Challenging priority : HIVE-2910 Improve the HWI interface

[Bertrand Dechoux <de...@gmail.com>]

Hi,

Here is my stand. Hive provides a dsl to easily explore data contained in
hadoop with limited experience with java and MapReduce.
And Hive Web Interface provides an easy exposition : users need only a
browser and the hadoop cluster can be well 'fire-walled' because the
communication is only through http.

I saw the work of HIVE-2910. Of course, it could still be improved. But it
was needed and it is a great first step. I don't see it as a minor
improvement. That is a great way for promoting the HWI.
As such, I think it is a shame the patch was only applied for the 0.10
version and not for the 0.9.1. Is there any reason for that? Is there any
change that it could be added to the 0.9.0? Do you have any estimate for
the release date of the 0.9.1 version?

Of course, there is at least an alternative : Beeswax within Hue but CDH is
a big dependency that can not just be added because the end user requires a
decent interface.

Thanks in advance for all feedback

Bertrand

Re: Challenging priority : HIVE-2910 Improve the HWI interface

[Bertrand Dechoux <de...@gmail.com>]

And now I have to apologize. I was one version late for hive (0.8.1). And
the version 0.9 does include HWI with bootstrap. The jira must be
misleading or I don't understand what the issue is about...

Bertrand

https://issues.apache.org/jira/browse/HIVE-2910

On Wed, Jul 25, 2012 at 6:15 PM, Bertrand Dechoux <de...@gmail.com>wrote:

> Great answer. Thanks a lot.
>
> 1) I understand the concern with branches but I quickly reviewed the
> change for 0.9.1 and not everything seemed to be a bug patch.
> So I thought : why not ask about HIVE-2910.
>
> 2) I wasn't sure about that, it seems logical though. That's a great news.
> I will definitely try using the upgraded war. And that indeed would solve
> my problem.
>
> 3) The demography of users have changed, maybe not the persons themselves.
> Right now I prefer the CLI but I can't allow everybody to do the same.
>
> 4) The priority depends on the users. I understand your point, however my
> concern now is to allow multiple users to 'poke' huge files easily. With
> that point of view, HWI is critical.
>
> Thanks again for the well constructed feedback.
>
> Bertrand
>
>
> On Wed, Jul 25, 2012 at 5:54 PM, Edward Capriolo <ed...@gmail.com>wrote:
>
>> Generally we only apply patches to trunk. Thus maintaining branches
>> becomes to much trouble for us. You have to remember that most hive
>> major versions have no actual major changes. Most everything is hidden
>> behind a query language. The only changes that have to be done
>> carefully are changes to the metastore that add things like views or
>> security.
>>
>> As for HIVE-2910, again this change made no changes to any core code.
>> HWI tries hard not to touch internals. As result I am almost 100%
>> positive you can replace hive-hwi-0.9.0.war with hive-hwi-0.10.0.war
>> in your hive lib directory, so you really do not need to wait for a
>> release just to get the new HWI.
>>
>> As for improving HWI, hwi is a funny story. The open source hive had
>> no web interface, at the time the type of users that were into hive
>> and hadoop (since it was so new) where hardcore coders and distributed
>> systems experts. Those types do not typically use web interfaces. Fast
>> forward to now and EVERYONE and their grandmother uses hadoop and
>> suddenly web interfaces are actually an important selling point!
>>
>> (I wrote hwi btw) In hive I have moved onto writing UDFs,
>> StorageHandlers, InputFormats, and doing patch reviews. The web
>> interface is low hanging fruit. If anyone wants to add a feature to
>> HWI (like the bootstrap one or others), open a jira and I will
>> promptly review and commit. Much like the features in the CLI
>> (autocomplete and .hiverc) they did not get better magically, but they
>> are again easy low hanging fruit that almost any coded can take care
>> of.
>>
>> On Wed, Jul 25, 2012 at 11:25 AM, Bertrand Dechoux <de...@gmail.com>
>> wrote:
>> > Hi,
>> >
>> > Here is my stand. Hive provides a dsl to easily explore data contained
>> in
>> > hadoop with limited experience with java and MapReduce.
>> > And Hive Web Interface provides an easy exposition : users need only a
>> > browser and the hadoop cluster can be well 'fire-walled' because the
>> > communication is only through http.
>> >
>> > I saw the work of HIVE-2910. Of course, it could still be improved. But
>> it
>> > was needed and it is a great first step. I don't see it as a minor
>> > improvement. That is a great way for promoting the HWI.
>> > As such, I think it is a shame the patch was only applied for the 0.10
>> > version and not for the 0.9.1. Is there any reason for that? Is there
>> any
>> > change that it could be added to the 0.9.0? Do you have any estimate
>> for the
>> > release date of the 0.9.1 version?
>> >
>> > Of course, there is at least an alternative : Beeswax within Hue but
>> CDH is
>> > a big dependency that can not just be added because the end user
>> requires a
>> > decent interface.
>> >
>> > Thanks in advance for all feedback
>> >
>> > Bertrand
>>
>
>
>
> --
> Bertrand Dechoux
>



-- 
Bertrand Dechoux

Re: Challenging priority : HIVE-2910 Improve the HWI interface

[Bertrand Dechoux <de...@gmail.com>]

Great answer. Thanks a lot.

1) I understand the concern with branches but I quickly reviewed the change
for 0.9.1 and not everything seemed to be a bug patch.
So I thought : why not ask about HIVE-2910.

2) I wasn't sure about that, it seems logical though. That's a great news.
I will definitely try using the upgraded war. And that indeed would solve
my problem.

