Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Accumulo >> mail # dev >> GSoC proposal: Ambari for Accumulo - forever : )

Copy link to this message
Re: GSoC proposal: Ambari for Accumulo - forever : )
On 4/25/13 11:40 AM, Andres Danter wrote:
> Excellent feedback, Keith. Thanks.
> I should have mentioned the REST API in the proposal.  I'm not sure if I
> can use it for the CLI.  I will certainly try, since I would like to adhere
> to the Ambari architecture as much as possible.
> As for your other points:
> *3.1.1 loggers do not exists in 1.5 ... so would only need to consider them
> if working w/ 1.4 .  I suppose you would want to work Accumulo 1.4, since
> that the current stable release.*
> Yes. I was going to work with Accumulo 1.4, but it is good to know that the
> loggers are not supported in 1.5.
Not so much "not supported" in 1.5, but more "not necessary". The
write-ahead log (mechanism to ensure data is not lost during ingest) is
backed by HDFS and managed by TServers in >=1.5 instead of a separate
Logger process (used local fs and handled replication internally) which
<=1.4 has.
> *3.1.3
>    Accumulo supports reading system config from xml local file and then
> zookeeper (whats set in zookeeper takes precedence).  The nice thing about
> setting something a config zookeeper, is that process restart is not
> required.*
> Good to know this too.  Does this only affect Zookeeper?  I should
> distinguish between processes that may need restarts to load a
> configuration vs those that do not.  Is the Zookeeper process the only one
> like that?
> *For section 3.2, are you proposing adding features to the existing Ambari
> GUI?  If so, are the Accumulo specific?*
> I am proposing to add features to the existing Ambari GUI.  The idea here
> is that something like user access controls can be generalized so that any
> Hadoop component being managed by Ambari is able to utilize them.  I'm
> hoping that the Ambari PMC is OK with me doing this.  I do not want to
> introduce any Accumulo-specific features that would make no sense for other
> components.
> *
> It seems that Ambari relies on RPMs.  Its possible you may have to spend
> time creating Accumulo RPMs that are useful for Ambari. You may want to
> think about allocating some time for this.  Christopher and John can give
> you more info about the current state of Accumulo RPMs.*
> This is a very good point.  While building RPMs is not difficult, it does
> require time to setup the configuration and test the builds.  I will
> include time for this in the proposal.
> Thanks again, Keith.
> Andres
> On Thu, Apr 25, 2013 at 10:54 AM, Keith Turner <[EMAIL PROTECTED]> wrote:
>> About the REST API, I meant to ask if this could be used from the command
>> line currently?  Or is too cumbersome?
>> On Thu, Apr 25, 2013 at 10:09 AM, Andres Danter <[EMAIL PROTECTED]> wrote:
>>> Yup.  Looks like that's the way to go.  Here is the link to the document.
>>> Let me know if you have any problems accessing it.
>> https://docs.google.com/file/d/0BxBY29P7A1RgVnZscGcycExDbEU/edit?usp=sharing
>>> On Thu, Apr 25, 2013 at 9:51 AM, Keith Turner <[EMAIL PROTECTED]> wrote:
>>>> Nothing.  I suppose the mailing list is stripping it off.  Could post
>> it
>>> as
>>>> an attachment to the ticket or put it on google drive or dropbox and
>>> send a
>>>> link.
>>>> On Thu, Apr 25, 2013 at 9:44 AM, Andres Danter <[EMAIL PROTECTED]>
>>> wrote:
>>>>> Weird.  I see the attachment on the email I sent, which means that it
>>>>> might have been stripped off.  Here it is again.  Let me know if you
>>>> still
>>>>> don't see it.
>>>>> Thanks,
>>>>> Andres
>>>>> On Thu, Apr 25, 2013 at 9:18 AM, Keith Turner <[EMAIL PROTECTED]>
>>> wrote:
>>>>>> was there supposed to be an attachment?
>>>>>> On Thu, Apr 25, 2013 at 2:19 AM, Andres Danter <[EMAIL PROTECTED]>
>>>> wrote:
>>>>>>> Hi Billie,
>>>>>>> Here is a draft of my proposal.  I've yet to write the technical
>>>>>> approach
>>>>>>> section (have not idea what I'm going to put there) and the "about
>>> me"
>>>>>>> section.  I definitely appreciate any feedback from you and anyone