Merge pull request #77 from ARGOeu/devel

Version 0.4.4
ARGOeu · Jun 19, 2018 · 0eb56ea · 0eb56ea
2 parents 4e578b2 + a9bd4a5
commit 0eb56ea
Show file tree

Hide file tree

Showing 11 changed files with 145 additions and 39 deletions.
diff --git a/argo-ncg.spec b/argo-ncg.spec
@@ -5,7 +5,7 @@
 
 Summary: ARGO Nagios config generator
 Name: argo-ncg
-Version: 0.4.3
+Version: 0.4.4
 Release: 1%{?dist}
 License: ASL 2.0
 Group: Network/Monitoring
@@ -14,6 +14,10 @@ Obsoletes: grid-monitoring-config-gen-nagios grid-monitoring-config-gen ncg-metr
 BuildRoot: %{_tmppath}/%{name}-%{version}-%{release}-root
 BuildArch: noarch
 Requires: perl-libwww-perl > 5.833-2
+Requires: psmisc
+%if 0%{?el7:1}
+Requires: perl(LWP::Protocol::https)
+%endif
 
 %description
 (NULL)

diff --git a/config/ncg-metric-config.d/argo.conf b/config/ncg-metric-config.d/argo.conf
@@ -0,0 +1,39 @@
+{
+   "argo.AMSPublisher-Check" : {
+      "parameter" : {
+         "-s" : "/var/run/argo-nagios-ams-publisher/sock",
+         "-q" : "'w:metrics+g:published360' -c 8000 -q 'w:alarms+g:published360' -c 1 -q 'w:metricsdevel+g:published360' -c 8000"
+      },
+      "config" : {
+         "interval" : 360,
+         "maxCheckAttempts" : 1,
+         "path" : "/usr/libexec/argo-monitoring/probes/argo",
+         "retryInterval" : 1,
+         "timeout" : 120
+      },
+      "flags" : {
+         "NOHOSTNAME" : 1,
+         "NOTIMEOUT" : 1
+      },
+      "probe" : "ams-publisher-probe"
+   },
+   "org.nagios.AmsDirSize" : {
+      "parameter" : {
+         "-d" : "/var/spool/argo-nagios-ams-publisher",
+         "-w" : 10000,
+         "-c" : 100000,
+         "-f" : 0
+      },
+      "config" : {
+         "interval" : 60,
+         "maxCheckAttempts" : 3,
+         "path" : "/usr/libexec/argo-monitoring/probes/nagiosexchange",
+         "retryInterval" : 5,
+         "timeout" : 15
+      },
+      "flags" : {
+         "NOHOSTNAME" : 1
+      },
+      "probe" : "check_dirsize.sh"
+   }
+}
diff --git a/config/ncg-metric-config.d/midmon.conf b/config/ncg-metric-config.d/midmon.conf
@@ -70,6 +70,30 @@
       },
       "probe" : "midmon/check_bdii_entries_num"
    },
+   "eu.egi.sec.dCache-3.0" : {
+      "attribute" : {
+         "SITE_BDII" : "-H",
+         "BDII_PORT" : "-p"
+      },
+      "parameter" : {
+         "-w" : "0:0",
+         "-f" : "\"(&(GlueSEImplementationName=dCache)(GlueSEImplementationVersion=3.0.*)(GlueSEUniqueID=*$HOSTNAME$*))\"",
+         "-b" : "O=grid"
+      },
+      "flags" : {
+         "NOHOSTNAME" : 1,
+         "OBSESS" :1
+      },
+      "docurl" : "https://wiki.egi.eu/wiki/MW_Nagios_tests",
+      "config" : {
+         "maxCheckAttempts" : 1,
+         "timeout" : 120,
+         "path" : "/usr/libexec/argo-monitoring/probes",
+         "retryInterval" : 15,
+         "interval" : 1440
+      },
+      "probe" : "midmon/check_bdii_entries_num"
+   },
    "org.bdii.GLUE2-Validate" : {
       "flags" : {
          "NRPE" : 1,

diff --git a/config/templates/commands.template b/config/templates/commands.template
@@ -102,7 +102,7 @@ define command{
 
