# Script: pgbouncer_follower.pl
# Copyright: 22/04/2012: v1.0.1 Glyn Astill <glyn@8kb.co.uk>
-# Requires: Perl 5.10.1+, PostgreSQL 9.0+ Slony-I 2.0+
+# Requires: Perl 5.10.1+, PostgreSQL 9.0+ Slony-I 2.0+ OR Streaming Replication
#
# This script is a command-line utility to monitor Slony-I clusters
# and reconfigure pgbouncer to follow replication sets.
use constant true => 1;
my $g_usage = 'Pass configuration file: pool_follower.pl -f <configuration_path> [-D] ';
+my $g_method = "slony";
my $g_debug = false;
my $g_pidfile = "/tmp/pgbouncer_follower_%mode.pid";
my $g_logfile = "/tmp/pgbouncer_follower_%mode.log";
if ($g_debug) {
printLogLn($g_logfile, "DEBUG: Logging to my '$g_logfile'");
- printLogLn($g_logfile, "\t Watching sets $g_clsets in Slony-I cluster '$g_clname' polling every ${g_poll_interval}ms");
- printLogLn($g_logfile, "\t Following " . ($g_all_databases ? "all databases" : "replicated database only") . " on an '$g_mode' node for the above replicated sets");
+ if ($g_method eq 'slony') {
+ printLogLn($g_logfile, "\t Watching sets $g_clsets in Slony-I cluster '$g_clname' polling every ${g_poll_interval}ms");
+ printLogLn($g_logfile, "\t Following " . ($g_all_databases ? "all databases" : "replicated database only") . " on an '$g_mode' node for the above replicated sets");
+ }
+ elsif ($g_method eq 'wal') {
+ printLogLn($g_logfile, "\t Watching streaming replication lag polling every ${g_poll_interval}ms");
+ }
+ else {
+ printLogLn($g_logfile, "\t ERROR: Unknown replication method: '$g_method'");
+ exit(1);
+ }
printLogLn($g_logfile, "\t Template config '$g_conf_template' Target config '$g_conf_target'");
printLogLn($g_logfile, "\t Reload command is '$g_reload_command'");
printLogLn($g_logfile, "\t Status stored in '$g_status_file'");
foreach my $conninfo (@g_conninfos) {
$conninfo_read++;
eval {
- @g_cluster = loadCluster($g_clname, $conninfo, $g_user, $g_pass, $g_addr, $g_clsets);
+ if ($g_method eq 'slony') {
+ @g_cluster = loadCluster($g_clname, $conninfo, $g_user, $g_pass, $g_addr, $g_clsets);
+ }
+ elsif ($g_method eq 'wal') {
+ @g_cluster = loadBinCluster($g_user, $g_pass);
+ }
if ($g_debug) {
- printLogLn($g_logfile, "DEBUG: ($conninfo_read) Cluster with " . scalar(@g_cluster) . " nodes read from conninfo: $conninfo");
+ printLogLn($g_logfile, "DEBUG: ($conninfo_read) " . (($g_method eq 'slony')?'Slony':'Streaming replication') . " cluster with " . scalar(@g_cluster) . " nodes " . (($g_method eq 'slony')?"read from conninfo: $conninfo":"provided via config conninfos"));
foreach (@g_cluster) {
printLogLn($g_logfile, "DEBUG: Node #" . @$_[0] . " DETAIL: " . @$_[1] . " " . @$_[2] . " " . (@$_[3] // "<NONE>") . " " . @$_[4] . " " . (@$_[5] // "<NONE>") . " " . @$_[6] . " " . @$_[7] . " " . @$_[8] . " " . @$_[9] . " " . @$_[10] . " " . @$_[11]);
}
my $target_auth = "";
if ($g_debug) {
- printLogLn($g_logfile, "DEBUG: All databases = " . ($g_all_databases ? 'true' : 'false'));
+ printLogLn($g_logfile, "DEBUG: All databases = " . ($all_databases ? 'true' : 'false'));
}
if ($g_auth_user ne "") {
if ($g_debug) {
printLogLn ($g_logfile, "DEBUG: Configuration for " . ($target_is_origin ? "origin" : "subscriber") . " of sets $target_sets node #$target_node_id $target_host:$target_port");
}
- if ($all_databases) {
+ if ($all_databases || $target_db eq '*') {
$_ =~ s/(\[databases\])/$1\n\* = host=$target_host port=$target_port$target_auth/;
}
else {
return @cluster;
}
+sub loadBinCluster {
+ my $dbuser = shift;
+ my $dbpass = shift;
+
+ my $dsn;
+ my $dbh;
+ my $sth;
+ my $query;
+ my $recovery;
+ my $xlog_location;
+ my $apply_lag;
+ my $primaries = 0;
+
+ my @parts;
+ my $timeline;
+ my $location;
+ my $primary_location;
+
+ my $hostname;
+ my $port;
+ my $database;
+ my @tmp_cluster;
+ my @cluster;
+ my $node_id = 1;
+ $g_best_config = true;
+
+ foreach my $conninfo (@g_conninfos) {
+ $dsn = "DBI:Pg:$conninfo};";
+
+ eval {
+ $dbh = DBI->connect($dsn, $dbuser, $dbpass, {RaiseError => 1});
+ # Check to see if the server is a secondary, and also pull the current xlog
+ # location and the apply lag using pg_last_xact_replay_timestamp to get the
+ # last commit timestamp from the primary applied on the secondary.
