seasharp
/
anvil
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Updated ocf:alteeve:server to better handle multi-peer DRBD configurations.

Browse Source 
Cleaned up some logging in DRBD->get_status.

Signed-off-by: Digimer <digimer@alteeve.ca>
main
Digimer 3 years ago
parent e62e5d7b0c
commit bc39c3fe5c
3 changed files with 52 additions and 61 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 11
      Anvil/Tools/DRBD.pm
  2. 98
      ocf/alteeve/server
  3. 4
      share/words.xml

11
Anvil/Tools/DRBD.pm
Unescape View File

@ -1648,7 +1648,7 @@ sub get_status
($output, $anvil->data->{drbd}{status}{$host}{return_code}) = $anvil->System->call({shell_call => $shell_call}); ($output, $anvil->data->{drbd}{status}{$host}{return_code}) = $anvil->System->call({shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
output => $output, output => $output,
"drbd::status::${host}::return_code" => $anvil->data->{drbd}{status}{return_code}, "drbd::status::${host}::return_code" => $anvil->data->{drbd}{status}{$host}{return_code},
}}); }});
} }
else else
@ -1666,7 +1666,7 @@ sub get_status
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
error => $error, error => $error,
output => $output, output => $output,
"drbd::status::${host}::return_code" => $anvil->data->{drbd}{status}{return_code}, "drbd::status::${host}::return_code" => $anvil->data->{drbd}{status}{$host}{return_code},
}}); }});
} }
@ -1913,12 +1913,14 @@ sub manage_resource
### can block startup, so to be safe, during start, we'll call adjust ### can block startup, so to be safe, during start, we'll call adjust
if ($task eq "up") if ($task eq "up")
{ {
# This generally brings up the resource
my $shell_call = $anvil->data->{path}{exe}{drbdadm}." adjust ".$resource; my $shell_call = $anvil->data->{path}{exe}{drbdadm}." adjust ".$resource;
my $output = ""; my $output = "";
my $return_code = 255; my $return_code = 255;
if ($anvil->Network->is_local({host => $target})) if ($anvil->Network->is_local({host => $target}))
{ {
# Local. # Local.
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { shell_call => $shell_call }});
($output, $return_code) = $anvil->System->call({shell_call => $shell_call}); ($output, $return_code) = $anvil->System->call({shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
output => $output, output => $output,
@ -1928,6 +1930,7 @@ sub manage_resource
else else
{ {
# Remote call. # Remote call.
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { shell_call => $shell_call }});
($output, my $error, $return_code) = $anvil->Remote->call({ ($output, my $error, $return_code) = $anvil->Remote->call({
debug => $debug, debug => $debug,
shell_call => $shell_call, shell_call => $shell_call,
@ -1944,12 +1947,15 @@ sub manage_resource
} }
} }
# If we 'adjust'ed abovem this will likely complain that the backing disk already exists, and that's
# fine.
my $shell_call = $anvil->data->{path}{exe}{drbdadm}." ".$task." ".$resource; my $shell_call = $anvil->data->{path}{exe}{drbdadm}." ".$task." ".$resource;
my $output = ""; my $output = "";
my $return_code = 255; my $return_code = 255;
if ($anvil->Network->is_local({host => $target})) if ($anvil->Network->is_local({host => $target}))
{ {
# Local. # Local.
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { shell_call => $shell_call }});
($output, $return_code) = $anvil->System->call({shell_call => $shell_call}); ($output, $return_code) = $anvil->System->call({shell_call => $shell_call});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
output => $output, output => $output,
@ -1959,6 +1965,7 @@ sub manage_resource
else else
{ {
# Remote call. # Remote call.
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { shell_call => $shell_call }});
($output, my $error, $return_code) = $anvil->Remote->call({ ($output, my $error, $return_code) = $anvil->Remote->call({
debug => $debug, debug => $debug,
shell_call => $shell_call, shell_call => $shell_call,

