Setting up an EPN-TAP service in EPN-2024

EPN2024-RI

EUROPLANET2024 Research Infrastructure 

H2020-INFRAIA-2019-1

Grant agreement no: 871149

Document:  VESPA- WP6-3-011-TD-v3.0(27)

doi:10.XXXX/abcd-1234

Setting up an EPN-TAP service in EPN-2024

Date: 2021-11-28

Responsible WP Leader: Stéphane Erard

Table of Contents

Reference documents

• [RD1] Demleitner, Greene, Le Sidaner, Plante (2014) The Virtual Observatory Registry, Astronomy and Computing, 7, 101-107. 2014A&C.....7..101D
• [RD2] Erard S., B. Cecconi, P. Le Sidaner, Demleitner M., Taylor M. (2021) EPN-TAP: Publishing Solar System Data to the Virtual Observatory. 
  IVOA Proposed Recommendation 2021-10-22, https://github.com/ivoa-std/EPNTAP
• [RD3] The UCD1+ controlled vocabulary Version 1.4
• [RD4] EPN-TAP installation for VESPA data providers / Tutorials:
  EPN-TAP Installation for VESPA Data Provider Tutorial
  Registering your VESPA EPN-TAP Server
• [RD5]  DaCHS GAVO / German Astronomical Virtual Observatory. Contains the documentation and a tutorial for installation and configuration.
  http://vo.ari.uni-heidelberg.de/docs/DaCHS
• [RD6] IVOA Registry interface
  http://ivoa.net/Documents/RegistryInterface/
• [RD7] VirtisPDS IDL/GDL library:
  http://voparis-europlanet.obspm.fr/othertool.shtml
• [RD8] Units in the VO:
  http://wiki.ivoa.net/twiki/bin/view/IVOA/VOUnitsRFC or
  http://www.ivoa.net/documents/VOUnits/
• [RD9] Data Access Layer Interface:
  https://www.ivoa.net/documents/DALI/
• [RD10] Data provider on-boarding process:
  VESPA Data provider on-boarding process

Acronym list

Introduction

This document provides practical guidelines to setup an EPN-TAP data service for non-specialists. It focuses on building the data service itself, not on software installation which is addressed in [RD4]. A simple example is used here to illustrate each step of the procedure: it will provide access to a series of individual files.

If you're in a hurry, a shorter example is available here: Setting-up an EPN-TAP service: Tutorial for Beginners

The procedure described here follows the recommendations from the DaCHS team. It involves the creation of a data table describing the service content, and the writing of a resource descriptor for DaCHS. Although this triggers the creation of a PostGreSQL database, no explicit SQL handling is required.

The database installed during this procedure can easily be used to feed a web site with a different interface, in particular through SPIP (procedure to be described).

DaCHS server

First of all, you need to use a VO framework which will handle the queries and answers to your service according to the TAP protocol. VESPA is supporting the use of DaCHS; although other VO frameworks (e. g., TAP library, Vollt, etc) support EPN-TAP services successfully, those may require more work on your side.

We assume you have installed DaCHS on a Virtual Machine as described in EPN-TAP Installation for VESPA Data Provider Tutorial (see [RD4]). In this configuration, you will implement your service on your server in the guest machine, although you can access it from your usual system (host machine) - see Figure 1. Similar considerations apply if you use an installation on Docker, see also DaCHS on Docker. In case of technical issues during the tutorial, some hints are provided in Annex I below.

Figure 1

It is assumed that this DaCHS server is used as a "development server" - i. e. an area not accessible to external users, where you can test new functions. If you already have services open to users, it is better to have both a "production server" (open to your users) and an independent "development server" (not registered, and preferably not on-line) where you can freely test new design solutions without perturbing the existing service.

You will need to enter the gavoadmin password of your server at some point - you can retrieve it this way:

more /var/gavo/etc/feed

Note the value on the line:

 password = xxxxxxxxxxxx

Example files

To get the reference example in working order, you can download the example files this way:

sudo apt-get install git
cd ~
git clone https://voparis-gitlab.obspm.fr/vespa/dachs/services/padc/voparis-tap-planeto/iks 
cd iks

This will clone the repository on your disk, including several examples of EPN-TAP services. The iks directory contains:

• dbiks2.pro - table building routine (IDL)
• indexiks.csv - basic data table in csv format (built by dbiks2.pro)
  
• catiksfiles.pro - specific PDS to VOTable conversion routine (IDL)

• IKS_mixin_q.rd - service resource descriptor
  

Note: beware that 2 versions of this service are currently available - branch master contains a basic version running on DaCHS v1, while branch dachs2 contains a modern version which requires Dachs ≥ 2.5. You should switch to the dachs2 branch unless you're using an older version of DaCHS: 

(from directory iks)
git checkout dachs2

VO Tools

To retrieve and visualize the data you will use VO tools such as TOPCAT, Aladin, or CASSIS, depending on the type of your data. These applications run in the Java Virtual Machine. You need to download their installer or jar files (some clients may provide a standard application interface or a webstart link). You can test a jar file by clicking on its icon or by issuing a shell command from their directory:

java -jar <application>.jar

• Aladin (for images)
  http://aladin.u-strasbg.fr/java/nph-aladin.pl?frame=downloading
• TOPCAT (for tabular data)
  http://www.star.bris.ac.uk/~mbt/topcat/#install
• CASSIS (for spectra)
  http://cassis.irap.omp.eu/?page=installation

In this tutorial, we will also use the VESPA query portal at Paris Observatory: http://vespa.obspm.fr

Data selection & service design

The very first step in building a service is to decide the content of this service: what are the data of interest, what is the proper organization of the data, is it consistent and correctly self-described?