+ # We will need to compare the current receive location to the primary xlog
+ # location, if they differ we can then use the apply_lag value; we'll have
+ # to post-process this.
+ # In 9.6+ we might want to pull the system identifier from pg_controldata view too
+ $query = "SELECT pg_is_in_recovery(),
+ CASE
+ WHEN pg_is_in_recovery() THEN pg_last_xlog_receive_location()
+ ELSE pg_current_xlog_location()
+ END,
+ COALESCE(extract(epoch from current_timestamp-pg_last_xact_replay_timestamp()),0)";
+ $sth = $dbh->prepare($query);
+ $sth->execute();
+ ($recovery, $xlog_location, $apply_lag) = $sth->fetchrow;
+ $sth->finish;
+
+ ($port) = $conninfo =~ m/port=(.+?)(\s|$)/g;
+ ($hostname) = $conninfo =~ m/host=(.+?)(\s|$)/g;
+ ($database) = $conninfo =~ m/dbname=(.+?)(\s|$)/g;
+ @parts = split('/', $xlog_location, 2);
+ $timeline = qtrim(trim($parts[0]));
+ $location = hex(qtrim(trim($parts[1])));
+
+ if ($g_debug) {
+ printLogLn($g_logfile, "DEBUG: Server: $hostname:$port " . ($recovery ? 'secondary' : 'primary') . " at $xlog_location ($timeline/$location)");
+ }
+
+ # For WAL replication we assume if we can contact a server then it is active,
+ # which isn't strictly true, but nodes that have fallen behind can be excluded
+ # via the max_ro_lag setting. We also substitute timeline+1 for the slony
+ # replication set.
+ if (!$recovery) {
+ $primaries++;
+ $primary_location = $xlog_location;
+ my @node=(1,'Primary',0,$timeline,"ACTIVE",($timeline+1),$conninfo,$database,$hostname,$port,1,$xlog_location,$apply_lag);
+ push(@tmp_cluster, \@node);
+ }
+ else {
+ $node_id++;
+ my @node=($node_id,"Secondary".($node_id-1),1,undef,"ACTIVE",($timeline+1),$conninfo,$database,$hostname,$port,$node_id,$xlog_location,$apply_lag);
+ push(@tmp_cluster, \@node);
+ }
+ };
+ if ($@) {
+ printLogLn($g_logfile, "ERROR: Could not connect to server with conninfo: $conninfo DETAIL: $@");
+ }
+ }
+
+ # Error if more than one primary discovered.
+ if ($primaries != 1) {
+ printLogLn($g_logfile, "ERROR: Invalid quantity of primaries: $primaries");
+ die "no primaries found";
+ }
+ # Do the post processing we mentioned above, once we know xlog locations differ
+ # then we can say the apply lag is correct, else it's just there has been no
+ # activity on the master.
+ else {
+ foreach (@tmp_cluster) {
+ $apply_lag = ((@$_[12] > 0 && @$_[11] != $primary_location) ? @$_[12] : 0);
+ my @node=(@$_[0],@$_[1],@$_[2],@$_[3],@$_[4],@$_[5],@$_[6],@$_[7],@$_[8],@$_[9],@$_[10],$apply_lag);
+ push(@cluster, \@node);
+ }
+ }
+
+ return @cluster;
+}
+
sub getConfig {
my @fields;
my $success = false;
@fields = split('=', $_, 2);
$value = qtrim(trim($fields[1]));
given(lc($fields[0])) {
+ when(/\breplication_method\b/i) {
+ $g_method = $value;
+ }
when(/\bdebug\b/i) {
$g_debug = checkBoolean($value);
}
when(/\blog_file\b/i) {
$g_logfile = $value;
}
- when(/\bslony_user\b/i) {
+ when(/\bslony_user\b/i) { # Depreciated
+ $g_user = $value;
+ }
+ when(/\bslony_pass\b/i) { # Depreciated
+ $g_pass = $value;
+ }
+ when(/\breplication_user\b/i) {
$g_user = $value;
}
- when(/\bslony_pass\b/i) {
+ when(/\breplication_pass\b/i) {
$g_pass = $value;
}
when(/\bslony_cluster_name\b/i) {
projects / slony-i / pgbouncer_follower / blobdiff