manage telegraf scripts with a defaults file assigned per node type

salt/telegraf/config.sls

@@ -32,17 +32,16 @@ tgrafetsdir:
     - name: /opt/so/conf/telegraf/scripts
     - makedirs: True
 
-tgrafsyncscripts:
+{% for script in TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]] %}
-  file.recurse:
+tgraf_sync_script_{{script}}:
-    - name: /opt/so/conf/telegraf/scripts
+  file.managed:
+    - name: /opt/so/conf/telegraf/scripts/{{script}}
     - user: root
     - group: 939
-    - file_mode: 770
+    - mode: 770
     - template: jinja
-    - source: salt://telegraf/scripts
+    - source: salt://telegraf/scripts/{{script}}
-    {% if GLOBALS.md_engine == 'SURICATA' %}
+{% endfor %}
-    - exclude_pat: zeekcaptureloss.sh
-    {% endif %}
 
 telegraf_sbin:
   file.recurse:

salt/telegraf/defaults.yaml

@@ -9,3 +9,82 @@ telegraf:
     flush_jitter: '0s'
     debug: 'false'
     quiet: 'false'
+  scripts:
+    eval:
+      - beatseps.sh
+      - checkfiles.sh
+      - influxdbsize.sh
+      - oldpcap.sh
+      - raid.sh
+      - redis.sh
+      - sostatus.sh
+      - stenoloss.sh
+      - suriloss.sh
+      - zeekcaptureloss.sh
+      - zeekloss.sh
+    standalone:
+      - beatseps.sh
+      - checkfiles.sh
+      - eps.sh
+      - influxdbsize.sh
+      - oldpcap.sh
+      - raid.sh
+      - redis.sh
+      - sostatus.sh
+      - stenoloss.sh
+      - suriloss.sh
+      - zeekcaptureloss.sh
+      - zeekloss.sh
+    manager:
+      - beatseps.sh
+      - influxdbsize.sh
+      - raid.sh
+      - redis.sh
+      - sostatus.sh
+    managersearch:
+      - beatseps.sh
+      - eps.sh
+      - influxdbsize.sh
+      - raid.sh
+      - redis.sh
+      - sostatus.sh
+    import:
+      - sostatus.sh
+    sensor:
+      - beatseps.sh
+      - checkfiles.sh
+      - oldpcap.sh
+      - raid.sh
+      - sostatus.sh
+      - stenoloss.sh
+      - suriloss.sh
+      - zeekcaptureloss.sh
+      - zeekloss.sh
+    heavynode:
+      - beatseps.sh
+      - checkfiles.sh
+      - eps.sh
+      - oldpcap.sh
+      - raid.sh
+      - redis.sh
+      - sostatus.sh
+      - stenoloss.sh
+      - suriloss.sh
+      - zeekcaptureloss.sh
+      - zeekloss.sh
+    idh:
+      - sostatus.sh
+    searchnode:
+      - beatseps.sh
+      - eps.sh
+      - raid.sh
+      - sostatus.sh
+    receiver:
+      - beatseps.sh
+      - eps.sh
+      - raid.sh
+      - redis.sh
+      - sostatus.sh
+    fleet:
+      - sostatus.sh
+    desktop: []

salt/telegraf/enabled.sls

@@ -7,6 +7,7 @@
 {% if sls.split('.')[0] in allowed_states %}
 {%   from 'vars/globals.map.jinja' import GLOBALS %}
 {%   from 'docker/docker.map.jinja' import DOCKER %}
+{%   from 'telegraf/map.jinja' import TELEGRAFMERGED %}
 
 
 include:
@@ -67,8 +68,10 @@ so-telegraf:
     {% endif %}
     - watch:
       - file: tgrafconf
-      - file: tgrafsyncscripts
       - file: node_config
+    {% for script in TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]] %}
+      - file: tgraf_sync_script_{{script}}
+    {% endfor %}
     - require: 
       - file: tgrafconf
       - file: node_config

