HurSungYun opened a new issue, #36356:
URL: https://github.com/apache/beam/issues/36356
### What happened?
When updating a Dataflow streaming pipeline, a NullPointerException: Null
topic occurs in KafkaSourceDescriptor.
### pipeline logic (Python)
```python
from apache_beam.io.kafka import ReadFromKafka
# wrapper class
class ReadFromPlaintextKafkaSource(ReadFromKafka):
def __init__(
self,
job_name: str,
broker_addresses: str,
topics: List[str],
redistribute: bool = False,
redistribute_num_keys: int = 0,
):
if not job_name:
raise ValueError("Kafka job name is required")
job_name = job_name.replace("-", "_")
if not broker_addresses:
raise ValueError("Kafka broker addresses are required")
if not topics:
raise ValueError("Kafka topics are required")
consumer_group_id = "some_group"
logger.info(f"Consumer group id: {consumer_group_id}")
consumer_client_id =
f"{consumer_group_id}.{''.join(random.choices(string.ascii_letters +
string.digits, k=6))}"
consumer_config = {
"allow.auto.create.topics": "false",
"auto.commit.interval.ms": "100",
"bootstrap.servers": broker_addresses,
"group.id": consumer_group_id,
"enable.auto.commit": "false",
"auto.offset.reset": "latest",
"client.id": consumer_client_id,
"fetch.max.bytes": "104857600", # 100MB
"max.poll.records": "3000",
"max.partition.fetch.bytes": "104857600", # 100MB
"fetch.min.bytes": "1048576", # 1MB
"request.timeout.ms": "25000", # 25 seconds timeout
"session.timeout.ms": "25000", # 25 seconds timeout
"default.api.timeout.ms": "25000", # 25 seconds timeout
"connections.max.idle.ms": "30000", # 30 seconds
"max.poll.interval.ms": "300000", # 5 minutes
"send.buffer.bytes": "52428800", # 50MB
"receive.buffer.bytes": "52428800", # 50MB
}
super().__init__(
consumer_config=consumer_config,
topics=topics,
with_metadata=False,
commit_offset_in_finalize=True,
redistribute=redistribute,
redistribute_num_keys=np.int32(redistribute_num_keys),
)
# pipeline logic
pipeline = BasePipeline(pipeline_options)
input_collection = pipeline | "Read from Kafka" >>
ReadFromPlaintextKafkaSource(
broker_addresses=pipeline_options.broker_address,
topics=["some_topic"],
job_name=pipeline_options.job_name,
)
```
When triggered with [update
method](https://cloud.google.com/dataflow/docs/guides/updating-a-pipeline),
errors occurred like this.
### stacktrace
<img width="1245" height="618" alt="Image"
src="https://github.com/user-attachments/assets/3a675e25-c7e7-4c07-b7f7-c22bd0af1453";
/>
```
Error message from worker: generic::unknown: java.lang.NullPointerException:
Null topic
org.apache.beam.sdk.io.kafka.AutoValue_KafkaSourceDescriptor.<init>(AutoValue_KafkaSourceDescriptor.java:36)
org.apache.beam.sdk.io.kafka.KafkaSourceDescriptor.create(KafkaSourceDescriptor.java:122)
org.apache.beam.sdk.io.kafka.SchemaUserTypeCreator$SchemaCodeGen$PDZ5LUq0.create(Unknown
Source)
org.apache.beam.sdk.schemas.FromRowUsingCreator.apply(FromRowUsingCreator.java:102)
org.apache.beam.sdk.schemas.FromRowUsingCreator.apply(FromRowUsingCreator.java:46)
org.apache.beam.sdk.schemas.SchemaCoder.decode(SchemaCoder.java:126)
org.apache.beam.sdk.coders.Coder.decode(Coder.java:159)
org.apache.beam.sdk.coders.LengthPrefixCoder.decode(LengthPrefixCoder.java:64)
org.apache.beam.sdk.coders.KvCoder.decode(KvCoder.java:83)
org.apache.beam.sdk.coders.KvCoder.decode(KvCoder.java:78)
org.apache.beam.sdk.values.WindowedValues$FullWindowedValueCoder.decode(WindowedValues.java:602)
org.apache.beam.sdk.values.WindowedValues$FullWindowedValueCoder.decode(WindowedValues.java:593)
org.apache.beam.sdk.values.WindowedValues$FullWindowedValueCoder.decode(WindowedValues.java:539)
org.apache.beam.sdk.fn.data.BeamFnDataInboundObserver.multiplexElements(BeamFnDataInboundObserver.java:231)
org.apache.beam.fn.harness.control.ProcessBundleHandler.processBundle(ProcessBundleHandler.java:527)
```
### Reproduction Steps
1. Create a streaming pipeline using Python Beam with Kafka IO
2. Run the pipeline on Dataflow
3. Update the pipeline (Update, not Drain)
4. Confirm NPE
### Analysis
1. AutoValueSchema field order inconsistency
* ReflectUtils.getMethods() may return different orders depending on
JVM implementation
* Mismatch between KafkaSourceDescriptor.create() parameter order and
the field order determined by AutoValueSchema
2. Occurs only during Dataflow Job Update
* Schema conversion is required during state restoration
* Drain does not restore state, so the issue does not occur
### Description
I doubt this error occurs because of this temporary code.
https://github.com/apache/beam/blob/a0831e0d4b1a36f8dd1d9c16ef388c02c6620e1a/sdks/java/io/kafka/src/main/java/org/apache/beam/sdk/io/kafka/KafkaSourceDescriptor.java#L110-L130
### Issue Priority
Priority: 2 (default / most bugs should be filed as P2)
### Issue Components
- [x] Component: Python SDK
- [x] Component: Java SDK
- [ ] Component: Go SDK
- [ ] Component: Typescript SDK
- [ ] Component: IO connector
- [ ] Component: Beam YAML
- [ ] Component: Beam examples
- [ ] Component: Beam playground
- [ ] Component: Beam katas
- [ ] Component: Website
- [ ] Component: Infrastructure
- [ ] Component: Spark Runner
- [ ] Component: Flink Runner
- [ ] Component: Samza Runner
- [ ] Component: Twister2 Runner
- [ ] Component: Hazelcast Jet Runner
- [x] Component: Google Cloud Dataflow Runner
--
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]
