|
 |
0c4b27 |
diff --color -uNr a/doc/man/Makefile.am b/doc/man/Makefile.am
|
|
|
0c4b27 |
--- a/doc/man/Makefile.am 2021-04-12 12:51:56.831835953 +0200
|
|
|
0c4b27 |
+++ b/doc/man/Makefile.am 2021-04-13 13:38:14.198361848 +0200
|
|
|
0c4b27 |
@@ -154,6 +154,7 @@
|
|
|
0c4b27 |
ocf_heartbeat_ovsmonitor.7 \
|
|
|
0c4b27 |
ocf_heartbeat_pgagent.7 \
|
|
|
0c4b27 |
ocf_heartbeat_pgsql.7 \
|
|
|
0c4b27 |
+ ocf_heartbeat_pgsqlms.7 \
|
|
|
0c4b27 |
ocf_heartbeat_pingd.7 \
|
|
|
0c4b27 |
ocf_heartbeat_podman.7 \
|
|
|
0c4b27 |
ocf_heartbeat_portblock.7 \
|
|
|
0c4b27 |
diff --color -uNr a/heartbeat/Makefile.am b/heartbeat/Makefile.am
|
|
|
0c4b27 |
--- a/heartbeat/Makefile.am 2021-04-12 12:51:56.831835953 +0200
|
|
|
0c4b27 |
+++ b/heartbeat/Makefile.am 2021-04-13 13:37:45.741292178 +0200
|
|
|
0c4b27 |
@@ -149,6 +149,7 @@
|
|
|
0c4b27 |
ovsmonitor \
|
|
|
0c4b27 |
pgagent \
|
|
|
0c4b27 |
pgsql \
|
|
|
0c4b27 |
+ pgsqlms \
|
|
|
0c4b27 |
pingd \
|
|
|
0c4b27 |
podman \
|
|
|
0c4b27 |
portblock \
|
|
|
0c4b27 |
@@ -209,7 +210,10 @@
|
|
|
0c4b27 |
mysql-common.sh \
|
|
|
0c4b27 |
nfsserver-redhat.sh \
|
|
|
0c4b27 |
findif.sh \
|
|
|
0c4b27 |
- ocf.py
|
|
|
0c4b27 |
+ ocf.py \
|
|
|
0c4b27 |
+ OCF_Directories.pm \
|
|
|
0c4b27 |
+ OCF_Functions.pm \
|
|
|
0c4b27 |
+ OCF_ReturnCodes.pm
|
|
|
0c4b27 |
|
|
|
0c4b27 |
# Legacy locations
|
|
|
0c4b27 |
hbdir = $(sysconfdir)/ha.d
|
|
|
0c4b27 |
diff --color -uNr a/heartbeat/OCF_Directories.pm b/heartbeat/OCF_Directories.pm
|
|
|
0c4b27 |
--- a/heartbeat/OCF_Directories.pm 1970-01-01 01:00:00.000000000 +0100
|
|
|
0c4b27 |
+++ b/heartbeat/OCF_Directories.pm 2021-04-13 13:37:35.621267404 +0200
|
|
|
0c4b27 |
@@ -0,0 +1,139 @@
|
|
|
0c4b27 |
+#!/usr/bin/perl
|
|
|
0c4b27 |
+# This program is open source, licensed under the PostgreSQL License.
|
|
|
0c4b27 |
+# For license terms, see the LICENSE file.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# Copyright (C) 2016-2020: Jehan-Guillaume de Rorthais and Mael Rimbault
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 NAME
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+OCF_Directories - Binaries and binary options for use in Resource Agents
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SYNOPSIS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ use FindBin;
|
|
|
0c4b27 |
+ use lib "$FindBin::RealBin/../../lib/heartbeat/";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ use OCF_Directories;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 DESCRIPTION
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+This module has been ported from the ocf-directories shell script of the
|
|
|
0c4b27 |
+resource-agents project. See L<https://github.com/ClusterLabs/resource-agents/>.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 VARIABLES
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Here are the variables exported by this module:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=over
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $INITDIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_DIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_RCDIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_CONFDIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_CF
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_VARLIB
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_RSCTMP
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_RSCTMP_OLD
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_FIFO
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_BIN
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_SBIN_DIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_DATEFMT
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_DEBUGLOG
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_RESOURCEDIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_DOCDIR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $__SCRIPT_NAME
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_VARRUN
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $HA_VARLOCK
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $ocf_prefix
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $ocf_exec_prefix
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=back
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+package OCF_Directories;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use strict;
|
|
|
0c4b27 |
+use warnings;
|
|
|
0c4b27 |
+use 5.008;
|
|
|
0c4b27 |
+use File::Basename;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+BEGIN {
|
|
|
0c4b27 |
+ use Exporter;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ our $VERSION = 'v2.3.0';
|
|
|
0c4b27 |
+ our @ISA = ('Exporter');
|
|
|
0c4b27 |
+ our @EXPORT = qw(
|
|
|
0c4b27 |
+ $INITDIR
|
|
|
0c4b27 |
+ $HA_DIR
|
|
|
0c4b27 |
+ $HA_RCDIR
|
|
|
0c4b27 |
+ $HA_CONFDIR
|
|
|
0c4b27 |
+ $HA_CF
|
|
|
0c4b27 |
+ $HA_VARLIB
|
|
|
0c4b27 |
+ $HA_RSCTMP
|
|
|
0c4b27 |
+ $HA_RSCTMP_OLD
|
|
|
0c4b27 |
+ $HA_FIFO
|
|
|
0c4b27 |
+ $HA_BIN
|
|
|
0c4b27 |
+ $HA_SBIN_DIR
|
|
|
0c4b27 |
+ $HA_DATEFMT
|
|
|
0c4b27 |
+ $HA_DEBUGLOG
|
|
|
0c4b27 |
+ $HA_RESOURCEDIR
|
|
|
0c4b27 |
+ $HA_DOCDIR
|
|
|
0c4b27 |
+ $__SCRIPT_NAME
|
|
|
0c4b27 |
+ $HA_VARRUN
|
|
|
0c4b27 |
+ $HA_VARLOCK
|
|
|
0c4b27 |
+ $ocf_prefix
|
|
|
0c4b27 |
+ $ocf_exec_prefix
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ our @EXPORT_OK = ( @EXPORT );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+our $INITDIR = ( $ENV{'INITDIR'} || '/etc/init.d' );
|
|
|
0c4b27 |
+our $HA_DIR = ( $ENV{'HA_DIR'} || '/etc/ha.d' );
|
|
|
0c4b27 |
+our $HA_RCDIR = ( $ENV{'HA_RCDIR'} || '/etc/ha.d/rc.d' );
|
|
|
0c4b27 |
+our $HA_CONFDIR = ( $ENV{'HA_CONFDIR'} || '/etc/ha.d/conf' );
|
|
|
0c4b27 |
+our $HA_CF = ( $ENV{'HA_CF'} || '/etc/ha.d/ha.cf' );
|
|
|
0c4b27 |
+our $HA_VARLIB = ( $ENV{'HA_VARLIB'} || '/var/lib/heartbeat' );
|
|
|
0c4b27 |
+our $HA_RSCTMP = ( $ENV{'HA_RSCTMP'} || '/run/resource-agents' );
|
|
|
0c4b27 |
+our $HA_RSCTMP_OLD = ( $ENV{'HA_RSCTMP_OLD'} || '/var/run/heartbeat/rsctmp' );
|
|
|
0c4b27 |
+our $HA_FIFO = ( $ENV{'HA_FIFO'} || '/var/lib/heartbeat/fifo' );
|
|
|
0c4b27 |
+our $HA_BIN = ( $ENV{'HA_BIN'} || '/usr/libexec/heartbeat' );
|
|
|
0c4b27 |
+our $HA_SBIN_DIR = ( $ENV{'HA_SBIN_DIR'} || '/usr/sbin' );
|
|
|
0c4b27 |
+our $HA_DATEFMT = ( $ENV{'HA_DATEFMT'} || '%b %d %T ' );
|
|
|
0c4b27 |
+our $HA_DEBUGLOG = ( $ENV{'HA_DEBUGLOG'} || '/dev/null' );
|
|
|
0c4b27 |
+our $HA_RESOURCEDIR = ( $ENV{'HA_RESOURCEDIR'}|| '/etc/ha.d/resource.d' );
|
|
|
0c4b27 |
+our $HA_DOCDIR = ( $ENV{'HA_DOCDIR'} || '/usr/share/doc/heartbeat' );
|
|
|
0c4b27 |
+our $__SCRIPT_NAME = ( $ENV{'__SCRIPT_NAME'} || fileparse($0) );
|
|
|
0c4b27 |
+our $HA_VARRUN = ( $ENV{'HA_VARRUN'} || '/var/run' );
|
|
|
0c4b27 |
+our $HA_VARLOCK = ( $ENV{'HA_VARLOCK'} || '/var/lock/subsys' );
|
|
|
0c4b27 |
+our $ocf_prefix = '/usr';
|
|
|
0c4b27 |
+our $ocf_exec_prefix = '/usr';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 COPYRIGHT AND LICENSE
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Copyright (C) 2016: Jehan-Guillaume de Rorthais and Mael Rimbault.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Licensed under the PostgreSQL License.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
diff --color -uNr a/heartbeat/OCF_Functions.pm b/heartbeat/OCF_Functions.pm
|
|
|
0c4b27 |
--- a/heartbeat/OCF_Functions.pm 1970-01-01 01:00:00.000000000 +0100
|
|
|
0c4b27 |
+++ b/heartbeat/OCF_Functions.pm 2021-04-13 13:37:35.621267404 +0200
|
|
|
0c4b27 |
@@ -0,0 +1,631 @@
|
|
|
0c4b27 |
+#!/usr/bin/perl
|
|
|
0c4b27 |
+# This program is open source, licensed under the PostgreSQL License.
|
|
|
0c4b27 |
+# For license terms, see the LICENSE file.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# Copyright (C) 2016-2020: Jehan-Guillaume de Rorthais and Mael Rimbault
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 NAME
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+OCF_Functions - helper subroutines for OCF agent
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SYNOPSIS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ use FindBin;
|
|
|
0c4b27 |
+ use lib "$FindBin::RealBin/../../lib/heartbeat/";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ use OCF_Functions;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 DESCRIPTION
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+This module has been ported from the ocf-shellfuncs shell script of the
|
|
|
0c4b27 |
+resource-agents project. See L<https://github.com/ClusterLabs/resource-agents/>.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 VARIABLE
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The only variable exported by this module is C<__OCF_ACTION>.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SUBROUTINES
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Here are the subroutines ported from ocf-shellfuncs and exported by this module:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=over
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ha_debug
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ha_log
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item hadate
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_is_clone
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_is_ms
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_is_probe
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_is_root
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_is_true
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_is_ver
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_local_nodename
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_log
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_exit_reason
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_maybe_random
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_ver2num
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_ver_complete_level
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_ver_level
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_version_cmp
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item set_logtag
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=back
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Here are the subroutines only existing in the perl module but not in the
|
|
|
0c4b27 |
+ocf-shellfuncs script:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=over
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item ocf_notify_env
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=back
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+package OCF_Functions;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use strict;
|
|
|
0c4b27 |
+use warnings;
|
|
|
0c4b27 |
+use 5.008;
|
|
|
0c4b27 |
+use POSIX qw( strftime setlocale LC_ALL );
|
|
|
0c4b27 |
+use English;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use FindBin;
|
|
|
0c4b27 |
+use lib "$FindBin::RealBin/../../lib/heartbeat/";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use OCF_ReturnCodes;
|
|
|
0c4b27 |
+use OCF_Directories;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+BEGIN {
|
|
|
0c4b27 |
+ use Exporter;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ our $VERSION = 'v2.3.0';
|
|
|
0c4b27 |
+ our @ISA = ('Exporter');
|
|
|
0c4b27 |
+ our @EXPORT = qw(
|
|
|
0c4b27 |
+ $__OCF_ACTION
|
|
|
0c4b27 |
+ ocf_is_root
|
|
|
0c4b27 |
+ ocf_maybe_random
|
|
|
0c4b27 |
+ ocf_is_true
|
|
|
0c4b27 |
+ hadate
|
|
|
0c4b27 |
+ set_logtag
|
|
|
0c4b27 |
+ ha_log
|
|
|
0c4b27 |
+ ha_debug
|
|
|
0c4b27 |
+ ocf_log
|
|
|
0c4b27 |
+ ocf_exit_reason
|
|
|
0c4b27 |
+ ocf_is_probe
|
|
|
0c4b27 |
+ ocf_is_clone
|
|
|
0c4b27 |
+ ocf_is_ms
|
|
|
0c4b27 |
+ ocf_is_ver
|
|
|
0c4b27 |
+ ocf_ver2num
|
|
|
0c4b27 |
+ ocf_ver_level
|
|
|
0c4b27 |
+ ocf_ver_complete_level
|
|
|
0c4b27 |
+ ocf_version_cmp
|
|
|
0c4b27 |
+ ocf_local_nodename
|
|
|
0c4b27 |
+ ocf_notify_env
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ our @EXPORT_OK = ( @EXPORT );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+our $__OCF_ACTION;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_is_root {
|
|
|
0c4b27 |
+ return $EUID == 0;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_maybe_random {
|
|
|
0c4b27 |
+ return int( rand( 32767 ) );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_is_true {
|
|
|
0c4b27 |
+ my $v = shift;
|
|
|
0c4b27 |
+ return ( defined $v and $v =~ /^(?:yes|true|1|YES|TRUE|ja|on|ON)$/ );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub hadate {
|
|
|
0c4b27 |
+ return strftime( $HA_DATEFMT, localtime );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub set_logtag {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return if defined $ENV{'HA_LOGTAG'} and $ENV{'HA_LOGTAG'} ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ENV{'OCF_RESOURCE_INSTANCE'} and $ENV{'OCF_RESOURCE_INSTANCE'} ne '' ) {
|
|
|
0c4b27 |
+ $ENV{'HA_LOGTAG'} = "$__SCRIPT_NAME($ENV{'OCF_RESOURCE_INSTANCE'})[$PID]";
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ $ENV{'HA_LOGTAG'}="${__SCRIPT_NAME}[$PID]";
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub __ha_log {
|
|
|
0c4b27 |
+ my $ignore_stderr = 0;
|
|
|
0c4b27 |
+ my $loglevel = '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $_[0] eq '--ignore-stderr' ) {
|
|
|
0c4b27 |
+ $ignore_stderr = 1;
|
|
|
0c4b27 |
+ shift;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ENV{'HA_LOGFACILITY'} = '' if not defined $ENV{'HA_LOGFACILITY'}
|
|
|
0c4b27 |
+ or $ENV{'HA_LOGFACILITY'} eq 'none';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # if we're connected to a tty, then output to stderr
|
|
|
0c4b27 |
+ if ( -t STDERR ) {
|
|
|
0c4b27 |
+ # FIXME
|
|
|
0c4b27 |
+ # T.N.: this was ported with the bug on $loglevel being empty
|
|
|
0c4b27 |
+ # and never set before the test here...
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_debug'}
|
|
|
0c4b27 |
+ and $ENV{'HA_debug'} == 0
|
|
|
0c4b27 |
+ and $loglevel eq 'debug'
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ return 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $ignore_stderr ) {
|
|
|
0c4b27 |
+ # something already printed this error to stderr, so ignore
|
|
|
0c4b27 |
+ return 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_LOGTAG'} and $ENV{'HA_LOGTAG'} ne '' ) {
|
|
|
0c4b27 |
+ printf STDERR "%s: %s\n", $ENV{'HA_LOGTAG'}, join ' ', @ARG;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ printf STDERR "%s\n", join ' ', @ARG;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ return 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ set_logtag();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_LOGD'} and $ENV{'HA_LOGD'} eq 'yes' ) {
|
|
|
0c4b27 |
+ system 'ha_logger', '-t', $ENV{'HA_LOGTAG'}, @ARG;
|
|
|
0c4b27 |
+ return 0 if ( $? >> 8 ) == 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( $ENV{'HA_LOGFACILITY'} eq '' ) {
|
|
|
0c4b27 |
+ # logging through syslog
|
|
|
0c4b27 |
+ # loglevel is unknown, use 'notice' for now
|
|
|
0c4b27 |
+ $loglevel = 'notice';
|
|
|
0c4b27 |
+ for ( "@ARG" ) {
|
|
|
0c4b27 |
+ if ( /ERROR/ ) {
|
|
|
0c4b27 |
+ $loglevel = 'err';
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( /WARN/ ) {
|
|
|
0c4b27 |
+ $loglevel = 'warning';
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif (/INFO|info/ ) {
|
|
|
0c4b27 |
+ $loglevel = 'info';
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ system 'logger', '-t', $ENV{'HA_LOGTAG'}, '-p',
|
|
|
0c4b27 |
+ "$ENV{'HA_LOGFACILITY'}.$loglevel", @ARG;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_LOGFILE'} and $ENV{'HA_LOGFILE'} ne '' ) {
|
|
|
0c4b27 |
+ # appending to $HA_LOGFILE
|
|
|
0c4b27 |
+ open my $logfile, '>>', $ENV{'HA_LOGFILE'};
|
|
|
0c4b27 |
+ printf $logfile "%s: %s %s\n", $ENV{'HA_LOGTAG'}, hadate(),
|
|
|
0c4b27 |
+ join (' ', @ARG);
|
|
|
0c4b27 |
+ close $logfile;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # appending to stderr
|
|
|
0c4b27 |
+ printf STDERR "%s %s\n", hadate(), join ' ', @ARG
|
|
|
0c4b27 |
+ if (not defined $ENV{'HA_LOGFACILITY'} or $ENV{'HA_LOGFACILITY'} eq '')
|
|
|
0c4b27 |
+ and (not defined $ENV{'HA_LOGFILE'} or $ENV{'HA_LOGFILE'} eq '' )
|
|
|
0c4b27 |
+ and not $ignore_stderr;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_DEBUGLOG'} and $ENV{'HA_DEBUGLOG'} ne ''
|
|
|
0c4b27 |
+ and $ENV{'HA_LOGFILE'} ne $ENV{'HA_DEBUGLOG'}
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ # appending to $HA_DEBUGLOG
|
|
|
0c4b27 |
+ open my $logfile, '>>', $ENV{'HA_DEBUGLOG'};
|
|
|
0c4b27 |
+ printf $logfile "%s: %s %s\n", $ENV{'HA_LOGTAG'}, hadate(),
|
|
|
0c4b27 |
+ join (' ', @ARG);
|
|
|
0c4b27 |
+ close $logfile;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ha_log {
|
|
|
0c4b27 |
+ return __ha_log( @ARG );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ha_debug {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 0 if defined $ENV{'HA_debug'} and $ENV{'HA_debug'} == 0;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( -t STDERR ) {
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_LOGTAG'} and $ENV{'HA_LOGTAG'} ne '' ) {
|
|
|
0c4b27 |
+ printf STDERR "%s: %s\n", $ENV{'HA_LOGTAG'}, join ' ', @ARG;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ printf STDERR "%s\n", join ' ', @ARG;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ set_logtag();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_LOGD'} and $ENV{'HA_LOGD'} eq 'yes' ) {
|
|
|
0c4b27 |
+ system 'ha_logger', '-t', $ENV{'HA_LOGTAG'}, '-D', 'ha-debug', @ARG;
|
|
|
0c4b27 |
+ return 0 if ( $? >> 8 ) == 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ENV{'HA_LOGFACILITY'} = '' if not defined $ENV{'HA_LOGFACILITY'}
|
|
|
0c4b27 |
+ or $ENV{'HA_LOGFACILITY'} eq 'none';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( $ENV{'HA_LOGFACILITY'} eq '' ) {
|
|
|
0c4b27 |
+ # logging through syslog
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ system 'logger', '-t', $ENV{'HA_LOGTAG'}, '-p',
|
|
|
0c4b27 |
+ "$ENV{'HA_LOGFACILITY'}.debug", @ARG;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ENV{'HA_DEBUGLOG'} and -f $ENV{'HA_DEBUGLOG'} ) {
|
|
|
0c4b27 |
+ my $logfile;
|
|
|
0c4b27 |
+ # appending to $HA_DEBUGLOG
|
|
|
0c4b27 |
+ open $logfile, '>>', $ENV{'HA_DEBUGLOG'};
|
|
|
0c4b27 |
+ printf $logfile "%s: %s %s\n", $ENV{'HA_LOGTAG'}, hadate(),
|
|
|
0c4b27 |
+ join (' ', @ARG);
|
|
|
0c4b27 |
+ close $logfile;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # appending to stderr
|
|
|
0c4b27 |
+ printf STDERR "%s: %s %s\n", $ENV{'HA_LOGTAG'}, hadate(), join ' ', @ARG
|
|
|
0c4b27 |
+ if (not defined $ENV{'HA_LOGFACILITY'} or $ENV{'HA_LOGFACILITY'} eq '')
|
|
|
0c4b27 |
+ and (not defined $ENV{'HA_DEBUGLOG'} or $ENV{'HA_DEBUGLOG'} eq '' );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# ocf_log: log messages from the resource agent
|
|
|
0c4b27 |
+# This function is slightly different from its equivalent in ocf-shellfuncs.in
|
|
|
0c4b27 |
+# as it behaves like printf.
