Merge branch 'main' into stable

files/scriptreciver/sysctl-d-gofasta.conf

@@ -0,0 +1,6 @@
+net.core.rmem_max=67108864
+net.core.wmem_max=67108864
+net.ipv4.tcp_rmem=4096 87380 33554432
+net.ipv4.tcp_wmem=4096 87380 33554432
+net.core.default_qdisc=fq
+net.ipv4.tcp_congestion_control=bbr

manifests/app_type.pp

@@ -101,6 +101,23 @@ define sunetdrive::app_type (
       content => template('sunetdrive/application/get_containers'),
       mode    => '0744',
     }
+    if ($nodenumber == 3) {
+      file { '/usr/lib/nagios/plugins/check_nextcloud_mounts.py':
+        ensure  => present,
+        owner   => 'root',
+        group   => 'root',
+        content => template('sunetdrive/application/check_nextcloud_mounts.py'),
+        mode    => '0744',
+      }
+      sunet::sudoer {'nagios_run_nextcloud_mounts_command':
+        user_name    => 'nagios',
+        collection   => 'nrpe_nextcloud_mounts_check',
+        command_line => '/usr/lib/nagios/plugins/check_nextcloud_mounts.py'
+      }
+      sunet::nagios::nrpe_command {'check_nextcloud_mounts':
+        command_line => '/usr/bin/sudo /usr/lib/nagios/plugins/check_nextcloud_mounts.py'
+      }
+    }
     if ($nodenumber == 3) {
       file { '/usr/local/bin/scan_external_mounts':
         ensure  => present,
@@ -388,5 +405,4 @@ define sunetdrive::app_type (
       }
     }
   }
-
 }

manifests/multinode.pp

@@ -46,6 +46,21 @@ class sunetdrive::multinode (
     content => template('sunetdrive/application/get_containers'),
     mode    => '0744',
   }
+  file { '/usr/lib/nagios/plugins/check_nextcloud_mounts.py':
+    ensure  => present,
+    owner   => 'root',
+    group   => 'root',
+    content => template('sunetdrive/application/check_nextcloud_mounts.py'),
+    mode    => '0744',
+  }
+  sunet::sudoer {'nagios_run_nextcloud_mounts_command':
+    user_name    => 'nagios',
+    collection   => 'nrpe_nextcloud_mounts_check',
+    command_line => '/usr/lib/nagios/plugins/check_nextcloud_mounts.py'
+  }
+  sunet::nagios::nrpe_command {'check_nextcloud_mounts':
+    command_line => '/usr/bin/sudo /usr/lib/nagios/plugins/check_nextcloud_mounts.py'
+  }
   file { '/usr/local/bin/scan_external_mounts':
     ensure  => present,
     force   => true,

manifests/scriptreceiver.pp

@@ -5,7 +5,19 @@ class sunetdrive::scriptreceiver()
   sunet::system_user {'script': username => 'script', group => 'script', managehome => true, shell => '/bin/bash' }
 
   # These tasks correspond to a ${task}.erb.sh template
-  $tasks = ['list_users', 'list_files_for_user', 'create_bucket', 'backup_db', 'purge_backups', 'maintenancemode', 'restart_sunet_service', 'start_sentinel', 'stop_sentinel', 'removeswap', 'backup_multinode_db']
+  $tasks = [
+    'list_users',
+    'list_files_for_user',
+    'create_bucket',
+    'backup_db',
+    'purge_backups',
+    'maintenancemode',
+    'restart_sunet_service',
+    'start_sentinel',
+    'stop_sentinel',
+    'removeswap',
+    'backup_multinode_db'
+  ]
 
