You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Javed, Haseeb" <ja...@buckeyemail.osu.edu> on 2018/03/16 20:22:10 UTC
Kafka Consumer v1.0 terminates prematurely
Hello,
I have a test cluster for Kafka consisting of one broker, one producer and one consumer, each on a separate node. Kafka broker and producer are v1.0.0. I am able to use the ProducerPerformance benchmark to write records to the broker. However, when I try to read the records using ConsumerPerformance benchmark with v1.0.0, the benchmark terminates prematurely with output suggesting there were no records to be read, as shown below:
start.time, end.time, data.consumed.in.MB, MB.sec, data.consumed.in.nMsg, nMsg.sec
2018-02-23 16:13:51:748, 2018-02-23 16:13:52:904, 0.0000, 0.0000, 0, 0.0000, 24, 1132, 0.0000, 0.0000
I tried to debug the issue further by observing what API Requests are received from the consumer by the broker and they goes something like this:
API_VERSIONS
METADATA
FIND_COORDINATOR
API_VERSIONS
JOIN_GROUP
SYNC_GROUP
OFFSET_FETCH
API_VERSIONS
LIST_OFFSETS
LIST_OFFSETS
OFFSET_COMMIT
LEAVE_GROUP
From what I understand, there are no FETCH requests being sent by the Consumer and therefore no records sent by the broker as a response. For some odd reason, the Consumer thinks that it does not need to read any records. However, when I run the same Consumer Performance benchmark with v0.11.1.0, I am successfully able to read data. The sequence of API requests receied at the broker end correctly indicate that a number of FETCH requests were being sent by the Consumer after it had already received the LIST_OFFSETS response from the broker, as shown by API requests sequence received at the broker end below:
API_VERSIONS
METADATA
FIND_COORDINATOR
API_VERSIONS
JOIN_GROUP
SYNC_GROUP
OFFSET_FETCH
API_VERSIONS
LIST_OFFSETS
FETCH
...
FETCH
OFFSET_COMMIT
LEAVE_GROUP
What could be causing this issue?
- Haseeb