|
|
|
0c4b27 |
+# Arguments:
|
|
|
0c4b27 |
+# * __OCF_PRIO: log level
|
|
|
0c4b27 |
+# * __OCF_MSG: printf-like format string
|
|
|
0c4b27 |
+# * all other arguments are values for the printf-like format string
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub ocf_log {
|
|
|
0c4b27 |
+ my $__OCF_PRIO;
|
|
|
0c4b27 |
+ my $__OCF_MSG;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # TODO: Revisit and implement internally.
|
|
|
0c4b27 |
+ if ( scalar @ARG < 2 ) {
|
|
|
0c4b27 |
+ ocf_log ( 'err', "Not enough arguments [%d] to ocf_log", scalar @ARG );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $__OCF_PRIO = shift;
|
|
|
0c4b27 |
+ $__OCF_MSG = shift;
|
|
|
0c4b27 |
+ $__OCF_MSG = sprintf $__OCF_MSG, @ARG;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ for ( $__OCF_PRIO ) {
|
|
|
0c4b27 |
+ if ( /crit/ ) { $__OCF_PRIO = 'CRIT' }
|
|
|
0c4b27 |
+ elsif ( /err/ ) { $__OCF_PRIO = 'ERROR' }
|
|
|
0c4b27 |
+ elsif ( /warn/ ) { $__OCF_PRIO = 'WARNING' }
|
|
|
0c4b27 |
+ elsif ( /info/ ) { $__OCF_PRIO = 'INFO' }
|
|
|
0c4b27 |
+ elsif ( /debug/ ) { $__OCF_PRIO = 'DEBUG' }
|
|
|
0c4b27 |
+ else { $__OCF_PRIO =~ tr/[a-z]/[A-Z]/ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $__OCF_PRIO eq 'DEBUG' ) {
|
|
|
0c4b27 |
+ ha_debug( "$__OCF_PRIO: $__OCF_MSG");
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ ha_log( "$__OCF_PRIO: $__OCF_MSG");
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# ocf_exit_reason: print exit error string to stderr and log
|
|
|
0c4b27 |
+# Usage: Allows the OCF script to provide a string
|
|
|
0c4b27 |
+# describing why the exit code was returned.
|
|
|
0c4b27 |
+# Arguments: reason - required, The string that represents
|
|
|
0c4b27 |
+# why the error occured.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub ocf_exit_reason {
|
|
|
0c4b27 |
+ my $cookie = $ENV{'OCF_EXIT_REASON_PREFIX'} || 'ocf-exit-reason:';
|
|
|
0c4b27 |
+ my $fmt;
|
|
|
0c4b27 |
+ my $msg;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # No argument is likely not intentional.
|
|
|
0c4b27 |
+ # Just one argument implies a printf format string of just "%s".
|
|
|
0c4b27 |
+ # "Least surprise" in case some interpolated string from variable
|
|
|
0c4b27 |
+ # expansion or other contains a percent sign.
|
|
|
0c4b27 |
+ # More than one argument: first argument is going to be the format string.
|
|
|
0c4b27 |
+ ocf_log ( 'err', 'Not enough arguments [%d] to ocf_exit_reason',
|
|
|
0c4b27 |
+ scalar @ARG ) if scalar @ARG < 1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $fmt = shift;
|
|
|
0c4b27 |
+ $msg = sprintf $fmt, @ARG;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ print STDERR "$cookie$msg\n";
|
|
|
0c4b27 |
+ __ha_log( '--ignore-stderr', "ERROR: $msg" );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# returns true if the CRM is currently running a probe. A probe is
|
|
|
0c4b27 |
+# defined as a monitor operation with a monitoring interval of zero.
|
|
|
0c4b27 |
+sub ocf_is_probe {
|
|
|
0c4b27 |
+ return ( $__OCF_ACTION eq 'monitor'
|
|
|
0c4b27 |
+ and $ENV{'OCF_RESKEY_CRM_meta_interval'} == 0 );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# returns true if the resource is configured as a clone. This is
|
|
|
0c4b27 |
+# defined as a resource where the clone-max meta attribute is present,
|
|
|
0c4b27 |
+# and set to greater than zero.
|
|
|
0c4b27 |
+sub ocf_is_clone {
|
|
|
0c4b27 |
+ return ( defined $ENV{'OCF_RESKEY_CRM_meta_clone_max'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_RESKEY_CRM_meta_clone_max'} > 0 );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# returns true if the resource is configured as a multistate
|
|
|
0c4b27 |
+# (master/slave) resource. This is defined as a resource where the
|
|
|
0c4b27 |
+# master-max meta attribute is present, and set to greater than zero.
|
|
|
0c4b27 |
+sub ocf_is_ms {
|
|
|
0c4b27 |
+ return ( defined $ENV{'OCF_RESKEY_CRM_meta_master_max'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_RESKEY_CRM_meta_master_max'} > 0 );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# version check functions
|
|
|
0c4b27 |
+# allow . and - to delimit version numbers
|
|
|
0c4b27 |
+# max version number is 999
|
|
|
0c4b27 |
+# letters and such are effectively ignored
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub ocf_is_ver {
|
|
|
0c4b27 |
+ return $ARG[0] =~ /^[0-9][0-9.-]*[0-9]$/;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_ver2num {
|
|
|
0c4b27 |
+ my $v = 0;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $v = $v * 1000 + $1 while $ARG[0] =~ /(\d+)/g;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $v;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_ver_level {
|
|
|
0c4b27 |
+ my $v = () = $ARG[0] =~ /(\d+)/g;
|
|
|
0c4b27 |
+ return $v;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_ver_complete_level {
|
|
|
0c4b27 |
+ my $ver = shift;
|
|
|
0c4b27 |
+ my $level = shift;
|
|
|
0c4b27 |
+ my $i = 0;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ for ( my $i = 0; $i < $level; $i++ ) {
|
|
|
0c4b27 |
+ $ver .= "$ver.0";
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $ver;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# usage: ocf_version_cmp VER1 VER2
|
|
|
0c4b27 |
+# version strings can contain digits, dots, and dashes
|
|
|
0c4b27 |
+# must start and end with a digit
|
|
|
0c4b27 |
+# returns:
|
|
|
0c4b27 |
+# 0: VER1 smaller (older) than VER2
|
|
|
0c4b27 |
+# 1: versions equal
|
|
|
0c4b27 |
+# 2: VER1 greater (newer) than VER2
|
|
|
0c4b27 |
+# 3: bad format
|
|
|
0c4b27 |
+sub ocf_version_cmp {
|
|
|
0c4b27 |
+ my $v1 = shift;
|
|
|
0c4b27 |
+ my $v2 = shift;
|
|
|
0c4b27 |
+ my $v1_level;
|
|
|
0c4b27 |
+ my $v2_level;
|
|
|
0c4b27 |
+ my $level_diff;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 3 unless ocf_is_ver( $v1 );
|
|
|
0c4b27 |
+ return 3 unless ocf_is_ver( $v2 );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $v1_level = ocf_ver_level( $v1 );
|
|
|
0c4b27 |
+ $v2_level = ocf_ver_level( $v2 );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $v1_level < $v2_level ) {
|
|
|
0c4b27 |
+ $level_diff = $v2_level - $v1_level;
|
|
|
0c4b27 |
+ $v1 = ocf_ver_complete_level( $v1, $level_diff );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $v1_level > $v2_level ) {
|
|
|
0c4b27 |
+ $level_diff = $v1_level - $v2_level;
|
|
|
0c4b27 |
+ $v2 = ocf_ver_complete_level( $v2, $level_diff );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $v1 = ocf_ver2num( $v1 );
|
|
|
0c4b27 |
+ $v2 = ocf_ver2num( $v2 );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $v1 == $v2 ) { return 1; }
|
|
|
0c4b27 |
+ elsif ( $v1 < $v2 ) { return 0; }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 2; # -1 would look funny in shell ;-) ( T.N. not in perl ;) )
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_local_nodename {
|
|
|
0c4b27 |
+ # use crm_node -n for pacemaker > 1.1.8
|
|
|
0c4b27 |
+ my $nodename;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ qx{ which pacemakerd > /dev/null 2>&1 };
|
|
|
0c4b27 |
+ if ( $? == 0 ) {
|
|
|
0c4b27 |
+ my $version;
|
|
|
0c4b27 |
+ my $ret = qx{ pacemakerd -\$ };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ret =~ /Pacemaker ([\d.]+)/;
|
|
|
0c4b27 |
+ $version = $1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( ocf_version_cmp( $version, '1.1.8' ) == 2 ) {
|
|
|
0c4b27 |
+ qx{ which crm_node > /dev/null 2>&1 };
|
|
|
0c4b27 |
+ $nodename = qx{ crm_node -n } if $? == 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ # otherwise use uname -n
|
|
|
0c4b27 |
+ $nodename = qx { uname -n };
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ chomp $nodename;
|
|
|
0c4b27 |
+ return $nodename;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Parse and returns the notify environment variables in a convenient structure
|
|
|
0c4b27 |
+# Returns undef if the action is not a notify
|
|
|
0c4b27 |
+# Returns undef if the resource is neither a clone or a multistate one
|
|
|
0c4b27 |
+sub ocf_notify_env {
|
|
|
0c4b27 |
+ my $i;
|
|
|
0c4b27 |
+ my %notify_env;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return undef unless $__OCF_ACTION eq 'notify';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return undef unless ocf_is_clone() or ocf_is_ms();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ %notify_env = (
|
|
|
0c4b27 |
+ 'type' => $ENV{'OCF_RESKEY_CRM_meta_notify_type'} || '',
|
|
|
0c4b27 |
+ 'operation' => $ENV{'OCF_RESKEY_CRM_meta_notify_operation'} || '',
|
|
|
0c4b27 |
+ 'active' => [ ],
|
|
|
0c4b27 |
+ 'inactive' => [ ],
|
|
|
0c4b27 |
+ 'start' => [ ],
|
|
|
0c4b27 |
+ 'stop' => [ ],
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ for my $action ( qw{ active start stop } ) {
|
|
|
0c4b27 |
+ next unless
|
|
|
0c4b27 |
+ defined $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_resource"}
|
|
|
0c4b27 |
+ and defined $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_uname"};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $i = 0;
|
|
|
0c4b27 |
+ $notify_env{ $action }[$i++]{'rsc'} = $_ foreach split /\s+/ =>
|
|
|
0c4b27 |
+ $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_resource"};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $i = 0;
|
|
|
0c4b27 |
+ $notify_env{ $action }[$i++]{'uname'} = $_ foreach split /\s+/ =>
|
|
|
0c4b27 |
+ $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_uname"};
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # notify_nactive_uname doesn't exists. See:
|
|
|
0c4b27 |
+ # http://lists.clusterlabs.org/pipermail/developers/2017-January/000406.html
|
|
|
0c4b27 |
+ if ( defined $ENV{"OCF_RESKEY_CRM_meta_notify_inactive_resource"} ) {
|
|
|
0c4b27 |
+ $i = 0;
|
|
|
0c4b27 |
+ $notify_env{'inactive'}[$i++]{'rsc'} = $_ foreach split /\s+/ =>
|
|
|
0c4b27 |
+ $ENV{"OCF_RESKEY_CRM_meta_notify_inactive_resource"};
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # exit if the resource is not a mutistate one
|
|
|
0c4b27 |
+ return %notify_env unless ocf_is_ms();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ for my $action ( qw{ master slave promote demote } ) {
|
|
|
0c4b27 |
+ $notify_env{ $action } = [ ];
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ next unless
|
|
|
0c4b27 |
+ defined $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_resource"}
|
|
|
0c4b27 |
+ and defined $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_uname"};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $i = 0;
|
|
|
0c4b27 |
+ $notify_env{ $action }[$i++]{'rsc'} = $_ foreach split /\s+/ =>
|
|
|
0c4b27 |
+ $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_resource"};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $i = 0;
|
|
|
0c4b27 |
+ $notify_env{ $action }[$i++]{'uname'} = $_ foreach split /\s+/ =>
|
|
|
0c4b27 |
+ $ENV{"OCF_RESKEY_CRM_meta_notify_${action}_uname"};
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Fix active and inactive fields for Pacemaker version < 1.1.16
|
|
|
0c4b27 |
+ # ie. crm_feature_set < 3.0.11
|
|
|
0c4b27 |
+ # See http://lists.clusterlabs.org/pipermail/developers/2016-August/000265.html
|
|
|
0c4b27 |
+ # and git commit a6713c5d40327eff8549e7f596501ab1785b8765
|
|
|
0c4b27 |
+ if (
|
|
|
0c4b27 |
+ ocf_version_cmp( $ENV{"OCF_RESKEY_crm_feature_set"}, '3.0.11' ) == 0
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ $notify_env{ 'active' } = [
|
|
|
0c4b27 |
+ @{ $notify_env{ 'master' } },
|
|
|
0c4b27 |
+ @{ $notify_env{ 'slave' } }
|
|
|
0c4b27 |
+ ];
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return %notify_env;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+$__OCF_ACTION = $ARGV[0];
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Return to sanity for the agents...
