backend hangs at sendto() and can't be terminated

one of our pg user send a select * from 10 million table without limit from
psql，before get the return data，he realize the mistake and quite from
psql.but after 2 hours,the sql still alive:

postgres=# SELECT * from pg_stat_activity where procpid = 8243;
-[ RECORD 1 ]----+---------------------------------
datid | 758972
datname | xxx
procpid | 8243
usesysid | 661846
usename | test
application_name | psql
client_addr | 10.136.4.90
client_hostname |
client_port | 6382
backend_start | 2013-07-08 14:11:00.942293+08
xact_start | 2013-07-08 14:31:11.157681+08
query_start | 2013-07-08 14:31:11.157681+08
waiting | f
current_query | select * from yyyy;

pg_terminate_backend return t but the backend still there.

strace the pid show the process hang at sendto() function call:
[postgres(at)xxx ~]$ strace -tv -p 8243
Process 8243 attached - interrupt to quit
17:02:26 sendto(10, "70804\0\0\0\f{2012070804}D\0\0\0`\0\5\0\0\0\1"...,
8152, 0, NULL, 0^C <unfinished ...>
Process 8243 detached

os is centos 6 x86-64,pg version is 9.1.9.

the process still there,how can I help to debug the problem?

Jov
blog: http:amutu.com/blog <http://amutu.com/blog>