In EPN-TAP, the data are presented in a table that describes data "granules" with several parameters, which are used to select data of interest for the user. "Granules" are the smallest data chunks described and accessible in the service; this granularity level must be identified from the start. In the common case where the data consist in a series of data files (e.g., telescopic images), granules usually are the individual files. When the data consist in series of scalar numbers describing objects (e.g., a list of orbital parameters), granules correspond to these objects and may be more difficult to identify (they can be targets, observations, etc… depending on the context).

• In EPN-TAP (v.2), all data products are described as distinct granules - they can be either individual data files (linked from the table) or sets of scalar parameters (included in the table). There is therefore at most one link to a file per granule. The only exception is the possibility to associate a granule to a small-size thumbnail, which is used to provide a quick-look in the search interface. "Previews" are more elaborate products which are treated as independent granules (e. g., reduced resolution images, a graphic representation of a table, etc).

• Some services contain related data products, e. g., a calibrated image and a map-projected version of the image, spectra and derived spectral parameters, tabular data and a graphic preview, etc. In such cases, data products derived from the same measurement are introduced as independent granules but are related through the same "observation". Instead, data products of similar nature (e.g., all calibrated spectra) belong to the same "group". All granules are therefore described separately, but may be associated in observations or groups (the grouping scheme is left to the provider's choice).

In practice, you have to define three ID parameters for each granule:

Explicit strings are recommended in these fields, because they may be used as search criteria by your users. Standard values of granule_gid are suggested in the table above.

Please remind that granule_uid must be unique for the service to work: otherwise the DaCHS server will issue fatal errors when ingesting the q.rd file later in the process.

An important part of service design is therefore to identify the groups of granules, and there is no unique way to do this. Group IDs are a handy way to separate different levels of data which are difficult to distinguish otherwise. A rule of thumb is to use different groups for similar granules that you usually do not want to retrieve together as an answer to a query: several kinds of map projections, measurements vs associated geometry, different channel measurements, etc — especially when these files are not easily identified from other EPN-TAP parameters (e.g., images and spectra for instance are easily identified from the dataproduct_type parameter).

Gathering information on the data

The next step is to document the granules as accurately as possible, so that the user will be able to locate them from a query based on observational or operational parameters. The current tutorial describes a common situation where the information is first gathered in a table describing the content of the database, and saved in csv format (one row per granule / file). This data table will then be ingested to build the service table used with the EPN-TAP protocol. The initial data table can be written from IDL or a similar computing environment, or even manually; this is best done on your usual (host) machine to benefit from a familiar interface (keyboard layout, editor, etc). For situations where the table is already available in SQL, or should be retrieved from a collection of fits files, see the other tutorials.

Describing your data

The EPN-TAP protocol [RD2] is based on a set of mandatory parameters describing all databases uniformly. Although most parameters can be left empty when not relevant, they must be informed whenever possible to insure the accuracy of the search functions.

Remember that parameters with constant values can be initialized in the next step (q.rd file), as well as duplicate rows (e.g., for files provided in two different formats). This may simplify the current step.

Constraints on numerical parameters

Most numerical parameters have to be converted in standard units at some point, to make uniform queries possible on all databases. The space, time, and spectral coordinates are often of particular importance to identify data of interest inside a database - beware that they must honor specific conventions in the data description so that your service can be searchable. Those are listed in the EPN-TAP parameter documents ([RD2] or EPN-TAP V2.0 parameters).

• Spatial coordinates can be provided in various systems (Cartesian, Body-fixed, Celestial, etc). Each system follows specific conventions, e.g., longitude must follow the eastward convention in a body-fixed frame.
• Time must be provided as UTC at the observer location, and formatted as Julian days, in double precision
• Spectral ranges must be provided as frequency in Hz in the service table.
  
• Floating-point parameters are best provided as double precision.
  

Constraints on string parameters

• Values of some parameters have to be selected from predefined lists (see [RD2] or EPN-TAP V2.0 parameters)
  
• In free format strings, the # character is not authorized 
• Heading and trailing spaces are forbidden (they cause nasty errors in the interfaces, and must be checked). Spaces are best avoided in general.
  
• In names, acronyms, etc, upper cases are allowed when they follow official naming schemes (e.g., IAU target names); lower cases are required otherwise
  
• Multiple values are supported for some parameters; lists are sequences of values separated by # (with no extra space)
• If you need to include "special" characters, UTF-8 encoding should do. Please check there are interpreted correctly at the end of the process.
  

• In addition to the granule identification parameters discussed above, some other EPN-TAP parameters must contain a value (see [RD2] or EPN-TAP V2.0 parameters) - those are mostly related to the service itself (service_title, dates of products, and dataproduct_type). Special care should be taken when selecting dataproduct_type, which is used essentially to identify the type of data (e.g., spectra from images) in a generic sense (a spectral plot is more a spectrum than an image). Its value must be selected from a predefined list.