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+undef $ENV{'LC_ALL'};
|
|
|
0c4b27 |
+$ENV{'LC_ALL'} = 'C';
|
|
|
0c4b27 |
+setlocale( LC_ALL, 'C' );
|
|
|
0c4b27 |
+undef $ENV{'LANG'};
|
|
|
0c4b27 |
+undef $ENV{'LANGUAGE'};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+$ENV{'OCF_ROOT'} = '/usr/lib/ocf'
|
|
|
0c4b27 |
+ unless defined $ENV{'OCF_ROOT'} and $ENV{'OCF_ROOT'} ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# old
|
|
|
0c4b27 |
+undef $ENV{'OCF_FUNCTIONS_DIR'}
|
|
|
0c4b27 |
+ if defined $ENV{'OCF_FUNCTIONS_DIR'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_FUNCTIONS_DIR'} eq "$ENV{'OCF_ROOT'}/resource.d/heartbeat";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Define OCF_RESKEY_CRM_meta_interval in case it isn't already set,
|
|
|
0c4b27 |
+# to make sure that ocf_is_probe() always works
|
|
|
0c4b27 |
+$ENV{'OCF_RESKEY_CRM_meta_interval'} = 0
|
|
|
0c4b27 |
+ unless defined $ENV{'OCF_RESKEY_CRM_meta_interval'};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Strip the OCF_RESKEY_ prefix from this particular parameter
|
|
|
0c4b27 |
+unless ( defined $ENV{'$OCF_RESKEY_OCF_CHECK_LEVEL'}
|
|
|
0c4b27 |
+ and $ENV{'$OCF_RESKEY_OCF_CHECK_LEVEL'} ne ''
|
|
|
0c4b27 |
+) {
|
|
|
0c4b27 |
+ $ENV{'OCF_CHECK_LEVEL'} = $ENV{'$OCF_RESKEY_OCF_CHECK_LEVEL'};
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+else {
|
|
|
0c4b27 |
+ ENV{'OCF_CHECK_LEVEL'} = 0;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+unless ( -d $ENV{'OCF_ROOT'} ) {
|
|
|
0c4b27 |
+ ha_log( "ERROR: OCF_ROOT points to non-directory $ENV{'OCF_ROOT'}." );
|
|
|
0c4b27 |
+ $! = $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ die;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+$ENV{'OCF_RESOURCE_TYPE'} = $__SCRIPT_NAME
|
|
|
0c4b27 |
+ unless defined $ENV{'OCF_RESOURCE_TYPE'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_RESOURCE_TYPE'} ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+unless ( defined $ENV{'OCF_RA_VERSION_MAJOR'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_RA_VERSION_MAJOR'} ne ''
|
|
|
0c4b27 |
+) {
|
|
|
0c4b27 |
+ # We are being invoked as an init script.
|
|
|
0c4b27 |
+ # Fill in some things with reasonable values.
|
|
|
0c4b27 |
+ $ENV{'OCF_RESOURCE_INSTANCE'} = 'default';
|
|
|
0c4b27 |
+ return 1;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+$ENV{'OCF_RESOURCE_INSTANCE'} = "undef" if $__OCF_ACTION eq 'meta-data';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+unless ( defined $ENV{'OCF_RESOURCE_INSTANCE'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_RESOURCE_INSTANCE'} ne ''
|
|
|
0c4b27 |
+) {
|
|
|
0c4b27 |
+ ha_log( "ERROR: Need to tell us our resource instance name." );
|
|
|
0c4b27 |
+ $! = $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ die;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 COPYRIGHT AND LICENSE
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Copyright (C) 2016: Jehan-Guillaume de Rorthais and Mael Rimbault.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Licensed under the PostgreSQL License.
|
|
|
0c4b27 |
diff --color -uNr a/heartbeat/OCF_ReturnCodes.pm b/heartbeat/OCF_ReturnCodes.pm
|
|
|
0c4b27 |
--- a/heartbeat/OCF_ReturnCodes.pm 1970-01-01 01:00:00.000000000 +0100
|
|
|
0c4b27 |
+++ b/heartbeat/OCF_ReturnCodes.pm 2021-04-13 13:37:35.621267404 +0200
|
|
|
0c4b27 |
@@ -0,0 +1,97 @@
|
|
|
0c4b27 |
+#!/usr/bin/perl
|
|
|
0c4b27 |
+# This program is open source, licensed under the PostgreSQL License.
|
|
|
0c4b27 |
+# For license terms, see the LICENSE file.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# Copyright (C) 2016-2020: Jehan-Guillaume de Rorthais and Mael Rimbault
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 NAME
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+OCF_ReturnCodes - Common varibales for the OCF Resource Agents supplied by
|
|
|
0c4b27 |
+heartbeat.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SYNOPSIS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ use FindBin;
|
|
|
0c4b27 |
+ use lib "$FindBin::RealBin/../../lib/heartbeat/";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ use OCF_ReturnCodes;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 DESCRIPTION
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+This module has been ported from the ocf-retrurncodes shell script of the
|
|
|
0c4b27 |
+resource-agents project. See L<https://github.com/ClusterLabs/resource-agents/>.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 VARIABLES
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Here are the variables exported by this module:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=over
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_SUCCESS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_ERR_GENERIC
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_ERR_ARGS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_ERR_UNIMPLEMENTED
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_ERR_PERM
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_ERR_INSTALLED
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_ERR_CONFIGURED
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_NOT_RUNNING
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_RUNNING_MASTER
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item $OCF_FAILED_MASTER
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=back
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+package OCF_ReturnCodes;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use strict;
|
|
|
0c4b27 |
+use warnings;
|
|
|
0c4b27 |
+use 5.008;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+BEGIN {
|
|
|
0c4b27 |
+ use Exporter;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ our $VERSION = 'v2.3.0';
|
|
|
0c4b27 |
+ our @ISA = ('Exporter');
|
|
|
0c4b27 |
+ our @EXPORT = qw(
|
|
|
0c4b27 |
+ $OCF_SUCCESS
|
|
|
0c4b27 |
+ $OCF_ERR_GENERIC
|
|
|
0c4b27 |
+ $OCF_ERR_ARGS
|
|
|
0c4b27 |
+ $OCF_ERR_UNIMPLEMENTED
|
|
|
0c4b27 |
+ $OCF_ERR_PERM
|
|
|
0c4b27 |
+ $OCF_ERR_INSTALLED
|
|
|
0c4b27 |
+ $OCF_ERR_CONFIGURED
|
|
|
0c4b27 |
+ $OCF_NOT_RUNNING
|
|
|
0c4b27 |
+ $OCF_RUNNING_MASTER
|
|
|
0c4b27 |
+ $OCF_FAILED_MASTER
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ our @EXPORT_OK = ( @EXPORT );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+our $OCF_SUCCESS = 0;
|
|
|
0c4b27 |
+our $OCF_ERR_GENERIC = 1;
|
|
|
0c4b27 |
+our $OCF_ERR_ARGS = 2;
|
|
|
0c4b27 |
+our $OCF_ERR_UNIMPLEMENTED = 3;
|
|
|
0c4b27 |
+our $OCF_ERR_PERM = 4;
|
|
|
0c4b27 |
+our $OCF_ERR_INSTALLED = 5;
|
|
|
0c4b27 |
+our $OCF_ERR_CONFIGURED = 6;
|
|
|
0c4b27 |
+our $OCF_NOT_RUNNING = 7;
|
|
|
0c4b27 |
+our $OCF_RUNNING_MASTER = 8;
|
|
|
0c4b27 |
+our $OCF_FAILED_MASTER = 9;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 COPYRIGHT AND LICENSE
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Copyright (C) 2016: Jehan-Guillaume de Rorthais and Mael Rimbault.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Licensed under the PostgreSQL License.
|
|
|
0c4b27 |
diff --color -uNr a/heartbeat/pgsqlms b/heartbeat/pgsqlms
|
|
|
0c4b27 |
--- a/heartbeat/pgsqlms 1970-01-01 01:00:00.000000000 +0100
|
|
|
0c4b27 |
+++ b/heartbeat/pgsqlms 2021-04-13 13:37:40.934280411 +0200
|
|
|
0c4b27 |
@@ -0,0 +1,2308 @@
|
|
|
0c4b27 |
+#!/usr/bin/perl
|
|
|
0c4b27 |
+# This program is open source, licensed under the PostgreSQL License.
|
|
|
0c4b27 |
+# For license terms, see the LICENSE file.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# Copyright (C) 2016-2020: Jehan-Guillaume de Rorthais and Mael Rimbault
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 NAME
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ocf_heartbeat_pgsqlms - A PostgreSQL multi-state resource agent for Pacemaker
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SYNOPSIS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+B<pgsqlms> [start | stop | monitor | promote | demote | notify | reload | methods | meta-data | validate-all]
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 DESCRIPTION
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Resource script for PostgreSQL in replication. It manages PostgreSQL servers using streaming replication as an HA resource.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use strict;
|
|
|
0c4b27 |
+use warnings;
|
|
|
0c4b27 |
+use 5.008;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use POSIX qw(locale_h);
|
|
|
0c4b27 |
+use Scalar::Util qw(looks_like_number);
|
|
|
0c4b27 |
+use File::Spec;
|
|
|
0c4b27 |
+use File::Temp;
|
|
|
0c4b27 |
+use Data::Dumper;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+my $OCF_FUNCTIONS_DIR;
|
|
|
0c4b27 |
+BEGIN {
|
|
|
0c4b27 |
+ $OCF_FUNCTIONS_DIR = defined $ENV{'OCF_FUNCTIONS_DIR'} ? "$ENV{'OCF_FUNCTIONS_DIR'}" : "$ENV{'OCF_ROOT'}/lib/heartbeat";
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+use lib "$OCF_FUNCTIONS_DIR";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+use OCF_ReturnCodes;
|
|
|
0c4b27 |
+use OCF_Directories;
|
|
|
0c4b27 |
+use OCF_Functions;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+our $VERSION = 'v2.3.0';
|
|
|
0c4b27 |
+our $PROGRAM = 'pgsqlms';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# OCF environment
|
|
|
0c4b27 |
+my $OCF_RESOURCE_INSTANCE = $ENV{'OCF_RESOURCE_INSTANCE'};
|
|
|
0c4b27 |
+my $OCF_RUNNING_SLAVE = $OCF_SUCCESS;
|
|
|
0c4b27 |
+my %OCF_NOTIFY_ENV = ocf_notify_env() if $__OCF_ACTION eq 'notify';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Default parameters values
|
|
|
0c4b27 |
+my $system_user_default = "postgres";
|
|
|
0c4b27 |
+my $bindir_default = "/usr/bin";
|
|
|
0c4b27 |
+my $pgdata_default = "/var/lib/pgsql/data";
|
|
|
0c4b27 |
+my $pghost_default = "/tmp";
|
|
|
0c4b27 |
+my $pgport_default = 5432;
|
|
|
0c4b27 |
+my $start_opts_default = "";
|
|
|
0c4b27 |
+my $maxlag_default = "0";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Set default values if not found in environment
|
|
|
0c4b27 |
+my $system_user = $ENV{'OCF_RESKEY_system_user'} || $system_user_default;
|
|
|
0c4b27 |
+my $bindir = $ENV{'OCF_RESKEY_bindir'} || $bindir_default;
|
|
|
0c4b27 |
+my $pgdata = $ENV{'OCF_RESKEY_pgdata'} || $pgdata_default;
|
|
|
0c4b27 |
+my $datadir = $ENV{'OCF_RESKEY_datadir'} || $pgdata;
|
|
|
0c4b27 |
+my $pghost = $ENV{'OCF_RESKEY_pghost'} || $pghost_default;
|
|
|
0c4b27 |
+my $pgport = $ENV{'OCF_RESKEY_pgport'} || $pgport_default;
|
|
|
0c4b27 |
+my $start_opts = $ENV{'OCF_RESKEY_start_opts'} || $start_opts_default;
|
|
|
0c4b27 |
+my $maxlag = $ENV{'OCF_RESKEY_maxlag'} || $maxlag_default;
|
|
|
0c4b27 |
+my $recovery_tpl = $ENV{'OCF_RESKEY_recovery_template'}
|
|
|
0c4b27 |
+ || "$pgdata/recovery.conf.pcmk";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# PostgreSQL commands path
|
|
|
0c4b27 |
+my $POSTGRES = "$bindir/postgres";
|
|
|
0c4b27 |
+my $PGCTL = "$bindir/pg_ctl";
|
|
|
0c4b27 |
+my $PGPSQL = "$bindir/psql";
|
|
|
0c4b27 |
+my $PGCTRLDATA = "$bindir/pg_controldata";
|
|
|
0c4b27 |
+my $PGISREADY = "$bindir/pg_isready";
|
|
|
0c4b27 |
+my $PGWALDUMP = "$bindir/pg_waldump";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# pacemaker commands path
|
|
|
0c4b27 |
+my $CRM_MASTER = "$HA_SBIN_DIR/crm_master --lifetime forever";
|
|
|
0c4b27 |
+my $CRM_NODE = "$HA_SBIN_DIR/crm_node";
|
|
|
0c4b27 |
+my $CRM_RESOURCE = "$HA_SBIN_DIR/crm_resource";
|
|
|
0c4b27 |
+my $ATTRD_PRIV = "$HA_SBIN_DIR/attrd_updater --private --lifetime reboot";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Global vars
|
|
|
0c4b27 |
+my $nodename;
|
|
|
0c4b27 |
+my $exit_code = 0;
|
|
|
0c4b27 |
+# numeric pgsql versions
|
|
|
0c4b27 |
+my $PGVERNUM;
|
|
|
0c4b27 |
+my $PGVER_93 = 90300;
|
|
|
0c4b27 |
+my $PGVER_10 = 100000;
|
|
|
0c4b27 |
+my $PGVER_12 = 120000;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Run a query using psql.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# This function returns an array with psql return code as first element and
|
|
|
0c4b27 |
+# the result as second one.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _query {
|
|
|
0c4b27 |
+ my $query = shift;
|
|
|
0c4b27 |
+ my $res = shift;
|
|
|
0c4b27 |
+ my $connstr = "dbname=postgres";
|
|
|
0c4b27 |
+ my $RS = chr(30); # ASCII RS (record separator)
|
|
|
0c4b27 |
+ my $FS = chr(3); # ASCII ETX (end of text)
|
|
|
0c4b27 |
+ my $postgres_uid = getpwnam( $system_user );
|
|
|
0c4b27 |
+ my $oldeuid = $>;
|
|
|
0c4b27 |
+ my $tmpfile;
|
|
|
0c4b27 |
+ my @res;
|
|
|
0c4b27 |
+ my $ans;
|
|
|
0c4b27 |
+ my $pid;
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( defined $res and defined $query and $query ne '' ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_query: wrong parameters!' );
|
|
|
0c4b27 |
+ return -1;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( $tmpfile = File::Temp->new(
|
|
|
0c4b27 |
+ TEMPLATE => 'pgsqlms-XXXXXXXX',
|
|
|
0c4b27 |
+ DIR => $HA_RSCTMP
|
|
|
0c4b27 |
+ ) )
|
|
|
0c4b27 |
+ {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not create or write in a temp file' );
|
|
|
0c4b27 |
+ exit $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ print $tmpfile $query;
|
|
|
0c4b27 |
+ chmod 0644, $tmpfile;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_query: %s', $query );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Change the effective user to the given system_user so after forking
|
|
|
0c4b27 |
+ # the given uid to the process should allow psql to connect w/o password
|
|
|
0c4b27 |
+ $> = $postgres_uid;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Forking + piping
|
|
|
0c4b27 |
+ $pid = open(my $KID, "-|");
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $pid == 0 ) { # child
|
|
|
0c4b27 |
+ exec $PGPSQL, '--set', 'ON_ERROR_STOP=1', '-qXAtf', $tmpfile,
|
|
|
0c4b27 |
+ '-R', $RS, '-F', $FS, '--port', $pgport, '--host', $pghost,
|
|
|
0c4b27 |
+ $connstr;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # parent
|
|
|
0c4b27 |
+ $> = $oldeuid;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ {
|
|
|
0c4b27 |
+ local $/;
|
|
|
0c4b27 |
+ $ans = <$KID>;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ close $KID;
|
|
|
0c4b27 |
+ $rc = $? >> 8;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_query: psql return code: %d', $rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( defined $ans ) {
|
|
|
0c4b27 |
+ chop $ans;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ push @{ $res }, [ split(chr(3) => $_, -1) ]
|
|
|
0c4b27 |
+ foreach split (chr(30) => $ans, -1);
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_query: @res: %s',
|
|
|
0c4b27 |
+ Data::Dumper->new( [ $res ] )->Terse(1)->Dump );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Possible return codes:
|
|
|
0c4b27 |
+ # -1: wrong parameters
|
|
|
0c4b27 |
+ # 0: OK
|
|
|
0c4b27 |
+ # 1: failed to get resources (memory, missing file, ...)
|
|
|
0c4b27 |
+ # 2: unable to connect
|
|
|
0c4b27 |
+ # 3: query failed
|
|
|
0c4b27 |
+ return $rc;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Get the last received location on a standby
|
|
|
0c4b27 |
+# if the first argument is true, returns the value as decimal
|
|
|
0c4b27 |
+# if the first argument is false, returns the value as LSN
|
|
|
0c4b27 |
+# Returns undef if query failed
|
|
|
0c4b27 |
+sub _get_last_received_lsn {
|
|
|
0c4b27 |
+ my ( $dec ) = @_;
|
|
|
0c4b27 |
+ my $pg_last_wal_receive_lsn = 'pg_last_wal_receive_lsn()';
|
|
|
0c4b27 |
+ my $pg_wal_lsn_diff = 'pg_wal_lsn_diff';
|
|
|
0c4b27 |
+ my $query;
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my @rs;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $PGVERNUM < $PGVER_10 ) {
|
|
|
0c4b27 |
+ $pg_last_wal_receive_lsn = 'pg_last_xlog_receive_location()';
|
|
|
0c4b27 |
+ $pg_wal_lsn_diff = 'pg_xlog_location_diff';
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $dec ) {
|
|
|
0c4b27 |
+ $query = "SELECT $pg_wal_lsn_diff( $pg_last_wal_receive_lsn, '0/0' )";
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ $query = "SELECT $pg_last_wal_receive_lsn";
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = _query( $query, \@rs );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $rs[0][0] if $rc == 0 and $rs[0][0];
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'err', 'Could not query last received LSN (%s)', $rc ) if $rc != 0;
|
|
|
0c4b27 |
+ ocf_log( 'err', 'No values for last received LSN' )
|
|
|
0c4b27 |
+ if $rc == 0 and not $rs[0][0];
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return undef;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Get the master score for each connected standby
|
|
|
0c4b27 |
+# Returns directly the result set of the query or exit with an error.
|
|
|
0c4b27 |
+# Exits with OCF_ERR_GENERIC if the query failed
|
|
|
0c4b27 |
+sub _get_lag_scores {
|
|
|
0c4b27 |
+ my $pg_current_wal_lsn = 'pg_current_wal_lsn()';
|
|
|
0c4b27 |
+ my $pg_wal_lsn_diff = 'pg_wal_lsn_diff';
|
|
|
0c4b27 |
+ my $write_lsn = 'write_lsn';
|
|
|
0c4b27 |
+ my $query;
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my @rs;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $PGVERNUM < $PGVER_10 ) {
|
|
|
0c4b27 |
+ $pg_current_wal_lsn = 'pg_current_xlog_location()';
|
|
|
0c4b27 |
+ $pg_wal_lsn_diff = 'pg_xlog_location_diff';
|
|
|
0c4b27 |
+ $write_lsn = 'write_location';
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # We check locations of connected standbies by querying the
|
|
|
0c4b27 |
+ # "pg_stat_replication" view.
|
|
|
0c4b27 |
+ # The row_number applies on the result set ordered on write_location ASC so
|
|
|
0c4b27 |
+ # the highest row_number should be given to the closest node from the
|
|
|
0c4b27 |
+ # master, then the lowest node name (alphanumeric sort) in case of equality.
|
|
|
0c4b27 |
+ # The result set itself is order by priority DESC to process best known
|
|
|
0c4b27 |
+ # candidate first.
