dhis2-users team mailing list archive

[Jason Pickering <jason.p.pickering@xxxxxxxxx>]

Yes, in this context, a materialized view is simply a table which is
produced from an SQL view. The advantage of course is that materialized
views can be indexed, etc and are normally much faster to serve than
dynamic SQL views.

The one thing I did not mention, is the refreshment of the materialized
views. As Randy pointed out, we have scheduled the refresh of the views as
a cron job. Ideally, this would be something which DHIS2 would do. If we
were to represent the 20 or so views as normal DHIS2 SQL views, there is no
native way to refresh them, and this needs to occur each night after the
data mart runs. So, ideally, scheduling of the regeneration of the SQL
views (and resource tables) should be an option as well. There is a work
around using CURL to force the regeneration, but again ,this is a bit of a
kludge as well. Hope this can get included at some point in time.


I think the real fix in this case would be.

1) Implement scheduling of SQL views.
2) If users want to use custom Postgresql functions, they could do this,
and they could be triggered by DHIS with "SELECT
my_custom_stored_procedure();" with the scheduler. This would eliminate the
need for a cron job from the system.
3) Allow for users to choose certain tables which should be available as
resources, for instance, "_orgunitstructure"  (Bob's point) or
"_my_custom_table", which might be some external table or materialized
view. DHIS would not do anything other than to produce JSON, HTML, Excel or
whatever from this table. Not really sure why this is not RESTFUL? Isn't
this the same thing as what is being done with the SQL views themselves?



 Regards,
Jason





On Thu, Dec 20, 2012 at 10:51 PM, Lars Helge Øverland
<larshelge@xxxxxxxxx>wrote:

