Updated to add all subnode nets to /etc/hosts before forming cluster

Signed-off-by: digimer <mkelly@alteeve.ca>
main
digimer 11 months ago
parent 760d9f53d8
commit d8ceb7fbf4
  1. 6
      share/words.xml
  2. 277
      tools/anvil-join-anvil

@ -2709,6 +2709,12 @@ The file: [#!variable!file!#] needs to be updated. The difference is:
<key name="log_0827">The physical volume: [#!variable!pv_name!#] has been resized!</key> <key name="log_0827">The physical volume: [#!variable!pv_name!#] has been resized!</key>
<key name="log_0828">The user answered: [#!variable!answer!#]</key> <key name="log_0828">The user answered: [#!variable!answer!#]</key>
<key name="log_0829">Failed to connect to any database, waiting before trying to connect again. We will reboot if we do not connect in: [#!variable!time_left!#] second.</key> <key name="log_0829">Failed to connect to any database, waiting before trying to connect again. We will reboot if we do not connect in: [#!variable!time_left!#] second.</key>
<key name="log_0830">The host: [#!variable!host_name!#] was found in the '/etc/hosts' file with the expected IP: [#!variable!ip_address!#]!</key>
<key name="log_0831">The host: [#!variable!host_name!#] was found in the '/etc/hosts' file, but the expected IP: [#!variable!ip_address!#] doesn't match the found IP: [#!variable!found_ip!#]! Ignoring this entry.</key>
<key name="log_0832">The host: [#!variable!host_name!#] was found not found in the '/etc/hosts' file! We'll wait a few seconds and check again.</key>
<key name="log_0833">All host names were found in '/etc/hosts', ready to proceed!</key>
<key name="log_0834">One or more hosts are not yet in the '/etc/hosts' file with expected IPs. We'll wait a short bit and check again.</key>
<key name="log_0835">There are entries we need to add to the '/etc/hosts' file before we can form the cluster, updating it now.</key>
<!-- Messages for users (less technical than log entries), though sometimes used for logs, too. --> <!-- Messages for users (less technical than log entries), though sometimes used for logs, too. -->
<key name="message_0001">The host name: [#!variable!target!#] does not resolve to an IP address.</key> <key name="message_0001">The host name: [#!variable!target!#] does not resolve to an IP address.</key>

@ -55,6 +55,9 @@ if (not $anvil->data->{sys}{database}{connections})
# Get the job details # Get the job details
load_job($anvil); load_job($anvil);
# Make sure the hosts file has entries for all nets for both subnodes
wait_for_etc_hosts($anvil);
# Hold until both subnodes are marked as configured and not in maintenance mode. # Hold until both subnodes are marked as configured and not in maintenance mode.
wait_for_subnodes($anvil); wait_for_subnodes($anvil);
@ -67,6 +70,7 @@ check_local_network($anvil);
# Wait until we can ping our peer on all networks. # Wait until we can ping our peer on all networks.
wait_for_access($anvil); wait_for_access($anvil);
### TODO: Change corosync.conf to use IPs, including MN if it exists.
# (wait for out peer and) Configure pacemaker # (wait for out peer and) Configure pacemaker
configure_pacemaker($anvil); configure_pacemaker($anvil);
@ -102,6 +106,271 @@ $anvil->nice_exit({exit_code => 0});
# Functions # # Functions #
############################################################################################################# #############################################################################################################
# Make sure the hosts file has entries for all nets for both subnodes
sub wait_for_etc_hosts
{
my ($anvil) = @_;
my $anvil_uuid = $anvil->data->{sys}{anvil_uuid};
my $node1_host_uuid = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_node1_host_uuid};
my $node2_host_uuid = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_node2_host_uuid};
my $manifest_uuid = $anvil->data->{sys}{manifest_uuid};
my $i_am = $anvil->data->{sys}{machine};
my $peer_is = $i_am eq "node1" ? "node2" : "node1";
my $peer_host_uuid = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{"anvil_".$peer_is."_host_uuid"};
my $peer_short_host_name = $anvil->data->{hosts}{host_uuid}{$peer_host_uuid}{short_host_name};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:anvil_uuid' => $anvil_uuid,
's2:manifest_uuid' => $manifest_uuid,
's3:node1_host_uuid' => $node1_host_uuid,
's4:node2_host_uuid' => $node2_host_uuid,
's5:i_am' => $i_am,
's6:peer_is' => $peer_is,
's7:peer_host_uuid' => $peer_host_uuid,
's8:peer_short_host_name' => $peer_short_host_name,
}});
my $problem = $anvil->Striker->load_manifest({debug => 2, manifest_uuid => $manifest_uuid});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
manifest_uuid => $anvil->data->{sys}{manifest_uuid},
problem => $problem,
}});
if ($problem)
{
# Something went wrong, fatally. Abort the job.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "job_0076", variables => { uuid => $anvil->data->{sys}{manifest_uuid} }});
update_progress($anvil, 100, "job_0076,!!uuid!".$anvil->data->{sys}{manifest_uuid}."!!");
$anvil->nice_exit({exit_code => 2});
