[
https://issues.apache.org/jira/browse/SPARK-2106?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14028699#comment-14028699
]
Cheng Hao commented on SPARK-2106:
----------------------------------
Oh, I see your point.
Actually I was suggesting is we can make the HiveContext more generic, hence
the other applications (SharkCLI) can use the HiveContext directly without
creating the sub class of HiveContext to override things (SessionState creating
etc.).
In order to achieving that, we may need to:
# Provide contruction method & arguments(for example:SessionState, this can be
also with default value)
# Make the QueryExecution visible. (currently most of methods in
HiveContext.QueryExecution are protected methods and they also tightly coupled)
# Refactor the API interfaces
{code:title=HiveContext.QueryExecution|borderStyle=solid}
runHive(cmd: String, ..): Seq[String] = {
...
case _ =>
sessionState.out.println(tokens(0) + " " + cmd_1)
Seq(proc.run(cmd_1).getResponseCode.toString) // will the (errCode,
Seq[String]) be more reasonable?
...
}
{code}
Of course, this is in lower priority, actually the [Add Shark CLI
Support|https://github.com/amplab/shark/pull/337] is done with creating the sub
class of HiveContext.
> Unify the HiveContext
> ---------------------
>
> Key: SPARK-2106
> URL: https://issues.apache.org/jira/browse/SPARK-2106
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Reporter: Cheng Hao
>
> I've been working on CLI for Catalyst, and from the CLI point of view,
> HiveContext may be required some changes as:
> - SessionState management
> `SessionState` instance creation & initialization should be done within the
> wrappers, e.g. in `SharkCliDriver` or `CLIService` etc. cause they know
> better how to load the user configuration and logger redirection etc. And in
> HiveContext, it can retrieve SessionState by calling `SessionState.get()`.
> - HiveContext API may not be enough for CLI
> 1) Retrieving the schema from the output of `SELECT`; but the internal class
> `QueryExecution` is hidden for CLI.
> 2) Retriveing the HQL result in CLI, besides the string based result, CLI
> also need to know the error code, as well as the call stack if exceptions
> thrown.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
