pandorafms/pandora_server/util/pandora_ha.pl

423 lines
13 KiB
Perl
Executable File

#!/usr/bin/perl
###############################################################################
# Pandora FMS Daemon Watchdog
###############################################################################
# Copyright (c) 2018-2021 Artica Soluciones Tecnologicas S.L
###############################################################################
use strict;
use warnings;
use DBI;
use Getopt::Std;
use POSIX qw(setsid strftime :sys_wait_h);
use threads;
use threads::shared;
use File::Path qw(rmtree);
# Default lib dir for Pandora FMS RPM and DEB packages.
use lib '/usr/lib/perl5';
use PandoraFMS::Tools;
use PandoraFMS::DB;
use PandoraFMS::Core;
use PandoraFMS::Config;
use Data::Dumper;
$Data::Dumper::Sortkeys = 1;
# Pandora server configuration.
my %Conf;
# Command line options.
my %Opts;
# Run as a daemon.
my $DAEMON = 0;
# Avoid retry old processing orders.
my $First_Cleanup = 1;
# PID file.
my $PID_FILE = '/var/run/pandora_ha.pid';
# Server service handler.
my $Pandora_Service;
# Controlled exit
my $Running = 0;
########################################################################
# Print the given message with a preceding timestamp.
########################################################################
sub log_message($$$;$) {
my ($conf, $source, $message, $verbosity_level) = @_;
my $level = $verbosity_level;
$level = 5 unless defined($level);
if (ref($conf) eq "HASH") {
logger($conf, 'HA (' . $source . ') ' . "$message", $level);
}
if ($source eq '') {
print $message;
}
else {
print strftime("%H:%M:%S", localtime()) . ' [' . $source . '] ' . "$message\n";
}
}
########################################################################
# Run as a daemon in the background.
########################################################################
sub ha_daemonize($) {
my ($pa_config) = @_;
$PID_FILE = $pa_config->{'ha_pid_file'} if defined($pa_config->{'ha_pid_file'});
open STDIN, "$DEVNULL" or die "Can't read $DEVNULL: $!";
open STDOUT, ">>$DEVNULL" or die "Can't write to $DEVNULL: $!";
open STDERR, ">>$DEVNULL" or die "Can't write to $DEVNULL: $!";
chdir '/tmp' or die "Can't chdir to /tmp: $!";
# Fork!
defined(my $pid = fork) or die "Can't fork: $!";
if ($pid) {
# Store PID of this process in file presented by config token
if ($PID_FILE ne "") {
if ( -e $PID_FILE && open (FILE, $PID_FILE)) {
$pid = <FILE> + 0;
close FILE;
# Check if pandora_ha is running.
die "[ERROR] pandora_ha is already running with pid: $pid." if (kill (0, $pid));
}
umask 0022;
open (FILE, '>', $PID_FILE) or die "[FATAL] $!";
print FILE $pid;
close (FILE);
}
exit;
}
setsid or die "Can't start a new session: $!";
}
########################################################################
# Check command line parameters.
########################################################################
sub ha_init_pandora($) {
my $conf = shift;
log_message($conf, '', "\nPandora FMS Daemon Watchdog " . $PandoraFMS::Tools::VERSION . " Copyright (c) Artica ST\n");
getopts('dp:', \%Opts);
# Run as a daemon.
$DAEMON = 1 if (defined($Opts{'d'}));
# PID file.
$PID_FILE = $Opts{'p'} if (defined($Opts{'p'}));
# Load config file from command line.
help_screen () if ($#ARGV != 0);
$conf->{'_pandora_path'} = $ARGV[0];
}
########################################################################
# Read external configuration file.
########################################################################
sub ha_load_pandora_conf($) {
my $conf = shift;
# Set some defaults.
$conf->{"servername"} = `hostname`;
chomp($conf->{"servername"});
$conf->{"ha_file"} = '/etc/pandora/pandora_ha.bin' unless defined $conf->{"ha_file"};
pandora_init($conf, 'Pandora HA');
pandora_load_config ($conf);
# Check conf tokens.
foreach my $param ('dbuser', 'dbpass', 'dbname', 'dbhost', 'log_file') {
die ("[ERROR] Bad config values. Make sure " . $conf->{'_pandora_path'} . " is a valid config file.\n\n") unless defined ($conf->{$param});
}
$conf->{'dbengine'} = 'mysql' unless defined ($conf->{'dbengine'});
$conf->{'dbport'} = '3306' unless defined ($conf->{'dbport'});
$conf->{'ha_interval'} = 10 unless defined ($conf->{'ha_interval'});
$conf->{'ha_monitoring_interval'} = 60 unless defined ($conf->{'ha_monitoring_interval'});