• Some EPN-TAP parameters describe the target, the origin of the data and the type of measurements, and must be informed when setting up the service. In general a standard value is required, from the reference sources indicated in the EPN-TAP documentation [RD2]. Those include:

• Special care should be taken with target names: the names should follow the exact IAU spelling, including case, or be left in lower cases. This is particularly sensitive for exoplanets, comets, and asteroids (e. g., names like xP for periodic comets, main denomination for asteroids). The alt_target_name parameter can be used to store alternative target names/spelling.

• Some parameters must take values selected from a predifined list, e. g., dataproduct_type is encoded according to the EPN-TAP protocol definition [RD2].

• A good practice is to start informing the EPN-TAP mandatory parameters with the available information. If other important quantities need to be stored, look for optional parameters which can accommodate them unambiguously. It is important to follow standard usage, which ensures consistent responses from related services.

In last resort, specific parameters can also be included in addition to the standard EPN-TAP ones. They will be available to search inside an individual service, but they obviously cannot be used to perform searches across several services. Ideally, you should contact the VESPA team if you need to define new parameters.

• It is important to realize that the TAP mechanism can only query the service table, not its header. All relevant parameters must therefore be present in the service table with proper units, even when constant.

Linking data files

An important parameter is the access_url which points to the files. When present, it must be associated with other parameters describing the file: access_format and access_estsize. However, this set of parameters is optional, and can be replaced by the data themselves in the case of a small data table of scalar values - allowing searches on the data themselves.

Note that the access_url parameter can point to a script, rather than a file. This may be a handy way to support exotic or non-standard formats, and to perform conversions on the fly. For instance, scripts are used in several services to convert csv or ascii files to VOTable. Scripts may also be used to extract the data of interest from another database, from larger files, or to send queries on servers via specific protocols (e.g., WMS, das2stream, etc). The access_format parameter always refers to the returned product, and is used to pick up a visualization tool for the data.

Providing footprints

Footprints provide significant added value to services distributing spatially extended data (images or spectral cubes), either on the sky or on planetary surfaces. The C1/C2 parameters provide footprints as a simple bounding boxe (e.g., min/max longitudes and latitudes) for quick searches. However, this system produces many false alarms and is not adequate for accurate searches.

In addition, EPN-TAP uses the s_region parameter to provide 2D footprints as ADQL polygons. Such footprints can be used for spatial searches and can be displayed in plotting tools (e.g., Aladin). To define footprints, you first have to sample the contour of your data products - at least the corners are required for a polygon, but more points along the edges would provide more accurate results. Points must be listed in sequence and in direct / anticlockwise sense. If your data are on the sky, use RA and Dec values in degrees; on a planetary body, use eastward longitudes. Points and circles can also be used as footprints, as per DALI standard [RD9].

You then enter the list of points in the parameter s_region of every granule with syntax: '{(lon1,lat1), (lon2,lat2), … }' (with no quotes). Arguments lon and lat must be provided as: 10.d, 5d, etc, where character d stands for degrees.

Queries for inclusions and intersections with a search box can be issued in TAP, including from the VEPSA portal (using its Query mode). Comparison between footprints of various data products will be also be available in the future.

Building the resource descriptor

Once the data are selected and formally described in a csv table, you are ready to set up your service as described below. A shorter and technical-only tutorial is available in the DaCHS documentation: http://docs.g-vo.org/DaCHS/tutorial.html#epn-tap

The service will be installed in a database "schema". The schema name is that of the service. The EPN-TAP protocol expects a service table called <schema>.epn_core (see Figure 2) - this table will be queried by the user interfaces (TAP clients), and has to be formatted according to the EPNCore standard.

DaCHS uses a "resource descriptor" to set up services, an xml file always called q.rd. This file calls a "mixin", which provides the complete definition of standard EPN-TAP parameters, and uses a "grammar" to handle data ingestion. Some python code can also be included to modify the original data table, e. g., to convert units or to duplicate rows when variations of the same granules are provided in the service. Upon import, the q.rd file will create the database under postgreSQL, the service table, and the service itself. Again, this file is best written in your usual (host) machine, then transferred to the virtual (guest) machine.

Note that while normal SQL identifiers are case-insensitive, mixed-case identifiers will often cause trouble, especially when provided between double quotes. To avoid any issue, provide your schema name (as well as any names for extra columns) in lowercase ascii and underscore exclusively, with no quotes.

“Special” (accentuated / diacritic) characters must be avoided in this file, even in descriptions (they will be displayed by various tools, which often expect ascii formatting).

Figure 2

In the present context, the q.rd file is made of 4 parts: general metadata; table structure; ingestion rules; regression tests.

You can get a first template on the guest machine where the DaCHS server is installed. In the directory of your service, type:

dachs start epn-tap

to generate an empty template with comments - you may want to copy this on your host machine to edit it in a more familiar environment. Alternatively, you can use the IKS_mixin_q.rd example file as a template and adapt it to your needs.