|
|
|
0c4b27 |
+ $query = qq{
|
|
|
0c4b27 |
+ SELECT application_name, priority, location, state, current_lag
|
|
|
0c4b27 |
+ FROM (
|
|
|
0c4b27 |
+ SELECT application_name,
|
|
|
0c4b27 |
+ (1000 - (
|
|
|
0c4b27 |
+ row_number() OVER (
|
|
|
0c4b27 |
+ PARTITION BY state IN ('startup', 'backup')
|
|
|
0c4b27 |
+ ORDER BY location ASC, application_name ASC
|
|
|
0c4b27 |
+ ) - 1
|
|
|
0c4b27 |
+ ) * 10
|
|
|
0c4b27 |
+ ) * CASE WHEN ( $maxlag > 0
|
|
|
0c4b27 |
+ AND current_lag > $maxlag)
|
|
|
0c4b27 |
+ THEN -1
|
|
|
0c4b27 |
+ ELSE 1
|
|
|
0c4b27 |
+ END AS priority,
|
|
|
0c4b27 |
+ location, state, current_lag
|
|
|
0c4b27 |
+ FROM (
|
|
|
0c4b27 |
+ SELECT application_name, $write_lsn AS location, state,
|
|
|
0c4b27 |
+ $pg_wal_lsn_diff($pg_current_wal_lsn, $write_lsn) AS current_lag
|
|
|
0c4b27 |
+ FROM pg_stat_replication
|
|
|
0c4b27 |
+ ) AS s2
|
|
|
0c4b27 |
+ ) AS s1
|
|
|
0c4b27 |
+ ORDER BY priority DESC
|
|
|
0c4b27 |
+ };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = _query( $query, \@rs );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc != 0 ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Query to get standby locations failed (%d)', $rc );
|
|
|
0c4b27 |
+ exit $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return \@rs;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# get the timeout for the current action given from environment var
|
|
|
0c4b27 |
+# Returns timeout as integer
|
|
|
0c4b27 |
+# undef if unknown
|
|
|
0c4b27 |
+sub _get_action_timeout {
|
|
|
0c4b27 |
+ my $timeout = $ENV{'OCF_RESKEY_CRM_meta_timeout'} / 1000;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_get_action_timeout: known timeout: %s',
|
|
|
0c4b27 |
+ defined $timeout ? $timeout : 'undef' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $timeout if defined $timeout and $timeout =~ /^\d+$/;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return undef;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Get, parse and return the value of the given private attribute name
|
|
|
0c4b27 |
+# Returns an empty string if not found.
|
|
|
0c4b27 |
+sub _get_priv_attr {
|
|
|
0c4b27 |
+ my ( $name, $node ) = @_;
|
|
|
0c4b27 |
+ my $val = '';
|
|
|
0c4b27 |
+ my $node_arg = '';
|
|
|
0c4b27 |
+ my $ans;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $node = '' unless defined $node;
|
|
|
0c4b27 |
+ $name = "$name-$OCF_RESOURCE_INSTANCE";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $node_arg= "--node $node" if $node ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ans = qx{ $ATTRD_PRIV --name "$name" --query $node_arg };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ans =~ m/^name=".*" host=".*" value="(.*)"$/;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $val = $1 if defined $1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_get_priv_attr: value of "%s"%s is "%s"', $name,
|
|
|
0c4b27 |
+ ( $node ? " on \"$node\"": ""),
|
|
|
0c4b27 |
+ $val );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $val;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Set the given private attribute name to the given value
|
|
|
0c4b27 |
+# As setting an attribute is asynchronous, this will return as soon as the
|
|
|
0c4b27 |
+# attribute is really set by attrd and available.
|
|
|
0c4b27 |
+sub _set_priv_attr {
|
|
|
0c4b27 |
+ my ( $name, $val ) = @_;
|
|
|
0c4b27 |
+ my $name_instance = "$name-$OCF_RESOURCE_INSTANCE";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_set_priv_attr: set "%s=%s"...', $name_instance, $val );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ qx{ $ATTRD_PRIV --name "$name_instance" --update "$val" };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # give attr name without the resource instance name as _get_priv_attr adds
|
|
|
0c4b27 |
+ # it as well
|
|
|
0c4b27 |
+ while ( _get_priv_attr( $name ) ne $val ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_set_priv_attr: waiting attrd ack for "%s"...', $name_instance );
|
|
|
0c4b27 |
+ select( undef, undef, undef, 0.1 );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Delete the given private attribute.
|
|
|
0c4b27 |
+# As setting an attribute is asynchronous, this will return as soon as the
|
|
|
0c4b27 |
+# attribute is really deleted by attrd.
|
|
|
0c4b27 |
+sub _delete_priv_attr {
|
|
|
0c4b27 |
+ my ( $name ) = @_;
|
|
|
0c4b27 |
+ my $name_instance = "$name-$OCF_RESOURCE_INSTANCE";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_delete_priv_attr: delete "%s"...', $name_instance );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ qx{ $ATTRD_PRIV --name "$name_instance" --delete };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # give attr name without the resource instance name as _get_priv_attr adds
|
|
|
0c4b27 |
+ # it as well
|
|
|
0c4b27 |
+ while ( _get_priv_attr( $name ) ne '' ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_delete_priv_attr: waiting attrd ack for "%s"...',
|
|
|
0c4b27 |
+ $name_instance );
|
|
|
0c4b27 |
+ select( undef, undef, undef, 0.1 );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Get, parse and return the resource master score on given node.
|
|
|
0c4b27 |
+# Returns an empty string if not found.
|
|
|
0c4b27 |
+# Returns undef on crm_master call on error
|
|
|
0c4b27 |
+sub _get_master_score {
|
|
|
0c4b27 |
+ my ( $node ) = @_;
|
|
|
0c4b27 |
+ my $node_arg = '';
|
|
|
0c4b27 |
+ my $score;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $node_arg = sprintf '--node "%s"', $node if defined $node and $node ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $score = qx{ $CRM_MASTER --quiet --get-value $node_arg 2> /dev/null };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return '' unless $? == 0 and defined $score;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ chomp $score;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $score;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Set the master score of the local node or the optionally given node.
|
|
|
0c4b27 |
+# As setting an attribute is asynchronous, this will return as soon as the
|
|
|
0c4b27 |
+# attribute is really set by attrd and available everywhere.
|
|
|
0c4b27 |
+sub _set_master_score {
|
|
|
0c4b27 |
+ my ( $score, $node ) = @_;
|
|
|
0c4b27 |
+ my $node_arg = '';
|
|
|
0c4b27 |
+ my $tmp;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $node_arg = sprintf '--node "%s"', $node if defined $node and $node ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ qx{ $CRM_MASTER $node_arg --quiet --update "$score" };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ while ( ( $tmp = _get_master_score( $node ) ) ne $score ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_set_master_score: waiting to set score to "%s" (currently "%s")...',
|
|
|
0c4b27 |
+ $score, $tmp );
|
|
|
0c4b27 |
+ select(undef, undef, undef, 0.1);
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# _master_score_exists
|
|
|
0c4b27 |
+# This subroutine checks if a master score is set for one of the relative clones
|
|
|
0c4b27 |
+# in the cluster and the score is greater or equal of 0.
|
|
|
0c4b27 |
+# Returns 1 if at least one master score >= 0 is found.
|
|
|
0c4b27 |
+# Returns 0 otherwise
|
|
|
0c4b27 |
+sub _master_score_exists {
|
|
|
0c4b27 |
+ my @partition_nodes = split /\s+/ => qx{ $CRM_NODE --partition };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ foreach my $node ( @partition_nodes ) {
|
|
|
0c4b27 |
+ my $score = _get_master_score( $node );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 1 if defined $score and $score ne '' and $score > -1;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 0;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check if the current transiation is a recover of a master clone on given node.
|
|
|
0c4b27 |
+sub _is_master_recover {
|
|
|
0c4b27 |
+ my ( $n ) = @_;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return (
|
|
|
0c4b27 |
+ scalar grep { $_->{'uname'} eq $n } @{ $OCF_NOTIFY_ENV{'master'} }
|
|
|
0c4b27 |
+ and scalar grep { $_->{'uname'} eq $n } @{ $OCF_NOTIFY_ENV{'promote'} }
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check if the current transition is a recover of a slave clone on given node.
|
|
|
0c4b27 |
+sub _is_slave_recover {
|
|
|
0c4b27 |
+ my ( $n ) = @_;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return (
|
|
|
0c4b27 |
+ scalar grep { $_->{'uname'} eq $n } @{ $OCF_NOTIFY_ENV{'slave'} }
|
|
|
0c4b27 |
+ and scalar grep { $_->{'uname'} eq $n } @{ $OCF_NOTIFY_ENV{'start'} }
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# check if th current transition is a switchover to the given node.
|
|
|
0c4b27 |
+sub _is_switchover {
|
|
|
0c4b27 |
+ my ( $n ) = @_;
|
|
|
0c4b27 |
+ my $old = $OCF_NOTIFY_ENV{'master'}[0]{'uname'};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 0 if scalar @{ $OCF_NOTIFY_ENV{'master'} } != 1
|
|
|
0c4b27 |
+ or scalar @{ $OCF_NOTIFY_ENV{'demote'} } != 1
|
|
|
0c4b27 |
+ or scalar @{ $OCF_NOTIFY_ENV{'promote'} } != 1;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return (
|
|
|
0c4b27 |
+ scalar grep { $_->{'uname'} eq $old } @{ $OCF_NOTIFY_ENV{'demote'} }
|
|
|
0c4b27 |
+ and scalar grep { $_->{'uname'} eq $n } @{ $OCF_NOTIFY_ENV{'slave'} }
|
|
|
0c4b27 |
+ and scalar grep { $_->{'uname'} eq $n } @{ $OCF_NOTIFY_ENV{'promote'} }
|
|
|
0c4b27 |
+ and not scalar grep { $_->{'uname'} eq $old } @{ $OCF_NOTIFY_ENV{'stop'} }
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Run the given command as the "system_user" given as parameter.
|
|
|
0c4b27 |
+# It basically forks and seteuid/setuid away from root.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _runas {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my $pid;
|
|
|
0c4b27 |
+ my @cmd = @_;
|
|
|
0c4b27 |
+ my (undef, undef, $postgres_uid, $postgres_gid ) = getpwnam( $system_user );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $pid = fork;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $pid == 0 ) { # in child
|
|
|
0c4b27 |
+ $) = "$postgres_gid $postgres_gid";
|
|
|
0c4b27 |
+ while ( my ( undef, undef, $gid, $members ) = getgrent ) {
|
|
|
0c4b27 |
+ $) .= " $gid" if grep { $system_user eq $_ } split /\s+/, $members
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ $( = $postgres_gid;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $< = $> = $postgres_uid;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ exec @cmd;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_runas: launching as "%s" command "%s"', $system_user,
|
|
|
0c4b27 |
+ join(' ', @cmd) );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ waitpid $pid, 0;
|
|
|
0c4b27 |
+ $rc = $? >> 8;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $rc;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check if instance is listening on the given host/port.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _pg_isready {
|
|
|
0c4b27 |
+ # Add 60s to the timeout or use a 24h timeout fallback to make sure
|
|
|
0c4b27 |
+ # Pacemaker will give up before us and take decisions
|
|
|
0c4b27 |
+ my $timeout = ( _get_action_timeout() || 60*60*24 ) + 60;
|
|
|
0c4b27 |
+ my $rc = _runas( $PGISREADY, '-h', $pghost, '-p', $pgport, '-d', 'postgres', '-t', $timeout );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Possible error codes:
|
|
|
0c4b27 |
+ # 1: ping rejected (usually when instance is in startup, in crash
|
|
|
0c4b27 |
+ # recovery, in warm standby, or when a shutdown is in progress)
|
|
|
0c4b27 |
+ # 2: no response, usually means the instance is down
|
|
|
0c4b27 |
+ # 3: no attempt, probably a syntax error, should not happen
|
|
|
0c4b27 |
+ return $rc;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check the postmaster.pid file and the postmaster process.
|
|
|
0c4b27 |
+# WARNING: we do not distinguish the scenario where postmaster.pid does not
|
|
|
0c4b27 |
+# exist from the scenario where the process is still alive. It should be ok
|
|
|
0c4b27 |
+# though, as this is considered a hard error from monitor.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _pg_ctl_status {
|
|
|
0c4b27 |
+ my $rc = _runas( $PGCTL, '--pgdata', $pgdata, 'status' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Possible error codes:
|
|
|
0c4b27 |
+ # 3: postmaster.pid file does not exist OR it does but the process
|
|
|
0c4b27 |
+ # with the PID found in the file is not alive
|
|
|
0c4b27 |
+ return $rc;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Start the local instance using pg_ctl
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _pg_ctl_start {
|
|
|
0c4b27 |
+ # Add 60s to the timeout or use a 24h timeout fallback to make sure
|
|
|
0c4b27 |
+ # Pacemaker will give up before us and take decisions
|
|
|
0c4b27 |
+ my $timeout = ( _get_action_timeout() || 60*60*24 ) + 60;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ my @cmd = ( $PGCTL, '--pgdata', $pgdata, '-w', '--timeout', $timeout, 'start' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ push @cmd => ( '-o', $start_opts ) if $start_opts ne '';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return _runas( @cmd );
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Enable the Standby mode.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# Up to v11, creates the recovery.conf file based on the given template.
|
|
|
0c4b27 |
+# Since v12, creates standby.signal.
|
|
|
0c4b27 |
+sub _enable_recovery {
|
|
|
0c4b27 |
+ my $fh;
|
|
|
0c4b27 |
+ my $content = '';
|
|
|
0c4b27 |
+ my $standby_file = "$datadir/standby.signal";
|
|
|
0c4b27 |
+ my (undef, undef, $uid, $gid) = getpwnam($system_user);
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $PGVERNUM < $PGVER_12 ) {
|
|
|
0c4b27 |
+ $standby_file = "$datadir/recovery.conf";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_enable_recovery: get replication configuration from the template file "%s"',
|
|
|
0c4b27 |
+ $recovery_tpl );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Create the recovery.conf file to start the instance as a secondary.
|
|
|
0c4b27 |
+ # NOTE: the recovery.conf is supposed to be set up so the secondary can
|
|
|
0c4b27 |
+ # connect to the primary instance, eg. using a virtual IP address.
|
|
|
0c4b27 |
+ # As there is no primary instance available at startup, secondaries will
|
|
|
0c4b27 |
+ # complain about failing to connect.
|
|
|
0c4b27 |
+ # As we can not reload a recovery.conf file on a standby without restarting
|
|
|
0c4b27 |
+ # it, we will leave with this.
|
|
|
0c4b27 |
+ # FIXME how would the reload help us in this case ?
|
|
|
0c4b27 |
+ unless ( defined open( $fh, '<', $recovery_tpl ) ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not open file "%s": %s', $recovery_tpl, $! );
|
|
|
0c4b27 |
+ exit $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Copy all parameters from the template file
|
|
|
0c4b27 |
+ while (my $line = <$fh>) {
|
|
|
0c4b27 |
+ chomp $line;
|
|
|
0c4b27 |
+ $content .= "$line\n";
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ close $fh;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_enable_recovery: write the standby file "%s"', $standby_file );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( open( $fh, '>', $standby_file ) ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not open file "%s": %s', $standby_file, $! );
|
|
|
0c4b27 |
+ exit $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Write the recovery.conf file using configuration from the template file
|
|
|
0c4b27 |
+ print $fh $content;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ close $fh;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( chown $uid, $gid, $standby_file ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not set owner of "%s"', $standby_file );
|
|
|
0c4b27 |
+ exit $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+ };
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Parse and return various informations about the local PostgreSQL instance as
|
|
|
0c4b27 |
+# reported by its controldata file.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# WARNING: the status is NOT updated in case of crash.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# This sub exit the script with an error on failure
|
|
|
0c4b27 |
+sub _get_controldata {
|
|
|
0c4b27 |
+ my %controldata;
|
|
|
0c4b27 |
+ my $ans;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ans = qx{ $PGCTRLDATA "$datadir" 2>/dev/null };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Parse the output of pg_controldata.
|
|
|
0c4b27 |
+ # This output is quite stable between pg versions, but we might need to sort
|
|
|
0c4b27 |
+ # it at some point if things are moving in there...
|
|
|
0c4b27 |
+ $ans =~ m{
|
|
|
0c4b27 |
+ # get the current state
|
|
|
0c4b27 |
+ ^\QDatabase cluster state\E:\s+(.*?)\s*$
|
|
|
0c4b27 |
+ .*
|
|
|
0c4b27 |
+ # Get the latest known REDO location
|
|
|
0c4b27 |
+ ^\QLatest checkpoint's REDO location\E:\s+([/0-9A-F]+)\s*$
|
|
|
0c4b27 |
+ .*
|
|
|
0c4b27 |
+ # Get the latest known TL
|
|
|
0c4b27 |
+ ^\QLatest checkpoint's TimeLineID\E:\s+(\d+)\s*$
|
|
|
0c4b27 |
+ .*
|
|
|
0c4b27 |
+ # Get the wal level
|
|
|
0c4b27 |
+ # NOTE: pg_controldata output changed with PostgreSQL 9.5, so we need to
|
|
|
0c4b27 |
+ # account for both syntaxes
|
|
|
0c4b27 |
+ ^(?:\QCurrent \E)?\Qwal_level setting\E:\s+(.*?)\s*$
|
|
|
0c4b27 |
+ }smx;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $controldata{'state'} = $1 if defined $1;
|
|
|
0c4b27 |
+ $controldata{'redo'} = $2 if defined $2;
|
|
|
0c4b27 |
+ $controldata{'tl'} = $3 if defined $3;
|
|
|
0c4b27 |
+ $controldata{'wal_level'} = $4 if defined $4;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ "_get_controldata: found: %s",
|
|
|
0c4b27 |
+ Data::Dumper->new( [ \%controldata ] )->Terse(1)->Dump );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return %controldata if defined $controldata{'state'}
|
|
|
0c4b27 |
+ and defined $controldata{'tl'}
|
|
|
0c4b27 |
+ and defined $controldata{'redo'}
|
|
|
0c4b27 |
+ and defined $controldata{'wal_level'};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not read all datas from controldata file for "%s"',
|
|
|
0c4b27 |
+ $datadir );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ "_get_controldata: controldata file: %s",
|
|
|
0c4b27 |
+ Data::Dumper->new( [ \%controldata ] )->Terse(1)->Dump, $ans );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ exit $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Pead major version from datadir/PG_VERSION and return it as numeric version
|
|
|
0c4b27 |
+sub _get_pg_version {
|
|
|
0c4b27 |
+ my $fh;
|
|
|
0c4b27 |
+ my $PGVERSION;
|
|
|
0c4b27 |
+ my $PGVERNUM;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check PG_VERSION
|
|
|
0c4b27 |
+ if ( ! -s "$datadir/PG_VERSION" ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'PG_VERSION does not exist in "%s"', $datadir );
|
|
|
0c4b27 |
+ exit $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( open( $fh, '<', "$datadir/PG_VERSION" ) ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( "Could not open file \"$datadir/PG_VERSION\": $!" );
|
|
|
0c4b27 |
+ exit $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ read( $fh, $PGVERSION, 32 );
|
|
|
0c4b27 |
+ close $fh;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ chomp $PGVERSION;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $PGVERSION =~ /^(\d+)(?:\.(\d+))?$/;
|
|
|
0c4b27 |
+ $PGVERNUM = $1 * 10000;
|
|
|
0c4b27 |
+ $PGVERNUM += $2 * 100 if $1 < 10; # no 2nd num in the major version from v10
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $PGVERNUM;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Use pg_controldata to check the state of the PostgreSQL server. This
|
|
|
0c4b27 |
+# function returns codes depending on this state, so we can find whether the
|
|
|
0c4b27 |
+# instance is a primary or a secondary, or use it to detect any inconsistency
|
|
|
0c4b27 |
+# that could indicate the instance has crashed.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _controldata_to_ocf {
|
|
|
0c4b27 |
+ my %cdata = _get_controldata();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ while ( 1 ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug', '_controldata: instance "%s" state is "%s"',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $cdata{'state'} );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Instance should be running as a primary.
|
|
|
0c4b27 |
+ return $OCF_RUNNING_MASTER if $cdata{'state'} eq "in production";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Instance should be running as a secondary.
|
|
|
0c4b27 |
+ # This state includes warm standby (rejects connections attempts,
|
|
|
0c4b27 |
+ # including pg_isready)
|
|
|
0c4b27 |
+ return $OCF_SUCCESS if $cdata{'state'} eq "in archive recovery";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # The instance should be stopped.
|
|
|
0c4b27 |
+ # We don't care if it was a primary or secondary before, because we
|
|
|
0c4b27 |
+ # always start instances as secondaries, and then promote if necessary.
|
|
|
0c4b27 |
+ return $OCF_NOT_RUNNING if $cdata{'state'} eq "shut down"
|
|
|
0c4b27 |
+ or $cdata{'state'} eq "shut down in recovery";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # The state is "in crash recovery", "starting up" or "shutting down".
|
|
|
0c4b27 |
+ # This state should be transitional, so we wait and loop to check if
|
|
|
0c4b27 |
+ # it changes.
|
|
|
0c4b27 |
+ # If it does not, pacemaker will eventually abort with a timeout.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_controldata: waiting for transitionnal state "%s" to finish',
|
|
|
0c4b27 |
+ $cdata{'state'} );
|
|
|
0c4b27 |
+ sleep 1;
|
|
|
0c4b27 |
+ %cdata = _get_controldata();
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If we reach this point, something went really wrong with this code or
|
|
|
0c4b27 |
+ # pg_controldata.