98
ocf/alteeve/server
Unescape View File

@ -792,14 +792,11 @@ sub start_drbd_resource
my $local_host = $anvil->Get->short_host_name(); my $local_host = $anvil->Get->short_host_name();
my $server = $anvil->data->{environment}{OCF_RESKEY_name}; my $server = $anvil->data->{environment}{OCF_RESKEY_name};
my $host = $anvil->Get->short_host_name; my $host = $anvil->Get->short_host_name;
my $peer = $anvil->data->{drbd}{config}{$host}{peer};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
server => $server, server => $server,
host => $host, host => $host,
peer => $peer,
}}); }});
### TODO: Local start up below, move the peer check to have local startup is handled
# Do we need startup? # Do we need startup?
my $local_startup_needed = 0; my $local_startup_needed = 0;
$anvil->DRBD->get_status({debug => 3}); $anvil->DRBD->get_status({debug => 3});
@ -827,76 +824,65 @@ sub start_drbd_resource
} }
} }
# Do I need to start the DRBD resource locally? If so, do so.
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { local_startup_needed => $local_startup_needed }}); $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { local_startup_needed => $local_startup_needed }});
if ($local_startup_needed) if ($local_startup_needed)
{ {
foreach my $resource (sort {$a cmp $b} keys %{$anvil->data->{server}{$local_host}{$server}{resource}}) foreach my $resource (sort {$a cmp $b} keys %{$anvil->data->{server}{$local_host}{$server}{resource}})
{ {
my $peer_ip = $anvil->data->{drbd}{config}{$host}{resource}{$resource}{connection}{$peer}{ip_address};
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0419", variables => {
server => $server,
peer => $peer,
peer_ip => $peer_ip,
resource => $resource,
}});
# Bring the local resource up # Bring the local resource up
$anvil->DRBD->manage_resource({ $anvil->DRBD->manage_resource({
debug => 2,
resource => $resource, resource => $resource,
task => "up", task => "up",
}); });
# Bring the peer's resource up. # Now wait for it to come up.
$anvil->DRBD->manage_resource({ my $waiting = 1;
resource => $resource, my $wait_until = time + 10;
task => "up", $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
target => $peer_ip, 's1:time' => time,
}); 's2:wait_until' => $wait_until,
}});
# Now wait for it to be connected or UpToDate...
my $waiting = 1;
while($waiting) while($waiting)
{ {
$anvil->DRBD->get_status({debug => 3}); $anvil->DRBD->get_status({debug => 3});
print "==] ".$local_host." [==] ".$resource." [==] ".$peer." [==\n"; my $all_up = 1;
print Dumper $anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{connection}{$peer};
print "=========================================================\n";
my $connection_state = $anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{connection}{$peer}{'connection-state'};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { connection_state => $connection_state }});
my $all_ready = 1;
foreach my $volume (sort {$a cmp $b} keys %{$anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{devices}{volume}}) foreach my $volume (sort {$a cmp $b} keys %{$anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{devices}{volume}})
{ {
my $disk_state = $anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{devices}{volume}{$volume}{'disk-state'}; my $disk_state = lc($anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{devices}{volume}{$volume}{'disk-state'});
my $replication_state = $anvil->data->{drbd}{status}{$local_host}{resource}{$resource}{connection}{$peer}{volume}{$volume}{'replication-state'}; $disk_state = "" if not defined $disk_state;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
disk_state => $disk_state, 's1:resource' => $resource,
replication_state => $replication_state, 's2:volume' => $volume,
's3:disk_state' => $disk_state,
}}); }});
# Is the peer isn't connected (directly or by being in Sync), or this volume if (($disk_state ne "inconsistent") &&
# isn't UpToDate, we need to keep waiting. ($disk_state ne "outdated") &&
if ((lc($disk_state) ne "uptodate") && ($replication_state !~ /^Sync/i) && (lc($connection_state) ne "connected")) ($disk_state ne "consistent") &&
($disk_state ne "uptodate"))
{ {
$all_ready = 0; $all_up = 0;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { all_ready => $all_ready }}); $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { all_up => $all_up }});
} }
} }
die; if ($all_up)
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { all_ready => $all_ready }});
if ($all_ready)
{ {
$waiting = 0; $waiting = 0;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { waiting => $waiting }}); $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { waiting => $waiting }});
} }
if ($waiting) if ($waiting)
{ {
sleep 1; sleep 2;
}
elsif (time > $wait_until)
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 0, priority => "alert", key => "warning_0138"});
$waiting = 0;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { waiting => $waiting }});
} }
} }
} }
@ -920,7 +906,7 @@ sub start_drbd_resource
} }
# See if we're inconsistent and, if so, if we can connect our peers. # See if we're inconsistent and, if so, if we can connect our peers.
sleep 5; sleep 2;
$anvil->DRBD->get_status({debug => 3}); $anvil->DRBD->get_status({debug => 3});
my $peer_startup_needed = 0; my $peer_startup_needed = 0;
foreach my $resource (sort {$a cmp $b} keys %{$anvil->data->{server}{$local_host}{$server}{resource}}) foreach my $resource (sort {$a cmp $b} keys %{$anvil->data->{server}{$local_host}{$server}{resource}})
@ -986,20 +972,16 @@ sub start_drbd_resource
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { access => $access }}); $anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { access => $access }});
if ($access) if ($access)
{ {
my ($output, $error, $return_code) = $anvil->Remote->call({
target => $host,
shell_call => $anvil->data->{path}{exe}{drbdadm}." up ".$resource,
});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
output => $output,
error => $error,
return_code => $return_code,
}});
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0695", variables => { $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0695", variables => {
return_code => $return_code, host => $host,
error => $error, resource => $resource,
output => $output,
}}); }});
$anvil->DRBD->manage_resource({
debug => 2,
resource => $resource,
task => "up",
target => $host,
});
} }
else else
{ {

