Difference between revisions of "WCS Wrapper Configuration for Point Data"

From Earth Science Information Partners (ESIP)
Line 21: Line 21:
 
=== Location Table ===
 
=== Location Table ===
  
The common thing between different databases is, that they need to have a location table.
+
The common thing between different databases is, that they need to have a location table. The current implementation is based on time series from stationary locations.
  
 
     table location
 
     table location

Revision as of 11:52, July 27, 2010

Back to WCS Wrapper

Back to WCS Wrapper Configuration

Project on SourceForge

Storing Point Data in a Relational Database

Provider point is an example how to configure this service to use SQL database to serve point data.

Point data is often stored in SQL databases. There's no standard schema, like CF-1.0 convention for NetCDF files, so it is not possible to just connect and start serving. You have to create the configuration description.

Therefore, the WCS query processor needs to know what to select and join. This information must be edited into the configuration script.

Notes on SQL

One of the most powerful ideas in relational database design is the concept of a view. You don't need to change the existing data tables, creating a view that makes your DB to look like the one needed is usually enough. This is by far the easiest way to configure your WCS.

It is better to design a normalized schema and only optimize with benchmarks available. Especially filtering small lat/lon ranges is much more efficient to do on a normalized location table rather than denormalized data table.

Location Table

The common thing between different databases is, that they need to have a location table. The current implementation is based on time series from stationary locations.

   table location
   +----------+-------------------------+
   | loc_code | lat   | lon     | elev  |
   +----------+-------------------------+
   | KMOD     | 37.63 | -120.95 |  30.0 |
   | KSTL     | 38.75 |  -90.37 | 172.0 |
   | KUGN     | 42.42 |  -87.87 | 222.0 |
   |...       |       |         |       |
   +----------+-------------------------+

Here loc_code is the primary key and lat,lon is the location. Optional fields can be added. CIRA VIEWS database has a location table, but it's called Site and it spells full longitude. The datafed browser uses standard names loc_code, loc_name, lat and lon for browsing. For plug-and-play compatibility we recommend using these names. In the CIRA VIEWS database, the view creation would be:

   create view location as 
   select 
       SiteCode as loc_code, 
       Latitude as lat, 
       Longitude as lon 
   from Site

The primary key is loc_code, being unique for all the locations.

If the fields have different names they can be aliased in the configuration.

Some Different DB Schema types

In this documentation three different schemas are presented. Each of them have good and bad points.

One Big Data Table

In this case, all the data is in the same table:

   +----------+------------+------+------+------+
   | loc_code | datetime   | TEMP | DEWP | VIS  |
   +----------+------------+------+------+------+
   | KMOD     | 2009-06-01 | 87.8 | 51.4 | 10   |
   | KMOD     | 2009-06-02 | 82.3 | 51.4 | NULL |
   | KSTL     | 2009-06-01 | 78.6 | 34.9 | 18   |
   | ...      |            |      |      |      |
   +----------+------------+------+------+------+

The foreign key to location table is loc_code. The primary key is (loc_code, datetime)

Strengths: Simple, No joining when querying all the fields.

Downsides: Needs nulls for missing data, querying just one field is inefficient.

Long And Skinny Table

In this case, all the data is in the same table:

   +----------+------------+------+-------+
   | loc_code | datetime   | data | param |
   +----------+------------+------+-------+
   | KMOD     | 2009-06-01 | 87.8 | TEMP  |
   | KMOD     | 2009-06-02 | 82.3 | TEMP  |
   | KSTL     | 2009-06-01 | 78.6 | TEMP  |
   | KMOD     | 2009-06-01 | 51.4 | DEWP  |
   | KMOD     | 2009-06-02 | 51.4 | DEWP  |
   | KSTL     | 2009-06-01 | 34.9 | DEWP  |
   | KMOD     | 2009-06-01 | 10   | VIS   |
   | KMOD     | 2009-06-02 | 10   | VIS   |
   | KSTL     | 2009-06-01 | 18   | VIS   |
   | ...      |            |      |       |
   +----------+------------+------+------+

Strengths: No nulls, Easy to add fields.

Downsides: Querying requires extra filtering with parameter index, slower than others.

One Data Table For Each Param

Each parameter has its own data table. In this case there's no need for nulls, and is the fastest for one parameter query.

   +----------+------------+------+
   | loc_code | datetime   | TEMP |
   +----------+------------+------+
   | KMOD     | 2009-06-01 | 87.8 |
   | KMOD     | 2009-06-02 | 82.3 |
   | KSTL     | 2009-06-01 | 78.6 |
   | ...      |            |      |
   +----------+------------+------+


   +----------+------------+------+
   | loc_code | datetime   | DEWP |
   +----------+------------+------+
   | KMOD     | 2009-06-01 | 51.4 |
   | KMOD     | 2009-06-02 | 51.4 |
   | KSTL     | 2009-06-01 | 34.9 |
   | ...      |            |      |
   +----------+------------+------+


   +----------+------------+-----+
   | loc_code | datetime   | VIS |
   +----------+------------+-----+
   | KMOD     | 2009-06-01 | 10  |
   | KMOD     | 2009-06-02 | 10  |
   | KSTL     | 2009-06-01 | 18  |
   | ...      |            |     |
   +----------+------------+-----+


Strengths: No nulls, Easy to add tables, easy to add heterogenous flag fields, fastest queries for single parameter.

