Sam Meder created KAFKA-1062:
--------------------------------
Summary: Reading topic metadata from zookeeper leads to
incompatible ordering of partition list
Key: KAFKA-1062
URL: https://issues.apache.org/jira/browse/KAFKA-1062
Project: Kafka
Issue Type: Bug
Components: consumer
Affects Versions: 0.8
Reporter: Sam Meder
Assignee: Neha Narkhede
The latest consumer changes to read data from Zookeeper during rebalance have
made the consumer rebalance code incompatible with older versions (making
rolling upgrades without downtime hard). The problem relates to how partitions
are ordered. The old code seems to have returned the partitions sorted:
... rebalancing the following partitions: ArrayBuffer(0, 1, 2, 3, 4, 5, 6, 7,
8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19) for topic produce-indexable-views
with consumers: ...
the new code instead uses:
... rebalancing the following partitions: List(0, 5, 10, 14, 1, 6, 9, 13, 2,
17, 12, 7, 3, 18, 16, 11, 8, 19, 4, 15) for topic produce-indexable-views with
consumers: ...
This causes new consumers and old consumers to claim the same partitions. I
realize that this may not be a big deal (although painful for us since it
disagrees with our deployment automation) since the code wasn't officially
released, but it seems simple enough to sort the partitions if you'd take such
a patch.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