}
##############################################################################
# Print a help screen and exit.
##############################################################################
sub help_screen {
log_message(undef, '', "Usage: $0 [options] <path to pandora_server.conf>\n\nOptions:\n\t-p <PID file> Write the PID of the process to the specified file.\n\t-d Run in the background.\n\n");
exit 1;
}
##############################################################################
# Keep server running
##############################################################################
sub ha_keep_pandora_running($$) {
my ($conf, $dbh) = @_;
my $OSNAME = $^O;
my $control_command;
$conf->{'pandora_service_cmd'} = 'service pandora_server' unless defined($conf->{'pandora_service_cmd'});
# Check if all servers are running
# Restart if crashed or keep interval is over.
my $component_last_contact = get_db_value(
$dbh,
'SELECT count(*) AS "delayed"
FROM tserver
WHERE ((status = -1) OR ( (unix_timestamp() - unix_timestamp(keepalive)) > (server_keepalive+1) AND status != 0 ))
AND server_type NOT IN (?, ?) AND name = ?',
PandoraFMS::Tools::SATELLITESERVER,
PandoraFMS::Tools::MFSERVER,
$conf->{'servername'}
);
my $nservers = get_db_value ($dbh, 'SELECT count(*) FROM tserver where name = ?', $conf->{'servername'});
$Pandora_Service = $conf->{'pandora_service_cmd'};
# Check if service is running
$control_command = "status-server";
if ($OSNAME eq "freebsd") {
$control_command = "status_server";
}
my $pid = `$Pandora_Service $control_command | awk '{print \$NF*1}' | tr -d '\.'`;
if ( ($pid > 0) && ($component_last_contact > 0)) {
# service running but not all components
log_message($conf, 'LOG', 'Pandora service running but not all components.');
print ">> service running but delayed...\n";
$control_command = "restart-server";
if ($OSNAME eq "freebsd") {
$control_command = "restart_server";
}
`$Pandora_Service $control_command 2>/dev/null`;
} elsif ($pid == 0) {
# service not running
log_message($conf, 'LOG', 'Pandora service not running.');
print ">> service not running...\n";
$control_command = "start-server";
if ($OSNAME eq "freebsd") {
$control_command = "start_server";
}
`$Pandora_Service $control_command 2>/dev/null`;
} elsif ($pid > 0
&& $nservers == 0
) {
my @server_list = get_enabled_servers($conf);
my $nservers = $#server_list;
# Process running but no servers active, restart.
# Try to restart pandora_server if no servers are found.
# Do not restart if is a configuration issue.
log_message($conf, 'LOG', 'Pandora service running without servers ['.$nservers.'].');
if ($nservers >= 0) {
log_message($conf, 'LOG', 'Restarting Pandora service...');
$control_command = "restart-server";
if ($OSNAME eq "freebsd") {
$control_command = "restart_server";
}
`$Pandora_Service $control_command 2>/dev/null`;
}
}
}
###############################################################################
# Update pandora services.
###############################################################################
sub ha_update_server($$) {
my ($config, $dbh) = @_;
my $OSNAME = $^O;
my $repoServer = pandora_get_tconfig_token(
$dbh, 'remote_config', '/var/spool/pandora/data_in'
);
$repoServer .= '/updates/server/';
my $lockFile = $repoServer.'/'.$config->{'servername'}.'.installed';
my $workDir = $config->{"temporal"}.'/server_update/';
my $versionFile = $repoServer.'version.txt';
return if (-e $lockFile) || (!-e $versionFile);
log_message($config, 'LOG', 'Detected server update: '.`cat "$versionFile"`);
if(!-e "$workDir" && !mkdir ($workDir)) {
log_message($config, 'ERROR', 'Server update failed: '.$!);
return;
}
my $r = `cd "$workDir/" && tar xzf "$repoServer/pandorafms_server.tar.gz" 2>&1`;
if ($? ne 0) {
log_message($config, 'ERROR', 'Failed to uncompress file: '.$r);
return;
}
$r = `cd "$workDir/pandora_server/" && ./pandora_server_installer --install 2>&1 >/dev/null`;
if ($? ne 0) {
log_message($config, 'ERROR', 'Failed to install server update: '.$r);
return;
} else {
log_message($config, 'LOG', 'Server update '.`cat "$versionFile"`.' installed');
}
# Cleanup
rmtree($workDir);
# Restart service
$config->{'pandora_service_cmd'} = 'service pandora_server'
unless defined($config->{'pandora_service_cmd'});
my $control_command = "restart-server";
if ($OSNAME eq "freebsd") {
$control_command = "restart_server";
}
`$config->{'pandora_service_cmd'} $control_command 2>/dev/null`;
`touch "$lockFile"`;
