@ -38,7 +38,13 @@ my $anvil = Anvil::Tools->new();
# Read switches (target ([user@]host[:port]) and the file with the target's password. If the password is
# passed directly, it will be used. Otherwise, the password will be read from the database.
$anvil->Get->switches;
$anvil->Get->switches({list => [
"as-machine",
"join",
"manifest",
"rejoin",
], man => $THIS_FILE});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => $anvil->data->{switches}});
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 2, key => "log_0115", variables => { program => $THIS_FILE }});
$anvil->Database->connect();
@ -52,6 +58,13 @@ if (not $anvil->data->{sys}{database}{connections})
$anvil->nice_exit({exit_code => 1});
}
# If there isn't a job UUID, see if the user is asking to rejoin this subnode to a cluster.
if ($anvil->data->{switches}{rejoin})
{
process_rejoin($anvil);
$anvil->nice_exit({exit_code => 0});
}
# Get the job details
load_job($anvil);
@ -102,10 +115,293 @@ $anvil->Database->insert_or_update_variables({
$anvil->nice_exit({exit_code => 0});
#############################################################################################################
# Functions #
#############################################################################################################
# Rejoin this host to an existing subnode.
sub process_rejoin
{
my ($anvil) = @_;
print "-=] Rejoin this host: [".$anvil->Get->short_host_name."] as a subnode in an existing Anvil! node.\n";
my $host_uuid = $anvil->Get->host_uuid();
my $as_machine = $anvil->data->{switches}{'as-machines'};
my $manifest = $anvil->data->{switches}{manifest};
my $manifest_name = "";
my $manifest_uuid = "";
my $anvil_uuid = "";
my $rebuild = 0;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:host_uuid' => $host_uuid,
's2:as_machine' => $as_machine,
's3:manifest' => $manifest,
}});
# Load data
$anvil->Database->get_hosts();
$anvil->Database->get_anvils();
$anvil->Database->get_manifests();
# In case we're being re-installed, see if we can find our own data.
my $old_manifest_uuid = "";
if (exists $anvil->data->{anvils}{host_uuid}{$host_uuid})
{
# Found it.
my $anvil_name = $anvil->data->{anvils}{host_uuid}{$host_uuid}{anvil_name};
$anvil_uuid = $anvil->data->{anvils}{host_uuid}{$host_uuid}{anvil_uuid};
my $old_role = $anvil->data->{anvils}{host_uuid}{$host_uuid}{role};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:anvil_name' => $anvil_name,
's2:anvil_uuid' => $anvil_uuid,
}});
print "This host used to be in the Anvil! [".$anvil_name."] as: [".$old_role."]\n";
if (exists $anvil->data->{manifests}{manifest_name}{$anvil_name})
{
# Found the manifest.
$old_manifest_uuid = $anvil->data->{manifests}{manifest_name}{$anvil_name}{manifest_uuid};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { old_manifest_uuid => $old_manifest_uuid }});
print "The manifest was found with the UUID: [".$old_manifest_uuid."]\n";
}
else
{
# Didn't find the manifest, something is wrong.
print "The manifest was NOT found (was it deleted?).\n";
print "- The manifes must be created for this host to rejoin the Anvil! node.\n";
print "- Alternatively, this host must be removed for the Anvil!\n";
print "[ Error ] - Unable to proceed at this time.\n";
$anvil->nice_exit({exit_code => 1});
}
# Does it match?
if (($manifest) && (($manifest ne $anvil_name) && ($manifest ne $old_manifest_uuid)))
{
# They asked for this machine to be joined to a different Anvil!.
print "[ Error ] - You asked to join the Anvil! node: [".$manifest."].\n";
print "[ Error ] - This host must be removed from: [".$anvil_name."] first.\n";
print "[ Error ] - Unable to proceed at this time.\n";
$anvil->nice_exit({exit_code => 1});
}
if (($as_machine) && ($as_machine ne $old_role))
{
print "[ Error ] - You asked to join the Anvil! node: [".$manifest."] as: [".$as_machine."]\n";
print "[ Error ] - This host was previously: [".$old_role."], so it needs to be that role again.\n";
print "[ Error ] - Alternatively, remove this host from that Anvil! and try again.\n";
print "[ Error ] - Unable to proceed at this time.\n";
$anvil->nice_exit({exit_code => 1});
}
$manifest = $anvil_name;
$as_machine = $old_role;
$rebuild = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:manifest' => $manifest,
's2:as_machine' => $as_machine,
's3:rebuild' => $rebuild,
}});
}
if ($manifest)