 define command {
         command_name handle_service_check
-        command_line /usr/libexec/argo-msg-nagios/handle_service_check  --role=<NAGIOS_ROLE> --tenant=<TENANT> --send-to-msg=<SEND_TO_MSG> && /usr/bin/ams-metric-to-queue --queue /var/spool/argo-nagios-ams-publisher/metrics/ /var/spool/argo-nagios-ams-publisher/metricsdevel/ --service "$_SERVICESERVICE_FLAVOUR$" --hostname "$HOSTNAME$" --metric "$_SERVICEMETRIC_NAME$" --status "$SERVICESTATE$" --summary "$SERVICEOUTPUT$" --message "$LONGSERVICEOUTPUT$" --vofqan "$_SERVICEVO_FQAN$" --voname "$_SERVICEVO$" --roc "$_SERVICEROC$" --servicestatetype "$SERVICESTATETYPE$"
+        command_line /usr/libexec/argo-msg-nagios/handle_service_check  --role=<NAGIOS_ROLE> --tenant=<TENANT> --send-to-msg=<SEND_TO_MSG> && /usr/bin/ams-metric-to-queue --queue /var/spool/argo-nagios-ams-publisher/metrics/ /var/spool/argo-nagios-ams-publisher/metricsdevel/ --service "$_SERVICESERVICE_FLAVOUR$" --hostname "$HOSTNAME$" --metric "$_SERVICEMETRIC_NAME$" --status "$SERVICESTATE$" --summary "$SERVICEOUTPUT$" --message "$LONGSERVICEOUTPUT$" --vofqan "$_SERVICEVO_FQAN$" --voname "$_SERVICEVO$" --roc "$_SERVICEROC$" --servicestatetype "$SERVICESTATETYPE$" --actual_data "$SERVICEPERFDATA$"
 }
 
 define command {

diff --git a/src/modules/NCG/LocalMetrics/Hash.pm b/src/modules/NCG/LocalMetrics/Hash.pm
@@ -426,19 +426,6 @@ $WLCG_SERVICE->{'org.nagios.MsgDirSize'}->{parameter}->{'-w'} = '10000';
 $WLCG_SERVICE->{'org.nagios.MsgDirSize'}->{parameter}->{'-c'} = '100000';
 $WLCG_SERVICE->{'org.nagios.MsgDirSize'}->{parameter}->{'-f'} = '';
 
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{probe} = "check_dirsize.sh";
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{config}->{timeout} = 15;
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{config}->{interval} = 60;
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{config}->{retryInterval} = 5;
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{config}->{maxCheckAttempts} = 3;
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{config}->{path} = '/usr/libexec/argo-monitoring/probes/nagiosexchange';
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{flags}->{NOHOSTNAME} = 1;
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{flags}->{PNP} = 1;
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{parameter}->{'-d'} = '/var/spool/argo-nagios-ams-publisher';
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{parameter}->{'-w'} = '10000';
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{parameter}->{'-c'} = '100000';
-$WLCG_SERVICE->{'org.nagios.AmsDirSize'}->{parameter}->{'-f'} = '';
-
 $WLCG_SERVICE->{'org.nagios.ProcessMsgToHandler'}->{probe} = 'check_procs';
 $WLCG_SERVICE->{'org.nagios.ProcessMsgToHandler'}->{config}->{path} = $NCG::NCG_PROBES_PATH_NAGIOS;
 $WLCG_SERVICE->{'org.nagios.ProcessMsgToHandler'}->{config}->{interval} = 15;
@@ -602,6 +589,7 @@ $WLCG_NODETYPE->{internal}->{"NAGIOS"} = [
 'emi.wms.WMS-JobMonit',
 'org.nordugrid.ARC-CE-monitor',
 'org.nordugrid.ARC-CE-clean',
+'argo.AMSPublisher-Check',
 ];
 
 $WLCG_NODETYPE->{internal}->{"NRPE"} = [

diff --git a/src/modules/NCG/LocalMetricsAttrs/Active.pm b/src/modules/NCG/LocalMetricsAttrs/Active.pm
@@ -379,6 +379,10 @@ sub _analyzeURLs {
     }
 