General metadata

These fields are intended to provide reference to the service and describe it in the registry (longer description here: Building the resource descriptor for your EPN-TAP service in DaCHS)

The first line provides the name of the schema, and must be replaced by that of your service (case sensitive).
The description of the service that follows must be adapted — lines starting with <meta name =… >. Only text is allowed here (no macro, e. g. \metaString). Use a generic service for contact.email. Some of these elements are only allowed to appear in one instance (e. g. contact.name, contributor.name) - include a list of values if required. 
The utype element provides reference to the protocol in use.
Subject values are best taken from the Unified Astronomy Thesaurus (IVOA variation). They must include at least one of the following top-level values for VESPA services: Solar system astronomy, Solar physics, Exoplanet astronomy (see also this page for all possible values and subdivisions).

<resource schema="**iks**">
  <meta name="title">**Full title**</meta>
  <meta name="description" format="plain">**Full description (~ 5 lines), possibly with bib reference.**</meta>
  <meta name="creationDate">2017-12-11T19:42:00Z</meta>
  <meta name="subject">comet</meta>
  <meta name="subject">spectroscopy</meta>
  <meta name="subject">1P Halley</meta>
  <meta name="subject">Solar system astronomy</meta>
    <meta name="creator.name">**Your name**</meta>
    <meta name="contact.name">**Somebody responsive**</meta>
    <meta name="contact.email">**mail of a generic service**</meta>
    <meta name="contact.address">**Postal address**</meta>
    <meta name="instrument">**Instrument name/acronym**</meta>
    <meta name="facility">**Instrument host name**</meta>
  <meta name="source">**bibcode of main reference**</meta>
  <meta name="contentLevel">General</meta>
  <meta name="contentLevel">University</meta>
  <meta name="contentLevel">Research</meta>
  <meta name="utype">ivo://vopdc.obspm/std/epncore#schema-2.0</meta>

    <meta name="coverage">
        <meta name="waveband">Infrared</meta>
        <meta name="Profile">none</meta>
    </meta>

The element referenceURL is used to link to an existing web site describing the service (specify that of your institute if no such web site exists).

Table structure

The table element declares the service table name, then introduces the epntap2 mixin that provides a complete definition of all standard EPN-TAP parameters. Optional parameters in use need to be declared in the second line, again with definition provided by the mixin (only declare those used in your service). Finally, non-standard parameters must be declared with all their metadata in successive blocks, as shown in the example (acquisition_id).
• The mention of spatial_frame_type="none" is a default that applies when no spatial_frame_type is applicable - it must be replaced by a relevant, applicable value, e.g., "body" for body-fixed coordinates (longitude and latitude).
• When defining non-standard parameters: all floating point parameters must be double precision; units must conform to the “Units in the VO” document [RD8]; an adequate UCD must be identified [RD3].

    <table id="epn_core" onDisk="true" adql="True">
        <mixin spatial_frame_type="none"
        optional_columns= "access_url access_format access_estsize access_md5 alt_target_name publisher bib_reference" >//epntap2#table-2_0</mixin>

        <column name="acquisition_id" type="text" 
            tablehead="Acquisition_id"
              description="Extra: ID of the data file in the original archive"
              ucd="meta.id" 
            verbLevel="2"/>
    </table>

Ingestion rules

The "data" element of the q.rd provides:

• A reference to the data source, in this case the path to the csv data file;
• The csvGrammar element contains the global ingestion rules:
  • A first rowfilter element providing the name of the service table and a global assignement of the csv file columns to EPN-TAP parameters (when the csv columns have parameter names);
  • A second rowfilter element handling special conditions may appear here. This contains a python code executed upon ingestion.
• The table element contains a rowmaker element providing extra ingestion rules:
  • <var> elements define intermediate variables for future use.
  • Mandatory parameters associated to table columns bearing a different name, or modified from the data table, must be associated using <bind> elements. Modifications can be introduced in the bind element (python style) or in the rowfilter code.
  • Optional and non-standard parameters must be defined in the <var> element of rowfilter code in general. Depending on the epntap2 mixin version in use, they may require to be defined in the <bind> element (check messages upon import).

        <data id="import">
        <sources>data/indexiks.csv</sources>
        <csvGrammar>
            <rowfilter procDef="//products#define">
                <bind name="table">"\schema.epn_core"</bind>
            </rowfilter>

            <rowfilter name="addExtraProducts">
                <!-- Duplicate input granules to provide 2 alternative versions -->
                <code>
                        # whatever processing, see below 
                </code>
            </rowfilter>
        </csvGrammar>

        <make table="epn_core">
            <rowmaker idmaps="*">
                <var key="alt_target_name" source="target"/>
                <var key="bib_reference" source="ref"/> 

