AMBARI-5990. Nagios templates for 1.x and 2.x stacks contain irrelevant services (dlysnichenko)

ambari-server/src/main/resources/stacks/HDP/1.3.2/services/NAGIOS/package/scripts/params.py

@@ -56,8 +56,6 @@ namenode_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.htt
 snamenode_port = get_port_from_url(config['configurations']['hdfs-site']["dfs.secondary.http.address"])
 
 hbase_master_rpc_port = "60000"
-rm_port = get_port_from_url(config['configurations']['yarn-site']['yarn.resourcemanager.webapp.address'])
-nm_port = "8042"
 hs_port = get_port_from_url(config['configurations']['mapred-site']['mapreduce.history.server.http.address'])
 journalnode_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.journalnode.http-address'])
 datanode_port = get_port_from_url(config['configurations']['hdfs-site']['dfs.datanode.http.address'])
@@ -121,8 +119,6 @@ _slave_hosts = default("/clusterHostInfo/slave_hosts", None)
 _tt_hosts = default("/clusterHostInfo/mapred_tt_hosts", [])
 _journalnode_hosts = default("/clusterHostInfo/journalnode_hosts", None)
 _zkfc_hosts = default("/clusterHostInfo/zkfc_hosts", None)
-_rm_host = default("/clusterHostInfo/rm_host", None)
-_nm_hosts = default("/clusterHostInfo/nm_hosts", None)
 _hs_host = default("/clusterHostInfo/hs_host", None)
 _zookeeper_hosts = default("/clusterHostInfo/zookeeper_hosts", None)
 _flume_hosts = default("/clusterHostInfo/flume_hosts", None)
@@ -159,8 +155,6 @@ hostgroup_defs = {
     'oozie-server' : _oozie_server,
     'webhcat-server' : _webhcat_server_host,
     'hue-server' : _hue_server_host,
-    'resourcemanager' : _rm_host,
-    'nodemanagers' : _nm_hosts,
     'historyserver2' : _hs_host,
     'jobhistory': _hs_host,
     'journalnodes' : _journalnode_hosts

ambari-server/src/main/resources/stacks/HDP/1.3.2/services/NAGIOS/package/templates/hadoop-servicegroups.cfg.j2

@@ -50,13 +50,7 @@ define servicegroup {
   alias  MAPREDUCE Checks
 }
 {% endif %}
-{%if hostgroup_defs['resourcemanager'] or
-  hostgroup_defs['nodemanagers'] %}
-define servicegroup {
-  servicegroup_name  YARN
-  alias  YARN Checks
-}
-{% endif %}
+
 {%if hostgroup_defs['flume-servers'] %}
 define servicegroup {
   servicegroup_name  FLUME
@@ -114,4 +108,4 @@ define servicegroup {
   servicegroup_name  HUE
   alias  HUE Checks
 }
-{% endif %}
+{% endif %}

ambari-server/src/main/resources/stacks/HDP/1.3.2/services/NAGIOS/package/templates/hadoop-services.cfg.j2

@@ -188,20 +188,6 @@ define service {
 {% endfor %}
 {% endif %}
 
-{% if hostgroup_defs['resourcemanager'] %}
-{% for hostname in hostgroup_defs['resourcemanager'] %}
-define service {
-        host_name	        {{ hostname }}
-        use                     hadoop-service
-        service_description     GANGLIA::Ganglia Monitor process for ResourceManager
-        servicegroups           GANGLIA
-        check_command           check_tcp_wrapper!{{ ganglia_collector_rm_port }}!-w 1 -c 1
-        normal_check_interval   0.25
-        retry_check_interval    0.25
-        max_check_attempts      4
-}
-{% endfor %}
-{% endif %}
 
 {% if hostgroup_defs['historyserver2'] %}
 {% for hostname in hostgroup_defs['historyserver2'] %}
@@ -422,145 +408,6 @@ define service {
 {% endif %}
 {% endif %}
 
