache.cassandra.config.CFMetaData.fromSchemaNoColumnsNoTriggers(CFMetaData.java:1555)
at org.apache.cassandra.config.CFMetaData.fromSchema(CFMetaData.java:1642)
at
org.apache.cassandra.config.KSMetaData.deserializeColumnFamilies(KSMetaData.java:305)
at org.apache.cassandra.db.DefsTables.mergeColumnFamilies(DefsTables.java:270)
at org.apache.cassandra.db.DefsTables.mergeSchema(DefsTables.java:183)
at
org.apache.cassandra.service.MigrationTask$1.response(MigrationTask.java:66)
at
org.apache.cassandra.net.ResponseVerbHandler.doVerb(ResponseVerbHandler.java:46)
at
org.apache.cassandra.net.MessageDeliveryTask.run(MessageDeliveryTask.java:56)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
at java.lang.Thread.run(Thread.java:748)
{code}
Root cause: incompatible data
In the `CFMetaData` class of version 2.0.0, there is a boolean field named
`replicate_on_write`. In the same class of version 2.1.0, however, this field
no longer exists. When serializing this class in function
`toSchemaNoColumnsNoTriggers`, it will first write all of its fields into a
`RowMutation` (in 2.0.0) / `Mutation` (in 2.1.0) class, and then serialize this
“Mutation” like class in the same way. In 2.0.0 the `replicate_on_write` field
gets serialized at
[https://github.com/apache/cassandra/blob/03045ca22b11b0e5fc85c4fabd83ce6121b5709b/src/java/org/apache/cassandra/config/CFMetaData.java#L1514]
.
When deserializing this class in function `fromSchemaNoColumnsNoTriggers`, it
reads all its fields from a map-like class `UntypedResultSet.Row`. In 2.0.0 the
`replicate_on_write` field gets deserialized at
[https://github.com/apache/cassandra/blob/03045ca22b11b0e5fc85c4fabd83ce6121b5709b/src/java/org/apache/cassandra/config/CFMetaData.java#L1555]
.
The problem is that the existence of the key is not checked, and the map
returns a `null` value because the message from 2.1.0 doesn’t contain the
`replicate_on_write` key, which leads to the NullPointerException.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