                <apply procDef="//epntap2#populate-2_0" name="fillepn">
                    <bind key="time_min">dateTimeToJdn(parseISODT(@time_obs))</bind>
                    <bind key="time_max">dateTimeToJdn(parseISODT(@time_obs))</bind>
                    <bind key="time_scale">"UTC"</bind>
					<bind key="spectral_resolution_max"> float(@sp_max) / float(@sp_res_max) </bind>
					<bind key="spectral_resolution_min"> float(@sp_min) / float(@sp_res_min) </bind>
					<bind key="spectral_sampling_step_min">2.99792458E14 *(float(@sp_step_min)+float(@sp_step_max))/ 2 / float(@sp_max)**2</bind>
					<bind key="spectral_sampling_step_max">2.99792458E14 *(float(@sp_step_min)+float(@sp_step_max))/ 2 / float(@sp_min)**2</bind>
                    <bind key="spectral_range_min">2.99792458E14 /float(@sp_max)</bind>
                    <bind key="spectral_range_max">2.99792458E14 /float(@sp_min)</bind>
                    <bind key="time_min">dateTimeToJdn(parseISODT(@time_obs))</bind>
                    <bind key="time_max">dateTimeToJdn(parseISODT(@time_obs))</bind>
                </apply>
            </rowmaker>
        </make>
    </data>

The python code in the rowfilter element allows for post-processing of the input table. In the case of IKS, the two blocks duplicate the granule entry and adjust the parameters for the two versions of the spectra (VOTable and PDS formats). Parameters are referred to with the @ prefix. Processing can also be triggered by conditions on the input parameters, see DaCHS doc or other examples.

                <code>
                        # VOTable version 
                        @granule_uid = @rootname+"C"
                        @granule_gid = "corrected"
                        @access_format = "application/x-votable+xml"
                        @access_estsize = "19"
                        @access_url = @a_url
                        @file_name = @rootname+".xml"
                        @thumbnail_url = @a_url+".png"
                        @creation_date="2013-11-17T10:41:00.00"
                        @modification_date="2016-04-28T15:43:00.00"
                        @release_date="2013-11-17T10:41:00.00"
                        yield row.copy()

                        # native version
                        @granule_uid = @rootname+"A"
                        @granule_gid = "archived"
                        @access_format = "text/plain"
                        @access_estsize = "4"
                        @access_url = @o_url
                        @file_name = @rootname+".tab"
                        @thumbnail_url = ""
                        @creation_date="1993-11-10T07:54:00.00"
                        @modification_date="1993-11-10T07:54:00.00"
                        @release_date="1993-11-10T07:54:00.00"
                        yield row
                </code>

Regression tests

DaCHS has a framework for running automated tests against your tables. The idea is that after each operational change (updating the software or the server, edits in the RD, etc), you can run:

dachs test -vc ALL

and the machine will complain if anything you test has broken.  In the case of EPN-TAP on DaCHS, the TAP part and whatever the mixin specifies (e.g., UCDs of columns) are already tested in the upstream software, so all you need to make sure is that your service works at all and that your ingestion rules still do what you think they do.  If you have ancillary datalink (or other) services, you should prepare tests for those, too. 

In the RD, the tests sit in one or more regTest elements within a regSuite element, like this:

	  <regSuite>
		  <regTest title="MPC EPN-TAP gives expected data">
			  <url lang="ADQL" 
			    query="select * from mpc.epn_core
			  	  where granule_uid='0000457'"
				  >/tap/sync</url>
			  <code>
				  row = self.getFirstVOTableRow()
				  self.assertAlmostEqual(row["semi_major_axis"],  3.09, places=1)
				  self.assertTrue(isinstance(row['creation_date'],  datetime.datetime))
				  self.assertEqual(row["target_name"], '(457) Alleghenia')
			  </code>
		  </regTest>
    </regSuite>

The example shows all you need for a basic test against an EPN-TAP service:

• Give the test a title that will later tell you roughly what failed; this is what will be given in the reports.
• The url element tells DaCHS what to query; here, it's encoding a TAP query, which you can see in the query attribute.  XML attributes can span several lines, so feel free to nicely format your queries, in particular where they go beyond just retrieving a single piece of granule metadata.
• The code element contains normal Python code with pyunit assertions.  In your case, follow the pattern: Read the first line of the result, and then see if some expected values are there using self.assertEqual and self.assertAlmostEqual (for floating-point data). There is no need to check each and every column, but it is a good idea to check anything that involves code written by you.

  If you let DaCHS deliver datasets, it might be a good idea to see if these are accessible.  For these, simply using DaCHS' custom assertHasStrings method comes in handy, perhaps like this:

	  <regTest title="THISSERVICE product delivery"
		  url="/getproduct/maidanak/data/Q2237p0305/Johnson_R/
			  red_kk050001.fits.gz">
		  <code>
			  self.assertHasStrings(b'\\x1f\\x8b\\x08\\x08')
		  </code>
	  </regTest>

    -- note the "b" to indicate you want to compare binary strings and the double backslashes which are necessary because DaCHS already looks at backslashes before passing them on to Python.

To learn more about DaCHS' testing framework, refer to  http://docs.g-vo.org/DaCHS/ref.html#regression-testing

Converting the data files (optional)

In many instances, the original data files are not available in a format easily handled in the VO environment. The usual VO formats are:

• FITS images and tables are OK if their headers are sufficiently informative.
• VOTables are xml files handled by all VO applications, and are particularly useful for tables, spectra, and catalogues.
• Tables can be stored as ascii or CSV files but may require specific extensions (.asc or .csv in general). In practice, these files are not self-described and are therefore difficult to use in the VO. Besides, files downloaded individually will contain no mention of origin, therefore this is not a preferred choice.
• CDF files are supported only in TOPCAT.
• Standard image formats such as PNG, JPEG, etc are usually supported by VO tools. However, compressed image formats are in general acceptable only for low accuracy previews and thumbnails, not for a science usage.