> Okay. So with materialized view you are referring to a plain db table,
> right? I thought you meant creating a view with sql something like "create
> materialized view" but I can see that has not been implemented directly in
> postgres yet.
>
> Anyway I think creating a dhis sql view per table is a reasonable
> work-around. It requires you only to create 20 objects, one time.
>
> One alternative would be to create some sort of dhis sql report. But it
> would be very similar in nature and require some development for 2.11.
>  Another alternative would be just to expose database tables through the
> web api but that does not sound very nice or restful to me.
>
> So if you are willing to create a sql view per table you can achieve 1.
>
> Re 2 you can simply point directly to the sql views through the web api,
> for instance to the demo like this, for xml:
>
> http://apps.dhis2.org/demo/api/sqlViews/dI68mLkP1wN/data.xml<http://apps.dhis2.org/demo/api/sqlViews/dI68mLkP1wN/data>
>
> for csv:
>
> http://apps.dhis2.org/demo/api/sqlViews/dI68mLkP1wN/data.csv
>
> I have committed and backported to 2.10 representations for Excel and
> HTML, you can reach them after upgrading by using the extensions .xls and
> .html .
>
> For web pages you can use plain links like this:
>
> <a href="http://apps.dhis2.org/demo/api/sqlViews/dI68mLkP1wN/data.xml<http://apps.dhis2.org/demo/api/sqlViews/dI68mLkP1wN/data>">Download
> as XML</a> |
> <a href="http://apps.dhis2.org/demo/api/sqlViews/dI68mLkP1wN/data.csv";>CSV</a>
> | etc
>
> For 3, yes you can easily include these as resources. Just add a new one,
> give it a name, select "external URL" as type and point to the sql view
> URLs above.
>
>
> For authentication you might get some tips from here:
>
> http://dhis2.org/doc/snapshot/en/implementer/html/ch08s02.html#d5e600
>
>
> You can download 2.10 latest from here in half an hour:
>
>
> http://apps.dhis2.org/ci/job/dhis-web-2.10/lastSuccessfulBuild/artifact/dhis-2/dhis-web/dhis-web-portal/target/dhis.war
>
>
>
>
>
> On Thu, Dec 20, 2012 at 5:28 PM, Jason Pickering <
> jason.p.pickering@xxxxxxxxx> wrote:
>
>> Hi Randy and Lars,
>> The fundemental problem is that view which they were using are linked to
>> the resource tables. These tables need to be regenerated, and when they are
>> regenerated, the tables are dropped and recreated. If views are linked to
>> these tables, the resource table regeneration cannot take place, because
>> the database forbids the table to be dropped, because there are linked
>> views. DHIS2 fails silently and only says "Resource tables generating..."
>> or something like that. I think we have described this problem on the list
>> before and I am sure you are aware of it.
>>
>> So, to get around this problem, we use stored procedures
>> to materialize the views, and then as Randy indicates, users can get to
>> them through ODBC.
>>
>> I guess it could be done Randy by creating an SQL View of the
>> materialized view. This does not really make a whole lot of sense, but it
>> could be done. So, if you create an SQL view in DHIS2 of "SELECT * FROM
>> _materalized view;" you will get a new materialized view available, as Lars
>> points out, through the API. A bit wasteful really but certainly possible.
>>
>> The "elegant" part which Randy mentions is that we do not define many
>> seperate views, but dynamically define the tables based on data element
>> groups. The views are very similar, they just differ in the data elements
>> which should be presented to the users. So, instead of maintaining like
>> 20 separate similar views, we have just done the same thing with one single
>> stored procedure instead. Second option then would be to have
>> these separate 20 (or whatever the number is) views as DHIS2 SQL views.
>> Again, not very elegant.
>>
>> Third option which I see (which I think is the best option) is to reflect
>> SQL views out of DHIS2 without ever materializing them. I think there
>> should be an option for "materialize" for each SQL view. If it is not
>> checked, the view would be produced dynamically at the time of request.
>> Again, not a very elegant solution for this problem, because we would still
>> need 20 separate "SELECT * FROM _materialized_view1", "SELECT * FROM
>> _materialized_view2", but at least we would not write the materalized view
>> twice.
>>
>> Fourth option (maybe a bit limiting) is to create the same SQL View type
>> of functionality for tables, lets call it a table reflector, which would
>> simply reflect a table as JSON, XML, CSV or whatever. The user could be
>> presented with a list of tables in the system, and simply choose the one
>> you want (in our case, one of the materialized views) and then they would
>> be available somehow through the API.
>>
>> Fifth option, is some external report. In Nigeria, we use R for this
>> purpose actually, and it is relatively straightforward. Other things like
>> PHP, Python or Birt could be used pretty easily as well.
>>
>> Just a bit of a grab-bag, but maybe there will be some possible solution
>> in there.
>>
>> Regards,
>> Jason
>>
>>
>>
>>
>> On Thu, Dec 20, 2012 at 1:58 PM, Wilson,Randy <rwilson@xxxxxxx> wrote:
>>
>>>  In Rwanda we have found report tables and MyDatamart inadequate for
>>> the needs of many of the central level health programs.  What they want is
>>> a dump of all the datavalues in *specific* *dataelement and indicator
>>> groups* along with the related orgunit hierarchy.  We originally were
>>> doing this using SQL views, under Data Administration, however some of the
>>> views are built on one another so when the Datamart and Resource table
>>> procedures run they is unable to drop the views before re-creating them….
>>> And consequently neither process runs.****
>>>
>>> ** **
>>>
>>> Jason Pickering has helped us develop an elegant function that gets
>>> around the view issue, and creates materialized views for all dataelement
>>> groups: special tables that are refreshed every night through a chron job
>>> (just like the datamart).  This gives us exactly what the users want – they
>>> generally link to the tables with ODBC connection using an excel pivot
>>> table – and because they are only retrieving the dataelement groups they
>>> want the file is not too big.  Unfortunately, this only works on the local
>>> area network – refreshing the pivot tables over the network is likely to be
>>> too slow, and our security settings don’t enable a remote ODBC connection.
>>> ****
>>>
>>> ** **
>>>
>>> We are now beginning to explore the DHIS-2 API to create a portal for
>>> users to access DHIS-2 objects.  I see that SQL views are exposed through
>>> this API, but for the reasons stated above we can’t rely on them.  Here is
>>> where we need help:****
>>>
>>> **1.       **Is there a way to expose other tables in the API? ****
>>>
>>> **2.       **Does anyone have some sample HTML code that we could use
>>> to list these special tables (they all start with _*view*_) in the
>>> portal and download them as xls, csv or xml?  Similar to what is shown in
>>> the DHIS-2 portal demo for downloading tables.****
>>>
>>> **3.       **Could we include the links to these tables as resources in
>>> users’ dashboards?****
>>>
>>> ** **
>>>
>>> *Randy Wilson*****
>>>
>>> Senior HMIS and Data Use Advisor****
>>>
>>> Rwanda/IHSSP****
>>>
>>> Management Sciences for Health****
>>>
>>> BOX 371
>>> Kigali, Rwanda
>>> +250788308835 (mobile)
>>> Skype name (wilsonrandy_us)****
>>>
>>> www.msh.org<http://www.msh.org/?utm_source=2012-10-15+MSH+internal+announce+email+signature&utm_campaign=iemailsig&utm_medium=email>
>>>  ****
>>>
>>> *Stronger health systems. Greater health impact.*****
>>>
>>> ** **
>>>
>>> _______________________________________________
>>> Mailing list: https://launchpad.net/~dhis2-users
>>> Post to     : dhis2-users@xxxxxxxxxxxxxxxxxxx
>>> Unsubscribe : https://launchpad.net/~dhis2-users
>>> More help   : https://help.launchpad.net/ListHelp
>>>
>>>
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~dhis2-users
>> Post to     : dhis2-users@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~dhis2-users
>> More help   : https://help.launchpad.net/ListHelp
>>
>>
>