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Unable get instance "%s" state using pg_controldata',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED ;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check the write_location of all secondaries, and adapt their master score so
|
|
|
0c4b27 |
+# that the instance closest to the master will be the selected candidate should
|
|
|
0c4b27 |
+# a promotion be triggered.
|
|
|
0c4b27 |
+# NOTE: This is only a hint to pacemaker! The selected candidate to promotion
|
|
|
0c4b27 |
+# actually re-check it is the best candidate and force a re-election by failing
|
|
|
0c4b27 |
+# if a better one exists. This avoid a race condition between the call of the
|
|
|
0c4b27 |
+# monitor action and the promotion where another slave might have catchup faster
|
|
|
0c4b27 |
+# with the master.
|
|
|
0c4b27 |
+# NOTE: we cannot directly use the write_location, neither a lsn_diff value as
|
|
|
0c4b27 |
+# promotion score as Pacemaker considers any value greater than 1,000,000 as
|
|
|
0c4b27 |
+# INFINITY.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# This sub must be executed from a master monitor action.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _check_locations {
|
|
|
0c4b27 |
+ my $partition_nodes;
|
|
|
0c4b27 |
+ my $node_score;
|
|
|
0c4b27 |
+ my $row_num;
|
|
|
0c4b27 |
+ my $row;
|
|
|
0c4b27 |
+ my @rs;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Set the master score if not already done
|
|
|
0c4b27 |
+ $node_score = _get_master_score();
|
|
|
0c4b27 |
+ _set_master_score( '1001' ) unless $node_score eq '1001';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Ask crm_node what nodes are present in our current cluster partition
|
|
|
0c4b27 |
+ $partition_nodes = qx{ $CRM_NODE --partition };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ @rs = @{ _get_lag_scores() };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $row_num = scalar @rs;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If no lag are reported at this point, it means that there is no
|
|
|
0c4b27 |
+ # secondary instance connected.
|
|
|
0c4b27 |
+ ocf_log( 'warning', 'No secondary connected to the master' )
|
|
|
0c4b27 |
+ if $row_num == 0;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # For each standby connected, set their master score based on the following
|
|
|
0c4b27 |
+ # rule: the first known node/application, with the highest priority and
|
|
|
0c4b27 |
+ # an acceptable state.
|
|
|
0c4b27 |
+ while ( $row = shift @rs ) {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $partition_nodes !~ /$row->[0]/ ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Ignoring unknown application_name/node "%s"',
|
|
|
0c4b27 |
+ $row->[0] );
|
|
|
0c4b27 |
+ next;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $row->[0] eq $nodename ) {
|
|
|
0c4b27 |
+ ocf_log( 'warning', 'Streaming replication with myself!' );
|
|
|
0c4b27 |
+ next;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $node_score = _get_master_score( $row->[0] );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $row->[3] =~ /^\s*(?:startup|backup)\s*$/ ) {
|
|
|
0c4b27 |
+ # We exclude any standby being in state backup (pg_basebackup) or
|
|
|
0c4b27 |
+ # startup (new standby or failing standby)
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Forbidding promotion on "%s" in state "%s"',
|
|
|
0c4b27 |
+ $row->[0], $row->[3] );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ _set_master_score( '-1', $row->[0] ) unless $node_score eq '-1';
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_check_locations: checking "%s" promotion ability (current_score: %s, priority: %s, location: %s, lag: %s)',
|
|
|
0c4b27 |
+ $row->[0], $node_score, $row->[1], $row->[2], $row->[4] );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $node_score ne $row->[1] ) {
|
|
|
0c4b27 |
+ if ( $row->[1] < -1 ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Update score of "%s" from %s to %s because replication lag (%s) is higher than given maxlag (%s).',
|
|
|
0c4b27 |
+ $row->[0], $node_score, $row->[1], $row->[4], $maxlag );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Update score of "%s" from %s to %s because of a change in the replication lag (%s).',
|
|
|
0c4b27 |
+ $row->[0], $node_score, $row->[1], $row->[4] );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ _set_master_score( $row->[1], $row->[0] );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_check_locations: "%s" keeps its current score of %s',
|
|
|
0c4b27 |
+ $row->[0], $row->[1] );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Remove this node from the known nodes list.
|
|
|
0c4b27 |
+ $partition_nodes =~ s/(?:^|\s)$row->[0](?:\s|$)/ /g;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $partition_nodes =~ s/(?:^\s+)|(?:\s+$)//g;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If there are still nodes in "partition_nodes", it means there is no
|
|
|
0c4b27 |
+ # corresponding line in "pg_stat_replication".
|
|
|
0c4b27 |
+ # Exclude these nodes that are not part of the cluster at this
|
|
|
0c4b27 |
+ # point.
|
|
|
0c4b27 |
+ foreach my $node (split /\s+/ => $partition_nodes) {
|
|
|
0c4b27 |
+ # Exclude the current node.
|
|
|
0c4b27 |
+ next if $node eq $nodename;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # do not warn if the master score is already set to -1000.
|
|
|
0c4b27 |
+ # this avoid log flooding (gh #138)
|
|
|
0c4b27 |
+ $node_score = _get_master_score( $node );
|
|
|
0c4b27 |
+ next if $node_score eq '-1000';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'warning', '"%s" is not connected to the primary', $node );
|
|
|
0c4b27 |
+ _set_master_score( '-1000', $node );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# _check_switchover
|
|
|
0c4b27 |
+# check if the pgsql switchover to the localnode is safe.
|
|
|
0c4b27 |
+# This is supposed to be called **after** the master has been stopped or demoted.
|
|
|
0c4b27 |
+# This sub checks if the local standby received the shutdown checkpoint from the
|
|
|
0c4b27 |
+# old master to make sure it can take over the master role and the old master
|
|
|
0c4b27 |
+# will be able to catchup as a standby after.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+# Returns 0 if switchover is safe
|
|
|
0c4b27 |
+# Returns 1 if swithcover is not safe
|
|
|
0c4b27 |
+# Returns 2 for internal error
|
|
|
0c4b27 |
+sub _check_switchover {
|
|
|
0c4b27 |
+ my $has_sht_chk = 0;
|
|
|
0c4b27 |
+ my $last_redo;
|
|
|
0c4b27 |
+ my $last_lsn;
|
|
|
0c4b27 |
+ my $ans;
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my $tl;
|
|
|
0c4b27 |
+ my %cdata;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $PGWALDUMP = "$bindir/pg_xlogdump" if $PGVERNUM < $PGVER_10;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Switchover in progress from "%s" to "%s".'
|
|
|
0c4b27 |
+ .' Need to check the last record in WAL',
|
|
|
0c4b27 |
+ $OCF_NOTIFY_ENV{'demote'}[0]{'uname'}, $nodename );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check if we received the shutdown checkpoint of the master during its
|
|
|
0c4b27 |
+ # demote process.
|
|
|
0c4b27 |
+ # We need the last local checkpoint LSN and the last received LSN from
|
|
|
0c4b27 |
+ # master to check in the WAL between these adresses if we have a
|
|
|
0c4b27 |
+ # "checkpoint shutdown" using pg_xlogdump/pg_waldump.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # Force a checkpoint to make sure the controldata shows the very last TL
|
|
|
0c4b27 |
+ # and the master's shutdown checkpoint
|
|
|
0c4b27 |
+ _query( q{ CHECKPOINT }, {} );
|
|
|
0c4b27 |
+ %cdata = _get_controldata();
|
|
|
0c4b27 |
+ $tl = $cdata{'tl'};
|
|
|
0c4b27 |
+ $last_redo = $cdata{'redo'};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Get the last received LSN from master
|
|
|
0c4b27 |
+ $last_lsn = _get_last_received_lsn();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( defined $last_lsn ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not fetch last received LSN!' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 2;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $ans = qx{ $PGWALDUMP --path "$datadir" --timeline "$tl" \\
|
|
|
0c4b27 |
+ --start "$last_redo" --end "$last_lsn" 2>&1 };
|
|
|
0c4b27 |
+ $rc = $?;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_check_switchover: %s rc: "%s", tl: "%s", last_chk: %s, last_lsn: %s, output: "%s"',
|
|
|
0c4b27 |
+ $PGWALDUMP, $rc, $tl, $last_redo, $last_lsn, $ans
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc == 0 and
|
|
|
0c4b27 |
+ $ans =~ m{^rmgr: XLOG.*desc: (?i:checkpoint)(?::|_SHUTDOWN) redo [0-9A-F/]+; tli $tl;.*; shutdown$}m
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Slave received the shutdown checkpoint' );
|
|
|
0c4b27 |
+ return 0;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Did not receive the shutdown checkpoint from the old master!' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return 1;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check to confirm if the instance is really started as _pg_isready stated and
|
|
|
0c4b27 |
+# check if the instance is primary or secondary.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _confirm_role {
|
|
|
0c4b27 |
+ my $is_in_recovery;
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my @rs;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = _query( "SELECT pg_is_in_recovery()", \@rs );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $is_in_recovery = $rs[0][0];
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc == 0 ) {
|
|
|
0c4b27 |
+ # The query was executed, check the result.
|
|
|
0c4b27 |
+ if ( $is_in_recovery eq 't' ) {
|
|
|
0c4b27 |
+ # The instance is a secondary.
|
|
|
0c4b27 |
+ ocf_log( 'debug', "_confirm_role: instance $OCF_RESOURCE_INSTANCE is a secondary");
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $is_in_recovery eq 'f' ) {
|
|
|
0c4b27 |
+ # The instance is a primary.
|
|
|
0c4b27 |
+ ocf_log( 'debug', "_confirm_role: instance $OCF_RESOURCE_INSTANCE is a primary");
|
|
|
0c4b27 |
+ # Check lsn diff with current slaves if any
|
|
|
0c4b27 |
+ _check_locations() if $__OCF_ACTION eq 'monitor';
|
|
|
0c4b27 |
+ return $OCF_RUNNING_MASTER;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # This should not happen, raise a hard configuration error.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Unexpected result from query to check if "%s" is a primary or a secondary: "%s"',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $is_in_recovery );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc == 1 or $rc == 2 ) {
|
|
|
0c4b27 |
+ # psql cound not connect to the instance.
|
|
|
0c4b27 |
+ # As pg_isready reported the instance was listening, this error
|
|
|
0c4b27 |
+ # could be a max_connection saturation. Just report a soft error.
|
|
|
0c4b27 |
+ ocf_exit_reason( 'psql could not connect to instance "%s"',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # The query failed (rc: 3) or bad parameters (rc: -1).
|
|
|
0c4b27 |
+ # This should not happen, raise a hard configuration error.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'The query to check if instance "%s" is a primary or a secondary failed (rc: %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Check to confirm if the instance is really stopped as _pg_isready stated
|
|
|
0c4b27 |
+# and if it was propertly shut down.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub _confirm_stopped {
|
|
|
0c4b27 |
+ my $pgctlstatus_rc;
|
|
|
0c4b27 |
+ my $controldata_rc;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Check the postmaster process status.
|
|
|
0c4b27 |
+ $pgctlstatus_rc = _pg_ctl_status();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $pgctlstatus_rc == 0 ) {
|
|
|
0c4b27 |
+ # The PID file exists and the process is available.
|
|
|
0c4b27 |
+ # That should not be the case, return an error.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Instance "%s" is not listening, but the process referenced in postmaster.pid exists',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # The PID file does not exist or the process is not available.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_confirm_stopped: no postmaster process found for instance "%s"',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( -f "$datadir/backup_label" ) {
|
|
|
0c4b27 |
+ # We are probably on a freshly built secondary that was not started yet.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_confirm_stopped: backup_label file exists: probably on a never started secondary',
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ return $OCF_NOT_RUNNING;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Continue the check with pg_controldata.
|
|
|
0c4b27 |
+ $controldata_rc = _controldata_to_ocf();
|
|
|
0c4b27 |
+ if ( $controldata_rc == $OCF_RUNNING_MASTER ) {
|
|
|
0c4b27 |
+ # The controldata has not been updated to "shutdown".
|
|
|
0c4b27 |
+ # It should mean we had a crash on a primary instance.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Instance "%s" controldata indicates a running primary instance, the instance has probably crashed',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_FAILED_MASTER;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $controldata_rc == $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ # The controldata has not been updated to "shutdown in recovery".
|
|
|
0c4b27 |
+ # It should mean we had a crash on a secondary instance.
|
|
|
0c4b27 |
+ # There is no "FAILED_SLAVE" return code, so we return a generic error.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Instance "%s" controldata indicates a running secondary instance, the instance has probably crashed',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $controldata_rc == $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ # The controldata state is consistent, the instance was probably
|
|
|
0c4b27 |
+ # propertly shut down.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ '_confirm_stopped: instance "%s" controldata indicates that the instance was propertly shut down',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_NOT_RUNNING;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Something went wrong with the controldata check.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Could not get instance "%s" status from controldata (returned: %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $controldata_rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+############################################################
|
|
|
0c4b27 |
+#### OCF FUNCS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SUPPORTED PARAMETERS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=over
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<pgdata>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Location of the PGDATA of your instance
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, string, default "/var/lib/pgsql/data")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<pghost>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The socket directory or IP address to use to connect to the local instance
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, string, default "/tmp")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<pgport>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The port to connect to the local instance
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, integer, default "5432")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<bindir>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Location of the PostgreSQL binaries.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, string, default "/usr/bin")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<system_user>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The system owner of your instance's process
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, string, default "postgres")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<recovery_template>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+B<ONLY> for PostgreSQL 11 and bellow.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The local template that will be copied as the C<PGDATA/recovery.conf> file.
|
|
|
0c4b27 |
+This template file must exists on all node.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+With PostgreSQL 12 and higher, the cluster will refuse to start if this
|
|
|
0c4b27 |
+parameter is set or a template file is found.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, string, default "$PGDATA/recovery.conf.pcmk")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<maxlag>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Maximum lag allowed on a standby before we set a negative master score on it.
|
|
|
0c4b27 |
+The calculation is based on the difference between the current xlog location on
|
|
|
0c4b27 |
+the master and the write location on the standby.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, integer, default "0" disables this feature)
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<datadir>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Path to the directory set in C<data_directory> from your postgresql.conf file.
|
|
|
0c4b27 |
+This parameter has same default than PostgreSQL itself: the C<pgdata> parameter
|
|
|
0c4b27 |
+value.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Unless you have a special PostgreSQL setup and you understand this parameter,
|
|
|
0c4b27 |
+B<ignore it>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optional, string, default to the value of C<pgdata>)
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<start_opts>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Additional arguments given to the postgres process on startup. See
|
|
|
0c4b27 |
+"postgres --help" for available options. Useful when the postgresql.conf file
|
|
|
0c4b27 |
+is not in the data directory (PGDATA), eg.:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ -c config_file=/etc/postgresql/9.3/main/postgresql.conf
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+(optinal, string, default "")
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=back
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_meta_data {
|
|
|
0c4b27 |
+ print qq{
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <resource-agent name="pgsqlms">
|
|
|
0c4b27 |
+ <version>1.0</version>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Resource script for PostgreSQL in replication. It manages PostgreSQL servers using streaming replication as an HA resource.
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Manages PostgreSQL servers in replication</shortdesc>
|
|
|
0c4b27 |
+ <parameters>
|
|
|
0c4b27 |
+ <parameter name="system_user" unique="0" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ System user account used to run the PostgreSQL server
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">PostgreSQL system User</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="$system_user_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="bindir" unique="0" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Path to the directory storing the PostgreSQL binaries. The agent uses psql, pg_isready, pg_controldata and pg_ctl.
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Path to the PostgreSQL binaries</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="$bindir_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="pgdata" unique="1" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Path to the data directory, e.g. PGDATA
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Path to the data directory</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="$pgdata_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="datadir" unique="1" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Path to the directory set in data_directory from your postgresql.conf file. This parameter
|
|
|
0c4b27 |
+ has the same default than PostgreSQL itself: the pgdata parameter value. Unless you have a
|
|
|
0c4b27 |
+ special PostgreSQL setup and you understand this parameter, ignore it.
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Path to the directory set in data_directory from your postgresql.conf file</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="PGDATA" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="pghost" unique="0" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Host IP address or unix socket folder the instance is listening on.
