Wildcard DB configuration lost on reload

[darix]

my pgbouncer.ini without comments:

[databases]
* = host=::1 port=6432 auth_user=pgbouncer pool_size=100
[pgbouncer]
logfile = /var/log/pgbouncer/pgbouncer.log
pidfile = /var/run/pgbouncer/pgbouncer.pid
listen_addr = ::1, 127.0.0.1, 172.16.16.2
listen_port = 5432
client_tls_sslmode=allow
client_tls_sslmode=allow
client_tls_ca_file=/etc/ssl/ca-bundle.pem
client_tls_key_file = /etc/pgbouncer/cert+key.pem
client_tls_cert_file = /etc/pgbouncer/cert+key.pem
client_tls_protocols = secure
client_tls_dheparams = auto
server_tls_sslmode = require
auth_type = hba
auth_file = /etc/pgbouncer/userlist.txt
auth_hba_file = /etc/pgbouncer/pg_hba.conf
auth_query=select * from pgbouncer.user_lookup($1);
admin_users = pgadmin
stats_users = pgadmin
pool_mode = session
server_reset_query = DISCARD ALL
max_client_conn = 1024
default_pool_size = 100

This config works fine after a fresh start, after sending the reload signal. it starts to Pooler Error: login rejected for connections that were working until the restart. could this be caused by calling config_postprocess() in load_config()?

This is especially annoying as you would normally use sighup for reopening the log files after logrotate. For openSUSE I created this small patch to have a different signal for the logrotate job until this issue is fixed.

Index: pgbouncer-1.8.1/src/main.c
===================================================================
--- pgbouncer-1.8.1.orig/src/main.c
+++ pgbouncer-1.8.1/src/main.c
@@ -437,6 +437,7 @@ static void handle_sigint(int sock, shor
 static struct event ev_sigusr1;
 static struct event ev_sigusr2;
 static struct event ev_sighup;
+static struct event ev_sigttin;
 
 static void handle_sigusr1(int sock, short flags, void *arg)
 {
@@ -476,6 +477,12 @@ static void handle_sighup(int sock, shor
        log_info("Got SIGHUP re-reading config");
        load_config();
 }
+
+static void handle_sigttin(int sock, short flags, void *arg)
+{
+       log_info("Got SIGTTIN re-open logfiles");
+       reset_logging();
+}
 #endif
 
 static void signal_setup(void)
@@ -508,6 +515,11 @@ static void signal_setup(void)
        err = signal_add(&ev_sighup, NULL);
        if (err < 0)
                fatal_perror("signal_add");
+
+       signal_set(&ev_sigttin, SIGTTIN, handle_sigttin, NULL);
+       err = signal_add(&ev_sigttin, NULL);
+       if (err < 0)
+               fatal_perror("signal_add");
 #endif
        signal_set(&ev_sigterm, SIGTERM, handle_sigterm, NULL);
        err = signal_add(&ev_sigterm, NULL);

[petere]

I can't reproduce that with a standard setup. Maybe it has something to do with the authentication setup you have. Can you play around with that a bit and see what makes a difference?

[darix]

what do you want me to test? the set up we use is basically the example with the SQL query.

[justinpryzby]

@darix is this still an issue with recent versions of pgbouncer ?

[darix]

I am still using the patch described above. So I havent tested that in a long long time.

I think having a signal for just reopening log files makes the whole logrotate bit cheaper anyway.