[
https://issues.apache.org/jira/browse/FLINK-13400?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17350916#comment-17350916
]
Jark Wu commented on FLINK-13400:
---------------------------------
Hi [~twalthr], I think this will be a large PR/task if we want to fix the hive
dependency. Currently, we tested some Hive connector features in SQL Client. On
the other hand, SQL Client also relies on Hive catalog/module to testing
statements about catalog and modules. The tests are mainly located in
{{flink-table/flink-sql-client/src/test/resources/sql/}} and {{DependencyTest}}
, etc...
If we want to remove Hive dependency, we may first need to:
1. migrate the Hive-related tests into Hive connector (Hive connector will have
a flink-sql-client dependency with test scope).
2. support testing catalog and modules, e.g. FLINK-17909, and replace them with
the hive-related tests in flink-sql-client.
3. remove hive dependency in flink-sql-client.
So this would be a huge work and would be better break it into sub-tasks.
I'm fine with this refactoring. However, I just have some concerns that this
would cost us a lot of time but the benefit maybe small. From my point of view,
the problem of this issue is just the mess Hive dependencies in
flink-sql-client. Could we just update all Hive dependencies to test scope, so
we don't worry about packaging them into sql client jar by mistake. And all the
"exclusions" list can be removed then (ideally). In this way, the list of Hive
dependencies will be small and easy to maintain. What do you think [~twalthr]?
cc [~lirui]
> Remove Hive and Hadoop dependencies from SQL Client
> ---------------------------------------------------
>
> Key: FLINK-13400
> URL: https://issues.apache.org/jira/browse/FLINK-13400
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Client
> Reporter: Timo Walther
> Priority: Critical
> Labels: auto-unassigned, stale-critical
> Fix For: 1.14.0
>
>
> 340/550 lines in the SQL Client {{pom.xml}} are just around Hive and Hadoop
> dependencies. Hive has nothing to do with the SQL Client and it will be hard
> to maintain the long list of exclusion there. Some dependencies are even in
> a {{provided}} scope and not {{test}} scope.
> We should remove all dependencies on Hive/Hadoop and replace catalog-related
> tests by a testing catalog. Similar to how we tests source/sinks.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
