hello all,

i just wanted to point out a potential issue in kafka-clients 0.10.1.1

i was using spark-sql-kafka-0-10, which is spark structured streaming
integration for kafka. it depends on kafka-clients 0.10.0.1 but since my
kafka servers are 0.10.1.1 i decided to upgrade kafka-clients to 0.10.1.1
also. i was not able to read from kafka in spark reliably. the issue seemed
to be that the kafka consumer got the latest offsets wrong. after
downgrading kafka-clients back to 0.10.0.1 it all worked correctly again.

did the behavior of KafkaConsumer.seekToEnd change between 0.10.0.1 and
0.10.1.1?

for the original discussion see here:
https://www.mail-archive.com/user@spark.apache.org/msg61290.html


i think the relevant code in spark is here:
https://github.com/apache/spark/blob/master/external/kafka-0-10-sql/src/main/scala/org/apache/spark/sql/kafka010/KafkaSource.scala#L399

best,
koert

Reply via email to