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Instance IP or unix socket folder</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="$pghost_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="pgport" unique="0" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Port the instance is listening on.
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Instance port</shortdesc>
|
|
|
0c4b27 |
+ <content type="integer" default="$pgport_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="maxlag" unique="0" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Maximum lag allowed on a standby before we set a negative master score on it. The calculation
|
|
|
0c4b27 |
+ is based on the difference between the current LSN on the master and the LSN
|
|
|
0c4b27 |
+ written on the standby.
|
|
|
0c4b27 |
+ This parameter must be a valid positive number as described in PostgreSQL documentation.
|
|
|
0c4b27 |
+ See: https://www.postgresql.org/docs/current/static/sql-syntax-lexical.html#SQL-SYNTAX-CONSTANTS-NUMERIC
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Maximum write lag before we mark a standby as inappropriate to promote</shortdesc>
|
|
|
0c4b27 |
+ <content type="integer" default="$maxlag_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="recovery_template" unique="1" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Path to the recovery.conf template. This file is simply copied to \$PGDATA
|
|
|
0c4b27 |
+ before starting the instance as slave.
|
|
|
0c4b27 |
+ ONLY for PostgreSQL 11 and bellow. This parameter is IGNORED for
|
|
|
0c4b27 |
+ PostgreSQL 12 and higher. The cluster will refuse to start if a template
|
|
|
0c4b27 |
+ file is found.
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Path to the recovery.conf template for PostgreSQL 11 and older.</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="PGDATA/recovery.conf.pcmk" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ <parameter name="start_opts" unique="0" required="0">
|
|
|
0c4b27 |
+ <longdesc lang="en">
|
|
|
0c4b27 |
+ Additionnal arguments given to the postgres process on startup.
|
|
|
0c4b27 |
+ See "postgres --help" for available options. Usefull when the
|
|
|
0c4b27 |
+ postgresql.conf file is not in the data directory (PGDATA), eg.:
|
|
|
0c4b27 |
+ "-c config_file=/etc/postgresql/9.3/main/postgresql.conf".
|
|
|
0c4b27 |
+ </longdesc>
|
|
|
0c4b27 |
+ <shortdesc lang="en">Additionnal arguments given to the postgres process on startup.</shortdesc>
|
|
|
0c4b27 |
+ <content type="string" default="$start_opts_default" />
|
|
|
0c4b27 |
+ </parameter>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ </parameters>
|
|
|
0c4b27 |
+ <actions>
|
|
|
0c4b27 |
+ <action name="start" timeout="60" />
|
|
|
0c4b27 |
+ <action name="stop" timeout="60" />
|
|
|
0c4b27 |
+ <action name="reload" timeout="20" />
|
|
|
0c4b27 |
+ <action name="promote" timeout="30" />
|
|
|
0c4b27 |
+ <action name="demote" timeout="120" />
|
|
|
0c4b27 |
+ <action name="monitor" depth="0" timeout="10" interval="15"/>
|
|
|
0c4b27 |
+ <action name="monitor" depth="0" timeout="10" interval="15" role="Master"/>
|
|
|
0c4b27 |
+ <action name="monitor" depth="0" timeout="10" interval="16" role="Slave"/>
|
|
|
0c4b27 |
+ <action name="notify" timeout="60" />
|
|
|
0c4b27 |
+ <action name="meta-data" timeout="5" />
|
|
|
0c4b27 |
+ <action name="validate-all" timeout="5" />
|
|
|
0c4b27 |
+ <action name="methods" timeout="5" />
|
|
|
0c4b27 |
+ </actions>
|
|
|
0c4b27 |
+ </resource-agent>
|
|
|
0c4b27 |
+ };
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SUPPORTED ACTIONS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+This resource agent supports the following actions (operations):
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=over
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<start>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Starts the resource. Suggested minimum timeout: 60.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<stop>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Stops the resource. Suggested minimum timeout: 60.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<reload>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Suggested minimum timeout: 20.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<promote>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Promotes the resource to the Master role. Suggested minimum timeout: 30.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<demote>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Demotes the resource to the Slave role. Suggested minimum timeout: 120.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<monitor (Master role)>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Performs a detailed status check. Suggested minimum timeout: 10.
|
|
|
0c4b27 |
+Suggested interval: 15.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<monitor (Slave role)>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Performs a detailed status check. Suggested minimum timeout: 10.
|
|
|
0c4b27 |
+Suggested interval: 16.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<notify>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Suggested minimum timeout: 60
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<meta-data>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Retrieves resource agent metadata (internal use only).
|
|
|
0c4b27 |
+Suggested minimum timeout: 5.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<methods>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Suggested minimum timeout: 5.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=item B<validate-all>
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Performs a validation of the resource configuration.
|
|
|
0c4b27 |
+Suggested minimum timeout: 5.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=back
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub ocf_methods {
|
|
|
0c4b27 |
+ print q{
|
|
|
0c4b27 |
+ start
|
|
|
0c4b27 |
+ stop
|
|
|
0c4b27 |
+ reload
|
|
|
0c4b27 |
+ promote
|
|
|
0c4b27 |
+ demote
|
|
|
0c4b27 |
+ monitor
|
|
|
0c4b27 |
+ notify
|
|
|
0c4b27 |
+ methods
|
|
|
0c4b27 |
+ meta-data
|
|
|
0c4b27 |
+ validate-all
|
|
|
0c4b27 |
+ };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+############################################################
|
|
|
0c4b27 |
+#### RA FUNCS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+sub pgsql_validate_all {
|
|
|
0c4b27 |
+ my $fh;
|
|
|
0c4b27 |
+ my $ans = '';
|
|
|
0c4b27 |
+ my %cdata;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless (
|
|
|
0c4b27 |
+ ocf_version_cmp( $ENV{"OCF_RESKEY_crm_feature_set"}, '3.0.9' ) == 2
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'PAF %s is compatible with Pacemaker 1.1.13 and greater',
|
|
|
0c4b27 |
+ $VERSION
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check notify=true
|
|
|
0c4b27 |
+ $ans = qx{ $CRM_RESOURCE --resource "$OCF_RESOURCE_INSTANCE" \\
|
|
|
0c4b27 |
+ --meta --get-parameter notify 2>/dev/null };
|
|
|
0c4b27 |
+ chomp $ans;
|
|
|
0c4b27 |
+ unless ( lc($ans) =~ /^true$|^on$|^yes$|^y$|^1$/ ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'You must set meta parameter notify=true for your master resource'
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check master-max=1
|
|
|
0c4b27 |
+ unless (
|
|
|
0c4b27 |
+ defined $ENV{'OCF_RESKEY_CRM_meta_master_max'}
|
|
|
0c4b27 |
+ and $ENV{'OCF_RESKEY_CRM_meta_master_max'} eq '1'
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'You must set meta parameter master-max=1 for your master resource'
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $PGVERNUM >= $PGVER_12 ) {
|
|
|
0c4b27 |
+ # check PostgreSQL setup: checks related to v12 and after
|
|
|
0c4b27 |
+ my $guc;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # recovery.conf template must not exists
|
|
|
0c4b27 |
+ if ( -f $recovery_tpl ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Recovery template file "%s" is forbidden for PostgreSQL 12 and above',
|
|
|
0c4b27 |
+ $recovery_tpl );
|
|
|
0c4b27 |
+ exit $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # WARNING: you MUST put -C as first argument to bypass the root check
|
|
|
0c4b27 |
+ $guc = qx{ $POSTGRES -C recovery_target_timeline -D "$pgdata" $start_opts};
|
|
|
0c4b27 |
+ chomp $guc;
|
|
|
0c4b27 |
+ unless ( $guc eq 'latest' ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ q{Parameter "recovery_target_timeline" MUST be set to 'latest'. } .
|
|
|
0c4b27 |
+ q{It is currently set to '%s'}, $guc );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $guc = qx{ $POSTGRES -C primary_conninfo -D "$pgdata" $start_opts};
|
|
|
0c4b27 |
+ unless ($guc =~ /\bapplication_name='?$nodename'?\b/) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ q{Parameter "primary_conninfo" MUST contain 'application_name=%s'. }.
|
|
|
0c4b27 |
+ q{It is currently set to '%s'}, $nodename, $guc );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ my @content;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check recovery template
|
|
|
0c4b27 |
+ if ( ! -f $recovery_tpl ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Recovery template file "%s" does not exist',
|
|
|
0c4b27 |
+ $recovery_tpl );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check content of the recovery template file
|
|
|
0c4b27 |
+ unless ( open( $fh, '<', $recovery_tpl ) ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Could not open file "%s": %s', $recovery_tpl, $! );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ @content = <$fh>;
|
|
|
0c4b27 |
+ close $fh;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( grep /^\s*standby_mode\s*=\s*'?on'?\s*$/, @content ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Recovery template file must contain "standby_mode = on"' );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( grep /^\s*recovery_target_timeline\s*=\s*'?latest'?\s*$/, @content ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ "Recovery template file must contain \"recovery_target_timeline = 'latest'\""
|
|
|
0c4b27 |
+ );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless (
|
|
|
0c4b27 |
+ grep /^\s*primary_conninfo\s*=.*['\s]application_name=$nodename['\s]/,
|
|
|
0c4b27 |
+ @content
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Recovery template file must contain in primary_conninfo parameter "application_name=%s"',
|
|
|
0c4b27 |
+ $nodename );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( looks_like_number($maxlag) ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'maxlag is not a number: "%s"', $maxlag );
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check system user
|
|
|
0c4b27 |
+ unless ( defined getpwnam $system_user ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'System user "%s" does not exist', $system_user );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # require 9.3 minimum
|
|
|
0c4b27 |
+ if ( $PGVERNUM < $PGVER_93 ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( "Require 9.3 and more" );
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check binaries
|
|
|
0c4b27 |
+ unless ( -x $PGCTL and -x $PGPSQL and -x $PGCTRLDATA and -x $PGISREADY
|
|
|
0c4b27 |
+ and ( -x $PGWALDUMP or -x "$bindir/pg_xlogdump")
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ "Missing one or more binary. Check following path: %s, %s, %s, %s, %s or %s",
|
|
|
0c4b27 |
+ $PGCTL, $PGPSQL, $PGCTRLDATA, $PGISREADY, $PGWALDUMP, "$bindir/pg_xlogdump" );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # require wal_level >= hot_standby
|
|
|
0c4b27 |
+ %cdata = _get_controldata();
|
|
|
0c4b27 |
+ unless ( $cdata{'wal_level'} =~ m{hot_standby|logical|replica} ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'wal_level must be one of "hot_standby", "logical" or "replica"' );
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Start the PostgreSQL instance as a *secondary*
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_start {
|
|
|
0c4b27 |
+ my $rc = pgsql_monitor();
|
|
|
0c4b27 |
+ my %cdata = _get_controldata();
|
|
|
0c4b27 |
+ my $prev_state = $cdata{'state'};
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Instance must be running as secondary or being stopped.
|
|
|
0c4b27 |
+ # Anything else is an error.
|
|
|
0c4b27 |
+ if ( $rc == $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Instance "%s" already started',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc != $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Unexpected state for instance "%s" (returned %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # From here, the instance is NOT running for sure.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_start: instance "%s" is not running, starting it as a secondary',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Must start as a standby, so enable recovery.
|
|
|
0c4b27 |
+ _enable_recovery();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Start the instance as a secondary.
|
|
|
0c4b27 |
+ $rc = _pg_ctl_start();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc == 0 ) {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Wait for the start to finish.
|
|
|
0c4b27 |
+ sleep 1 while ( $rc = pgsql_monitor() ) == $OCF_NOT_RUNNING;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc == $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Instance "%s" started', $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Check if a master score exists in the cluster.
|
|
|
0c4b27 |
+ # During the very first start of the cluster, no master score will
|
|
|
0c4b27 |
+ # exists on any of the existing slaves, unless an admin designated
|
|
|
0c4b27 |
+ # one of them using crm_master. If no master exists the cluster will
|
|
|
0c4b27 |
+ # not promote a master among the slaves.
|
|
|
0c4b27 |
+ # To solve this situation, we check if there is at least one master
|
|
|
0c4b27 |
+ # score existing on one node in the cluster. Do nothing if at least
|
|
|
0c4b27 |
+ # one master score is found among the clones of the resource. If no
|
|
|
0c4b27 |
+ # master score exists, set a score of 1 only if the resource was a
|
|
|
0c4b27 |
+ # shut downed master before the start.
|
|
|
0c4b27 |
+ if ( $prev_state eq "shut down" and not _master_score_exists() ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'No master score around. Set mine to 1' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ _set_master_score( '1' );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Instance "%s" is not running as a slave (returned %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Instance "%s" failed to start (rc: %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Stop the PostgreSQL instance
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_stop {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my $state;
|
|
|
0c4b27 |
+ my $pidfile = "$datadir/postmaster.pid";
|
|
|
0c4b27 |
+ # Add 60s to the timeout or use a 24h timeout fallback to make sure
|
|
|
0c4b27 |
+ # Pacemaker will give up before us and take decisions
|
|
|
0c4b27 |
+ my $timeout = ( _get_action_timeout() || 60*60*24 ) + 60;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Instance must be running as secondary or primary or being stopped.
|
|
|
0c4b27 |
+ # Anything else is an error.
|
|
|
0c4b27 |
+ $rc = pgsql_monitor();
|
|
|
0c4b27 |
+ if ( $rc == $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Instance "%s" already stopped',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc != $OCF_SUCCESS and $rc != $OCF_RUNNING_MASTER ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Unexpected state for instance "%s" (returned %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # From here, the instance is running for sure.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_stop: instance "%s" is running, stopping it',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Try to quit with proper shutdown.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = _runas( $PGCTL, '--pgdata', $pgdata, '-w', '--timeout', $timeout,
|
|
|
0c4b27 |
+ '-m', 'fast', 'stop' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc == 0 ) {
|
|
|
0c4b27 |
+ # Wait for the stop to finish.
|
|
|
0c4b27 |
+ sleep 1 while ( $rc = pgsql_monitor() ) != $OCF_NOT_RUNNING ;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Instance "%s" stopped', $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Instance "%s" failed to stop', $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Monitor the PostgreSQL instance
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_monitor {
|
|
|
0c4b27 |
+ my $pgisready_rc;
|
|
|
0c4b27 |
+ my $controldata_rc;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_monitor: monitor is a probe' ) if ocf_is_probe();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # First check, verify if the instance is listening.
|
|
|
0c4b27 |
+ $pgisready_rc = _pg_isready();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $pgisready_rc == 0 ) {
|
|
|
0c4b27 |
+ # The instance is listening.
|
|
|
0c4b27 |
+ # We confirm that the instance is up and return if it is a primary or a
|
|
|
0c4b27 |
+ # secondary
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_monitor: instance "%s" is listening',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return _confirm_role();
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $pgisready_rc == 1 ) {
|
|
|
0c4b27 |
+ # The attempt was rejected.
|
|
|
0c4b27 |
+ # This could happen in several cases:
|
|
|
0c4b27 |
+ # - at startup
|
|
|
0c4b27 |
+ # - during shutdown
|
|
|
0c4b27 |
+ # - during crash recovery
|
|
|
0c4b27 |
+ # - if instance is a warm standby
|
|
|
0c4b27 |
+ # Except for the warm standby case, this should be a transitional state.
|
|
|
0c4b27 |
+ # We try to confirm using pg_controldata.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_monitor: instance "%s" rejects connections - checking again...',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ $controldata_rc = _controldata_to_ocf();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $controldata_rc == $OCF_RUNNING_MASTER
|
|
|
0c4b27 |
+ or $controldata_rc == $OCF_SUCCESS
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ # This state indicates that pg_isready check should succeed.
|
|
|
0c4b27 |
+ # We check again.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_monitor: instance "%s" controldata shows a running status',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $pgisready_rc = _pg_isready();
|
|
|
0c4b27 |
+ if ( $pgisready_rc == 0 ) {
|
|
|
0c4b27 |
+ # Consistent with pg_controdata output.
|
|
|
0c4b27 |
+ # We can check if the instance is primary or secondary
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_monitor: instance "%s" is listening',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return _confirm_role();
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Still not consistent, raise an error.
|
|
|
0c4b27 |
+ # NOTE: if the instance is a warm standby, we end here.
|
|
|
0c4b27 |
+ # TODO raise an hard error here ?
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Instance "%s" controldata is not consistent with pg_isready (returned: %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $pgisready_rc );
|
|
|
0c4b27 |
+ ocf_log( 'info',
|
|
|
0c4b27 |
+ 'If this instance is in warm standby, this resource agent only supports hot standby',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $pgisready_rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $controldata_rc == $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ # This state indicates that pg_isready check should fail with rc 2.
|
|
|
0c4b27 |
+ # We check again.
|
|
|
0c4b27 |
+ $pgisready_rc = _pg_isready();
|
|
|
0c4b27 |
+ if ( $pgisready_rc == 2 ) {
|
|
|
0c4b27 |
+ # Consistent with pg_controdata output.
|
|
|
0c4b27 |
+ # We check the process status using pg_ctl status and check
|
|
|
0c4b27 |
+ # if it was propertly shut down using pg_controldata.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_monitor: instance "%s" is not listening',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return _confirm_stopped();
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ # Still not consistent, raise an error.
|
|
|
0c4b27 |
+ # TODO raise an hard error here ?
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Instance "%s" controldata is not consistent with pg_isready (returned: %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $pgisready_rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Something went wrong with the controldata check, hard fail.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Could not get instance "%s" status from controldata (returned: %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $controldata_rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_INSTALLED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ elsif ( $pgisready_rc == 2 ) {
|
|
|
0c4b27 |
+ # The instance is not listening.
|
|
|
0c4b27 |
+ # We check the process status using pg_ctl status and check
|
|
|
0c4b27 |
+ # if it was propertly shut down using pg_controldata.
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_monitor: instance "%s" is not listening',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return _confirm_stopped();
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ elsif ( $pgisready_rc == 3 ) {
|
|
|
0c4b27 |
+ # No attempt was done, probably a syntax error.
|
|
|
0c4b27 |
+ # Hard configuration error, we don't want to retry or failover here.
