IRC log of sparql on 2010-10-06

Timestamps are in UTC.

00:47:40 [LeeF]
LeeF has joined #sparql
02:41:18 [AlexPassant]
AlexPassant has joined #sparql
03:31:41 [iv_an_ru]
iv_an_ru has joined #sparql
07:38:50 [AndyS]
AndyS has joined #sparql
08:44:43 [AndyS]
AndyS has joined #sparql
08:45:54 [SteveH_]
SteveH_ has joined #sparql
09:15:58 [AndyS]
AndyS has joined #sparql
10:20:03 [AndyS]
SteveH_, good catch on xml:base.
10:20:15 [SteveH_]
AndyS, that bit is kinda nuts
10:20:34 [AndyS]
Don't panic ! :-) the doc is wrong.
10:21:12 [AndyS]
There can be many xml:base in a doc, it only applied to the XML element it is an attribute of.
10:21:34 [SteveH_]
that's what I thought, but wasn't 100% sure
10:29:32 [SteveH_]
the requirement for handling local URIs is a bit optimisitic as well, I bet not many systems can get that right relaibly
10:30:25 [AndyS]
What is the requirement?
10:30:48 [SteveH_]
it would be impossible in 4store, and we built our own HTTP server
10:31:07 [SteveH_]
...?graph=foo should reconstruct the correct URI based on the client's request URI
10:31:44 [SteveH_]
e.g. if the request was https://server1:8888/data/?graph=foo the graph URI will be https://server1:8888/data/foo
10:31:59 [SteveH_]
but in many cases the server has no way to correctly reconstruct that URI
10:32:19 [AndyS]
The resolve "foo" against http://example/rdf-graphs/service/?graph=foo ?
10:32:52 [SteveH_]
let me look what the example was
10:33:06 [SteveH_]
the example in the doc was a simple one, which doesn't exhibit the problem
10:33:07 [AndyS]
I think that proxies rely sufficient info to reconstruct the original URL but I'm just checking this.
10:33:16 [AndyS]
s/rely/relay/
10:33:21 [SteveH_]
they don't relay http v's https
10:33:23 [SteveH_]
I checked
10:33:34 [AndyS]
Ah. Bother. Tricky.
10:33:55 [SteveH_]
I'm not convinced they relay the clients requester domain name either, but not sure about that
10:34:02 [AndyS]
I'm checking what relayed servlets get to see.
10:34:23 [SteveH_]
s/requester/requested/
10:34:25 [AndyS]
I have no idea what is real-world behaviour.
10:34:52 [SteveH_]
we tend to use apache reverse proxies, not sure how much of their behaviour is req'd or conventional
10:35:14 [SteveH_]
I don't think relays/reverse proxies are goverened by any particular standard, but not sure about that
10:44:19 [AndyS]
Servlet getRequestURL rebuilds the original URL. I use apache to gateway "sparql.org" (httpd) to a more protected Jetty instance.
10:47:48 [AndyS]
I don't use any clever connector from Apache - it proxies:
10:48:36 [AndyS]
ServerName www.sparql.org
10:48:37 [AndyS]
ProxyRequests off
10:48:37 [AndyS]
ProxyPass / http://127.0.0.1:2020/
10:48:37 [AndyS]
ProxyPassReverse / http://127.0.0.1:2020/
10:48:37 [AndyS]
ProxyPreserveHost On
10:49:44 [AndyS]
Hmm - just had to restart httpd - think someone/thing is attacking (maybe by accidient) sparql.org exploiting a Jetty bug. Have now upgraded and hope it goes away.
10:51:41 [SteveH_]
ah, so Apache ProxyPass passes on the requested domain, so that's covered in that case, but not everything has the same featureset as apache, and it still doesn't pass http / https
10:52:07 [SteveH_]
wait, no, that's your config
10:52:45 [SteveH_]
but if getRequestURL rebuilds it, the data must be there somewhere
11:01:54 [AndyS]
time to get out wireshark
11:02:48 [SteveH_]
or nc :)
11:06:47 [AndyS]
(that's a cut down config extract)
11:07:24 [SteveH_]
right
11:08:00 [AndyS]
Not sure it passes http/https but if https -> https and http->http then OK and works as backend does the https (e.g. cert) not front public facing.
11:15:43 [SteveH_]
we do https -> http for 4/5store because our server doesn't speak https
11:16:08 [SteveH_]
we also do it for jetty in datapatrol because there's something wrong with jetty's https support (apparently)
11:16:24 [SteveH_]
I know nothing about jetty