3) The demography of users have changed, maybe not the persons themselves.
Right now I prefer the CLI but I can't allow everybody to do the same.

4) The priority depends on the users. I understand your point, however my
concern now is to allow multiple users to 'poke' huge files easily. With
that point of view, HWI is critical.

Thanks again for the well constructed feedback.

Bertrand

On Wed, Jul 25, 2012 at 5:54 PM, Edward Capriolo <ed...@gmail.com>wrote:

> Generally we only apply patches to trunk. Thus maintaining branches
> becomes to much trouble for us. You have to remember that most hive
> major versions have no actual major changes. Most everything is hidden
> behind a query language. The only changes that have to be done
> carefully are changes to the metastore that add things like views or
> security.
>
> As for HIVE-2910, again this change made no changes to any core code.
> HWI tries hard not to touch internals. As result I am almost 100%
> positive you can replace hive-hwi-0.9.0.war with hive-hwi-0.10.0.war
> in your hive lib directory, so you really do not need to wait for a
> release just to get the new HWI.
>
> As for improving HWI, hwi is a funny story. The open source hive had
> no web interface, at the time the type of users that were into hive
> and hadoop (since it was so new) where hardcore coders and distributed
> systems experts. Those types do not typically use web interfaces. Fast
> forward to now and EVERYONE and their grandmother uses hadoop and
> suddenly web interfaces are actually an important selling point!
>
> (I wrote hwi btw) In hive I have moved onto writing UDFs,
> StorageHandlers, InputFormats, and doing patch reviews. The web
> interface is low hanging fruit. If anyone wants to add a feature to
> HWI (like the bootstrap one or others), open a jira and I will
> promptly review and commit. Much like the features in the CLI
> (autocomplete and .hiverc) they did not get better magically, but they
> are again easy low hanging fruit that almost any coded can take care
> of.
>
> On Wed, Jul 25, 2012 at 11:25 AM, Bertrand Dechoux <de...@gmail.com>
> wrote:
> > Hi,
> >
> > Here is my stand. Hive provides a dsl to easily explore data contained in
> > hadoop with limited experience with java and MapReduce.
> > And Hive Web Interface provides an easy exposition : users need only a
> > browser and the hadoop cluster can be well 'fire-walled' because the
> > communication is only through http.
> >
> > I saw the work of HIVE-2910. Of course, it could still be improved. But
> it
> > was needed and it is a great first step. I don't see it as a minor
> > improvement. That is a great way for promoting the HWI.
> > As such, I think it is a shame the patch was only applied for the 0.10
> > version and not for the 0.9.1. Is there any reason for that? Is there any
> > change that it could be added to the 0.9.0? Do you have any estimate for
> the
> > release date of the 0.9.1 version?
> >
> > Of course, there is at least an alternative : Beeswax within Hue but CDH
> is
> > a big dependency that can not just be added because the end user
> requires a
> > decent interface.
> >
> > Thanks in advance for all feedback
> >
> > Bertrand
>



-- 
Bertrand Dechoux

Re: Challenging priority : HIVE-2910 Improve the HWI interface

[Edward Capriolo <ed...@gmail.com>]

Generally we only apply patches to trunk. Thus maintaining branches
becomes to much trouble for us. You have to remember that most hive
major versions have no actual major changes. Most everything is hidden
behind a query language. The only changes that have to be done
carefully are changes to the metastore that add things like views or
security.

As for HIVE-2910, again this change made no changes to any core code.
HWI tries hard not to touch internals. As result I am almost 100%
positive you can replace hive-hwi-0.9.0.war with hive-hwi-0.10.0.war
in your hive lib directory, so you really do not need to wait for a
release just to get the new HWI.

As for improving HWI, hwi is a funny story. The open source hive had
no web interface, at the time the type of users that were into hive
and hadoop (since it was so new) where hardcore coders and distributed
systems experts. Those types do not typically use web interfaces. Fast
forward to now and EVERYONE and their grandmother uses hadoop and
suddenly web interfaces are actually an important selling point!

(I wrote hwi btw) In hive I have moved onto writing UDFs,
StorageHandlers, InputFormats, and doing patch reviews. The web
interface is low hanging fruit. If anyone wants to add a feature to
HWI (like the bootstrap one or others), open a jira and I will
promptly review and commit. Much like the features in the CLI
(autocomplete and .hiverc) they did not get better magically, but they
are again easy low hanging fruit that almost any coded can take care
of.

On Wed, Jul 25, 2012 at 11:25 AM, Bertrand Dechoux <de...@gmail.com> wrote:
> Hi,
>
> Here is my stand. Hive provides a dsl to easily explore data contained in
> hadoop with limited experience with java and MapReduce.
> And Hive Web Interface provides an easy exposition : users need only a
> browser and the hadoop cluster can be well 'fire-walled' because the
> communication is only through http.
>
> I saw the work of HIVE-2910. Of course, it could still be improved. But it
> was needed and it is a great first step. I don't see it as a minor
> improvement. That is a great way for promoting the HWI.
> As such, I think it is a shame the patch was only applied for the 0.10
> version and not for the 0.9.1. Is there any reason for that? Is there any
> change that it could be added to the 0.9.0? Do you have any estimate for the
> release date of the 0.9.1 version?
>
> Of course, there is at least an alternative : Beeswax within Hue but CDH is
> a big dependency that can not just be added because the end user requires a
> decent interface.
>
> Thanks in advance for all feedback
>
> Bertrand