Merge branch 'improve_monitoring' into 0.6

13 files changed, 91 insertions, 10 deletions

diff --git a/puppet/modules/site_check_mk/files/agent/local_checks/all_hosts/run_node_tests.sh b/puppet/modules/site_check_mk/files/agent/local_checks/all_hosts/run_node_tests.sh
new file mode 100644
index 00000000..1dd0afc9
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/agent/local_checks/all_hosts/run_node_tests.sh
@@ -0,0 +1,5 @@
+#!/bin/sh
+#
+# runs node tests
+
+/srv/leap/bin/run_tests --checkmk
diff --git a/puppet/modules/site_check_mk/files/agent/local_checks/mx/check_leap_mx.sh b/puppet/modules/site_check_mk/files/agent/local_checks/mx/check_leap_mx.sh
new file mode 100755
index 00000000..2958d2ed
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/agent/local_checks/mx/check_leap_mx.sh
@@ -0,0 +1,33 @@
+#!/bin/bash
+
+
+WARN=1
+CRIT=5
+
+# in minutes
+MAXAGE=10
+
+STATUS[0]='OK'
+STATUS[1]='Warning'
+STATUS[2]='Critical'
+CHECKNAME='Leap_MX_Queue'
+
+WATCHDIR='/var/mail/vmail/Maildir/new/'
+
+
+total=`find $WATCHDIR -type f -mmin +$MAXAGE | wc -l`
+
+if [ $total -lt $WARN ]
+then
+  exitcode=0
+else
+  if [ $total -le $CRIT ]
+  then
+    exitcode=1
+  else
+      exitcode=2
+  fi
+fi
+
+echo "$exitcode $CHECKNAME stale_files=$total ${STATUS[exitcode]}: $total stale files (>=$MAXAGE min) in $WATCHDIR."
+
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg
index a1eb1312..3f6cc413 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/bigcouch.cfg
@@ -1,4 +1,9 @@
-/opt/bigcouch/var/log/bigcouch.log
+/opt/bigcouch/var/log/bigcouch.log nocontext=1
+# ignore requests that are fine
+ I undefined - -.*200$
+ I undefined - -.*201$
+ I 127.0.0.1 undefined.* ok
+ I 127.0.0.1 localhost:5984 .* ok
  C Uncaught error in HTTP request: {exit,
  C Uncaught error in HTTP request: {exit,normal}
  C Uncaught error in HTTP request: {error,
@@ -8,8 +13,4 @@
  C error
  C Connection attempt from disallowed node
  W Shutting down group server
-# ignore requests that are fine
- I undefined - -.*200$ 
- I undefined - -.*201$
- I 127.0.0.1 undefined.* ok  
- I 127.0.0.1 localhost:5984 .* ok 
+ W Apache CouchDB has started
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/leap_mx.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/leap_mx.cfg
new file mode 100644
index 00000000..c71c5392
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/leap_mx.cfg
@@ -0,0 +1,4 @@
+/var/log/leap_mx.log
+ W Don't know how to deliver mail
+ W No public key, stopping the processing chain
+
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg
index 54b782d3..623d1e46 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/soledad.cfg
@@ -1,4 +1,5 @@
 /var/log/soledad.log
  C WSGI application error
+ C Error
  C error
-
+ W Timing out client:
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/couchdb.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/couchdb.cfg
index 5f8d5b95..c92b5af7 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/couchdb.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/couchdb.cfg
@@ -1,2 +1,4 @@
  C /usr/local/bin/couch-doc-update.*failed
  C /usr/local/bin/couch-doc-update.*ERROR
+ W epmd: got partial packet only on file descriptor
+
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/stunnel.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/stunnel.cfg
index cf7ebca8..a4e428b4 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/syslog/stunnel.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog/stunnel.cfg
@@ -3,3 +3,4 @@
 # this is a temporary failure and happens very often, so we 
 # ignore it
  I stunnel:.*Connection reset by peer
+ I stunnel:.*Peer suddenly disconnected
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog_header.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog_header.cfg
new file mode 100644
index 00000000..f60d752b
--- /dev/null
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog_header.cfg
@@ -0,0 +1 @@
+/var/log/syslog
diff --git a/puppet/modules/site_check_mk/files/agent/logwatch/syslog.cfg b/puppet/modules/site_check_mk/files/agent/logwatch/syslog_tail.cfg
index f3505c1c..450b9e90 100644
--- a/puppet/modules/site_check_mk/files/agent/logwatch/syslog.cfg
+++ b/puppet/modules/site_check_mk/files/agent/logwatch/syslog_tail.cfg
@@ -1,4 +1,3 @@
-/var/log/syslog
 # some general patterns
  C panic
  C Oops
diff --git a/puppet/modules/site_check_mk/manifests/agent.pp b/puppet/modules/site_check_mk/manifests/agent.pp
index efb05b37..b752182e 100644
--- a/puppet/modules/site_check_mk/manifests/agent.pp
+++ b/puppet/modules/site_check_mk/manifests/agent.pp
@@ -18,6 +18,11 @@ class site_check_mk::agent {
     ensure  => directory,
   }
 