salt/telegraf/etc/telegraf.conf

@@ -193,7 +193,7 @@
   username = "{{ ES_USER }}"
   password = "{{ ES_PASS }}"
   insecure_skip_verify = true
-{%- elif grains['role'] in ['so-searchnode', 'so-hotnode', 'so-warmnode']  %}
+{%- elif grains['role'] in ['so-searchnode']  %}
 [[inputs.elasticsearch]]
   servers = ["https://{{ NODEIP }}:9200"]
   cluster_stats = false
@@ -244,6 +244,8 @@
 {%- endif %}
 
 # # Read metrics from one or more commands that can output to stdout
+{%- if 'sostatus.sh' in TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]] %}
+{%-   do TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]].remove('sostatus.sh') %}
 [[inputs.exec]]
    commands = [
       "/scripts/sostatus.sh"
@@ -251,122 +253,26 @@
    data_format = "influx"
    timeout = "15s"
    interval = "60s"
+{%- endif %}
 
-#   ## Commands array
+{%- if TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]] | length > 0 %}
-{% if grains['role'] in ['so-manager'] %}
 [[inputs.exec]]
    commands = [
-      "/scripts/redis.sh",
+{%-   for script in TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]] %}
-      "/scripts/influxdbsize.sh",
+      "/scripts/{{script}}"{% if not loop.last %},{% endif %}
-      "/scripts/raid.sh",
+{%-   endfor %}
-      "/scripts/beatseps.sh"
    ]
   data_format = "influx"
   ## Timeout for each command to complete.
   timeout = "15s"
-{% elif grains['role'] in ['so-managersearch'] %}
+{%- endif %}
-[[inputs.exec]]
-   commands = [
-      "/scripts/redis.sh",
-      "/scripts/influxdbsize.sh",
-      "/scripts/eps.sh",
-      "/scripts/raid.sh",
-      "/scripts/beatseps.sh"
-   ]
-  data_format = "influx"
-  ## Timeout for each command to complete.
-  timeout = "15s"
-{% elif grains['role'] in ['so-searchnode', 'so-receiver'] %}
-[[inputs.exec]]
-   commands = [
-      "/scripts/eps.sh",
-      "/scripts/raid.sh",
-  {% if grains.role == 'so-receiver' %}
-      "/scripts/redis.sh",
-  {% endif %}
-      "/scripts/beatseps.sh"
-   ]
-  data_format = "influx"
-  ## Timeout for each command to complete.
-  timeout = "15s"
-{% elif grains['role'] == 'so-sensor' %}
-[[inputs.exec]]
-   commands = [
-      "/scripts/stenoloss.sh",
-      "/scripts/suriloss.sh",
-      "/scripts/checkfiles.sh",
-  {%- if MDENGINE == 'ZEEK' and ZEEK_ENABLED %}
-      "/scripts/zeekloss.sh",
-      "/scripts/zeekcaptureloss.sh",
-  {%- endif %}
-      "/scripts/oldpcap.sh",
-      "/scripts/raid.sh",
-      "/scripts/beatseps.sh"
-   ]
-  data_format = "influx"
-  timeout = "15s"
-{% elif grains['role'] == 'so-heavynode' %}
-[[inputs.exec]]
-   commands = [
-      "/scripts/redis.sh",
-      "/scripts/stenoloss.sh",
-      "/scripts/suriloss.sh",
-      "/scripts/checkfiles.sh",
-  {%- if MDENGINE == 'ZEEK' and ZEEK_ENABLED %}
-      "/scripts/zeekloss.sh",
-      "/scripts/zeekcaptureloss.sh",
-  {%- endif %}
-      "/scripts/oldpcap.sh",
-      "/scripts/eps.sh",
-      "/scripts/raid.sh",
-      "/scripts/beatseps.sh"
-   ]
-  data_format = "influx"
-  timeout = "15s"
-{% elif grains['role'] == 'so-standalone' %}
-[[inputs.exec]]
-   commands = [
-      "/scripts/redis.sh",
-      "/scripts/influxdbsize.sh",
-      "/scripts/stenoloss.sh",
-      "/scripts/suriloss.sh",
-      "/scripts/checkfiles.sh",
-  {%- if MDENGINE == 'ZEEK' and ZEEK_ENABLED %}
-      "/scripts/zeekloss.sh",
-      "/scripts/zeekcaptureloss.sh",
-  {%- endif %}
-      "/scripts/oldpcap.sh",
-      "/scripts/eps.sh",
-      "/scripts/raid.sh",
-      "/scripts/beatseps.sh"
-   ]
-  data_format = "influx"
-  timeout = "15s"
-{% elif grains['role'] == 'so-eval' %}
-[[inputs.exec]]
-   commands = [
-      "/scripts/redis.sh",
-      "/scripts/stenoloss.sh",
-      "/scripts/suriloss.sh",
-      "/scripts/checkfiles.sh",
-  {%- if MDENGINE == 'ZEEK' and ZEEK_ENABLED %}
-      "/scripts/zeekloss.sh",
-      "/scripts/zeekcaptureloss.sh",
-  {%- endif %}
-      "/scripts/oldpcap.sh",
-      "/scripts/influxdbsize.sh",
-      "/scripts/raid.sh",
-      "/scripts/beatseps.sh"
-   ]
-  data_format = "influx"
-  timeout = "15s"
-{% endif %}
 