-{% if hostgroup_defs['resourcemanager'] %}
-# YARN::RESOURCEMANAGER Checks 
-define service {
-        hostgroup_name          resourcemanager
-        use                     hadoop-service
-        service_description     RESOURCEMANAGER::ResourceManager Web UI
-        servicegroups           YARN
-        check_command           check_webui!resourcemanager!{{ rm_port }}
-        normal_check_interval   1
-        retry_check_interval    1
-        max_check_attempts      3
-}
-
-{% if check_cpu_on %}
-define service {
-        hostgroup_name          resourcemanager
-        use                     hadoop-service
-        service_description     RESOURCEMANAGER::ResourceManager CPU utilization
-        servicegroups           YARN
-#        check_command           check_cpu!200%!250%
-        check_command           check_cpu!{{ rm_port }}!200%!250%!{{ str(hadoop_ssl_enabled).lower() }}!{{ nagios_keytab_path }}!{{ nagios_principal_name }}!{{ kinit_path_local }}!{{ str(security_enabled).lower() }}
-        normal_check_interval   5
-        retry_check_interval    2 
-        max_check_attempts      5
-}
-{% endif %}
-
-define service {
-        hostgroup_name          resourcemanager
-        use                     hadoop-service
-        service_description     RESOURCEMANAGER::ResourceManager RPC latency
-        servicegroups           YARN
-        check_command           check_rpcq_latency!ResourceManager!{{ rm_port }}!3000!5000!{{ str(hadoop_ssl_enabled).lower() }}!{{ nagios_keytab_path }}!{{ nagios_principal_name }}!{{ kinit_path_local }}!{{ str(security_enabled).lower() }}
-        normal_check_interval   5
-        retry_check_interval    1 
-        max_check_attempts      5
-}
-
-define service {
-        hostgroup_name          resourcemanager
-        use                     hadoop-service
-        service_description     RESOURCEMANAGER::ResourceManager process
-        servicegroups           YARN
-        check_command           check_tcp_wrapper!{{ rm_port }}!-w 1 -c 1
-        normal_check_interval   1
-        retry_check_interval    0.5
-        max_check_attempts      3
-}
-{%  endif %}
-
-{% if hostgroup_defs['nodemanagers'] %}
-# YARN::NODEMANAGER Checks
-define service {
-        hostgroup_name          nodemanagers
-        use                     hadoop-service
-        service_description     NODEMANAGER::NodeManager process
-        servicegroups           YARN
-        check_command           check_tcp_wrapper!{{ nm_port }}!-w 1 -c 1
-        normal_check_interval   1
-        retry_check_interval    0.5
-        max_check_attempts      3
-}
-
-define service {
-        hostgroup_name          nodemanagers
-        use                     hadoop-service
-        service_description     NODEMANAGER::NodeManager health
-        servicegroups           YARN
-        check_command           check_nodemanager_health!{{ nm_port }}!{{ str(security_enabled).lower() }}!{{ str(hadoop_ssl_enabled).lower() }}!{{ nagios_keytab_path }}!{{ nagios_principal_name }}!{{ kinit_path_local }}
-        normal_check_interval   1
-        retry_check_interval    1
-        max_check_attempts      3
-}
-define service {
-        hostgroup_name          nagios-server
-        use                     hadoop-service
-        service_description     NODEMANAGER::Percent NodeManagers live
-        servicegroups           YARN
-        check_command           check_aggregate!"NODEMANAGER::NodeManager process"!10%!30%
-        normal_check_interval   0.5
-        retry_check_interval    0.25
-        max_check_attempts      3
-}
-{%  endif %}
-
-{% if hostgroup_defs['historyserver2'] %}
-# MAPREDUCE::JOBHISTORY Checks
-define service {
-        hostgroup_name          historyserver2
-        use                     hadoop-service
-        service_description     JOBHISTORY::HistoryServer Web UI
-        servicegroups           MAPREDUCE
-        check_command           check_webui!historyserver2!{{ hs_port }}
-        normal_check_interval   1
-        retry_check_interval    1
-        max_check_attempts      3
-}
-
-{% if check_cpu_on %}
-define service {
-        hostgroup_name          historyserver2
-        use                     hadoop-service
-        service_description     JOBHISTORY::HistoryServer CPU utilization
-        servicegroups           MAPREDUCE
-#       check_command           check_cpu!200%!250%
-        check_command           check_cpu!{{ hs_port }}!200%!250%!{{ str(hadoop_ssl_enabled).lower() }}!{{ nagios_keytab_path }}!{{ nagios_principal_name }}!{{ kinit_path_local }}!{{ str(security_enabled).lower() }}
-        normal_check_interval   5
-        retry_check_interval    2 
-        max_check_attempts      5
-}
-{%  endif %}
-
-define service {
-        hostgroup_name          historyserver2
-        use                     hadoop-service
-        service_description     JOBHISTORY::HistoryServer process
-        servicegroups           MAPREDUCE
-        check_command           check_tcp_wrapper!{{ hs_port }}!-w 1 -c 1
-        normal_check_interval   1
-        retry_check_interval    0.5
-        max_check_attempts      3
-}
-
-{%  endif %}
-
-{% if hostgroup_defs['journalnodes'] %}
-# Journalnode checks
-define service {
-        hostgroup_name          journalnodes
-        use                     hadoop-service
-        service_description     JOURNALNODE::JournalNode process
-        servicegroups           HDFS
-        check_command           check_tcp_wrapper!{{ journalnode_port }}!-w 1 -c 1
-        normal_check_interval   1
-        retry_check_interval    0.5
-        max_check_attempts      3
-}
-
-{% endif %}
 