     if ($attr = $self->{SITEDB}->hostAttribute($hostname, "eu.egi.cloud.vm-management.occi_URL")) {
+        my $port;
+        if ($attr =~ /(\S+?:\/\/)?([-_.A-Za-z0-9]+):(\d+)/ ) {
+           $port = $3;
+        }
         eval {my $occiHash = {};
         my $occiurl = url($attr);
         $self->{SITEDB}->hostAttribute($hostname, 'OCCI_PORT', $occiurl->port);
@@ -392,7 +396,10 @@ sub _analyzeURLs {
             $occiHash->{$key} = $value;
         }
         $self->{SITEDB}->hostAttribute($hostname, 'OCCI_SCHEME', $occiurl->scheme);
-        $self->{SITEDB}->hostAttribute($hostname, 'OCCI_URL', $occiurl->scheme."://".$occiurl->host.":".$occiurl->port.$occiurl->epath);
+        my $occiurlAttr = $occiurl->scheme."://".$occiurl->host;
+        $occiurlAttr .= ":" . $port if ($port);
+        $occiurlAttr .= $occiurl->epath;
+        $self->{SITEDB}->hostAttribute($hostname, 'OCCI_URL', $occiurlAttr);
         if (!exists $occiHash->{OCCI_RESOURCE}) {
             if (!exists $occiHash->{OCCI_PLATFORM}) {
                 $self->{SITEDB}->hostAttribute($hostname, 'OCCI_RESOURCE', 'small');

diff --git a/src/modules/NCG/SiteContacts/GOCDB.pm b/src/modules/NCG/SiteContacts/GOCDB.pm
@@ -141,6 +141,11 @@ sub getData {
     }
 
     my $req = HTTP::Request->new(GET => $url);
+
+    if ( $self->{USERNAME} && $self->{PASSWORD} ) {
+        $req->authorization_basic($self->{USERNAME}, $self->{PASSWORD});
+    }
+
     my $res = $self->safeHTTPSCall($ua,$req);
     if (!$res->is_success) {
         $self->error("Could not get results from GOCDB: ".$res->status_line);
@@ -209,8 +214,14 @@ can contains following elements:
 
   ROC - roc name must be set in case when CONTACT_TYPE is set to 'roc'.
   
-  TIMEOUT - HTTP timeout,
-  (default: DEFAULT_HTTP_TIMEOUT inherited from NCG)
+  TIMEOUT - HTTP timeout
+          - default: DEFAULT_HTTP_TIMEOUT inherited from NCG
+
+  USERNAME - username for basic authentication
+           - default: undefined
+
+  PASSWORD - password for basic authentication
+           - default: undefined
 
 =item C<getData>
 

diff --git a/src/modules/NCG/SiteInfo/GOCDB.pm b/src/modules/NCG/SiteInfo/GOCDB.pm
@@ -90,6 +90,10 @@ sub getData {
         }
         my $req = HTTP::Request->new(GET => $url);
 
+        if ( $self->{USERNAME} && $self->{PASSWORD} ) {
+            $req->authorization_basic($self->{USERNAME}, $self->{PASSWORD});
+        }
+
         my $res = $self->safeHTTPSCall($ua,$req);
         if (!$res->is_success) {
             $self->error("Could not get results from GOCDB: ".$res->status_line);
@@ -208,6 +212,26 @@ sub getData {
                     }
                 }
             }
+            foreach $elem ($site->getElementsByTagName("EXTENSIONS")) {
+                foreach my $ext ($elem->getElementsByTagName("EXTENSION")) {
+                    my ($elemInt, $keyInt, $valueInt);
+                    foreach $elemInt ($ext->getElementsByTagName("KEY")) {
+                        my $value = $elemInt->getFirstChild->getNodeValue();
+                        if ($value) {
+                            $keyInt = $value;
+                        }
+                    }
+                    foreach $elemInt ($ext->getElementsByTagName("VALUE")) {
+                        my $value = $elemInt->getFirstChild->getNodeValue();
+                        if ($value) {
+                            $valueInt = $value;
+                        }
+                    }
+                    if ($keyInt && $valueInt) {
+                        $self->{SITEDB}->hostAttribute($hostname, $keyInt, $valueInt);
+                    }
+                }
+            }
         }
 
         if (!$self->{SITEDB}->siteCountry()) {
@@ -267,8 +291,14 @@ can contains following elements:
   SCOPE - scope of services (for possible values see GOCDB documentation)
         - default: undefined
              
