hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ashutosh Chauhan <hashut...@apache.org>
Subject Re: hooks in metastore functions
Date Thu, 10 Mar 2011 18:32:09 GMT
Cool. I will prepare a patch and post on HIVE-2038.

Ashutosh
On Wed, Mar 9, 2011 at 16:59, John Sichi <jsichi@fb.com> wrote:
> Yeah, thinking about it more, they're likely to end up looking different.  The listener
should cover most possible repository changes, whereas HiveMetaHook is focused on a narrower
set of object definitions.
>
> JVS
>
> On Mar 9, 2011, at 1:48 PM, Ashutosh Chauhan wrote:
>
>> It might be possible to extend and modify the HiveMetaHook interface.
>> But, I think keeping them separate is better because MetaHook and
>> MetaStoreListener are interfaces for two different functionalities.
>> MetaHook is for communicating with external system if there is a need
>> for it. MetaStoreListener observe changes on metastore and run some
>> logic in response to those changes. What do you think?
>>
>> Ashutosh
>>
>> On Wed, Mar 9, 2011 at 13:36, John Sichi <jsichi@fb.com> wrote:
>>> Couldn't we reuse HiveMetaHook for this new purpose (with an instance loaded
via global config vs associated with the table handler)?
>>>
>>> JVS
>>>
>>> On Mar 8, 2011, at 2:12 PM, Ashutosh Chauhan wrote:
>>>
>>>> Hi all,
>>>>
>>>> I have a requirement that every time some change on metastore takes
>>>> place, we have some logic which needs to be run. For example, if a new
>>>> table is getting created in metastore I want to send a message to a
>>>> message bus. Easiest way for this to work is to add the logic in
>>>> createTable(). Control it by a hiveConf param and turn it off by
>>>> default. Alternative way is via hooks. Have this extra logic in hook
>>>> and then load and fire the hook if its available. Does anyone has an
>>>> opinion which of these two is preferable. Second one requires new hook
>>>> loading and execution logic. I am currently interested in four
>>>> functions: createTable() dropTable() addPartition() dropPartition().
>>>> Current, HiveMetaHook which exists in createTable() doesn't perfectly
>>>> fit the bill, since it is fired only when user expresses it in his
>>>> create table statement (i.e., if he has specified a storage handler)
>>>> Instead I want to have this logic always run.
>>>> If it is unclear, let me know, I can post the code  which can
>>>> demonstrate my usecase.
>>>>
>>>> Ashutosh
>>>
>>>
>
>

Mime
View raw message