 {%- if salt['pillar.get']('healthcheck:enabled', False) %}
 [[inputs.file]]
   files = ["/host/nsm/zeek/logs/zeek_restart.log"]
   data_format = "influx"
 {%- endif %}
+
 [[inputs.file]]
   files = ["/etc/telegraf/node_config.json"]
   name_override = "node_config"

salt/telegraf/map.jinja

@@ -3,5 +3,13 @@
    https://securityonion.net/license; you may not use this file except in compliance with the
    Elastic License 2.0. #}
 
+{% from 'vars/globals.map.jinja' import GLOBALS %}
 {% import_yaml 'telegraf/defaults.yaml' as TELEGRAFDEFAULTS %}
 {% set TELEGRAFMERGED = salt['pillar.get']('telegraf', TELEGRAFDEFAULTS.telegraf, merge=True) %}
+{% from 'zeek/config.map.jinja' import ZEEKMERGED %}
+
+{# if the md engine isn't zeek or zeek is disabled, dont run the zeek scripts for telegraf #}
+{% if (GLOBALS.role in ['so-eval', 'so-standalone', 'so-sensor', 'so-heavynode']) and (GLOBALS.md_engine != 'ZEEK' or not ZEEKMERGED.enabled) %}
+{%   do TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]].remove('zeekloss.sh') %}
+{%   do TELEGRAFMERGED.scripts[GLOBALS.role.split('-')[1]].remove('zeekcaptureloss.sh') %}
+{% endif %}

salt/telegraf/soc_telegraf.yaml

@@ -42,4 +42,21 @@ telegraf:
       global: True
       advanced: True
       helpLink: telegraf.html
-    
+  scripts:
+    eval: &telegrafscripts
+      description: List of input.exec scripts to run for this node type. The script must be present in salt/telegraf/scripts.
+      forcedType: "[]string"
+      multiline: True
+      advanced: True
+      helpLink: telegraf.html
+    standalone: *telegrafscripts
+    manager: *telegrafscripts
+    managersearch: *telegrafscripts
+    import: *telegrafscripts
+    sensor: *telegrafscripts
+    heavynode: *telegrafscripts
+    idh: *telegrafscripts
+    searchnode: *telegrafscripts
+    receiver: *telegrafscripts
+    fleet: *telegrafscripts
+    desktop: *telegrafscripts