Difference between revisions of "WCS Wrapper Configuration for Point Data"
(New page: Back to WCS Wrapper [http://sourceforge.net/p/aq-ogc-services/home/ Project on SourceForge] Questions and comments should go to [http://sourceforge.net/...) |
|||
Line 1: | Line 1: | ||
[[WCS_Access_to_netCDF_Files| Back to WCS Wrapper]] | [[WCS_Access_to_netCDF_Files| Back to WCS Wrapper]] | ||
+ | |||
+ | [[WCS_Wrapper_Configuration| Back to WCS Wrapper Configuration]] | ||
[http://sourceforge.net/p/aq-ogc-services/home/ Project on SourceForge] | [http://sourceforge.net/p/aq-ogc-services/home/ Project on SourceForge] |
Revision as of 11:02, July 27, 2010
Back to WCS Wrapper Configuration
Questions and comments should go to sourceforge discussions, bug reports to sourceforge tickets. Urgent issues can be asked from Kari Hoijarvi 314-935-6099(w) or 314-843-6436(h)
Structure of OWS/web
OWS/web is for system developers only.
OWS/web/static contains static web content. You can put any documentation here and it will be served as a web page or download. The home page index.html is pretty much mandatory, and you shoud change favicon.ico to reflect your organization. We highly recommend, that you customize these to document your WCS service.
OWS/web/static/cache is a folder for temporary files. The service uses it for output files. Anything you put there will be deleted when space is needed.
The installation contains an example datasets OWS/web/static/testprovider and OWS/web/static/point. The testprovider is a demo NetCDF dataset, point is an example how to server point data from a SQL database. Every service will have a folder with the same name here.
You may now check the provider page http://localhost:8080/testprovider which is served as a static file. Any file under static becomes accessible
The Human Interface: Create the index.html Front Pages for Visitors.
If no query is present, the server gives a default page index.html. You should provide pages for your server and for all the providers.
The server index.html is at OWS/web/static/index.html, which will be displayed from url http://localhost:8080/, Index of an external server index.html is here.
Every provider folder should also have an index.html like OWS/web/static/testprovider\index.html which will be displayed from http://localhost:8080/testprovider, index of an external provider front page is here
Every provider should have wcs_capabilities.conf that lists keywords and contact information. The format is simple, copy one from the testprovider and edit it.
Windows Implementation Bug
Important There is a bug deep in python core libraries that make serving text files tricky. The files need to be encoded with unix style line ending convention '\n', instead of windows style '\r\n'.
To fix this, issue command:
python /OWS/web/owsadmin.py unix_nl filename.html
for every html file you serve.
Serving data from periodic collection of NetCDF files
Sometimes you have accumulated a huge number of small NetCDF files, like daily slices from a model output. You could combine those into one big cube, but you for a terabyte of files, that may not be an option.
Download our HTAP test package custom-netcdf-1.2.0.zip. It only has two days of data to make download small. Then read the custom provider page
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.
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': ...