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

* Finished Cluster->check_server_constraints() and added it to scan-cluster. This now makes sure servers don't roll back to their old host after it has been fenced and recovers.

Browse Source 
* Completely disabled Network->check_network(), it's causing more problems than it solves.

Signed-off-by: Digimer <digimer@alteeve.ca>
main
Digimer 4 years ago
parent 97d20ccd71
commit d3052c0229
5 changed files with 114 additions and 1 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 101
      Anvil/Tools/Cluster.pm
  2. 3
      Anvil/Tools/Network.pm
  3. 4
      scancore-agents/scan-cluster/scan-cluster
  4. 5
      share/words.xml
  5. 2
      tools/anvil-daemon

101
Anvil/Tools/Cluster.pm
Unescape View File

@ -768,7 +768,108 @@ sub check_server_constraints
my $debug = defined $parameter->{debug} ? $parameter->{debug} : 3;
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => $debug, key => "log_0125", variables => { method => "Cluster->check_server_constraints()" }});
# Are we a node?
my $host_type = $anvil->Get->host_type({debug => $debug});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { host_type => $host_type }});
if ($host_type ne "node")
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, priority => "alert", key => "warning_0125"});
return("!!error!!");
}
# Are we in the cluster?
my $problem = $anvil->Cluster->parse_cib({debug => $debug});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { problem => $problem }});
if ($problem)
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, priority => "alert", key => "warning_0126"});
return('!!error!!');
}
# Are we a full member?
if (not $anvil->data->{cib}{parsed}{'local'}{ready})
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, priority => "alert", key => "warning_0127"});
return('!!error!!');
}
# Is our peer offline? If it's online, do nothing
if ($anvil->data->{cib}{parsed}{peer}{ready})
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 2, key => "log_0640"});
return(0);
}
# Get the list of fence methods for my peer and I and make sure their configs are valid.
my $anvil_uuid = $anvil->Cluster->get_anvil_uuid({debug => $debug});
my $anvil_name = $anvil->Get->anvil_name_from_uuid({debug => $debug, anvil_uuid => $anvil_uuid });
my $local_node_name = $anvil->data->{cib}{parsed}{'local'}{name};
my $local_host_uuid = $anvil->Get->host_uuid();
my $peer_node_name = $anvil->data->{cib}{parsed}{peer}{name};
my $peer_host_uuid = $anvil->data->{cib}{parsed}{peer}{host_uuid};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
anvil_uuid => $anvil_uuid,
anvil_name => $anvil_name,
local_node_name => $local_node_name,
local_host_uuid => $local_host_uuid,
peer_node_name => $peer_node_name,
peer_host_uuid => $peer_host_uuid,
}});
foreach my $id (sort {$a cmp $b} keys %{$anvil->data->{cib}{parsed}{configuration}{constraints}{location}})
{
my $node_name = $anvil->data->{cib}{parsed}{configuration}{constraints}{location}{$id}{node};
my $resource = $anvil->data->{cib}{parsed}{configuration}{constraints}{location}{$id}{resource};
my $score = $anvil->data->{cib}{parsed}{configuration}{constraints}{location}{$id}{score};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
id => $id,
node_name => $node_name,
resource => $resource,
score => $score,
}});
$anvil->data->{location_constraint}{resource}{$resource}{node}{$node_name}{score} = $score;
$anvil->data->{location_constraint}{resource}{$resource}{node}{$node_name}{id} = $id;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
"location_constraint::resource::${resource}::node::${node_name}::score" => $anvil->data->{location_constraint}{resource}{$resource}{node}{$node_name}{score},
"location_constraint::resource::${resource}::node::${node_name}::id" => $anvil->data->{location_constraint}{resource}{$resource}{node}{$node_name}{id},
}});
}
# Higher score == preferred
foreach my $resource (sort {$a cmp $b} keys %{$anvil->data->{location_constraint}{resource}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => { resource => $resource }});
my $high_score = 0;
my $preferred_node = "";
foreach my $node_name (sort {$a cmp $b} keys %{$anvil->data->{location_constraint}{resource}{$resource}{node}})
{
my $score = $anvil->data->{location_constraint}{resource}{$resource}{node}{$node_name}{score};
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
node_name => $node_name,
score => $score,
}});
if ($score > $high_score)
{
$high_score = $score;
$preferred_node = $node_name;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
high_score => $high_score,
preferred_node => $preferred_node,
}});
}
if ($local_node_name ne $preferred_node)
{
# Make us the preferred node.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0641", variables => { server => $resource }});
$anvil->Cluster->_set_server_constraint({
server => $resource,
preferred_node => $local_node_name,
});
}
}
}
return(0);
}

3
Anvil/Tools/Network.pm
Unescape View File

@ -261,6 +261,9 @@ sub check_network
my $debug = defined $parameter->{debug} ? $parameter->{debug} : 3;
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => $debug, key => "log_0125", variables => { method => "Network->check_internet()" }});
# This is causing more problems than it solves. Disabled for the time being.
return(0);
my $heal = defined $parameter->{heal} ? $parameter->{heal} : "down_only";
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => $debug, list => {
heal => $heal,

4
scancore-agents/scan-cluster/scan-cluster
Unescape View File

@ -106,6 +106,10 @@ check_config($anvil);
# Check the fence delay
check_fence_delay($anvil);
# Check that the location constraint is sensible.
$anvil->Cluster->check_server_constraints();
$anvil->nice_exit({exit_code => 0});
#############################################################################################################

5
share/words.xml
Unescape View File

@ -1833,6 +1833,8 @@ The file: [#!variable!file!#] needs to be updated. The difference is:
<key name="log_0637">We've got: [#!variable!local_server_count!#] servers, and the peer has: [#!variable!peer_server_count!#] servers. Skipping fence delay preference checks for now.</key>
<key name="log_0638">We're hosting servers, and our peer is not. Making the fence delay favours this node.</key>
<key name="log_0639">The Anvil! daemon is in startup mode, and the job: [#!variable!job_uuid!#], command: [#!variable!job_command!#] is not a startup job, ignoring it for now.</key>
<key name="log_0640">Out peer is online, no need to check server location constraints.</key>
<key name="log_0641">The server: [#!variable!server!#] has a location constraint that preferres our peer, but our peer is offline. Updating the location constraint to prefer this node.</key>
<!-- 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>
@ -2817,6 +2819,9 @@ Read UUID: .... [#!variable!read_uuid!#]
<key name="warning_0122">[ Warning ] - Asked to find or set the fence delay, but this is not a node.</key>
<key name="warning_0123">[ Warning ] - Asked to find or set the fence delay, but node is not in a cluster.</key>
<key name="warning_0124">[ Warning ] - Asked to find or set the fence delay, but node is not fully in the cluster yet.</key>
<key name="warning_0125">[ Warning ] - Asked to check server location constraints, but this is not a node.</key>
<key name="warning_0126">[ Warning ] - Asked to check server location constraints, but this node is not in a cluster.</key>
<key name="warning_0127">[ Warning ] - Asked to check server location constraints, but this node is not fully in the cluster yet.</key>
<!-- The entries below here are not sequential, but use a key to find the entry. -->
<!-- Run 'striker-parse-os-list to find new entries. -->

2
tools/anvil-daemon
Unescape View File

@ -393,7 +393,7 @@ sub check_network
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { return_code => $return_code }});
}
$anvil->Network->check_network({heal => "all"});
#$anvil->Network->check_network({heal => "all"});
$anvil->data->{sys}{network}{initial_checks} = 1;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {

Write Preview
Loading…
Cancel
Save