4
share/words.xml
Unescape View File

@ -2087,7 +2087,8 @@ The file: [#!variable!file!#] needs to be updated. The difference is:
<key name="log_0692">The anvil-safe-stop job has completed and will now power off.</key> <key name="log_0692">The anvil-safe-stop job has completed and will now power off.</key>
<key name="log_0693">The anvil-configure-host tool is requesting a reboot.</key> <key name="log_0693">The anvil-configure-host tool is requesting a reboot.</key>
<key name="log_0694">The connection to: [#!variable!host!#] for the resource: [#!variable!resource!#] is in the connection state: [#!variable!connection_state!#]. Will try to connect to the peer and up the resource now.</key> <key name="log_0694">The connection to: [#!variable!host!#] for the resource: [#!variable!resource!#] is in the connection state: [#!variable!connection_state!#]. Will try to connect to the peer and up the resource now.</key>
<key name="log_0695">The request to start the resource had the return code: [#!variable!return_code!#]. Call output, if any, was: [#!variable!output!#]. Errors, if any, were: [#!variable!error!#].</key> <key name="log_0695">About to request the start of the resource: [#variable!resource!#] on: [#!variable!host!#].</key>
<key name="log_0696">The peer: [#!variable!peer!#] is defined in the resource: [#!variable!resource!#] but we don't connect to it, ignoring it.</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>
@ -3121,6 +3122,7 @@ We will sleep a bit and try again.
<key name="warning_0135">[ Warning ] - The program: [#!variable!program!#] was not found to be running.</key> <key name="warning_0135">[ Warning ] - The program: [#!variable!program!#] was not found to be running.</key>
<key name="warning_0136">[ Warning ] - Failed to connect to the host: [#!variable!host!#]! Unable to up the resource, so the server may not start. If the peer can't be recovered, manually forcing the local resource(s) to UpToDate may be required.</key> <key name="warning_0136">[ Warning ] - Failed to connect to the host: [#!variable!host!#]! Unable to up the resource, so the server may not start. If the peer can't be recovered, manually forcing the local resource(s) to UpToDate may be required.</key>
<key name="warning_0137">[ Warning ] - Timed out waiting for the connections to the peers, and the local resource(s) is not in 'UpToDate' state. Booting the server will likely fail.</key> <key name="warning_0137">[ Warning ] - Timed out waiting for the connections to the peers, and the local resource(s) is not in 'UpToDate' state. Booting the server will likely fail.</key>
<key name="warning_0138">[ Warning ] - Timed out waiting for the connections to the peers.</key>
<!-- The entries below here are not sequential, but use a key to find the entry. --> <!-- The entries below here are not sequential, but use a key to find the entry. -->
<!-- Run 'striker-parse-os-list to find new entries. --> <!-- Run 'striker-parse-os-list to find new entries. -->

Write Preview
Loading…
Cancel
Save