@ -11,12 +11,14 @@
# 5 = Problem parsing job data or loading manifest or anvil data using job data. # 5 = Problem parsing job data or loading manifest or anvil data using job data.
# #
# TODO: # TODO:
# # -
#
use strict; use strict;
use warnings; use warnings;
use Anvil::Tools; use Anvil::Tools;
use Data::Dumper; use Data::Dumper;
use String::ShellQuote;
my $THIS_FILE = ($0 =~ /^.*\/(.*)$/)[0]; my $THIS_FILE = ($0 =~ /^.*\/(.*)$/)[0];
my $running_directory = ($0 =~ /^(.*?)\/$THIS_FILE$/)[0]; my $running_directory = ($0 =~ /^(.*?)\/$THIS_FILE$/)[0];
@ -72,17 +74,37 @@ sub configure_pacemaker
my $manifest_uuid = $anvil->data->{sys}{manifest_uuid}; my $manifest_uuid = $anvil->data->{sys}{manifest_uuid};
### TODO: Move these to variables in the 'sys' hash ### TODO: Move these to variables in the 'sys' hash
my $anvil_name = $anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed}{name}; my $anvil_name = $anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed}{name};
my $anvil_uuid = $anvil->data->{sys}{anvil_uuid}; my $anvil_uuid = $anvil->data->{sys}{anvil_uuid};
my $host_name = $anvil->data->{sys}{host_name}; my $host_name = $anvil->data->{sys}{host_name};
my $new_password = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_password}; my $new_password = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_password};
my $node1_host_uuid = $anvil->data->{sys}{node1_host_uuid} = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_node1_host_uuid};
my $node1_host_name = $anvil->data->{hosts}{host_uuid}{$node1_host_uuid}{host_name};
$node1_host_name =~ s/\..*$//;
my $node2_host_uuid = $anvil->data->{sys}{node2_host_uuid} = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_node2_host_uuid};
my $node2_host_name = $anvil->data->{hosts}{host_uuid}{$node2_host_uuid}{host_name};
$node2_host_name =~ s/\..*$//;
my $peer_host_name = $anvil->Get->host_uuid() eq $node1_host_uuid ? $node2_host_name : $node1_host_name;
my $peer_host_uuid = $anvil->Get->host_uuid() eq $node1_host_uuid ? $node2_host_uuid : $node1_host_uuid;
my $escaped_password = shell_quote($new_password);
my $auth_shell_call = $anvil->data->{path}{exe}{pcs}." host auth ".$node1_host_name." ".$node2_host_name." -u hacluster -p ".$escaped_password;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
}});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:machine' => $machine, machine => $machine,
's1:anvil_name' => $anvil_name, anvil_uuid => $anvil_uuid,
's3:host_name' => $host_name, anvil_name => $anvil_name,
's4:manifest_uuid' => $manifest_uuid, host_name => $host_name,
's5:anvil_uuid' => $anvil_uuid, manifest_uuid => $manifest_uuid,
's6:new_password' => $anvil->Log->is_secure($new_password), node1_host_uuid => $node1_host_uuid,
node1_host_name => $node1_host_name,
node2_host_uuid => $node2_host_uuid,
node2_host_name => $node2_host_name,
peer_host_name => $peer_host_name,
peer_host_uuid => $peer_host_uuid,
new_password => $anvil->Log->is_secure($new_password),
escaped_password => $anvil->Log->is_secure($escaped_password),
auth_shell_call => $anvil->Log->is_secure($auth_shell_call),
}}); }});
# If this is a DR box, we don't use pacemaker. # If this is a DR box, we don't use pacemaker.
@ -126,14 +148,134 @@ sub configure_pacemaker
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0095,!!daemon!libvirtd!!"); update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0095,!!daemon!libvirtd!!");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0095", variables => { daemon => "libvirtd" }}); $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0095", variables => { daemon => "libvirtd" }});
# If there is no corosync.conf, see if the peer has it. If so, copy it. If not, we'll initialize the
# cluster shortly.
if (not -e $anvil->data->{path}{configs}{'corosync.conf'})
{
my $cluster_conf = $anvil->Storeage->read_file({
file => $anvil->data->{path}{configs}{'corosync.conf'},
target => $peer_host_name,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { cluster_conf => $cluster_conf }});
if ($cluster_conf ne "!!error!!")
