Error parsing host address when executing queries

[mmajis]

Describe the bug

Running a push query in ksqldb v0.7.1 errors out with Error parsing host address http://<docker network host address>:8088/. Expected format host:port.

Having had a quick look at the code, it looks like StreamsConfig.APPLICATION_SERVER_CONFIG has wrong format when executing the query.

To Reproduce

1. Clone this repo: https://github.com/mmajis/ksqldb-query-bug-repro
2. Put your confluent cloud api key and secret in a .env file inside the repo to configure Confluent Cloud connection:

CC_API_KEY=<your api key>
CC_API_SECRET=<your api secret>
CC_ADDRESS=<your confluent cloud bootstrap server address and port>

3. docker-compose up -d
4. docker-compose exec ksqldb-cli bash
5. LOG_DIR=. ksqldb http://ksql-server:8088
6. Create a stream:

  CREATE STREAM my_stream (my_value VARCHAR) WITH (
  VALUE_FORMAT = 'JSON',
  KAFKA_TOPIC = 'my_stream_topic', 
  PARTITIONS=1, 
  REPLICAS=3);

7. SELECT * FROM my_stream EMIT CHANGES;

Expected behavior

Expected the query SELECT * FROM my_stream EMIT CHANGES; to start running.

Actual behaviour

The ksqlDB-cli outputs this error: Error parsing host address http://<docker-compose_ksql-server_hostname>:8088/. Expected format host:port.

Stack trace in ksqldb-cli log:

[2020-02-27 18:53:05,739] ERROR Error parsing host address http://1130da4e15fb:8088/. Expected format host:port.
org.apache.kafka.streams.KafkaStreams.parseHostInfo(KafkaStreams.java:818)
org.apache.kafka.streams.KafkaStreams.<init>(KafkaStreams.java:707)
org.apache.kafka.streams.KafkaStreams.<init>(KafkaStreams.java:584)
io.confluent.ksql.query.KafkaStreamsBuilderImpl.buildKafkaStreams(KafkaStreamsBuilderImpl.java:43)
io.confluent.ksql.query.QueryExecutor.buildTransientQuery(QueryExecutor.java:165)
io.confluent.ksql.engine.EngineExecutor.executeQuery(EngineExecutor.java:129)
io.confluent.ksql.engine.KsqlEngine.executeQuery(KsqlEngine.java:215)
io.confluent.ksql.rest.server.resources.streaming.StreamedQueryResource.handlePushQuery(StreamedQueryResource.java:269)
io.confluent.ksql.rest.server.resources.streaming.StreamedQueryResource.handleStatement(StreamedQueryResource.java:208)
io.confluent.ksql.rest.server.resources.streaming.StreamedQueryResource.streamQuery(StreamedQueryResource.java:161)
sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
java.lang.reflect.Method.invoke(Method.java:498)
org.glassfish.jersey.server.model.internal.ResourceMethodInvocationHandlerFactory.lambda$static$0(ResourceMethodInvocationHandlerFactory.java:52)
...omitted tens of glassfish and jetty lines...

Additional context

This used to work with v0.6.0.

There's a separate branch local_kafka in the demo repo where the querying works. So the issue seems to be related to the configuration necessary to use Confluent Cloud.

[terryf82]

I also ran into the same issue, but while using ksqlDB 0.7

I'm not sure if it's relevant, but I've only seen it when creating streams that are JSON formatted like the example above. Have not encountered it when using a purely AVRO pipeline.

Use of Confluent Cloud seems to be the common thread.

[yfaykya]

Hi - Is there any work arounds? I don't use the cloud at all. I just spin a container of cp-ksql-server:5.4.1 and then connect with confluentinc/cp-ksql-cli

I can create streams but if I ever try SELECT * FROM my_stream EMIT CHANGES; it gives error as above with the ksql container ID as the host.

[yfaykya]

I sorted it by removing trailing slash from KSQL_LISTENERS.
I was passing in KSQL_LISTENERS=http://0.0.0.0:8088/

[mmajis]

Yes, it works without the trailing slash there.

Notably this page has some trailing slashes for KSQL_LISTENERS in the examples: https://github.com/confluentinc/ksql/blob/master/docs-md/operate-and-deploy/installation/install-ksqldb-with-docker.md

[terryf82]

Confirming that removing the trailing slash also worked here.

@rodesai this was the issue I mentioned via Slack last week. Seems like a documentation bug rather than a technical one.

[mmajis]

Since the value for KSQL_LISTENERS looks like a URI I think it'd be good for developer experience that the parsing could handle a trailing slash.

[agavra]

Looks like the fix didn't totally fix the issue - it fixed a similar issue