TAP query

A library to query TAP servers

Author: Markus Demleitner


TAP is a relatively complex protocol to execute potentially long-running (ADQL) queries on remote servers. As a part of the GAVO VOTable library, we provide a shallow client that speaks TAP. It is the intention of this document to keep the TAP spec from your reading list. If it doesn't do that, complain, and we'll try to fix it.

Obtaining the library

See the GAVO VOTable Library documentation

For the impatient

All you need to query a server is its access URL. What we need here is the root of the hierarchy, i.e., without any sync or async.

For the typcial case of a sync query, here's the pattern to retrieve a sequence of dictionaries, one per result line, using sync querying:

from gavo import votable

res = votable.ADQLSyncJob(

data, metadata = votable.load(res.openResult())
for row in metadata.iterDicts(data):
  print row

Here's a variation using async querying (meaning: your query can take a long time if necessary), and yielding a numpy record array:

from gavo import votable
from numpy import rec

accessURL = "http://dc.zah.uni-heidelberg.de/__system__/tap/run/tap"
query = "SELECT TOP 3 * FROM TAP_SCHEMA.tables"
job = votable.ADQLTAPJob(accessURL, query)
  data, metadata = votable.load(job.openResult())
ra = rec.fromrecords(data, dtype=votable.makeDtype(metadata))

Note, however, that with record arrays NULL values will cause the whole thing to fail. So, if you are after record arrays, our advice is to install astropy and write:

from cStringIO import StringIO

from astropy.table import Table
from gavo import votable

accessURL = "http://dc.zah.uni-heidelberg.de/__system__/tap/run/tap"
query = "SELECT TOP 3 * FROM TAP_SCHEMA.tables"
job = votable.ADQLTAPJob(accessURL, query)
  table = Table.read(StringIO(job.openResult().read()), format="votable")

print table

More on dataIterator and metadata can be found in the GAVO VOTable library documentation.


The class you will usually deal with is ADQLTAPJob, which is constructed with the endpoint URL and the query. The construction will access the, which means it may very well raise network-related exceptions.

You can pass a userParams dictionary to the constructor. This is intended for tho TAP parameters (in particular, FORMAT, MAXREC, RUNID or service-defined ones). You should not use userParams for UPLOAD but instead use the addUpload method described below.

You can also change the parameters later using setParameter(key, value). This again causes a server connection to be made, as are accesses to ADQLTAPJob's properties, viz.,

Running a job

Once you have constructed (and possibly modified) a job, call start() to tell the remote server to put it into its execution queue. You can then poll the job's phase (now and then):

job = votable.ADQLTAPJob(...)
job.executionDuration = 6000
while job.phase not in set([tapquery.ERROR, tapquery.COMPLETED]):

You can call abort() to kill a running job, and you can call delete() when done. As a matter of fact, although server operators will eventually destroy your job anyway, it's common courtesy to clean up behind you unless you have good reason to keep the result data. So the recommended pattern is:

job = votable.ADQLTAPJob(...)
  ... do your thing ...

Now, doing the polling by hand is tedious, and this there is a function waitForPhases(phases, ...) that takes a set of phases to wait for, and then polls at increasing intervals (that you could control though the method's further arguments; see API docs for details).

But really, the run method is what you'd usually use; it has the added advantage that it raises a RemoteError or a RemoteAbort exception if something went wrong on the server side. So, the standard pattern is:

job = votable.ADQLTAPJob(...)
  ... add uploads, set parameters if you must ...
  ... change UWS parameters (executionDuration, destruction, etc) ...
  ... read results ...

TAP parameters

The following parameters are defined by the TAP spec:

Getting results

Right now, this library heavily leans towards ADQL. When doing ADQL queries, there is only one result. You can access it using the openResult() method, which returns a file-like object (actually, it's what urllib.urlopen returns) that you can read your results from. Since successful ADQL queries return single-table VOTables, you can feed this directly to votable.load(f):

data, metadata = votable.load(job.openResult())

ADQL TAP results are simple web resources. To get their URLs, there's the getResultURL method. This is particularly useful with uploads (see below).

More complex query patterns could yield more results; in that case, you will have to inspect job.results, which is a list of triples of an access URL, an opaque string-typed id, and a UWS "result type" that you probably can safely ignore.


All exceptions originating in the library are subclasses of tapquery.Error. For simple applications, this is also available as votable.TAPQueryError.

The API docs list some exceptions you should expect; also, of course, all kinds of network-related exceptions could come out of the library. No serious attempt is being made to catch such exceptions and translate them. If something is wrong network-wise, we feel it's better to freely admit this.

TAP-level errors (like syntax errors in the query, timeouts, and the like) leave error information server-side. Rather than inspecting the exception objects, you should use the getErrorFromServer method on the ADQLTAPJob.

In this, proper TAP error messages (those coming back as VOTables) are parsed out, data that's not parseable as a TAP error message is returned as-is; thus the string returned may be long (e.g., some fancy 404 HTML page).


TAP allows uploads to servers, although not all services actually support this (you can use a service's capabilities to figure that out; we may add support for parsing those if there's demand).

Services that do support it let you at least upload using URLs to VOTables or upload VOTables inline. Both cases are supported in ADQLTAPJob's addUpload(name, data) method. name is the name that the table will be visible as later. Use something that works as a regular SQL identifier here.

data can be either a string (which is then interpreted as the URL to take the upload from) or a file-like object (that is then turned into an inline upload). In both cases, only VOTable is (reliably) supported as the upload format.

Note that URLs of previous results work as upload URLs. Here's an example of how that might look like, where this retrieves all objects in PPMXL in the vicinity of a detection of a neutrino:

from gavo import votable

accessURL = "http://dc.g-vo.org/tap"
job1 = votable.ADQLTAPJob(accessURL,
  "SELECT nualpha, nudelta FROM amanda.nucand WHERE nch>100")

job2 = votable.ADQLTAPJob(accessURL,
  "SELECT DISTINCT TOP 10000 raj2000, dej2000, pmRA, pmDE"
  " FROM ppmxl.main"
  " JOIN TAP_UPLOAD.nupos"
  "   ON (1= CONTAINS(POINT('ICRS',raj2000,dej2000),"
  "      CIRCLE('ICRS', nualpha , nudelta, 0.25)))")
job2.addUpload("nupos", job1.getResultURL())

dataIterator, metadata = votable.load(job2.openResult())
print list(dataIterator)

Sync Querying

TAP also supports a synchronous querying mode; for simple, quick queries, this is simpler and has less overhead. So, if you are certain that your query will run quickly and the result set is small, you can use the ADQLSyncJob. This works mostly like the ADQLTAPJob, except of course everything that deals with remote state management basically is a no-op. Instead, either run or start just query the remote server and return when the server is done. For convenience, they return the job itself, so that you can say things like:

print votable.ADQLSyncJob(

All "unpredictable" exceptions on sync jobs are raised from within run; these will usually be tapquery.WrongStatus or tapquery.NetworkError exceptions, for when the TAP server has complained or something was wrong on the way between the client and the server.

The WrongStatus instances have a payload attribute that contains any message body the server might have sent with the headers; frequently this contains explanations what may have gone wrong. Since no http headers are available, there's no saying what format the error message came in. Tell us if that bugs you.

The error associated with the exception object will usually not be particularly useful. Instead, obtain an error message from the server using the getErrorFromServer method like for ADQLTAPJobs.