From 7e0aef04d5397b6e78cb6ae0b7037c87cbdf8287 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?P=C3=A1ll=20Gu=C3=B0j=C3=B3n=20Sigur=C3=B0sson?=
 <palli@opensource.is>
Date: Mon, 1 Mar 2010 22:09:31 +0000
Subject: [PATCH] check_dataprotector added

---
 check_dataprotector/trunk/check_dp_backups    | 78 +++++++++++++++++++
 check_dataprotector/trunk/check_dp_idb        | 17 ++++
 .../trunk/check_dp_mountrequest               | 24 ++++++
 check_dataprotector/trunk/check_dp_pool       | 40 ++++++++++
 check_dataprotector/trunk/check_dp_services   | 12 +++
 check_dataprotector/trunk/check_dp_tablespace | 28 +++++++
 .../trunk/nrpe_local/dataprotector.cfg        |  7 ++
 7 files changed, 206 insertions(+)
 create mode 100755 check_dataprotector/trunk/check_dp_backups
 create mode 100755 check_dataprotector/trunk/check_dp_idb
 create mode 100755 check_dataprotector/trunk/check_dp_mountrequest
 create mode 100755 check_dataprotector/trunk/check_dp_pool
 create mode 100755 check_dataprotector/trunk/check_dp_services
 create mode 100755 check_dataprotector/trunk/check_dp_tablespace
 create mode 100644 check_dataprotector/trunk/nrpe_local/dataprotector.cfg

diff --git a/check_dataprotector/trunk/check_dp_backups b/check_dataprotector/trunk/check_dp_backups
new file mode 100755
index 0000000..5d3b399
--- /dev/null
+++ b/check_dataprotector/trunk/check_dp_backups
@@ -0,0 +1,78 @@
+#!/usr/bin/python
+# -*- coding: utf-8 -*-
+
+import subprocess
+import datetime
+import sys
+now = datetime.datetime.now()
+
+
+# return code 3 = No sessions matching the search criteria were found. 
+omnistat_returncode_no_sessions_found=3
+class GenericObject:
+  def __init__(self):
+    self.dict = {}
+
+def runCommand(command):
+  proc = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE,stderr=subprocess.PIPE,)
+  stdout, stderr = proc.communicate('through stdin to stdout')
+  if proc.returncode > 0 and proc.returncode != omnistat_returncode_no_sessions_found: 
+    print "Error running '%s' command:\n %s" % (command, stderr)
+    print proc.returncode
+    print stderr, stdout
+    sys.exit(1)
+  else:
+    return stdout
+
+def getSessions():
+  sessions = []
+  date = "%s/%s/%s" % ( now.year, now.month, now.day )
+  output = runCommand( '/opt/omni/bin/omnistat -previous -since %s' % date )
+  for line in output.split('\n')[1:]:
+    tmp = line.split()
+    if len(tmp) < 4: continue
+    sessionID = tmp[0]
+    sessions.append( sessionID )
+    #print sessionID
+  return sessions
+
+def parseSessions(session_list):
+  parsedSessions = []
+  for sessionID in session_list:
+    output = runCommand('/opt/omni/bin/omnidb -rpt %s -detail' % sessionID )
+    session = GenericObject()
+    parsedSessions.append ( session )
+    for line in output.split('\n'):
+      tmp = line.split(':')
+      if len(tmp) != 2: continue
+      key,value = tmp
+      key = key.strip()
+      value = value.strip()
+      session.dict[key] = value
+  return parsedSessions
+
+sessions = getSessions()
+parsedSessions = parseSessions( sessions )
+
+
+total_errors = 0
+total_warnings = 0
+total_size = 0
+total_sessions = 0
+for session in parsedSessions:
+	errors = int( session.dict['Number of errors'] )
+	warnings = int( session.dict['Number of warnings'] )
+	try:
+		size = 1024 * int( session.dict['Session size'].split()[0] )
+	except:
+		size = 0
+	total_errors = total_errors + errors
+	total_warnings = total_warnings + warnings
+	total_size = total_size + size
+	total_sessions += 1
+
+print "Everything is ok | errors=%d warnings=%d size=%dB num_sessions=%dc" % (total_errors,total_warnings,total_size,total_sessions)
+
+
+
+
diff --git a/check_dataprotector/trunk/check_dp_idb b/check_dataprotector/trunk/check_dp_idb
new file mode 100755
index 0000000..a283426
--- /dev/null
+++ b/check_dataprotector/trunk/check_dp_idb
@@ -0,0 +1,17 @@
+#!/bin/sh
+
+WARNING=5
+CRITICAL=2
+
+
+/opt/omni/bin/omnirpt -tab -report obj_lastbackup -days -$WARNING | grep -w IDB > /dev/null
+RESULT=$?