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Unknown error while checking if instance "%s" is listening (returned %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $pgisready_rc );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Unexpected result when checking instance "%s" status',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Demote the PostgreSQL instance from primary to secondary
|
|
|
0c4b27 |
+# To demote a PostgreSQL instance, we must:
|
|
|
0c4b27 |
+# * stop it gracefully
|
|
|
0c4b27 |
+# * create recovery.conf with standby_mode = on
|
|
|
0c4b27 |
+# * start it
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_demote {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = pgsql_monitor();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Running as primary. Normal, expected behavior.
|
|
|
0c4b27 |
+ if ( $rc == $OCF_RUNNING_MASTER ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_demote: "%s" currently running as a primary',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE ) ;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc == $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ # Already running as secondary. Nothing to do.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_demote: "%s" currently running as a secondary',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc == $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ # Instance is stopped. Nothing to do.
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_demote: "%s" currently shut down',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc == $OCF_ERR_CONFIGURED ) {
|
|
|
0c4b27 |
+ # We actually prefer raising a hard or fatal error instead of leaving
|
|
|
0c4b27 |
+ # the CRM abording its transition for a new one because of a soft error.
|
|
|
0c4b27 |
+ # The hard error will force the CRM to move the resource immediately.
|
|
|
0c4b27 |
+ return $OCF_ERR_CONFIGURED;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # TODO we need to make sure at least one slave is connected!!
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # WARNING if the resource state is stopped instead of master, the ocf ra dev
|
|
|
0c4b27 |
+ # rsc advises to return OCF_ERR_GENERIC, misleading the CRM in a loop where
|
|
|
0c4b27 |
+ # it computes transitions of demote(failing)->stop->start->promote actions
|
|
|
0c4b27 |
+ # until failcount == migration-threshold.
|
|
|
0c4b27 |
+ # This is a really ugly trick to keep going with the demode action if the
|
|
|
0c4b27 |
+ # rsc is already stopped gracefully.
|
|
|
0c4b27 |
+ # See discussion "CRM trying to demote a stopped resource" on
|
|
|
0c4b27 |
+ # developers@clusterlabs.org
|
|
|
0c4b27 |
+ unless ( $rc == $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ # Add 60s to the timeout or use a 24h timeout fallback to make sure
|
|
|
0c4b27 |
+ # Pacemaker will give up before us and take decisions
|
|
|
0c4b27 |
+ my $timeout = ( _get_action_timeout() || 60*60*24 ) + 60;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # WARNING the instance **MUST** be stopped gracefully.
|
|
|
0c4b27 |
+ # Do **not** use pg_stop() or service or systemctl here as these
|
|
|
0c4b27 |
+ # commands might force-stop the PostgreSQL instance using immediate
|
|
|
0c4b27 |
+ # after some timeout and return success, which is misleading.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = _runas( $PGCTL, '--pgdata', $pgdata, '--mode', 'fast', '-w',
|
|
|
0c4b27 |
+ '--timeout', $timeout , 'stop' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # No need to wait for stop to complete, this is handled in pg_ctl
|
|
|
0c4b27 |
+ # using -w option.
|
|
|
0c4b27 |
+ unless ( $rc == 0 ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Failed to stop "%s" using pg_ctl (returned %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Double check that the instance is stopped correctly.
|
|
|
0c4b27 |
+ $rc = pgsql_monitor();
|
|
|
0c4b27 |
+ unless ( $rc == $OCF_NOT_RUNNING ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ 'Unexpected "%s" state: monitor status (%d) disagree with pg_ctl return code',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # At this point, the instance **MUST** be stopped gracefully.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Note: We do not need to handle the recovery.conf file here as pgsql_start
|
|
|
0c4b27 |
+ # deal with that itself. Equally, no need to wait for the start to complete
|
|
|
0c4b27 |
+ # here, handled in pgsql_start.
|
|
|
0c4b27 |
+ $rc = pgsql_start();
|
|
|
0c4b27 |
+ if ( $rc == $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'pgsql_demote: "%s" started as a secondary',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # NOTE: No need to double check the instance state as pgsql_start already use
|
|
|
0c4b27 |
+ # pgsql_monitor to check the state before returning.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Starting "%s" as a standby failed (returned %d)',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE, $rc );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Promote the secondary instance to primary
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_promote {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my $cancel_switchover;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = pgsql_monitor();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $rc == $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ # Running as slave. Normal, expected behavior.
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_promote: "%s" currently running as a standby',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc == $OCF_RUNNING_MASTER ) {
|
|
|
0c4b27 |
+ # Already a master. Unexpected, but not a problem.
|
|
|
0c4b27 |
+ ocf_log( 'info', '"%s" already running as a primary',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ elsif ( $rc == $OCF_NOT_RUNNING ) { # INFO this is not supposed to happen.
|
|
|
0c4b27 |
+ # Currently not running. Need to start before promoting.
|
|
|
0c4b27 |
+ ocf_log( 'info', '"%s" currently not running, starting it',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = pgsql_start();
|
|
|
0c4b27 |
+ if ( $rc != $OCF_SUCCESS ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Failed to start the instance "%s"',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Unexpected error, cannot promote "%s"',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # At this point, the instance **MUST** be started as a secondary.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Cancel the switchover if it has been considered not safe during the
|
|
|
0c4b27 |
+ # pre-promote action
|
|
|
0c4b27 |
+ $cancel_switchover = _get_priv_attr('cancel_switchover');
|
|
|
0c4b27 |
+ if ( $cancel_switchover ) { # if not empty or not 0
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Switchover has been canceled from pre-promote action' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ _delete_priv_attr( 'cancel_switchover' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC if $cancel_switchover eq '1';
|
|
|
0c4b27 |
+ return $OCF_ERR_ARGS; # ban the resource from the node if we have an
|
|
|
0c4b27 |
+ # internal error during _check_switchover
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Do not check for a better candidate if we try to recover the master
|
|
|
0c4b27 |
+ # Recover of a master is detected during the pre-promote action. It sets the
|
|
|
0c4b27 |
+ # private attribute 'recover_master' to '1' if this is a master recover.
|
|
|
0c4b27 |
+ if ( _get_priv_attr( 'recover_master' ) eq '1' ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Recovering old master, no election needed');
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ else {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # The promotion is occurring on the best known candidate (highest
|
|
|
0c4b27 |
+ # master score), as chosen by pacemaker during the last working monitor
|
|
|
0c4b27 |
+ # on previous master (see pgsql_monitor/_check_locations subs).
|
|
|
0c4b27 |
+ # To avoid any race condition between the last monitor action on the
|
|
|
0c4b27 |
+ # previous master and the **real** most up-to-date standby, we
|
|
|
0c4b27 |
+ # set each standby location during the "pre-promote" action, and stored
|
|
|
0c4b27 |
+ # them using the "lsn_location" resource attribute.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # The best standby to promote would have the highest known LSN. If the
|
|
|
0c4b27 |
+ # current resource is not the best one, we need to modify the master
|
|
|
0c4b27 |
+ # scores accordingly, and abort the current promotion.
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_promote: checking if current node is the best candidate for promotion' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Exclude nodes that are known to be unavailable (not in the current
|
|
|
0c4b27 |
+ # partition) using the "crm_node" command
|
|
|
0c4b27 |
+ my @active_nodes = split /\s+/ => _get_priv_attr( 'nodes' );
|
|
|
0c4b27 |
+ my $node_to_promote = '';
|
|
|
0c4b27 |
+ my $ans;
|
|
|
0c4b27 |
+ my $max_tl;
|
|
|
0c4b27 |
+ my $max_lsn;
|
|
|
0c4b27 |
+ my $node_tl;
|
|
|
0c4b27 |
+ my $node_lsn;
|
|
|
0c4b27 |
+ my $wal_num;
|
|
|
0c4b27 |
+ my $wal_off;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Get the "lsn_location" attribute value for the current node, as set
|
|
|
0c4b27 |
+ # during the "pre-promote" action.
|
|
|
0c4b27 |
+ # It should be the greatest among the secondary instances.
|
|
|
0c4b27 |
+ $ans = _get_priv_attr( 'lsn_location' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ if ( $ans eq '' ) {
|
|
|
0c4b27 |
+ # This should not happen as the "lsn_location" attribute should have
|
|
|
0c4b27 |
+ # been updated during the "pre-promote" action.
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Can not get current node LSN location' );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ chomp $ans;
|
|
|
0c4b27 |
+ ( $max_tl, $max_lsn ) = split /#/, $ans;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', 'pgsql_promote: current node TL#LSN location: %s#%s',
|
|
|
0c4b27 |
+ $max_tl, $max_lsn );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Now we compare with the other available nodes.
|
|
|
0c4b27 |
+ foreach my $node ( @active_nodes ) {
|
|
|
0c4b27 |
+ # We exclude the current node from the check.
|
|
|
0c4b27 |
+ next if $node eq $nodename;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Get the "lsn_location" attribute value for the node, as set during
|
|
|
0c4b27 |
+ # the "pre-promote" action.
|
|
|
0c4b27 |
+ # This is implemented as a loop as private attributes are asynchronously
|
|
|
0c4b27 |
+ # available from other nodes.
|
|
|
0c4b27 |
+ # see: https://github.com/ClusterLabs/PAF/issues/131
|
|
|
0c4b27 |
+ # NOTE: if a node did not set its lsn_location for some reason, this will end
|
|
|
0c4b27 |
+ # with a timeout and the whole promotion will start again.
|
|
|
0c4b27 |
+ WAIT_FOR_LSN: {
|
|
|
0c4b27 |
+ $ans = _get_priv_attr( 'lsn_location', $node );
|
|
|
0c4b27 |
+ if ( $ans eq '' ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'pgsql_promote: waiting for LSN from %s', $node );
|
|
|
0c4b27 |
+ select( undef, undef, undef, 0.1 );
|
|
|
0c4b27 |
+ redo WAIT_FOR_LSN;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ chomp $ans;
|
|
|
0c4b27 |
+ ( $node_tl, $node_lsn ) = split /#/, $ans;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_promote: comparing with "%s": TL#LSN is %s#%s',
|
|
|
0c4b27 |
+ $node, $node_tl, $node_lsn );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If the node has a higher LSN, select it as a best candidate to
|
|
|
0c4b27 |
+ # promotion and keep looping to check the TL/LSN of other nodes.
|
|
|
0c4b27 |
+ if ( $node_tl > $max_tl
|
|
|
0c4b27 |
+ or ( $node_tl == $max_tl and $node_lsn > $max_lsn )
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ ocf_log( 'debug',
|
|
|
0c4b27 |
+ 'pgsql_promote: "%s" is a better candidate to promote (%s#%s > %s#%s)',
|
|
|
0c4b27 |
+ $node, $node_tl, $node_lsn, $max_tl, $max_lsn );
|
|
|
0c4b27 |
+ $node_to_promote = $node;
|
|
|
0c4b27 |
+ $max_tl = $node_tl;
|
|
|
0c4b27 |
+ $max_lsn = $node_lsn;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If any node has been selected, we adapt the master scores accordingly
|
|
|
0c4b27 |
+ # and break the current promotion.
|
|
|
0c4b27 |
+ if ( $node_to_promote ne '' ) {
|
|
|
0c4b27 |
+ ocf_exit_reason(
|
|
|
0c4b27 |
+ '%s is the best candidate to promote, aborting current promotion',
|
|
|
0c4b27 |
+ $node_to_promote );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Reset current node master score.
|
|
|
0c4b27 |
+ _set_master_score( '1' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Set promotion candidate master score.
|
|
|
0c4b27 |
+ _set_master_score( '1000', $node_to_promote );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # We fail the promotion to trigger another promotion transition
|
|
|
0c4b27 |
+ # with the new scores.
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Else, we will keep on promoting the current node.
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless (
|
|
|
0c4b27 |
+ # Promote the instance on the current node.
|
|
|
0c4b27 |
+ _runas( $PGCTL, '--pgdata', $pgdata, '-w', 'promote' ) == 0 )
|
|
|
0c4b27 |
+ {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'Error during promotion command' );
|
|
|
0c4b27 |
+ return $OCF_ERR_GENERIC;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # The instance promotion is asynchronous, so we need to wait for this
|
|
|
0c4b27 |
+ # process to complete.
|
|
|
0c4b27 |
+ while ( pgsql_monitor() != $OCF_RUNNING_MASTER ) {
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Waiting for the promote to complete' );
|
|
|
0c4b27 |
+ sleep 1;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Promote complete' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# This action is called **before** the actual promotion when a failing master is
|
|
|
0c4b27 |
+# considered unreclaimable, recoverable or a new master must be promoted
|
|
|
0c4b27 |
+# (switchover or first start).
|
|
|
0c4b27 |
+# As every "notify" action, it is executed almost simultaneously on all
|
|
|
0c4b27 |
+# available nodes.
|
|
|
0c4b27 |
+sub pgsql_notify_pre_promote {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my $node_tl;
|
|
|
0c4b27 |
+ my $node_lsn;
|
|
|
0c4b27 |
+ my %cdata;
|
|
|
0c4b27 |
+ my %active_nodes;
|
|
|
0c4b27 |
+ my $attr_nodes;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Promoting instance on node "%s"',
|
|
|
0c4b27 |
+ $OCF_NOTIFY_ENV{'promote'}[0]{'uname'} );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # No need to do an election between slaves if this is recovery of the master
|
|
|
0c4b27 |
+ if ( _is_master_recover( $OCF_NOTIFY_ENV{'promote'}[0]{'uname'} ) ) {
|
|
|
0c4b27 |
+ ocf_log( 'warning', 'This is a master recovery!' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ _set_priv_attr( 'recover_master', '1' )
|
|
|
0c4b27 |
+ if $OCF_NOTIFY_ENV{'promote'}[0]{'uname'} eq $nodename;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Environment cleanup!
|
|
|
0c4b27 |
+ _delete_priv_attr( 'lsn_location' );
|
|
|
0c4b27 |
+ _delete_priv_attr( 'recover_master' );
|
|
|
0c4b27 |
+ _delete_priv_attr( 'nodes' );
|
|
|
0c4b27 |
+ _delete_priv_attr( 'cancel_switchover' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # check for the last received entry of WAL from the master if we are
|
|
|
0c4b27 |
+ # the designated slave to promote
|
|
|
0c4b27 |
+ if ( _is_switchover( $nodename ) and scalar
|
|
|
0c4b27 |
+ grep { $_->{'uname'} eq $nodename } @{ $OCF_NOTIFY_ENV{'promote'} }
|
|
|
0c4b27 |
+ ) {
|
|
|
0c4b27 |
+ $rc = _check_switchover();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( $rc == 0 ) {
|
|
|
0c4b27 |
+ # Shortcut the election process as the switchover will be
|
|
|
0c4b27 |
+ # canceled
|
|
|
0c4b27 |
+ _set_priv_attr( 'cancel_switchover', $rc );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS; # return code is ignored during notify
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If the sub keeps going, that means the switchover is safe.
|
|
|
0c4b27 |
+ # Keep going with the election process in case the switchover was
|
|
|
0c4b27 |
+ # instruct to the wrong node.
|
|
|
0c4b27 |
+ # FIXME: should we allow a switchover to a lagging slave?
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # We need to trigger an election between existing slaves to promote the best
|
|
|
0c4b27 |
+ # one based on its current LSN location. Each node set a private attribute
|
|
|
0c4b27 |
+ # "lsn_location" with its TL and LSN location.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # During the following promote action, The designated standby for
|
|
|
0c4b27 |
+ # promotion use these attributes to check if the instance to be promoted
|
|
|
0c4b27 |
+ # is the best one, so we can avoid a race condition between the last
|
|
|
0c4b27 |
+ # successful monitor on the previous master and the current promotion.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # As we can not break the transition from a notification action, we check
|
|
|
0c4b27 |
+ # during the promotion if each node TL and LSN are valid.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Force a checpoint to make sure the controldata shows the very last TL
|
|
|
0c4b27 |
+ _query( q{ CHECKPOINT }, {} );
|
|
|
0c4b27 |
+ %cdata = _get_controldata();
|
|
|
0c4b27 |
+ $node_lsn = _get_last_received_lsn( 'in decimal' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ unless ( defined $node_lsn ) {
|
|
|
0c4b27 |
+ ocf_log( 'warning', 'Unknown current node LSN' );
|
|
|
0c4b27 |
+ # Return code are ignored during notifications...