   $environment = sunetdrive::get_environment()
   $config = hiera_hash($environment)
@@ -35,7 +47,9 @@ class sunetdrive::scriptreceiver()
     type   => 'ssh-ed25519',
     key    => $script_pub_key,
   }
-
+  file { '/etc/sysctl.d/gofasta.conf':
+    ensure  => 'absent',
+  }
   file { '/opt/rotate':
     ensure => directory,
     mode   => '0750',

manifests/sitemonitornaemon.pp

@@ -11,6 +11,9 @@ class sunetdrive::sitemonitornaemon() {
   $environment = sunetdrive::get_environment()
   $influx_passwd = safe_hiera('influx_passwd')
   $slack_url = safe_hiera('slack_url')
+  $extra_host_groups = {
+    node3_hosts => join($facts['configured_hosts_in_cosmos']['all'].filter |$host| { $host =~ /^node3\./ }, ',')
+  }
 
   file { '/usr/local/bin/slack_nagios.sh':
     ensure  => present,
@@ -45,6 +48,11 @@ class sunetdrive::sitemonitornaemon() {
     content => template('sunetdrive/monitor/sunetdrive_thruk_templates.conf.erb'),
     mode    => '0644',
   }
+  file { '/etc/naemon/conf.d/sunetdrive_extra_hostgroups.cfg':
+    ensure  => present,
+    content => template('sunetdrive/monitor/sunetdrive_extra_hostgroups.cfg.erb'),
+    mode    => '0644',
+  }
   nagioscfg::service {'check_scriptherder':
     hostgroup_name => ['sunetdrive::nrpe'],
     check_command  => 'check_nrpe_1arg_to300!check_scriptherder',
@@ -99,6 +107,12 @@ class sunetdrive::sitemonitornaemon() {
     description    => 'Status of sarimner interface',
     contact_groups => ['alerts']
   }
+  nagioscfg::service {'check_nextcloud_mounts':
+    hostgroup_name => ['node3_hosts','sunetdrive::multinode'],
+    check_command  => 'check_nrpe_1arg!check_nextcloud_mounts',
+    description    => 'S3 buckets with multiple Nextcloud mounts',
+    contact_groups => ['alerts']
+  }
 
 }
 

templates/application/check_nextcloud_mounts.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python3
+
+from collections import Counter
+import json
+import shlex
+import subprocess
+import sys
+
+exit = 0
+base_message = "OK: no duplicate mounts"
+long_message = ""
+
+get_containers = subprocess.Popen('/usr/local/bin/get_containers', stdout=subprocess.PIPE).stdout.read()
+containers = get_containers.decode().splitlines()
+
+for i, container in enumerate(containers, start=1):
+    buckets = []
+    list_command = f"/usr/local/bin/nocc {container} files_external:list --all --show-password --output json"
+    command = shlex.split(list_command)
+    mount_data_byte = subprocess.Popen(command, stdout=subprocess.PIPE).stdout.read()
+    try:
+        mount_data = json.loads(mount_data_byte.decode())
+    except json.decoder.JSONDecodeError as err:
+        if i == 1 or i != len(containers):
+            base_message = "WARNING: invalid json"
+        long_message +=  f"\ncontainer: {container} - json decode error: {err}"
+#        lets do exit 0 for now
+#        exit = 1
+        continue
+    for items in mount_data:
+        buckets.append(items["configuration"]["bucket"])
+    bucket_count = dict(Counter(buckets))
+    for k, v in bucket_count.items():
+        if v > 1:
+            base_message = "WARNING: buckets with multiple mounts"
+            long_message += f"\ncontainer: {container} - bucket: {k} - {v}"
+#            lets do exit 0 for now
+#            exit = 1
+print(base_message)
+if long_message != "":
+    print(long_message.lstrip())
+sys.exit(exit)

templates/monitor/sunetdrive_extra_hostgroups.cfg.erb

@@ -0,0 +1,8 @@
+<% @extra_host_groups.each do |group, members| -%>
+# <%= group %>
+define hostgroup {
+  hostgroup_name <%= group %>
+  alias <%= group %>
+  members <%= members %>
+}
+<% end -%>