}
# Create the list of host names we need to see in /etc/hosts and the hostnames to use in corosync.
foreach my $machine (sort {$a cmp $b} keys %{$anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed}{machine}})
{
my $this_host_uuid = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{"anvil_".$machine."_host_uuid"};
my $this_short_host_name = $anvil->data->{hosts}{host_uuid}{$this_host_uuid}{short_host_name};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { machine => $machine }});
foreach my $network_name (sort {$a cmp $b} keys %{$anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed}{machine}{$machine}{network}})
{
my $host_name = $this_short_host_name.".".$network_name;
my $ip_address = $anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}{parsed}{machine}{$machine}{network}{$network_name}{ip};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:network_name' => $network_name,
's2:host_name' => $host_name,
's3:ip_address' => $ip_address,
}});
$anvil->data->{networks}{$host_name}{ip_address} = $ip_address;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"s1:networks::${host_name}::ip_address" => $anvil->data->{networks}{$host_name}{ip_address},
}});
}
}
my $waiting = 1;
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0477");
while($waiting)
{
# Update the /etc/hosts file. Note that this doesn't add hosts that are not yet trusted, so
# anything not found will be inserted here.
$anvil->System->update_hosts({debug => 2});
# Now lets see if all expected hosts names are in the /etc/hosts file yet.
my $ready = 1;
my $hosts_file = $anvil->Storage->read_file({
file => $anvil->data->{path}{configs}{hosts},
force_read => 1,
cache => 0,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { hosts_file => $hosts_file }});
foreach my $host_name (sort {$a cmp $b} keys %{$anvil->data->{networks}})
{
my $found = 0;
my $ip_address = $anvil->data->{networks}{$host_name}{ip_address};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
's1:host_name' => $host_name,
's2:ip_address' => $ip_address,
}});
foreach my $line (split/\n/, $hosts_file)
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { line => $line }});
$line =~ s/#.*$//;
if ($line =~ /^(\d.*?)\s+(.*)$/)
{
my $this_ip = $1;
my $hosts = $2;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
this_ip => $this_ip,
hosts => $hosts,
}});
if ($anvil->Validate->ip({ip => $this_ip}))
{
foreach my $this_host (split/\s+/, $hosts)
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { this_host => $this_host }});
if ($host_name eq $this_host)
{
my $variables = {
host_name => $host_name,
ip_address => $ip_address,
found_ip => $this_ip,
};
if ($ip_address eq $this_ip)
{
# Found it
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0830", variables => $variables });
$found = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { found => $found }});
}
else
{
# Found the host, but the IP doesn't match.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "log_0831", variables => $variables });
}
}
last if $found;
}
}
}
last if $found;
}
if (not $found)
{
$ready = 0;
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "log_0832", variables => { host_name => $host_name }});
# Add the IP to be added to /etc/hosts.
$anvil->data->{to_add}{$host_name} = $ip_address;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { "to_add::${ip_address}" => $anvil->data->{to_add}{$host_name} }});
}
}
if ($ready)
