Josh Elser created HBASE-26599:
----------------------------------
Summary: Netty exclusion through ZooKeeper not effective as
intended
Key: HBASE-26599
URL: https://issues.apache.org/jira/browse/HBASE-26599
Project: HBase
Issue Type: Bug
Components: dependencies
Affects Versions: 2.4.8
Reporter: Josh Elser
Assignee: Josh Elser
Picking up where [~psomogyi] has been digging this week. We've been seeing an
issue where MiniDFS-based tests fail to start due to missing netty classes.
HBASE-25969 seems to have intended to remove transitive Netty but was
ineffective (at least for hadoop.profile=3.0). The dependency exclusion was for
{{io.netty:netty}} and {{io.netty:netty-all}} but ZooKeeper 3.5.7 transitively
depends on {{netty-handler}} and {{netty-transport-native-epoll}} (per
[https://search.maven.org/artifact/org.apache.zookeeper/zookeeper/3.5.7/jar)]
The funny part is that we _should_ have seen failures in any hbase unit test
using MiniDFS because we excluded netty and netty-all in HBASE-25969, but
because we missed the exclusions, we still keep running.
The intent of HBASE-25969 was good, but I think we need to revisit the
execution. We need netty dependencies on the scope=test classpath. We just want
to keep them off the scope=compile classpath (out of our client and server
jars).
disclaimer: I have not yet looked at 2.5.x or master yet to see if this also
affects them.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