 {% if hostgroup_defs['slaves'] %}
 # HDFS::DATANODE Checks

ambari-server/src/main/resources/stacks/HDP/2.0.6/services/NAGIOS/package/scripts/params.py

@@ -171,8 +171,6 @@ hostgroup_defs = {
     'namenode' : namenode_host,
     'snamenode' : _snamenode_host,
     'slaves' : _slave_hosts,
-    # HDP1
-    #'tasktracker-servers' : _mapred_tt_hosts,
     'agent-servers' : all_hosts,
     'nagios-server' : _nagios_server_host,
     'jobtracker' : _jtnode_host,

ambari-server/src/main/resources/stacks/HDP/2.0.6/services/NAGIOS/package/templates/hadoop-services.cfg.j2

@@ -173,21 +173,6 @@ define service {
 {% endfor %}
 {% endif %}
 
-{% if hostgroup_defs['jobtracker'] %}
-{% for hostname in hostgroup_defs['jobtracker'] %}
-define service {
-        host_name	        {{ hostname }}
-        use                     hadoop-service
-        service_description     GANGLIA::Ganglia Monitor process for JobTracker
-        servicegroups           GANGLIA
-        check_command           check_tcp_wrapper!{{ ganglia_collector_jobtracker_port }}!-w 1 -c 1
-        normal_check_interval   0.25
-        retry_check_interval    0.25
-        max_check_attempts      4
-}
-{% endfor %}
-{% endif %}
-
 {% if hostgroup_defs['hbasemasters'] %}
 {% for hostname in hostgroup_defs['hbasemasters'] %}
 define service {
@@ -427,7 +412,7 @@ define service {
         servicegroups           HDFS
         check_command           check_hdfs_blocks!$HOSTGROUPMEMBERS:namenode$!{{ namenode_port }}!{{ nn_metrics_property }}!{{ str(hadoop_ssl_enabled).lower() }}!{{ nagios_keytab_path }}!{{ nagios_principal_name }}!{{ kinit_path_local }}!{{ str(security_enabled).lower() }}
         normal_check_interval   2
-        retry_check_interval    1 
+        retry_check_interval    1
         max_check_attempts      1
 }
 
@@ -438,15 +423,12 @@ define service {
         servicegroups           HDFS
         check_command           check_hdfs_capacity!$HOSTGROUPMEMBERS:namenode$!{{ namenode_port }}!80%!90%!{{ str(hadoop_ssl_enabled).lower() }}!{{ nagios_keytab_path }}!{{ nagios_principal_name }}!{{ kinit_path_local }}!{{ str(security_enabled).lower() }}
         normal_check_interval   2
-        retry_check_interval    1 
+        retry_check_interval    1
         max_check_attempts      1
 }
 
 {% endif %}
 
-# MAPREDUCE Checks
-{# On HDP1 here are jobtracker and tasktracker alters #}
-
 {% if hostgroup_defs['resourcemanager'] %}
 # YARN::RESOURCEMANAGER Checks 
 define service {