{
# Write the file out.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0100"});
$anvil->Storage->write_file({
body => $cluster_conf,
file => $anvil->data->{path}{configs}{'corosync.conf'},
user => "root",
group => "root",
mode => "0644",
});
}
}
### Run on node 1 only. ### Run on node 1 only.
if ($machine eq "node2") if ($machine eq "node2")
{ {
# We loop until the peer finishes. # We loop until the peer finishes or the peer's job hit's 100 .
} }
else else
{ {
# Proceed with cluster setup. # Proceed with cluster setup.
my $waiting = 1;
my $warning_printed = 0;
while($waiting)
{
my ($output, $return_code) = $anvil->System->call({debug => 3, secure => 1, shell_call => $auth_shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => {
output => $output,
return_code => $return_code,
}});
if ($return_code)
{
# Something went wrong.
if (not $warning_printed)
{
# Update the job
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0097");
$warning_printed = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { warning_printed => $warning_printed }});
}
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0097"});
sleep 5;
}
else
{
# We're good.
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0098");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0098"});
$waiting = 0;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { waiting => $waiting }});
}
}
# If there is no corosync.conf, see if the peer has it. If so, copy it. If not, initialize
# the cluster.
if (not -e $anvil->data->{path}{configs}{'corosync.conf'})
{
# There's no cluster yet, initialize it.
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0099,!!anvil_name!".$anvil_name."!!");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0099", variables => { anvil_name => $anvil_name }});
my $shell_call = $anvil->data->{path}{exe}{pcs}." cluster setup ".$anvil_name." ".$node1_host_name." ".$node2_host_name;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => { shell_call => $shell_call }});
my ($output, $return_code) = $anvil->System->call({debug => 3, shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => {
output => $output,
return_code => $return_code,
}});
if ($return_code)
{
# Something went wrong
update_progress($anvil, 100, "job_0101,!!error!".$output."!!");
sleep 2;
$anvil->nice_exit({exit_code => 5});
}
}
# Now, if we can read the CIB, see where the setup is. If not, start by setting up the
# cluster.
my $cib_data = "";
my $cluster_started = 0;
until ($cib_data)
{
my $shell_call = $anvil->data->{path}{exe}{pcs}." cluster cib";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => { shell_call => $shell_call }});
($cib_data, my $return_code) = $anvil->System->call({debug => 3, shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => {
cib_data => $cib_data,
return_code => $return_code,
}});
if ($return_code)
{
if (not $cluster_started)
{
# Start the cluster.
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0102");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0102"});
$cluster_started = 1;
my $shell_call = $anvil->data->{path}{exe}{pcs}." cluster start --all";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => {
cluster_started => $cluster_started,
shell_call => $shell_call,
}});
my ($output, $return_code) = $anvil->System->call({debug => 3, shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 0, list => {
output => $output,
return_code => $return_code,
}});
}
}
die;
}
die;
} }
=cut =cut
$anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed}; $anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed};
@ -896,7 +1038,7 @@ sub check_local_network
{ {
# It's fine # It's fine
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0087,!!interface!".$in_iface."!!,!!mtu!".$mtu."!!"); update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0087,!!interface!".$in_iface."!!,!!mtu!".$mtu."!!");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2 , key => "job_0087", variables => { $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 3 , key => "job_0087", variables => {
interface => $in_iface, interface => $in_iface,
mtu => $mtu, mtu => $mtu,
}}); }});
@ -1037,7 +1179,10 @@ sub check_local_network
### TODO: Do we really need passwordless SSH anymore? ### TODO: Do we really need passwordless SSH anymore?
# Configure SSH by adding ours and our peer's SSH keys to ~/.ssh/known_hosts # Configure SSH by adding ours and our peer's SSH keys to ~/.ssh/known_hosts
$anvil->System->check_ssh_keys({debug => 2}); $anvil->System->check_ssh_keys({debug => 3});
# Update the hosts file.
$anvil->System->update_hosts({debug => 3});
# Setup IPMI, if needed. # Setup IPMI, if needed.
### TODO: Do this when on real hardware ### TODO: Do this when on real hardware
Digimer
5 years ago