# After apply update, permission over files are changed, allow group to
# modify/delete files.
`chmod 770 "$repoServer"`;
`chmod 770 "$repoServer/../"`;
`chmod 660 "$repoServer"/*`;
}
###############################################################################
# Connect to ha database, falling back to direct connection to db.
###############################################################################
sub ha_database_connect($) {
my $conf = shift;
my $dbh = enterprise_hook('ha_connect', [$conf]);
if (!defined($dbh)) {
$dbh = db_connect ('mysql', $conf->{'dbname'}, $conf->{'dbhost'}, $conf->{'dbport'}, $conf->{'dbuser'}, $conf->{'dbpass'});
}
return $dbh;
}
###############################################################################
# Main
###############################################################################
sub ha_main($) {
my ($conf) = @_;
# Set the PID file.
$conf->{'PID'} = $PID_FILE;
# Log to a separate file if needed.
$conf->{'log_file'} = $conf->{'ha_log_file'} if defined ($conf->{'ha_log_file'});
ha_daemonize($conf) if ($DAEMON == 1);
$Running = 1;
while ($Running) {
eval {
eval {
local $SIG{__DIE__};
# Load enterprise components.
enterprise_load($conf, 1);
# Register Enterprise logger
enterprise_hook('pandoraha_logger', [\&log_message]);
log_message($conf, 'LOG', 'Enterprise capabilities loaded');
};
if ($@) {
# No enterprise capabilities.
log_message($conf, 'LOG', 'No enterprise capabilities');
}
# Start the Pandora FMS server if needed.
log_message($conf, 'LOG', 'Checking the pandora_server service.');
# Connect to a DB.
my $dbh = ha_database_connect($conf);
if ($First_Cleanup == 1) {
log_message($conf, 'LOG', 'Cleaning previous unfinished actions');
enterprise_hook('pandoraha_cleanup_states', [$conf, $dbh]);
$First_Cleanup = 0;
}
# Check if there are updates pending.
ha_update_server($conf, $dbh);
# Keep pandora running
ha_keep_pandora_running($conf, $dbh);
# Are we the master?
pandora_set_master($conf, $dbh);
if (!pandora_is_master($conf)) {
log_message($conf, 'LOG', $conf->{'servername'} . ' is not the current master. Skipping DB-HA actions and monitoring.');
# Exit current eval.
return;
}
# Synchronize database.
enterprise_hook('pandoraha_sync_node', [$conf, $dbh]);
# Monitoring.
enterprise_hook('pandoraha_monitoring', [$conf, $dbh]);
# Pending actions.
enterprise_hook('pandoraha_process_queue', [$conf, $dbh, $First_Cleanup]);
# Cleanup and exit
db_disconnect ($dbh);
};
log_message($conf, 'WARNING', $@) if ($@);
log_message($conf, 'LOG', "Sleep.");
sleep($conf->{'ha_interval'});
}
}
################################################################################
# Stop pandora server
################################################################################
sub stop {
my $OSNAME = $^O;
if ($Running == 1) {
$Running = 0;
# cleanup and stop pandora_server
print ">> stopping server...\n";
my $control_command = "stop-server";
if ($OSNAME eq "freebsd") {
$control_command = "stop_server";
}
`$Pandora_Service $control_command 2>/dev/null`;
}
}
################################################################################
# END block.
################################################################################
END {
stop();
}
$SIG{INT} = \&stop;
$SIG{TERM} = \&stop;
# Init
ha_init_pandora(\%Conf);
# Read config file
ha_load_pandora_conf (\%Conf);
# Main
ha_main(\%Conf);
exit 0;