+  file { '/usr/lib/check_mk_agent/local/run_node_tests.sh':
+    source => 'puppet:///modules/site_check_mk/agent/local_checks/all_hosts/run_node_tests.sh',
+    mode   => '0755'
+  }
+
   include site_check_mk::agent::mrpe
   include site_check_mk::agent::logwatch
 }
diff --git a/puppet/modules/site_check_mk/manifests/agent/logwatch/syslog.pp b/puppet/modules/site_check_mk/manifests/agent/logwatch/syslog.pp
index ac3dda64..c927780d 100644
--- a/puppet/modules/site_check_mk/manifests/agent/logwatch/syslog.pp
+++ b/puppet/modules/site_check_mk/manifests/agent/logwatch/syslog.pp
@@ -5,9 +5,14 @@ class site_check_mk::agent::logwatch::syslog {
   }
 
   concat::fragment { 'syslog_header':
-    source => 'puppet:///modules/site_check_mk/agent/logwatch/syslog.cfg',
+    source => 'puppet:///modules/site_check_mk/agent/logwatch/syslog_header.cfg',
     target => '/etc/check_mk/logwatch.d/syslog.cfg',
     order  => '01';
   }
+  concat::fragment { 'syslog_tail':
+    source => 'puppet:///modules/site_check_mk/agent/logwatch/syslog_tail.cfg',
+    target => '/etc/check_mk/logwatch.d/syslog.cfg',
+    order  => '99';
+  }
 
 }
diff --git a/puppet/modules/site_check_mk/manifests/agent/mx.pp b/puppet/modules/site_check_mk/manifests/agent/mx.pp
new file mode 100644
index 00000000..2dd50400
--- /dev/null
+++ b/puppet/modules/site_check_mk/manifests/agent/mx.pp
@@ -0,0 +1,22 @@
+class site_check_mk::agent::mx {
+
+  # watch logs
+  file { '/etc/check_mk/logwatch.d/leap_mx.cfg':
+    source => 'puppet:///modules/site_check_mk/agent/logwatch/leap_mx.cfg',
+  }
+
+  # local nagios plugin checks via mrpe
+  file_line {
+    'Leap_MX_Procs':
+      line => 'Leap_MX_Procs  /usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -a leap_mx',
+      path => '/etc/check_mk/mrpe.cfg';
+  }
+
+
+  # check stale files in queue dir
+  file { '/usr/lib/check_mk_agent/local/check_leap_mx.sh':
+    source => 'puppet:///modules/site_check_mk/agent/local_checks/mx/check_leap_mx.sh',
+    mode   => '0755'
+  }
+
+}
diff --git a/puppet/modules/site_check_mk/manifests/agent/webapp.pp b/puppet/modules/site_check_mk/manifests/agent/webapp.pp
index 48fa0828..932d6bb9 100644
--- a/puppet/modules/site_check_mk/manifests/agent/webapp.pp
+++ b/puppet/modules/site_check_mk/manifests/agent/webapp.pp
@@ -1,7 +1,9 @@
 class site_check_mk::agent::webapp {
 
   # check webapp login
-  package { 'python-srp': ensure => installed }
+  package { [ 'python-srp', 'python-requests', 'python-yaml' ]:
+    ensure => installed
+  }
   file { '/usr/lib/check_mk_agent/local/nagios-webapp_login.py':
     ensure => link,
     target => '/srv/leap/webapp/test/nagios/webapp_login.py'