{
# Did we get a valid manifest?
if (exists $anvil->data->{manifests}{manifest_name}{$manifest})
{
$manifest_name = $manifest;
$manifest_uuid = $anvil->data->{manifests}{manifest_name}{$manifest}{manifest_uuid};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:manifest_name' => $manifest_name,
's2:manifest_uuid' => $manifest_uuid,
}});
}
elsif (exists $anvil->data->{manifests}{manifest_uuid}{$manifest})
{
$manifest_name = $anvil->data->{manifests}{manifest_uuid}{$manifest}{manifest_name};
$manifest_uuid = $manifest;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:manifest_name' => $manifest_name,
's2:manifest_uuid' => $manifest_uuid,
}});
}
else
{
print "[ Error ] - The manifest: [".$manifest."] was not found in the database.\n";
print "[ Error ] - Try again without this switch to see the available manifests.\n";
$anvil->nice_exit({exit_code => 1});
}
}
else
{
# Show the existing manifests.
my $count = keys %{$anvil->data->{manifests}{manifest_name}};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { count => $count }});
foreach my $this_manifest_name (sort {$a cmp $b} keys %{$anvil->data->{manifests}{manifest_name}})
{
my $this_manifest_uuid = $anvil->data->{manifests}{manifest_name}{$this_manifest_name}{manifest_uuid};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:this_manifest_name' => $this_manifest_name,
's2:this_manifest_uuid' => $this_manifest_uuid,
}});
if ($count == 1)
{
$manifest_name = $this_manifest_name;
$manifest_uuid = $this_manifest_uuid;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:manifest_name' => $manifest_name,
's2:manifest_uuid' => $manifest_uuid,
}});
}
else
{
print "- Manifest: [".$manifest_name."] (uuid: [".$manifest_uuid."])\n";
}
}
print "- Which manifest do you want to join this host to?\n";
my $answer = <STDIN>;
chomp($answer);
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { answer => $answer }});
# Did we get a valid manifest?
if (exists $anvil->data->{manifests}{manifest_name}{$answer})
{
$manifest_name = $answer;
$manifest_uuid = $anvil->data->{manifests}{manifest_name}{$answer}{manifest_uuid};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:manifest_name' => $manifest_name,
's2:manifest_uuid' => $manifest_uuid,
}});
}
elsif (exists $anvil->data->{manifests}{manifest_uuid}{$answer})
{
$manifest_name = $anvil->data->{manifests}{manifest_uuid}{$answer}{manifest_name};
$manifest_uuid = $answer;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:manifest_name' => $manifest_name,
's2:manifest_uuid' => $manifest_uuid,
}});
}
else
{
if ($answer)
{
print "[ Error ] - Your answer: [".$answer."] doesn't match a valid manifest.\n";
}
print "[ Error ] - Please try again (you can use --manifest <name or UUID> to avoind this prompt).\n";
$anvil->nice_exit({exit_code => 1});
}
}
if (not $as_machine)
{
print "- Will this node be 'node1' or 'node2'?\n";
my $answer = <STDIN>;
chomp($answer);
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { answer => $answer }});
if ($answer eq "node1")
{
$as_machine = "node1";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { as_machine => $as_machine }});
}
elsif ($answer eq "node2")
{
$as_machine = "node2";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { as_machine => $as_machine }});
}
else
{
print "[ Error ] - Please enter 'node1' or 'node2'.\n";
print "[ Error ] - Please try again (you can use '--as-manchine node{1,2}' to avoind this prompt).\n";
$anvil->nice_exit({exit_code => 1});
}
}
if (not $anvil_uuid)
{
$anvil_uuid = $anvil->data->{anvils}{anvil_name}{$manifest_name}{anvil_uuid};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { anvil_uuid => $anvil_uuid }});
if (not $anvil_uuid)
{
print "[ Error ] - Failed to find an Anvil! UUID for the node: [".$manifest_name."].\n";
print "[ Error ] - Has the Anvil! been deleted from the database?\n";
print "[ Error ] - Unable to proceed.\n";
$anvil->nice_exit({exit_code => 1});
}
}
print "\n";
print "-=] Joining: [".$manifest_name."] as: [".$as_machine."] (Anvil! UUID: [".$anvil_uuid."]\n";
if ($rebuild)
{
print "[ Note ] - This is a rebuild, the previous data recorded by this host will be\n";
print " preserved.\n";
}
else
{
print "[ Warning ] - This will replace the old subnode, and all the previous data\n";
print " associated with it!\n";
print "[ Warning ] - Be certain the old host will NOT come back! If it does, it can\n";
print " cause confusion with the Anvil! node!\n";
}
if (($anvil->data->{switches}{y}) or ($anvil->data->{switches}{yes}))
{
print "[ Note ] - Confirmed by switch, proceeding.\b";
}
else
{
print $anvil->Words->string({key => "message_0021"})." ";
my $answer = <STDIN>;
chomp($answer);
if ($answer !~ /^y/i)
{
print $anvil->Words->string({key => "message_0022"})."\n";
$anvil->nice_exit({exit_code => 0});
