Forum OpenACS Q&A: host / site-node map

Greenpeace uses a modified version of this but we always redirect the user to our "smart" URLs.  Bruno Mattarollo enhanced it slightly, but I don't remember the details.

Could you expand a bit on the bugs you've found?

What I have done is actually to modify it so that you could use a redirect from somehost.foo.com to mainsite.foo.org/some/url/

This has some disavantages, like if you want to preserve the hostname somehost.foo.com in the address bar of the users' browser, you can't do that with my modified version. Also, at this stage, my version only redirects domains, without taking care of path or query strings, i.e. you can't do something smart like somehost.foo.com/somepath/here?somequery=string to mainsite.foo.org/some/url/somepath/here?somequery=string. Why I didn't do this? because I think that it can be tricky and confusing and really, we didn't needed it @ Greenpeace.

Do you want to get the small modifications I did?

Ben asked

- is anyone extensively using host site node mapping? I've found many bugs in the way it works and am wondering if my initial fixes (specifically in directory redirection and index file redirection) are worth uploading, or if we should just let the host side node map stuff die (as it is quite horrendous as is).

We'll probably be using it for parts of Greenpeace Planet, if it doesn't misbehave too badly. Being able to map, say, www.greenpeace.nl to www.greenpeace.org/nederland solves a bunch of problems with how Google indexes regional subsites. Ben, if you're not going to commit your fixes, could you send them along my way?

I found a problem that perhaps Ben did too. There's a tweak in rp_filter that Don added right after the hostname-based subsites patch:

    # DRB: a bug in ns_conn causes urlc to be set to one and urlv to be set to
    # {} if you hit the site with the host name alone.  This confuses code that
    # expects urlc to be set to zero and the empty list.  This bug is probably due
    # to changes in list handling in Tcl 8x vs. Tcl 7x.

    if { [ad_conn urlc] == 1 && [lindex [ad_conn urlv] 0] == "" } {
        ad_conn -set urlc 0
        ad_conn -set urlv [list]
    }

It turns out that the same problem crops if you hit the site with a mapped hostname alone, so you need a different tweek. Something like this:

    if { [ad_conn urlc] == 1 && [lindex [ad_conn urlv] 0] == "" } {
        ad_conn -set urlc 0
        ad_conn -set urlv [list]

    } elseif { [ad_conn urlc] > 1 && [lindex [ad_conn urlv] end] == "" } {

        ad_conn -set urlc [expr [ad_conn urlc] - 1]
        ad_conn -set urlv [list [lreplace [ad_conn urlv] end end]]
    }

I haven't had a chance to test this extensively as of yet.

There's another issue with host-node mapping that may have been kicked around a little already: if you login to the site from one hostname, and then switch to a subsite using a different, "mapped" hostname, you loose your login... because the cookie has a different name. I remember looking into this briefly about a year ago, but only vaguely. I think the original specification for cookies says the server can set the hostname of the cookie it's accessing, but in practice it's a big security hole. I think depending on the browser security settings, you can't read/write cookies for a different host. Again, I'm dredging this up from memory.

Some people do that with a series of cookie-setting redirects going through all the possible domains. It sounds ugly, and it is ugly.

If I remember correctly this was implemented in openacs 3.x (which does not make it less ugly).

is anyone extensively using host site node mapping?

Not yet, but I might want to do so later.

In general I think the host-node mapping stuff is very interesting for small OpenACS sites. Having the possibility to quickly add a site with its own domain to an existing OpenACS installation is great, e.g. if you want to do your friends a favour who need a simple low-traffic site with some functionality. It would be sad if OpenACS lost this functionality - so: yes, your fixes are highly appreciated.

It also looks like there's a bug in the data model for the host-node map. Here's the table that stores the map:

create table host_node_map (
   host         varchar(200),
   node_id      integer not null
                constraint host_node_map_pk primary key
                constraint host_node_map_fk
                           references acs_objects (object_id)
);

Because the node_id is the primary key for the table, it's impossible to have multiple hostnames pointing to the same node (a restriction that makes no sense), but it is possible to have the same hostname appear multiple times, pointing to different node_ids (an ambiguous situation that would probably cause some pretty nasty breakage if it ever came to be).

I'm changing the table so that host, rather than node_id, is the primary key.

I've just started playing with site-node and I think I've been noticing some of the strangeness people mention in this thread. I was under the impression that if I mapped

/foo/

to

foo.mydomain.com

then that site node would exist on it's own -- have it's own login, sitemap, users and groups -- but that's not what I'm getting at:

http://cook.mytlc.com/

But when I go to:

http://cook.mytlc.com/register/logout

I don't get logged out. I have to go to:

http://www.mytlc.com/

to log out.

Can anyone spot what's wrong? Is the system intended to work this way? Is everything at

*.mydomain.com

intended to share the same login? Anyway, I would use these site nodes if they worked. One down side is that singleton applications can't work on all the nodes.

Eric