Package: postgresql-common
Version: 94
Severity: Serious
pg_ctlcluster conflicts with instances with the same configured port, but
different listen_addresses and unix_socket_directory:
$ grep port db*/postgresql*
db2/postgresql.conf:port = 5432
db/postgresql.conf:port = 5432
$ grep listen_addresses db*/postgresql*
db2/postgresql.conf:listen_addresses = '192.168.0.111'
db/postgresql.conf:listen_addresses = 'localhost'
$ grep unix_socket_directory db*/postgresql*
db2/postgresql.conf:unix_socket_directory = '/var/run/postgresql/8.3/db2'
db/postgresql.conf:unix_socket_directory = '/var/run/postgresql/8.3/db'
The culprit seems to be cluster_port_running(), called by pg_ctlcluster in
the subroutine start(), to make sure the specified unix socket is not
accidently used by another running instance:
[...]
# check conflicting port
foreach my $v (get_versions) {
foreach my $c (get_version_clusters $v) {
error("Port conflict: cluster $v/$c is already running on port
" .
$info{'port'}) if cluster_port_running $v, $c,
$info{'port'};
}
}
[...]
This makes it impossible to use several instances on different
listen_addresses with the same port number. It seems to me the whole check
is far too harsh anyways: it might be sufficient to check against the
current unix socket directory only. If the connect() to this unix socket
succeeds, then there is another instance using this unix socket already. I
propose the following patch to fix the behavior of pg_ctlcluster:
diff --git a/pg_ctlcluster b/pg_ctlcluster
index 531e6b2..335e4f4 100755
--- a/pg_ctlcluster
+++ b/pg_ctlcluster
@@ -244,11 +244,12 @@ sub start {
start_check_pid_file;
# check conflicting port
- foreach my $v (get_versions) {
- foreach my $c (get_version_clusters $v) {
- error("Port conflict: cluster $v/$c is already running on port
" .
- $info{'port'}) if cluster_port_running $v, $c,
$info{'port'};
- }
+ if (cluster_port_running $version, $cluster, $info{'port'}) {
+
+ my $sockdir = PgCommon::get_conf_value $version, $cluster,
'postgresql.conf',
+ 'unix_socket_directory';
+ error("Port conflict: another instance is already running on
".$sockdir." with port ".$info{'port'});
+
}
# get locale used by initdb
--
Regards
Bernd Helmle
--
To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