In any case, an important matter is that data files must be entirely self-documented, e. g., they must contain (in addition to orign and credits) all metadata providing important observation conditions (a subset of those contained in the table) — otherwise files downloaded individually would be hardly usable. This is routinely done using the PARAM elements in VOTable headers, or standard keywords in fits headers; relying only on the file name is a bad idea.

For this reason, it may be convenient to convert the data files to VOTable or fits format. This will not only permit self-documentation of the data files, but will also make them loadable directly by the VO tools; your service will then fully benefit from the VO interface of VESPA. As noted above, conversion can also be performed on the fly by using the access_url parameter to point to a script.

If you have to convert your files, notice that there is no need to convert the units used in the data files to the EPNCore standards - this is only required in the epn_core table to support queries.

When converting the data files to VOTable, special care must be taken to describe the data area: all quantities are associated with a name, unit, datatype, and UCD. UCDs must follow the VOTable/IVOA standard [RD3]. Tools to help selecting the adequate UCDs are available here:

http://dc.zah.uni-heidelberg.de/ucds/ui/ui/form

http://cds.unistra.fr/UCD/tools.htx

Installing your service

If you've edited the files in your host machine, you first need to copy them on the guest in the correct location. You first have to create a directory in the gavo tree, with the exact name of the schema. Respect the path given below (/var/gavo/inputs/iks/q.rd, where iks is replaced by your service schema name) and don't change directory names or move files after after a first import. You may need to adjust authorizations.

	On the host side: 
scp -P 2222 q.rd dachsroot@localhost:/var/gavo/inputs/iks/q.rd
scp -P 2222 indexiks.csv dachsroot@localhost:/var/gavo/inputs/iks/data/.

Once your q.rd file is ready and properly located check its validity using the command:

	Log in as dachsroot with the password retrieved earlier:
ssh dachsroot@localhost
	Then, on the guest side:
cd /var/gavo/inputs/iks
dachs val q.rd

If the syntax is correct, this will answer:

q.rd -- OK

If you need to edit the files, do it in your guest machine and scp again to the host.

Then import the file:

dachs imp q.rd

If import is successful, the output will indicate:

Making data import
Starting /var/gavo/inputs/iks2/data/indexiks.csv
Done /var/gavo/inputs/iks2/data/indexiks.csv, read 206
Shipped 206/206
Rows affected: 20

Any other answer is suspect. If you're logged in as dachsroot (or use the sudo command), you'll see possible error messages which may be hidden otherwise.

When import is done, restart the DaCHS server (sudo may be required here):

sudo dachs serve restart

You can now test your service in private mode. Make a test query through your web browser using the following URL (replace serveraddress and myschema by the correct values; in test mode, serveraddress should be 127.0.0.1:8080):

http://serveradress/__system__/tap/run/tap/sync?REQUEST=doQuery&LANG=ADQL&QUERY=SELECT * from myschema.epn_core&FORMAT=votable/td

Testing / querying your service

Checking your service output

The first test is to check your service output using the taplint validator associated to TOPCAT. Download the library stilts.jar from

http://andromeda.star.bris.ac.uk/releases/stilts/

cd <where stilts is located on your disk>
java -jar stilts.jar taplint tapurl=<http://yourserver.wherever/__system__/tap/run/tap> stages='tme epn' report=ew

(see stilts doc for further options)

Errors and warnings must be processed iteratively, as taplint filters recurring messages.

Alternative: the related on-line validator provided by PADC is aging, but will be updated in the short term

http://voparis-validator.obspm.fr/

• Select "EPN-TAP2" in the "Specification" menu, click LOAD
• Enter your service schema name in the corresponding field (e. g., iks.epn_core) - please check and remove any extra space in this field, this is important
• Provide the URL of your test server in the empty field “URL to validate” at the bottom, e.g. http://yourserver.wherever/__system__/tap/run/tap
• click VALIDATE

Messages are relative to the output VOTable from the server, which is formatted based on the definition of the epn_core table as described in your q.rd file. In case of problem, go back to the above definitions and check every step - in particular, check the extra parameters in your service, not predefined in the epntap2 mixin (warnings in the validator are OK).

Tests using various clients

A client is a piece of software allowing the user to write queries, and handling answers. Among possible clients, the VESPA portal is a web form optimized for EPN-TAP, available here: http://vespa.obspm.fr

The VESPA portal can access services which are not declared, provided they are on the internet with a fixed address.
The TOPCAT tool also includes a generic TAP client which may be used for testing, although the interface is more technical. TAPhandle (http://saada.unistra.fr/taphandle/) is another alternative.
When testing, check that all parameters appear as you wish with the proper units, and that "special" characters (preferably entered with UTF-8 encoding) display correctly in the interfaces.

Tests using DaCHS

The DaCHS server includes a low-level client which can access its own services (therefore, you don't have to specify a server address). It accepts the same ADQL requests as above.