{
# Ready!
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "log_0833"});
$waiting = 0;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { waiting => $waiting }});
}
else
{
# Not ready, wait a bit.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "log_0834"});
# Add the values to /etc/hosts.
my $new_hosts_body = "";
foreach my $line (split/\n/, $hosts_file)
{
# See if this line needs to be modified.
if ($line =~ /^(\d.*?)\s+(.*)$/)
{
my $this_ip = $1;
my $hosts = $2;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
this_ip => $this_ip,
hosts => $hosts,
}});
if ($anvil->Validate->ip({ip => $this_ip}))
{
my $changes = 0;
my $new_hosts = "";
my $comment = ($hosts =~ /(#.*)$/)[0];
$comment = "" if not defined $comment;
$hosts =~ s/#.*?//;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
comment => $comment,
hosts => $hosts,
}});
foreach my $host_name (split/\s+/, $hosts)
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { host_name => $host_name }});
if ((exists $anvil->data->{to_add}{$host_name}) &&
($anvil->data->{to_add}{$host_name}) &&
($this_ip ne $anvil->data->{to_add}{$host_name}))
{
# Remove this host.
$changes = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { changes => $changes }});
delete $anvil->data->{to_add}{$host_name};
next;
}
$new_hosts .= $host_name." ";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { new_hosts => $new_hosts }});
}
# Do we have any names to add to this IP?
foreach my $host_name (sort {$a cmp $b} keys %{$anvil->data->{to_add}})
{
if ($this_ip eq $anvil->data->{to_add}{$host_name})
{
# Add it.
$new_hosts .= $host_name." ";
$changes = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { changes => $changes }});
delete $anvil->data->{to_add}{$host_name};
}
}
if ($changes)
{
$new_hosts_body .= $this_ip."\t".$new_hosts.$comment."\n";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { changes => $changes }});
next;
}
}
}
# If we're alive here, just add the old line.
$new_hosts_body .= $line."\n";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { new_hosts_body => $new_hosts_body }});
}
# Add any hosts still not processed.
foreach my $host_name (sort {$a cmp $b} keys %{$anvil->data->{to_add}})
{
$new_hosts_body .= $anvil->data->{to_add}{$host_name}."\t".$host_name."\n";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { new_hosts_body => $new_hosts_body }});
delete $anvil->data->{to_add}{$host_name};
}
# If there's a difference, write the new hosts file.
my $difference = diff \$new_hosts_body, \$hosts_file, { STYLE => 'Unified' };
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { difference => $difference }});
if ($difference)
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 3, key => "job_0112"});
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0112");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "log_0835"});
my $failed = $anvil->Storage->write_file({
debug => 2,
overwrite => 1,
file => $anvil->data->{path}{configs}{hosts},
body => $new_hosts_body,
user => "root",
group => "root",
mode => "0644",
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { failed => $failed }});
}
sleep 1;
}
}
return(0);
}
sub wait_for_access sub wait_for_access
{ {
my ($anvil) = @_; my ($anvil) = @_;
@ -1329,7 +1598,7 @@ sub configure_pacemaker
} }
} }
# Update (if needed) corosync.conf to use the BCN1 and SN1 as knet networks. # Update (if needed) corosync.conf to use the BCN1, MN1 and SN1 as knet networks.
if ($machine eq "node1") if ($machine eq "node1")
{ {
update_progress($anvil, ($anvil->data->{job}{progress} += 1), "job_0344"); update_progress($anvil, ($anvil->data->{job}{progress} += 1), "job_0344");
@ -1363,6 +1632,7 @@ sub check_corosync
my $new_password = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_password}; my $new_password = $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}{anvil_password};
while ($waiting) while ($waiting)
{ {
### TODO: Add MN where appropriate
my $problem = $anvil->Cluster->parse_cib({debug => 3}); my $problem = $anvil->Cluster->parse_cib({debug => 3});
my $peer_ready = $anvil->data->{cib}{parsed}{peer}{ready}; my $peer_ready = $anvil->data->{cib}{parsed}{peer}{ready};
my $peer_name = $anvil->data->{cib}{parsed}{peer}{name}; my $peer_name = $anvil->data->{cib}{parsed}{peer}{name};
@ -2432,11 +2702,6 @@ sub check_local_network
} }
} }
# Update the hosts file.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 3, key => "job_0112"});
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0112");
$anvil->System->update_hosts({debug => 3});
# 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->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 3, key => "job_0113"}); $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 3, key => "job_0113"});
update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0113"); update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0113");

Loading…
Cancel
Save