Java Client: NullPointerException with GenericRecord.getField() on a sub-record

[epsteina16]

Describe the bug
Calling GenericRecord.getField(Field field) on a GenericRecord that is a sub-record, where the field is a string, the Pulsar client throws a NullPointerException:

java.lang.NullPointerException
at org.apache.pulsar.client.impl.schema.generic.GenericJsonRecord.isBinaryValue(GenericJsonRecord.java:110)
at org.apache.pulsar.client.impl.schema.generic.GenericJsonRecord.getField(GenericJsonRecord.java:87)
at org.apache.pulsar.client.api.schema.GenericRecord.getField(GenericRecord.java:53)

To Reproduce
Steps to reproduce the behavior:

1. Using the Pulsar Java Client create a consumer of type Schema.AUTO_CONSUME, to receive GenericRecords from the broker.
2. Fill the topic with messages with a subrecord with a string in it, in the form of a JsonSchema. This is an example schema:
   {
   "type": "JSON",
   "schema": "{"type":"record","namespace":"com.example","name":"MyJsonSchema","fields":[{"name":"subrecord_field","type":{"type":"record","name":"subrecord_field","fields":[{"name":"substring_field","type":"string"}]}}]}",
   "properties": {}
   }
3. When a message is received, get the subrecord by calling .getField(Field field) on the top level record.
4. The try to get the field for the substring field by calling .getField() on the subrecord.

Example code:
GenericRecord subRecord = record.getField("subrecord_field");
subRecord.getField("substring_field"); // This will throw a null pointer exception

Expected behavior
getField should return the Object that is the field. i.e. for this, we should get a String back.

Desktop (please complete the following information):

• OS: linux

Additional context
Pulsar client version 2.7.0

[codelipenghui]

@congbobo184 Could you please take a look at this issue?

[congbobo184]

@epsteina16 does the subrecord_field have value? I test the scenes, if it have value then it will not throw NPE. it don't have value it will throw NPE.

[epsteina16]

The subrecord_field contains fields. Whether or not substring_field is an empty string or has content, I see this. Pulsar should not have a NPE either way.

[codelipenghui]

@epsteina16 Agree, I notice substring_field is not allow null from the schema definition right? Can the writer write a message with substring_field=null?

@congbobo184 I think we need to fix this case, we should check if the GenericRecord has this field, if not, we should return null. You mentioned have value or don't have value, this means not null or null?

[eolivelli]

@vroyer PTAL
are you interested in providing a fix ?

[congbobo184]

Move this to the next release.

[eolivelli]

I believe that probably this has already been fixed

[michaeljmarshall]

@epsteina16 - can you confirm whether or not this bug is still present in recent releases of Pulsar? We are releasing 2.7.4 now, so I am going to remove this label. However, if the bug still exists, we should fix it. Thanks