r4624 and actually fixed it.

The module now supports recursive search of client address through the
chain of trusted proxies (closes #100), in the same scope as the geo
module.  Proxies are listed by the "geoip_proxy" directive, recursive
search is enabled by the "geoip_proxy_recursive" directive.  IPv6 is
partially supported: proxies may be specified with IPv6 addresses.

Example:
    geoip_country .../GeoIP.dat;
    geoip_proxy 127.0.0.1;
    geoip_proxy ::1;
    geoip_proxy 10.0.0.0/8;
    geoip_proxy_recursive on;

The module now supports recursive search of client address through
the chain of trusted proxies, controlled by the "proxy_recursive"
directive in the "geo" block.  It also gets partial IPv6 support:
now proxies may be specified with IPv6 addresses.

Example:
    geo $test {
        ...
        proxy 127.0.0.1;
        proxy ::1;
        proxy_recursive;
    }

There's also a slight change in behavior.  When original client
address (as specified by the "geo" directive) is one of the
trusted proxies, and the value of the X-Forwarded-For request
header cannot not be parsed as a valid address, an original client
address will be used for lookup.  Previously, 255.255.255.255 was
used in this case.

The module now supports recursive search of client address through
the chain of trusted proxies, controlled by the "real_ip_recursive"
directive (closes #2).  It also gets full IPv6 support (closes #44)
and canonical value of the $client_addr variable on address change.

Example:
    real_ip_header X-Forwarded-For;
    set_real_ip_from 127.0.0.0/8;
    set_real_ip_from ::1;
    set_real_ip_from unix:;
    real_ip_recursive on;

On input it takes an original address, string in the X-Forwarded-For format
and its length, list of trusted proxies, and a flag indicating to perform
the recursive search.  On output it returns NGX_OK and the "deepest" valid
address in a chain, or NGX_DECLINED.  It supports AF_INET and AF_INET6.
Additionally, original address and/or proxy may be specified as AF_UNIX.

Due to weight being set to 0 for down peers, order of peers after sorting
wasn't the same as without the "down" flag (with down peers at the end),
resulting in client rebalancing for clients on other servers.  The only
rebalancing which should happen after adding "down" to a server is one
for clients on the server.

The problem was introduced in r1377 (which fixed endless loop by setting
weight to 0 for down servers).  The loop is no longer possible with new
smooth algorithm, so preserving original weight is safe.

For edge case weights like { 5, 1, 1 } we now produce { a, a, b, a, c, a, a }
sequence instead of { c, b, a, a, a, a, a } produced previously.

Algorithm is as follows: on each peer selection we increase current_weight
of each eligible peer by its weight, select peer with greatest current_weight
and reduce its current_weight by total number of weight points distributed
among peers.

In case of { 5, 1, 1 } weights this gives the following sequence of
current_weight's:

     a  b  c
     0  0  0  (initial state)

     5  1  1  (a selected)
    -2  1  1

     3  2  2  (a selected)
    -4  2  2

     1  3  3  (b selected)
     1 -4  3

     6 -3  4  (a selected)
    -1 -3  4

     4 -2  5  (c selected)
     4 -2 -2

     9 -1 -1  (a selected)
     2 -1 -1

     7  0  0  (a selected)
     0  0  0

To preserve weight reduction in case of failures the effective_weight
variable was introduced, which usually matches peer's weight, but is
reduced temporarily on peer failures.

This change also fixes loop with backup servers and proxy_next_upstream
http_404 (ticket #47), and skipping alive upstreams in some cases if there
are multiple dead ones (ticket #64).

With r->filter_finalize set the ngx_http_finalize_connection() wasn't
called from ngx_http_finalize_request() called with NGX_OK, resulting in
r->main->count not being decremented, thus causing request hang in some
rare situations.

See here for more details:
http://mailman.nginx.org/pipermail/nginx-devel/2012-May/002190.html

Patch by Yichun Zhang (agentzh).