On your Virtual Machine, the DaCHS query form is located here:
http://127.0.0.1:8080/__system__/adql/query/form

Tests using TOPCAT

You can easily test the service access with TOPCAT.

A graphical tutorial about using TOPCAT to access EPN-TAP services is available here:  http://voparis-europlanet.obspm.fr/utilities/Tuto_TopCat.pdf

• open TOPCAT
• chose menu  VO->Table Access Protocol (TAP) Query
• write the url of your server in the field “TAP URL” at the bottom of the dialogue

To access the DaCHS server installed on a Virtual Machine in EPN-TAP Installation for VESPA Data Provider Tutorial (test mode), enter:

http://127.0.0.1:8080/__system__/tap/run/tap

then click “Use Service”.

TOPCAT will use the specifications of TAP to browse the server and will display all the tables available for query.

Select “iks.epn_core” from the list. All the fields from this table will display under Columns in the right panel.
You can retrieve all granules in TOPCAT by typing the query in the ADQL text field at the bottom:

select * from iks.epn_core

Type "Run Query", the table will load in the Table list.
Click on the 4th icon from the left in TOPCAT (“Display table cell data”). The complete service table will display in TOPCAT.
Go to column access_url, click 3 times on a cell to select and copy it. Select File->Load Table. Paste into location and click OK. The result file (a spectrum) is loaded into TOPCAT.
Select the icon “Scatter Plot”, a graph will plot. If multiple axes are present, you can easily select the axes to be plotted. The rainbow icon allows you to display a third column using colors.

Tests using the VESPA portal

Your service is not yet registered, but if it is installed on a machine directly connected to the internet (or during the implementation workshops) you can test it from the VESPA portal. Alternetively, you can install the VESPA client on your VM (see Installing a local VESPA client).

Click the button “Custom service” on the top of the VESPA page then fill the “Service URL” and “Schema name” fields with the proper indications. If the test IKS service is installed on your Virtual Machine, this is (using the IP number retrieved in the previous step):

Service URL = http://10.42.0.1:8080/tap 
Schema name = iks

You can now issue a test query related to your service in VESPA, using standard EPN-TAP parameters: time between 1986-03-05 and 1986-03-12, target type = comet, dataproduct type = sp (spectrum). VESPA will transform your entries into an ADQL query:

SELECT * FROM ... WHERE (dataproduct_type LIKE '%sp%') AND (target_class LIKE '%comet%') AND time_max <= 2446501.50000000 AND instrument_name = 'IKS' AND time_min >= 2446494.50000000

This query is then embedded into an EPN-TAP request sent to your server.

The “Query results” page describes the service in test and provides the number of results. If matches are found, the box is displayed in green (grey if no result, or red if the service is in error). Click on the title/link in this row to access a list of individual results.

The “Service results” page will display the main parameters of the results, including access_url if the data are linked from the epn_core table. Other parameters (including non-mandatory EPN-TAP parameters) can be displayed using the “Show all” button. From there, your data can be selected and sent to VO tools (see the Help button on this page for help on the interface capabilities).

From the Service results page (after a first query) you can use the “Other” menu on the left panel to access and query the non-mandatory parameters present in your service. URLs and filenames can also be queried this way.

Registering a service

"Registering your service" means to make it known to the Virtual Observatory Registry, a large set of metadata on all the services making up the VO. In case you are curious about the background, see [RD1].

Before proceeding to do publish your metadata, make sure that the general settings are in order. This means:

1. Put some sort of acronym for yourself into the [ivoa]authority in your /etc/gavo.rc (e.g., iks-paris). See below for additional constraints, though it doesn't really matter much if you publish through purx.
2. Review /var/gavo/etc/defaultmeta.txt; in particular try to have some halfway stable e-mail address in contact.email, preferably not your personal mail address.
3. Make sure that in particular the title and description in your RD are clear, concise, and to the point. Thinking about good subjects again might also be a good idea.
4. On DaCHS 2.4 and later, run dachs limits on your RD(s) to update column metadata and coverage information.

With this in place, your resource record should look reasonable. You can review it at http://localhost:8080/getRR/(schema)/q/epn_core, where (schema) is the name of your resource directory; in our example
it would be iks.

This resource record now needs to get into the Registry. Technically, this entails getting services called "full registries" to somehow get it, and there are several ways to make that happen. We will discuss two of them here: The simple way through purx, and the better way through the RofR.

The simple way: purx

Purx is a service run in Heidelberg that regularly pulls registry records you point it to and hands them through to the full registries, doing any identifier magic that needs to be done. With that, here is
what you need to do to get into the Registry:

1. go to http://dc.g-vo.org/purx/q/enroll/custom
2. paste the public URL corresponding to http://localhost:8080/getRR/__system__/tap/run into "VOResource URL" and hit "Next". You only need to do that once regardless of how many EPN tables you register later. If purx complains about validation problems, you'll need to fix the underlying problems (and probably tell dachs-support@g-vo.org about them) at this point.
3. fetch mails from the contact address you gave in defaultmeta.txt and click on the URL that purx has sent you.
4. go back to http://dc.g-vo.org/purx/q/enroll/custom and paste the public URL for http://localhost:8080/getRR/(schema)/q/epn_core. If you have multiple tables, repeat that for all of them.
5. again, fetch mails and click on the purx links to confirm.
6. wait a day or so until your record has propagated, and you should be in the Registry

After that, updates will be automatic once you change things in your service. There will be a delay of up to two days (one for purx to pick up your changes, one for the full registries to pick up changes from purx).

