Re: Small problem with... test harness? SPARQL processor?

Ivan Herman wrote:
> I installed my latest version of pyRdfa distiller which corresponds to
> the Last Call version. I ran the test harness and I got an error on
> tests #1, #29, #51. 

There is an issue with Virtuoso/SPARQLer sites crashing on random
requests when 60 near-simultaneous requests come from the same IP. Our
server logs give this:

HTTPError: HTTP Error 500: SPARQL Request Failed,
           referer: http://rdfa.digitalbazaar.com/rdfa-test-harness/

In general, if you see ERROR - this is what happened. We didn't have
enough test cases before where this was an issue... but since we have
many more test cases now, it is certainly an issue.

I'll try adding something to throttle the outgoing connections :)

> Well... what is weird is that if I run the test
> harness through Virtuoso, then this is what I get; if I use sparqler,
> the #1 is fine, and #29 and #51 are still errors.

#29 and #51 are failing due to the whitespace preservation. Your
implementation is probably doing the right thing - the SPARQL tests are
incorrect for those two. We will have to fix them.

> I wonder whether you already changed your implementation and you had (or
> not) experiences with it. I also wonder whether this is not something
> where we might have problems with the way sparql processors compare
> strings; with the non-normalized versions we have now I am not sure what
> exactly happens:-( (Which might also explain why virtuoso and sparqler
> behave differently...)

The problems boil down to two things:

- Both SPARQL endpoints are throwing up errors because we're slamming
  each site whenever we do a "Run All Tests".
- Test #29 and #51 assumed whitespace canonicalization, which we don't
  do anymore.

> P.S. One tiny feature request to the test harness: would it be possible
> to add a link to each individual test file from the test harness itself?

Yes, it would. I'll put this in whenever I get around to messing with
the test harness again (don't expect it soon, my plate is full!) :)

> When I get an 'error' I would of course like to see what the error is;
> if I get the URI of the test file itself (ie, the xhtml file) I can
> continue with testing right away, whereas I have to locate that file
> first now...

I'll try to add an area for error feedback - although, it's almost
always the same thing (the SPARQL endpoint died unexpectedly).

-- manu

-- 
Manu Sporny
President/CEO - Digital Bazaar, Inc.
blog: RDFa Basics in 8 minutes (video)
http://blog.digitalbazaar.com/2008/01/07/rdfa-basics/

Received on Thursday, 21 February 2008 13:32:39 UTC