[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: THREDDS API Question



Hi Nathan,

I don't agree that it is intractable or even cross purposes. We currently keep track of a number of things from the config catalog files including the catalogs, datasetScans, and NcML datasets. We do that because we need that information to provide the current services available from the TDS. We don't keep track of every dataset in the config catalogs because we haven't needed to (and I'm not sure how it would scale). But, we certainly can add the ability to keep track of new information. I've been trying to figure out how much work it would be to refactor things to provide a method for getting notified of configuration events. That way you could keep the information you want and decide how to search on it.

This doesn't solve the issue of having multiple datasets in the catalog tree with the same access URL. As for the issue of external references, we currently don't follow any catalogRefs that are not on the server and we could do the same with dataset access URLs.

However, I still think working on a new CrawlableDataset implementation might be a better way to deal with your problem.

Ethan


Nathan Potter wrote:


Ethan,


I've been thinking about it too. It is probably the case that this is an intractable issue. The THREDDS architecture benefits from the ability to write catalogs independent of the data source/archive. THREDDS is designed so that I can write a catalog and serve it on my system, while the data sources referenced in the catalog may exist on different systems - often far away and outside my sphere of influence.


The idea that the Access Link (the link that THREDDS generates that allows data access) can be used to back track into the THREDDS catalog is cross purposes to the THREDDS design. It would be possible for a particular server implementation to have very carefully constructed THREDDS catalogs that would allow it to do this, but a general case is simply not possible.

At least as far as I can see it.

Nathan



On Jun 11, 2007, at 4:09 PM, Ethan Davis wrote:

Hi Nathan,

Sorry this is taking awhile. I'm trying to figure out some of the trade offs and such involved in a variety of ways of handling this. I should have a more detailed response tomorrow.

Ethan


Nathan Potter wrote:

Ethan et al.,

After talking with Ethan on the phone today I think I can state the issue more clearly:

The current THREDDS Servlet Framework (TSF) does not allow the collection/dataset information to be retrieved via the request URL.

The API method DataRootHandler.getCatalog(java.lang.String path, java.net.URI baseURI) expects the "path" parameter to be the path in the THREDDS catalog to the catalog file. There is no restriction on the file name of the catalog file. The path in the THREDDS catalog to the file may be different that the access URL.

What this means is that when a servlet receives an access request, even one that comes from a valid access link in a THREDDS catalog(.html), the servlet only knows about the request URL, nothing more. If the servlet needs to get the THREDDS dataset/collection information (and associated metadata if any) then it has no recourse but to attempt to search the catalog from the highest level looking for a dataset with a matching "urlPath" attribute. This activity may fail if:

- The THREDDS catalog employs <catalogRef> elements.

- The "urlPath" is not unique within the catalog.


I think that the TSF API should be augmented with accessor methods that allow the DataRootHandler to return InvDataset an InvCatalog to be retrieved based on information that a servlet has access to at run time, i.e. data that can be retrieved from the HttpServletRequest object.




Nathan





On Jun 4, 2007, at 5:00 PM, Nathan Potter wrote:


On Jun 4, 2007, at 1:05 PM, Ethan Davis wrote:

Hi Nathan,

Can you explain the context for these questions. This is on the server side (in Hyrax)?


Yes, server side.



Nathan Potter wrote:
Greetings,

So I am using the THREDDS API in an attempt to get the <property> elements for a dataset. I've run into a couple of (possibly related) problems.

Just to clarify our terminology. When you say "THREDDS API" you mean both the thredds.catalog and thredds.servlet packages? I generally split those apart and call the thredds.catalog package the "THREDDS Catalog API" and call the thredds.servlet package the "THREDDS Servlet Framework" (TSF).


[Note: the TSF is probably only useful for those writing servers.]


I wasn't distinguishing. But since DataRootHandler is in the TSF then that is where I am suggesting an API change.





** 1) I can't get the dataset information without searching.

In the HttpServletRequest I have the URL for the dataset, say:

http://localhost:8080/opendap/wcs/MODIS/Grid/test.hdf.html

Is this URL for an OPeNDAP HTML response?


Right, but the requested response isn't really meaningful in this discussion since all I am really after is the THREDDS dataset information for the atom/leaf/dataset test.hdf



Are you trying to get the property from the THREDDS catalog so you can use it in the OPeNDAP response?

Well... In truth it's much more complex than that, but since I will have to do that too we can roll with that vision for the moment.





In order for me to get THREDDS to divulge the <property> elements for the dataset I have to:

- take the dataset name "wcs/MODIS/Grid/test.hdf.html" and back track to the
collection name, "wcs/MODIS/Grid/".
- ask the DataRootHandler for the InvCatalog for "wcs/MODIS/Grid/"
- Ask the InvCatalog for the InvDataset for "wcs/MODIS/Grid/"
- Search the child datasets of the "wcs/MODIS/Grid/" InvDataset for the
one whose name (lexically) matches "wcs/MODIS/Grid/test.hdf.set"
- Read the properties of that InvDataset


That seems awfully complex. (Of course there may a more straight forward way that I am not aware of.)