Downsides: More tables, querying all the parameters at once requires a massive join.

Configuring the WCS using SQL Views

This is demonstrated in the test provider point.

The demonstration is using sqlite, which is distributed with python by default. The project has following files:

  • pntdata.py: This script creates the test database and fills it with dummy data.
  • pntdata.db: The sqlite database file created by pntdata.py
  • point_config.py
    • WCS coverage information
    • Mapping the coverages and fields to SQL tables.
  • point_WCS.py
    • Loads the metadata. In this demo version, this is done by hardcoding the tables in point_config.py. In CIRA/VIEWS this is done by querying the parameter table.
    • Gets db connection. The metadata mappings allows the service to generate SQL on it's own.


Contents of point_config.py

All of these are python dictionaries and lists.

The syntax is simple, This is a list:

   ['loc_code', 'lat', 'lon']

This is a dictionary:

   {'key1':'value1', 'key2': 'value2 }

Since these are just python objects, they can be generated a database as well.

Location Table Configuration

First, the location table for coverage SURF_MET. Because WCS does not have a good place to describe a location table, we use WFS, Web Feature Service to do the same. Sample WFS Call.

   location_info = {
       'location':{
           'service':'WFS',
           'version':'1.0.0',

In the CIRA/VIEWS database, we're not authorized to create a view. So we need to map the 'Site' table and it's columns.

           'table_alias':'Site',
           'columns':{
               'loc_code':{'column_alias':'SiteCode'},
               'lat':{'column_alias':'Latitude'},
               'lon':{'column_alias':'Longitude'},
               }
           },
       }

Data Table Configuration using SQL View

   point_info = {

First key is the coverage information and it's descriptions:

       'SURF_MET':
           {
               'Title':'Surface Meteorological Observations',
               'Abstract':'Dummy test data.',

The covered area and time. The Time dimension is a true dimension here, but contrary to grid data, the X-Y dimensions for point data are not dimensions, but attributes of the location dimension.

               'axes':{
                   'X':(-180, 179.75), 
                   'Y':(-90, 89.383),
                   'T':iso_time.parse('2009-09-01T12:00:00/2009-09-03T12:00:00/PT1H'),
                   },

Then comes the description of the fields.


               'fields':{
                   'TEMP':{
                       'Title':'Temperature',
                       'datatype': 'float',
                       'units':'deg F',

The location table is a real dimension. In this case, the location table is shared, so we use the previously declared variable 'location_info' If the location tables are parameter specific, they can be specified individually.

                       'axes':location_info,

The access instructions. This configuration is using 'complete_view', so the administrator has created the view that joins together the location table and the temperature data table. The SQL query will typically look like select loc_code, lat, lon, datetime, temp, flag from TEMP_V where datetime = '2009-09-01 and (lat between 34 and 44) and (lon between -90 and -80). This is by far the easiest way to configure the WCS.

                       'complete_view':{
                           'view_alias':'TEMP_V',
                           'columns':['loc_code','lat','lon','datetime','temp', 'flag'],
                           },
                       },

The end of the first field.


Data Table Configuration using by Mapping Original Tables

For demonstration purposes, the next field is configured without a view.

                   'DEWP':{
                       'Title':'Dewpoint',
                       'datatype': 'float',
                       'units':'deg F',
                       'axes':location_info,

First we tell which table contains the data.

                       'table_alias':'DEWP_base',

The CIRA/VIEWS has FactDate, not datetime, so we have to map.

                       'datetime_alias':'FactDate',

If you have a simple join by loc_code, then the point the SQL processor can just generate 'inner join DEWP_base on DEWP_base.loc_code = location.loc_code'. In the CIRA/VIEWS, it's more complicated. We need to join the Site table for locations and since all the data is in the same table, we also need to filter by the parameter code, which now requires joining the parameter table. Also the foreign key is SiteID, not the loc_code alias SiteCode.

                       'joins':(
                           'inner join AirFact3 on AirFact3.SiteID = Site.SiteID ' +
                           'inner join Parameter on AirFact3.ParamID = Parameter.ParamID'),

The default common data filter is empty. Again, CIRA/VIEWS needs more:

  • Aggregation: some data is aggregated, we're interested in the raw data only.
  • Exclude null lat/lon and -999 in data.
  • Only get data from certain programs.
  • Finally: Since all the parameters are in the same table, filter by param code.


                       'common_data_filter':(
                           'AggregationID = 1 and Site.latitude is not null and Site.Longitude is not null ' +  
                           'and AirFact3.FactValue <> -999 ' + 
                           'and AirFact3.ProgramID in (10001, 10005, 20002) -- ('INA', 'IMPPRE', 'ASPD') ' + 
                           "and Parameter.ParamCode = 'MF'"),

Finally: All the data is in the 'FactValue', so we need to map the column.

                       'data_columns':[
                           {'name':'MF', 'column_alias':'FactValue'},
                           ],

In this point database, the above is not necessary since fields have defaults. But we still need the data column.

                       'data_columns':[
                           {'name':'dewp'},
                           ],
                       },

The end of field.

                   },
           },

If you have another coverage, add it here:

       'Cov-Name':
           {
               'Title': ...