}
}
# If this isn't a rebuild, purge the old host.
if (not $rebuild)
{
my $node_key = $as_machine eq "node1" ? "anvil_node1_host_uuid" : "anvil_node2_host_uuid";
my $old_host_uuid = $anvil->data->{anvils}{anvil_name}{$manifest_name}{$node_key} ? $anvil->data->{anvils}{anvil_name}{$manifest_name}{$node_key} : "";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:node_key' => $node_key,
's2:old_host_uuid' => $old_host_uuid,
}});
if ($old_host_uuid)
{
print "[ Note ] - Purging the old host: [".$old_host_uuid."] from the database.\n";
print "[ Note ] - Please be patient!\n";
my $shell_call = $anvil->data->{path}{exe}{'striker-purge-target'}." ".$anvil->Log->switches;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, 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, list => {
output => $output,
return_code => $return_code,
}});
}
}
# Register a job.
my ($job_uuid) = $anvil->Database->insert_or_update_jobs({
job_host_uuid => $host_uuid,
job_command => $anvil->data->{path}{exe}{'anvil-join-anvil'}.$anvil->Log->switches,
job_data => "as_machine=".$as_machine.",manifest_uuid=".$manifest_uuid.",anvil_uuid=".$anvil_uuid,
job_name => "join_anvil::".$as_machine,
job_title => "job_0072",
job_description => "job_0073",
job_progress => 0,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { job_uuid => $job_uuid }});
print "Registering a job. This host should (re)join the Anvil! node shortly.\n";
return(0);
}
# Make sure the hosts file has entries for all nets for both subnodes
sub wait_for_etc_hosts
{
@ -559,6 +855,30 @@ sub configure_pacemaker
return(0);
}
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { 'path::configs::corosync-authkey' => $anvil->data->{path}{configs}{'corosync-authkey'} }});
if (not -f $anvil->data->{path}{configs}{'corosync-authkey'})
{
# In case we're rebuilding, see if the peer already has the '/etc/corosync/authkey' file.
my $corosync_authkey = $anvil->Storage->read_file({
file => $anvil->data->{path}{configs}{'corosync-authkey'},
target => $peer_host_name,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { corosync_authkey => $corosync_authkey }});
if ($corosync_authkey ne "!!error!!")
{
# Write the file out.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0485"});
$anvil->Storage->write_file({
debug => 2,
body => $corosync_authkey,
file => $anvil->data->{path}{configs}{'corosync-authkey'},
user => "root",
group => "root",
mode => "0400",
});
}
}
### Run on both nodes.
# Enable pcsd and start the pcsd daemon.
my ($return_code) = $anvil->System->enable_daemon({daemon => "pcsd.service"});
@ -605,6 +925,7 @@ sub configure_pacemaker
# If there is no corosync.conf, see if the peer has it. If so, copy it. If not, we'll initialize the
# cluster shortly.
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { 'path::configs::corosync.conf' => $anvil->data->{path}{configs}{'corosync.conf'} }});
if (not -e $anvil->data->{path}{configs}{'corosync.conf'})
{
my $corosync_conf = $anvil->Storage->read_file({
@ -625,6 +946,13 @@ sub configure_pacemaker
mode => "0644",
});
}
# Restart corosync
my ($return_code) = $anvil->System->restart_daemon({daemon => "corosync.service"});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { return_code => $return_code }});
$return_code = undef;
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0095", variables => { daemon => "corosync.service" }});
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0095,!!daemon!corosync.service!!");
}
# Node 1 initializes, node 2 waits.
@ -637,6 +965,12 @@ sub configure_pacemaker
# We loop until the peer finishes or the peer's job hit's 100.
my $tried_starting = 0;
my $both_online = 0;
my $try_starting = time + 30;
my $delay = 5;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
delay => $delay,
try_starting => $try_starting,
}});
until($both_online)
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
@ -708,10 +1042,73 @@ sub configure_pacemaker
}
if (not $both_online)
{
my $delay = 5;
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0816", variables => { delay => $delay }});
sleep $delay;
if (time > $try_starting)
{
# Try starting pacemaker, in case we're rebuilding.
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0164");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0164"});
$try_starting += 60;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { try_starting => $try_starting }});
my $shell_call = $anvil->data->{path}{exe}{pcs}." cluster start";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, 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, list => {
output => $output,
return_code => $return_code,
}});
}
else
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0816", variables => { delay => $delay }});
sleep $delay;
}
}
}
### TODO: Left off here; parse 'pcs cluster status' and if there's 'Unable to authenticate', re-auth.
my $auth = 0;
my $shell_call = $anvil->data->{path}{exe}{pcs}." cluster status";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, 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, list => {
output => $output,
return_code => $return_code,
}});
my $in_pcsd = 0;
foreach my $line (split/\n/, $output)
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { line => $line }});
if ($in_pcsd)
{
if ($line =~ /Unable to authenticate/i)
{
$auth = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { auth => $auth }});
}
}
if ($line =~ /PCSD Status/i)
{
$in_pcsd = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { in_pcsd => $in_pcsd }});
}
last if $auth;
}
if ($auth)
{
# Also, re-auth. We need to run this on both hosts.
my $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, secure => 1, list => { shell_call => $shell_call }});
my ($output, $return_code) = $anvil->System->call({secure => 1, shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
output => $output,
return_code => $return_code,
}});
}
}
else
Digimer
8 months ago
committed by
GitHub