That is about as simple as it gets. Though I would suggest you make sure the THREDDS configuration (TSF) knows about this dataset first by getting the CrawlableDataset that matches the dataset URL:
DataRootHandler.getCrawlableDataset("wcs/MODIS/Grid/test.hdf")
// I dropped of the trailing ".html" assuming it was the OPeNDAP dataset URL extension


When I tried this I could only get CrawlableDataset objects for catalogs that were part of a <datasetScan>




Are you using InvDataset.findDatasetByName( String name) to find the child dataset?

No.


Also, depending on how you setup your dataset IDs, you could ask the catalog to find the dataset by ID, like


cat.findDatasetByID( "wcs/MODIS/Grid/test.hdf")

Ahhh... I just tried that and it works. So, that greatly simplifies that step, thanks!






** 2) When I ask for a catalog I have to know the name of the XML file in which it resides.

In the above example, when I ask the DataRootHandler for the InvCatalog I ask for: " wcs/MODIS/Grid/catalog.xml" Which is all well and good if all of the catalogs are stored in files called catalog.xml. Essentially this means that anyone configuring a THREDDS catalog has to create a hierarchy of directories that mimics the organizatiopn of the collections, and all of the THREDDS information must be stored in files called "catalog.xml".

Why do you need to create this hierarchy of directories mimicking the data collection hierarchy? The TSF should keep track of your config catalogs and the automatically generated catalogs.

Right, but if all of the THREDDS catalog files have the name "catalog.xml" they can't all be in the same directory, so they have to live in some kind of directory hierarchy - I just figured it made sense to mimic the collection organization, but that's not necessary.





THREDDS does not actually require this - I can make a complex hierarchy of collections by using either a single (complex) top level catalog.xml file, or a collection of XML files in a single directory that employ <catalogRef> elements to create their organizations.
However the API breaks down in both cases.


If the catalog is composed of a collection of XML files in a single directory that employ <catalogRef> elements to create their organizations, then in order to retrieve catalog information I would have to KNOW how the information was organized (file names, directory hierarchy , etc.) But I don't know - since the catalog may be created by a user after compile time (although THREDDS does know this since it parsed all of the catalog information at start up) - and I shouldn't have to know. For me to know would require that I parse the top level catalog.xml file and build the XML doc tree myself. At which point it I can get the elusive <property> elements from the XML doc in memory.

If the catalog is composed of a single (complex) top level catalog.xml file then I would have to know that and just ask for the top level catalog.

(Searching the entire catalog from the top down for my dataset doesn't seem to work either...)

I'm sorry, I'm having a hard time following here. What are you trying to do and why?

For any request that is looking for one of the OPeNDAP data responses I need to search the THREDDS catalog for the dataset, and if found, I need to extract any metadata that may in the catalog for that dataset.




Is the problem that you may not know if the dataset is contained in a catalog generated because of a datasetScan element or contained directly in one of the THREDDS config catalogs?

I think that's a separate issue.



All of these methods of writing and organizing catalogs are legitimate in THREDDS, and users writing THREDDS catalogs would likely employ one or more of these methods when writing their catalogs.


I propose that the THREDDS API be extended so that one can simply ask the DataRootHandler for an InvDataset or an InvCatalog. Like:


    InvDataset id = drh.getDataSet("wcs/MODIS/foo.nc");
    InvCatalog id = drh.getCatalog("wcs/MODIS/");

or possible the InvDataset that represents a collection:

    InvDataset id = drh.getDataSet("wcs/MODIS/");


If the DataRootHandler doesn't have it, return null.


Is that unreasonable?

I'll have to take a closer look at this.

Ethan


Nathan


= = = Nathan Potter ndp at opendap.org OPeNDAP, Inc. 541.752.1852




===============================================================================


To unsubscribe thredds, visit:
http://www.unidata.ucar.edu/mailing-list-delete-form.html
===============================================================================



--
Ethan R. Davis Telephone: (303) 497-8155
Software Engineer Fax: (303) 497-8690
UCAR Unidata Program Center E-mail: address@hidden
P.O. Box 3000
Boulder, CO 80307-3000 http://www.unidata.ucar.edu/
---------------------------------------------------------------------------





= = = Nathan Potter ndp at opendap.org OPeNDAP, Inc. 541.752.1852



= = = Nathan Potter ndp at opendap.org OPeNDAP, Inc. 541.752.1852


===============================================================================


To unsubscribe thredds, visit:
http://www.unidata.ucar.edu/mailing-list-delete-form.html
===============================================================================



--
Ethan R. Davis Telephone: (303) 497-8155
Software Engineer Fax: (303) 497-8690
UCAR Unidata Program Center E-mail: address@hidden
P.O. Box 3000
Boulder, CO 80307-3000 http://www.unidata.ucar.edu/
---------------------------------------------------------------------------





= = = Nathan Potter ndp at opendap.org OPeNDAP, Inc. 541.752.1852


-- Ethan R. Davis Telephone: (303) 497-8155 Software Engineer Fax: (303) 497-8690 UCAR Unidata Program Center E-mail: address@hidden P.O. Box 3000 Boulder, CO 80307-3000 http://www.unidata.ucar.edu/ ---------------------------------------------------------------------------


=============================================================================== To unsubscribe thredds, visit: http://www.unidata.ucar.edu/mailing-list-delete-form.html ===============================================================================


NOTE: All email exchanges with Unidata User Support are recorded in the Unidata inquiry tracking system and then made publicly available through the web. If you do not want to have your interactions made available in this way, you must let us know in each email you send to us.