-  TIMEOUT - HTTP timeout,
-  (default: DEFAULT_HTTP_TIMEOUT inherited from NCG)
+  TIMEOUT - HTTP timeout
+          - default: DEFAULT_HTTP_TIMEOUT inherited from NCG
+
+  USERNAME - username for basic authentication
+           - default: undefined
+
+  PASSWORD - password for basic authentication
+           - default: undefined
 
 =back
 

diff --git a/src/modules/NCG/SiteSet/GOCDB.pm b/src/modules/NCG/SiteSet/GOCDB.pm
@@ -111,6 +111,11 @@ sub getData
         }
 
         my $req = HTTP::Request->new(GET => $url);
+
+        if ( $self->{USERNAME} && $self->{PASSWORD} ) {
+            $req->authorization_basic($self->{USERNAME}, $self->{PASSWORD});
+        }
+
         my $res = $self->safeHTTPSCall($ua,$req);
         if (!$res->is_success) {
             $self->error("Could not get results from GOCDB: ".$res->status_line);
@@ -208,8 +213,14 @@ can contains following elements:
 
   SUBGRID_EXCLUDE - comma separated list of Subgrids to exclude.
               
-  TIMEOUT - HTTP timeout,
-  (default: DEFAULT_HTTP_TIMEOUT inherited from NCG)
+  TIMEOUT - HTTP timeout
+          - default: DEFAULT_HTTP_TIMEOUT inherited from NCG
+
+  USERNAME - username for basic authentication
+           - default: undefined
+
+  PASSWORD - password for basic authentication
+           - default: undefined
 
 =back
 

diff --git a/src/ncg.pl b/src/ncg.pl
@@ -262,8 +262,7 @@ sub getConfig {
     my %options = (-ConfigFile => $CONFIG_FILE, -AllowMultiOptions => 1,
                    -InterPolateVars => 1, -InterPolateEnv => 1,
                    -IncludeGlob => 1, -UseApacheInclude => 1, 
-                   -IncludeRelative => 1, -IncludeDirectories => 1, 
-                   -AutoTrue => 1);
+                   -IncludeRelative => 1, -IncludeDirectories => 1);
     my $config = new Config::General(%options);
     if (!$config) {
         alarm(0);

diff --git a/src/ncg.reload.sh b/src/ncg.reload.sh
@@ -1,6 +1,5 @@
 #!/bin/sh
 
-NAGIOS_RUNNING=1
 NCG_TIMEOUT=1800
 OUTPUT_DIR_TMP=/etc/nagios/argo-ncg.d.tmp.$$
 CONFIG_FILE_TMP=/etc/nagios/nagios.cfg.tmp.$$
@@ -20,17 +19,15 @@ revert_config_and_exit () {
   exit 1
 }
 
-# check if nagios is running at all
-service nagios status 2>&1 > /dev/null
-# status will return 1 if not running
-if [ $? -ne 0 ]; then
-    NAGIOS_RUNNING=0
-fi
+# kill all running ncg.pl instances
+killall -9 ncg.pl
+# clean any tmp generated by ncg.reload.sh
+rm -rf /etc/nagios/argo-ncg.d.tmp.*
 
 /usr/sbin/ncg.pl --timeout $NCG_TIMEOUT --output-dir=$OUTPUT_DIR_TMP --final-output-dir=/etc/nagios/argo-ncg.d $NCG_OPTIONS $NCG_BACKUP_OPTIONS || revert_config_and_exit
 
 sed "s|/etc/nagios/argo-ncg.d|$OUTPUT_DIR_TMP|" /etc/nagios/nagios.cfg > $CONFIG_FILE_TMP
-/usr/bin/nagios -v $CONFIG_FILE_TMP || revert_config_and_exit
+nagios -v $CONFIG_FILE_TMP || revert_config_and_exit
 
 # remove temp
 rm -rf $CONFIG_FILE_TMP
@@ -40,10 +37,6 @@ rm -rf /etc/nagios/argo-ncg.d.backup
 mv /etc/nagios/argo-ncg.d /etc/nagios/argo-ncg.d.backup
 mv $OUTPUT_DIR_TMP /etc/nagios/argo-ncg.d
 
-if [ $NAGIOS_RUNNING -eq 1 ]; then
-  /sbin/service nagios reload
-else
-  # here we try to start nagios, continue running even if it fails
-  echo "Nagios is not running, attempting to start it"
-  /sbin/service nagios start
-fi
+# Init script doesn't report status properly
+# safer way is to restart
+service nagios restart