On 2017/03/31 8:28, David Rowley wrote:
> create table t (a int, b int);
> insert into t1 select x/100,x/100 from generate_series(1,100000) x;
> create extension if not exists postgres_fdw;
> create server test_server foreign data wrapper postgres_fdw options
> (host 'localhost', port '5432', dbname 'postgres');
> create foreign table ft_t (a int,b int) server test_server;
> select 'create user mapping for current_user server test_server
> options(user ''' || current_user || ''');';
> \gexec
> select count(*) from pg_stat_Activity; -- > 6
> analyze ft_t;
> ERROR: could not connect to server "test_server"
> DETAIL: FATAL: sorry, too many clients already
> CONTEXT: Remote SQL command: DECLARE c1 CURSOR FOR SELECT a, b FROM
> public.ft_t
> Remote SQL command: SELECT a, b FROM public.ft_t
> Remote SQL command: SELECT a, b FROM public.ft_t
> Remote SQL command: SELECT a, b FROM public.ft_t
> (lots of these)
>
> select count(*) from pg_stat_Activity; --> 105
>
> I've not had a moment to check into what's going on.
IIUC, I think the cause would be that since the foreign table ft_t is
considered to be still foreign on the foreign server, which is actually
the same server, postgres_fdw recursively repeats the loopback access to
ft_t. (So, the same thing would happen for something like: select *
from ft_t.) If the analysis is right, ISTM that it's the user's fault.
Best regards,
Etsuro Fujita