The better way: run a publishing registry

DaCHS has everything needed to run your own publishing registry built in; running one is a particularly good idea if you plan to publish more than a few tables.

To make your resources visible to the VO, you will have to tell the Registry of Registries (RofR) about your registry, and you will need to arrange for some administrative records to be in there. You will also have to make sure your authority (the thing in [ivoa]authority in your /etc/gavo.rc) is not already taken by someone else.

Step by step:

(a) Make sure your authority is not used by anybody.

To do that, go to http://localhost:8080/wirr/q/ui/fixed?field0=ivoid&operator0=%3D&operand0=&field1=restype&operator1=%3D&operand1=vg%3Aauthority and enter your authority into the search field. Hit "Run Query", and if nothing matches, you're good. If there are matches, pick a different authority; you may want to re-import your data after that if you have publisher DIDs or similar in your tables.

(b) Publish the administrative records. That's

dachs pub //services
dachs pub //tap

(c) Publish your tables; in each resource directory, just run

dachs pub q

(d) Go to http://rofr.ivoa.net, hit "validate/register a registry" and paste the URL of your OAI endpoint, which is your public URL with "oai.xml" appended (for instance http://dc.g-vo.org/oai.xml). By the way, you can also point your browser to this latter URL, and if you follow the link to "All identifiers defined here", you can see what you are publishing.

(e) Hit "Validate". If there are problems, try to fix them, and contact dachs-support@g-vo.org if you don't know how.

(f) Once your endpoint validates, you will see a "Register" button. Hit it and you're done. You will have to wait ~one day until your resources become visible, because the full registries will only harvest you about once per day.

When you do changes to your RDs, say dachs pub q again to tell DaCHS that relevant changes have occurred. This will make it re-feed the records to the Registry. If you have to change your server's URL later, just send a mail to the RofR operators and notify them of this change.

Registering to VESPA 

The above procedure publishes your service in the IVOA registry, which makes it accessible from TOPCAT and other standard tools.

However, you also need to drop a mail to the VESPA team to make your service accessible from the VESPA portal with other validated services. The VESPA team will perform a manual review of the service and check consistency with other related services before publication.

Once the registry declaration is validated your service will seen by the VESPA portal, and coordinated queries will be possible. Use the default button “Form” on the left panel of VESPA portal's first page. The result will be a list of available services. Those containing answers to your query will be displayed in green and you will be able to select results from individual services.

Annex I : possible issues

• When copy/pasting the above examples, beware of special spaces! In case of problem, erase and replace them manually.

• Also remove any trailing and heading space in strings (especially in granule IDs), they can block the interfaces.

• The version of postgresql used in examples is 9.4. You may have to adapt some commands in the above examples to the version number in use.

• To log on your VM from the guest terminal, type

ssh -p 2222 <user>@localhost

• To copy files from your host machine to the guest, use:

scp -P 2222 *.sql <user>@localhost:~/iks/.

(note this is uppercase P — this is different from the ssh command)

• A possibly simpler solution to produce VOTables under IDL/GDL is to use the write_vot.pro routine, which is a wrapper routine of the stilts/TOPCAT library: https://github.com/epn-vespa/IDL_VOtable.

Again, VOTable handling is easier in python, using astropy.

• In some situations, you may need to erase a first try of your service and start again from a clean state (e. g., if you've changed the directory name, or see a persistent error with ascii encoding). To achieve this you need to first drop the service before importing the q.rd and restarting the server:

dachs drop -f q.rd
dachs val q.rd
dachs imp q.rd
sudo dachs serve restart

• When validating a service with the ObsParis validator, beware that the validator may be out of sync with the epntap2 mixin installed on older server versions — in this case "fatal errors" are actually benign.

(refer to previous versions of this doc)

• Some installations of java on OS X / Mac are known to be unstable - in particular v1.7 produces numerous graphic bugs and incompatibilities. TOPCAT used with the jdbc extension is particularly sensitive to this, as well as some other VO tools. Both versions 1.6 and 1.8 appear to work smoothly. If the interface looks unstable, keep the window size unchanged to minimize graphic mismatches.

• jdbc version must be consistent with your versions of both postgresql and java, see here: https://jdbc.postgresql.org/download.html.

It can be retrieved e.g. (check for the correct version) with: wget https://jdbc.postgresql.org/download/postgresql-9.4.1208.jre6.jar

• If the jdbc is ran inside a Virtual Machine, you first need to open an ssh tunnel in order to reach the distant database from the host:

ssh <user>@127.0.0.1 -p 2222 -L 5432:127.0.0.1:5432     

(see EPN-TAP Installation for VESPA Data Provider Tutorial — Part 1)

• Under Windows + Cygwin, lauching the jdbc requires a different syntax:

java -cp "topcat-full.jar;postgresql-9.1-903.jdbc4.jar" -Djdbc.drivers=org.postgresql.Driver uk.ac.starlink.topcat.Driver