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $node_lsn = "$cdata{'tl'}#$node_lsn";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Current node TL#LSN: %s', $node_lsn );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Set the "lsn_location" attribute value for this node so we can use it
|
|
|
0c4b27 |
+ # during the following "promote" action.
|
|
|
0c4b27 |
+ _set_priv_attr( 'lsn_location', $node_lsn );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'warning', 'Could not set the current node LSN' )
|
|
|
0c4b27 |
+ if $? != 0 ;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # If this node is the future master, keep track of the slaves that
|
|
|
0c4b27 |
+ # received the same notification to compare our LSN with them during
|
|
|
0c4b27 |
+ # promotion
|
|
|
0c4b27 |
+ if ( $OCF_NOTIFY_ENV{'promote'}[0]{'uname'} eq $nodename ) {
|
|
|
0c4b27 |
+ # Build the list of active nodes:
|
|
|
0c4b27 |
+ # master + slave + start - stop
|
|
|
0c4b27 |
+ # FIXME: Deal with rsc started during the same transaction but **after**
|
|
|
0c4b27 |
+ # the promotion ?
|
|
|
0c4b27 |
+ $active_nodes{ $_->{'uname'} }++ foreach @{ $OCF_NOTIFY_ENV{'active'} },
|
|
|
0c4b27 |
+ @{ $OCF_NOTIFY_ENV{'start'} };
|
|
|
0c4b27 |
+ $active_nodes{ $_->{'uname'} }-- foreach @{ $OCF_NOTIFY_ENV{'stop'} };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $attr_nodes = join " "
|
|
|
0c4b27 |
+ => grep { $active_nodes{$_} > 0 } keys %active_nodes;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ _set_priv_attr( 'nodes', $attr_nodes );
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# This action is called after a promote action.
|
|
|
0c4b27 |
+sub pgsql_notify_post_promote {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # We have a new master (or the previous one recovered).
|
|
|
0c4b27 |
+ # Environment cleanup!
|
|
|
0c4b27 |
+ _delete_priv_attr( 'lsn_location' );
|
|
|
0c4b27 |
+ _delete_priv_attr( 'recover_master' );
|
|
|
0c4b27 |
+ _delete_priv_attr( 'nodes' );
|
|
|
0c4b27 |
+ _delete_priv_attr( 'cancel_switchover' );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# This is called before a demote occurs.
|
|
|
0c4b27 |
+sub pgsql_notify_pre_demote {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my %cdata;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # do nothing if the local node will not be demoted
|
|
|
0c4b27 |
+ return $OCF_SUCCESS unless scalar
|
|
|
0c4b27 |
+ grep { $_->{'uname'} eq $nodename } @{ $OCF_NOTIFY_ENV{'demote'} };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = pgsql_monitor();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # do nothing if this is not a master recovery
|
|
|
0c4b27 |
+ return $OCF_SUCCESS unless _is_master_recover( $nodename )
|
|
|
0c4b27 |
+ and $rc == $OCF_FAILED_MASTER;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # in case of master crash, we need to detect if the CRM tries to recover
|
|
|
0c4b27 |
+ # the master clone. The usual transition is to do:
|
|
|
0c4b27 |
+ # demote->stop->start->promote
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # There are multiple flaws with this transition:
|
|
|
0c4b27 |
+ # * the 1st and 2nd actions will fail because the instance is in
|
|
|
0c4b27 |
+ # OCF_FAILED_MASTER step
|
|
|
0c4b27 |
+ # * the usual start action is dangerous as the instance will start with
|
|
|
0c4b27 |
+ # a recovery.conf instead of entering a normal recovery process
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # To avoid this, we try to start the instance in recovery from here.
|
|
|
0c4b27 |
+ # If it success, at least it will be demoted correctly with a normal
|
|
|
0c4b27 |
+ # status. If it fails, it will be catched up in next steps.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Trying to start failing master "%s"...',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Either the instance managed to start or it couldn't.
|
|
|
0c4b27 |
+ # We rely on the pg_ctk '-w' switch to take care of this. If it couldn't
|
|
|
0c4b27 |
+ # start, this error will be catched up later during the various checks
|
|
|
0c4b27 |
+ _pg_ctl_start();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ %cdata = _get_controldata();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'State is "%s" after recovery attempt', $cdata{'state'} );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# This is called before a stop occurs.
|
|
|
0c4b27 |
+sub pgsql_notify_pre_stop {
|
|
|
0c4b27 |
+ my $rc;
|
|
|
0c4b27 |
+ my %cdata;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # do nothing if the local node will not be stopped
|
|
|
0c4b27 |
+ return $OCF_SUCCESS unless scalar
|
|
|
0c4b27 |
+ grep { $_->{'uname'} eq $nodename } @{ $OCF_NOTIFY_ENV{'stop'} };
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $rc = _controldata_to_ocf();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # do nothing if this is not a slave recovery
|
|
|
0c4b27 |
+ return $OCF_SUCCESS unless _is_slave_recover( $nodename )
|
|
|
0c4b27 |
+ and $rc == $OCF_RUNNING_SLAVE;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # in case of slave crash, we need to detect if the CRM tries to recover
|
|
|
0c4b27 |
+ # the slaveclone. The usual transition is to do: stop->start
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # This transition can no twork because the instance is in
|
|
|
0c4b27 |
+ # OCF_ERR_GENERIC step. So the stop action will fail, leading most
|
|
|
0c4b27 |
+ # probably to fencing action.
|
|
|
0c4b27 |
+ #
|
|
|
0c4b27 |
+ # To avoid this, we try to start the instance in recovery from here.
|
|
|
0c4b27 |
+ # If it success, at least it will be stopped correctly with a normal
|
|
|
0c4b27 |
+ # status. If it fails, it will be catched up in next steps.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Trying to start failing slave "%s"...',
|
|
|
0c4b27 |
+ $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # Either the instance managed to start or it couldn't.
|
|
|
0c4b27 |
+ # We rely on the pg_ctk '-w' switch to take care of this. If it couldn't
|
|
|
0c4b27 |
+ # start, this error will be catched up later during the various checks
|
|
|
0c4b27 |
+ _pg_ctl_start();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ %cdata = _get_controldata();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'info', 'State is "%s" after recovery attempt', $cdata{'state'} );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Notify type actions, called on all available nodes before (pre) and after
|
|
|
0c4b27 |
+# (post) other actions, like promote, start, ...
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_notify {
|
|
|
0c4b27 |
+ my $type_op;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ocf_log( 'debug', "pgsql_notify: environment variables: %s",
|
|
|
0c4b27 |
+ Data::Dumper->new( [ \%OCF_NOTIFY_ENV ] )->Sortkeys(1)->Terse(1)->Dump );
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return unless %OCF_NOTIFY_ENV;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ $type_op = "$OCF_NOTIFY_ENV{'type'}-$OCF_NOTIFY_ENV{'operation'}";
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ for ( $type_op ) {
|
|
|
0c4b27 |
+ if ( /^pre-promote$/ ) { return pgsql_notify_pre_promote() }
|
|
|
0c4b27 |
+ elsif ( /^post-promote$/ ) { return pgsql_notify_post_promote() }
|
|
|
0c4b27 |
+ elsif ( /^pre-demote$/ ) { return pgsql_notify_pre_demote() }
|
|
|
0c4b27 |
+ elsif ( /^pre-stop$/ ) { return pgsql_notify_pre_stop() }
|
|
|
0c4b27 |
+ }
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Action used to allow for online modification of resource parameters value.
|
|
|
0c4b27 |
+#
|
|
|
0c4b27 |
+sub pgsql_reload {
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ # No action necessary, the action declaration is enough to inform pacemaker
|
|
|
0c4b27 |
+ # that the modification of any non-unique parameter can be applied without
|
|
|
0c4b27 |
+ # having to restart the resource.
|
|
|
0c4b27 |
+ ocf_log( 'info', 'Instance "%s" reloaded', $OCF_RESOURCE_INSTANCE );
|
|
|
0c4b27 |
+ return $OCF_SUCCESS;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+############################################################
|
|
|
0c4b27 |
+#### MAIN
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+exit ocf_meta_data() if $__OCF_ACTION eq 'meta-data';
|
|
|
0c4b27 |
+exit ocf_methods() if $__OCF_ACTION eq 'methods';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Avoid "could not change directory" when executing commands as "system-user".
|
|
|
0c4b27 |
+chdir File::Spec->tmpdir();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# mandatory sanity checks
|
|
|
0c4b27 |
+# check pgdata
|
|
|
0c4b27 |
+if ( ! -d $pgdata ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'PGDATA "%s" does not exist', $pgdata );
|
|
|
0c4b27 |
+ exit $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# check datadir
|
|
|
0c4b27 |
+if ( ! -d $datadir ) {
|
|
|
0c4b27 |
+ ocf_exit_reason( 'data_directory "%s" does not exist', $datadir );
|
|
|
0c4b27 |
+ exit $OCF_ERR_ARGS;
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Set PostgreSQL version
|
|
|
0c4b27 |
+$PGVERNUM = _get_pg_version();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Set current node name.
|
|
|
0c4b27 |
+$nodename = ocf_local_nodename();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+$exit_code = pgsql_validate_all();
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+exit $exit_code if $exit_code != $OCF_SUCCESS or $__OCF_ACTION eq 'validate-all';
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+# Run action
|
|
|
0c4b27 |
+for ( $__OCF_ACTION ) {
|
|
|
0c4b27 |
+ if ( /^start$/ ) { $exit_code = pgsql_start() }
|
|
|
0c4b27 |
+ elsif ( /^stop$/ ) { $exit_code = pgsql_stop() }
|
|
|
0c4b27 |
+ elsif ( /^monitor$/ ) { $exit_code = pgsql_monitor() }
|
|
|
0c4b27 |
+ elsif ( /^promote$/ ) { $exit_code = pgsql_promote() }
|
|
|
0c4b27 |
+ elsif ( /^demote$/ ) { $exit_code = pgsql_demote() }
|
|
|
0c4b27 |
+ elsif ( /^notify$/ ) { $exit_code = pgsql_notify() }
|
|
|
0c4b27 |
+ elsif ( /^reload$/ ) { $exit_code = pgsql_reload() }
|
|
|
0c4b27 |
+ else { $exit_code = $OCF_ERR_UNIMPLEMENTED }
|
|
|
0c4b27 |
+}
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+exit $exit_code;
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 EXAMPLE CRM SHELL
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The following is an example configuration for a pgsqlms resource using the
|
|
|
0c4b27 |
+crm(8) shell:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ primitive pgsqld pgsqlms \
|
|
|
0c4b27 |
+ params pgdata="/var/lib/postgresql/9.6/main" \
|
|
|
0c4b27 |
+ bindir="/usr/lib/postgresql/9.6/bin" \
|
|
|
0c4b27 |
+ pghost="/var/run/postgresql" \
|
|
|
0c4b27 |
+ recovery_template="/etc/postgresql/9.6/main/recovery.conf.pcmk" \
|
|
|
0c4b27 |
+ start_opts="-c config_file=/etc/postgresql/9.6/main/postgresql.conf" \
|
|
|
0c4b27 |
+ op start timeout=60s \
|
|
|
0c4b27 |
+ op stop timeout=60s \
|
|
|
0c4b27 |
+ op promote timeout=30s \
|
|
|
0c4b27 |
+ op demote timeout=120s \
|
|
|
0c4b27 |
+ op monitor interval=15s timeout=10s role="Master" \
|
|
|
0c4b27 |
+ op monitor interval=16s timeout=10s role="Slave" \
|
|
|
0c4b27 |
+ op notify timeout=60s
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ ms pgsql-ha pgsqld meta notify=true
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 EXAMPLE PCS
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+The following is an example configuration for a pgsqlms resource using pcs(8):
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ pcs resource create pgsqld ocf:heartbeat:pgsqlms \
|
|
|
0c4b27 |
+ bindir=/usr/pgsql-9.6/bin pgdata=/var/lib/pgsql/9.6/data \
|
|
|
0c4b27 |
+ op start timeout=60s \
|
|
|
0c4b27 |
+ op stop timeout=60s \
|
|
|
0c4b27 |
+ op promote timeout=30s \
|
|
|
0c4b27 |
+ op demote timeout=120s \
|
|
|
0c4b27 |
+ op monitor interval=15s timeout=10s role="Master" \
|
|
|
0c4b27 |
+ op monitor interval=16s timeout=10s role="Slave" \
|
|
|
0c4b27 |
+ op notify timeout=60s --master notify=true
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 SEE ALSO
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+http://clusterlabs.org/
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=head1 AUTHOR
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Jehan-Guillaume de Rorthais and Mael Rimbault.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+=cut
|
|
|
0c4b27 |
diff --color -uNr a/paf_LICENSE b/paf_LICENSE
|
|
|
0c4b27 |
--- a/paf_LICENSE 1970-01-01 01:00:00.000000000 +0100
|
|
|
0c4b27 |
+++ b/paf_LICENSE 2021-04-14 09:16:39.083555835 +0200
|
|
|
0c4b27 |
@@ -0,0 +1,19 @@
|
|
|
0c4b27 |
+Copyright (c) 2016-2020, Jehan-Guillaume de Rorthais, Mael Rimbault.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Permission to use, copy, modify, and distribute this software and its
|
|
|
0c4b27 |
+documentation for any purpose, without fee, and without a written agreement
|
|
|
0c4b27 |
+is hereby granted, provided that the above copyright notice and this
|
|
|
0c4b27 |
+paragraph and the following two paragraphs appear in all copies.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+IN NO EVENT SHALL THE AUTHOR OR DISTRIBUTORS BE LIABLE TO ANY PARTY FOR
|
|
|
0c4b27 |
+DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, INCLUDING
|
|
|
0c4b27 |
+LOST PROFITS, ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS
|
|
|
0c4b27 |
+DOCUMENTATION, EVEN IF THE AUTHOR OR DISTRIBUTORS HAVE BEEN ADVISED OF THE
|
|
|
0c4b27 |
+POSSIBILITY OF SUCH DAMAGE.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+THE AUTHOR AND DISTRIBUTORS SPECIFICALLY DISCLAIMS ANY WARRANTIES,
|
|
|
0c4b27 |
+INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
|
|
|
0c4b27 |
+AND FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
|
|
|
0c4b27 |
+ON AN "AS IS" BASIS, AND THE AUTHOR AND DISTRIBUTORS HAS NO OBLIGATIONS TO
|
|
|
0c4b27 |
+PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
diff --color -uNr a/paf_README.md b/paf_README.md
|
|
|
0c4b27 |
--- a/paf_README.md 1970-01-01 01:00:00.000000000 +0100
|
|
|
0c4b27 |
+++ b/paf_README.md 2021-04-14 09:18:57.450968048 +0200
|
|
|
0c4b27 |
@@ -0,0 +1,86 @@
|
|
|
0c4b27 |
+# PostgreSQL Automatic Failover
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+High-Availibility for Postgres, based on industry references Pacemaker and
|
|
|
0c4b27 |
+Corosync.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+## Description
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Pacemaker is nowadays the industry reference for High Availability. In the same
|
|
|
0c4b27 |
+fashion than for Systemd, all Linux distributions moved (or are moving) to this
|
|
|
0c4b27 |
+unique Pacemaker+Corosync stack, removing all other existing high availability
|
|
|
0c4b27 |
+stacks (CMAN, RGManager, OpenAIS, ...). It is able to detect failure on various
|
|
|
0c4b27 |
+services and automatically decide to failover the failing resource to another
|
|
|
0c4b27 |
+node when possible.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+To be able to manage a specific service resource, Pacemaker interact with it
|
|
|
0c4b27 |
+through a so-called "Resource Agent". Resource agents must comply to the OCF
|
|
|
0c4b27 |
+specification which define what they must implement (start, stop, promote,
|
|
|
0c4b27 |
+etc), how they should behave and inform Pacemaker of their results.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+PostgreSQL Automatic Failover is a new OCF resource Agent dedicated to
|
|
|
0c4b27 |
+PostgreSQL. Its original wish is to keep a clear limit between the Pacemaker
|
|
|
0c4b27 |
+administration and the PostgreSQL one, to keep things simple, documented and
|
|
|
0c4b27 |
+yet powerful.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+Once your PostgreSQL cluster built using internal streaming replication, PAF is
|
|
|
0c4b27 |
+able to expose to Pacemaker what is the current status of the PostgreSQL
|
|
|
0c4b27 |
+instance on each node: master, slave, stopped, catching up, etc. Should a
|
|
|
0c4b27 |
+failure occurs on the master, Pacemaker will try to recover it by default.
|
|
|
0c4b27 |
+Should the failure be non-recoverable, PAF allows the slaves to be able to
|
|
|
0c4b27 |
+elect the best of them (the closest one to the old master) and promote it as
|
|
|
0c4b27 |
+the new master. All of this thanks to the robust, feature-full and most
|
|
|
0c4b27 |
+importantly experienced project: Pacemaker.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+For information about how to install this agent, see `INSTALL.md`.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+## Setup and requirements
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+PAF supports PostgreSQL 9.3 and higher. It has been extensively tested under
|
|
|
0c4b27 |
+CentOS 6 and 7 in various scenario.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+PAF has been written to give to the administrator the maximum control
|
|
|
0c4b27 |
+over their PostgreSQL configuration and architecture. Thus, you are 100%
|
|
|
0c4b27 |
+responsible for the master/slave creations and their setup. The agent
|
|
|
0c4b27 |
+will NOT edit your setup. It only requires you to follow these pre-requisites:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ * slave __must__ be in hot_standby (accept read-only connections) ;
|
|
|
0c4b27 |
+ * the following parameters __must__ be configured in the appropriate place :
|
|
|
0c4b27 |
+ * `standby_mode = on` (for PostgreSQL 11 and before)
|
|
|
0c4b27 |
+ * `recovery_target_timeline = 'latest'`
|
|
|
0c4b27 |
+ * `primary_conninfo` wih `application_name` set to the node name as seen
|
|
|
0c4b27 |
+ in Pacemaker.
|
|
|
0c4b27 |
+ * these last parameters has been merged inside the instance configuration
|
|
|
0c4b27 |
+ file with PostgreSQL 12. For PostgreSQL 11 and before, you __must__
|
|
|
0c4b27 |
+ provide a `recovery.conf` template file.
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+When setting up the resource in Pacemaker, here are the available parameters you
|
|
|
0c4b27 |
+can set:
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+ * `bindir`: location of the PostgreSQL binaries (default: `/usr/bin`)
|
|
|
0c4b27 |
+ * `pgdata`: location of the PGDATA of your instance (default:
|
|
|
0c4b27 |
+ `/var/lib/pgsql/data`)
|
|
|
0c4b27 |
+ * `datadir`: path to the directory set in `data_directory` from your
|
|
|
0c4b27 |
+ postgresql.conf file. This parameter has same default than PostgreSQL
|
|
|
0c4b27 |
+ itself: the `pgdata` parameter value. Unless you have a special PostgreSQL
|
|
|
0c4b27 |
+ setup and you understand this parameter, __ignore it__
|
|
|
0c4b27 |
+ * `pghost`: the socket directory or IP address to use to connect to the
|
|
|
0c4b27 |
+ local instance (default: `/tmp` or `/var/run/postgresql` for DEBIAN)
|
|
|
0c4b27 |
+ * `pgport`: the port to connect to the local instance (default: `5432`)
|
|
|
0c4b27 |
+ * `recovery_template`: __only__ for PostgreSQL 11 and before. The local
|
|
|
0c4b27 |
+ template that will be copied as the `PGDATA/recovery.conf` file. This
|
|
|
0c4b27 |
+ file must not exist on any node for PostgreSQL 12 and after.
|
|
|
0c4b27 |
+ (default: `$PGDATA/recovery.conf.pcmk`)
|
|
|
0c4b27 |
+ * `start_opts`: Additional arguments given to the postgres process on startup.
|
|
|
0c4b27 |
+ See "postgres --help" for available options. Useful when the postgresql.conf
|
|
|
0c4b27 |
+ file is not in the data directory (PGDATA), eg.:
|
|
|
0c4b27 |
+ `-c config_file=/etc/postgresql/9.3/main/postgresql.conf`
|
|
|
0c4b27 |
+ * `system_user`: the system owner of your instance's process (default:
|
|
|
0c4b27 |
+ `postgres`)
|
|
|
0c4b27 |
+ * `maxlag`: maximum lag allowed on a standby before we set a negative master
|
|
|
0c4b27 |
+ score on it. The calculation is based on the difference between the current
|
|
|
0c4b27 |
+ xlog location on the master and the write location on the standby.
|
|
|
0c4b27 |
+ (default: 0, which disables this feature)
|
|
|
0c4b27 |
+
|
|
|
0c4b27 |
+For a demonstration about how to setup a cluster, see
|
|
|
0c4b27 |
+[http://clusterlabs.github.io/PAF/documentation.html](http://clusterlabs.github.io/PAF/documentation.html).
|
|
|
0c4b27 |
+
|