+
+
+if [ $RESULT -gt 0 ]; then
+	echo "Warning, Last Dataprotector backup is more than $WARNING days old"
+	exit 1
+fi
+
+echo "Last Dataprotector backup is less than $WARNING days old"
+
diff --git a/check_dataprotector/trunk/check_dp_mountrequest b/check_dataprotector/trunk/check_dp_mountrequest
new file mode 100755
index 0000000..f6c73ce
--- /dev/null
+++ b/check_dataprotector/trunk/check_dp_mountrequest
@@ -0,0 +1,24 @@
+#!/bin/sh
+
+
+OUTPUT=$(/opt/omni/bin/omnistat)
+RESULT=$?
+
+echo $OUTPUT | grep -q "No currently running sessions."
+RESULT=$?
+
+if [ $RESULT -eq 0 ]; then
+	echo "OK - No running sessions"
+	exit 0
+fi
+
+echo $OUTPUT | grep -q "Mount Request" 
+RESULT=$?
+
+if [ $RESULT -eq 0 ]; then
+	echo "Warning, Dataprotector has mount requests"
+	exit 1	
+fi
+
+echo -n $OUTPUT
+exit 0
diff --git a/check_dataprotector/trunk/check_dp_pool b/check_dataprotector/trunk/check_dp_pool
new file mode 100755
index 0000000..4bc35bb
--- /dev/null
+++ b/check_dataprotector/trunk/check_dp_pool
@@ -0,0 +1,40 @@
+#!/bin/sh
+
+POOL="$1"
+WARNING=5
+CRITICAL=2
+
+if [ "$POOL" = "" ]; then
+	echo "Unknown - No Media pool specified"
+	exit 3
+fi
+
+OUTPUT=$(/opt/omni/bin/omnimm -list_pool "$POOL")
+RESULT=$?
+
+if [ $RESULT -eq 3 ]; then
+	echo "Unknown - Media pool '$POOL' was not found"
+	exit 3
+fi
+
+if [ $RESULT -ne 0 ]; then
+	echo "Unknown - exit code from omnimm=$RESULT, output: $OUTPUT"
+	exit 3
+fi
+
+FREE_MEDIA=$(echo "$OUTPUT" | tail -n +4 | wc -l)
+
+PERFDATA="'$POOL'=$FREE_MEDIA"
+
+if [ $FREE_MEDIA -lt $CRITICAL ]; then
+	echo "Critical: $FREE_MEDIA available media in pool $POOL | $PERFDATA"
+	exit 2
+fi
+
+if [ $FREE_MEDIA -lt $WARNING ]; then
+	echo "Warning: $FREE_MEDIA available media in pool $POOL | $PERFDATA"
+	exit 1
+fi
+
+echo "OK: $FREE_MEDIA available media in pool $POOL | $PERFDATA"
+exit 0
diff --git a/check_dataprotector/trunk/check_dp_services b/check_dataprotector/trunk/check_dp_services
new file mode 100755
index 0000000..95edef9
--- /dev/null
+++ b/check_dataprotector/trunk/check_dp_services
@@ -0,0 +1,12 @@
+#!/bin/sh
+
+/opt/omni/sbin/omnisv -status | grep "Status: All Data Protector relevant processes/services up and running."
+
+RESULT=$?
+
+if [ $RESULT -gt 0 ]; then
+	echo "Critical, some Data Protector services are in stopped state"
+	exit 1	
+fi
+
+
diff --git a/check_dataprotector/trunk/check_dp_tablespace b/check_dataprotector/trunk/check_dp_tablespace
new file mode 100755
index 0000000..b5b0fd7
--- /dev/null
+++ b/check_dataprotector/trunk/check_dp_tablespace
@@ -0,0 +1,28 @@
+#!/bin/sh
+
+
+#/opt/omni/bin/omnirpt -report db_size 
+
+echo $USER > /tmp/check_dp_tablespace.debug
+echo $@ >> /tmp/check_dp_tablespace.debug
+
+
+OUTPUT=$(/opt/omni/bin/omnirpt -report db_size)
+RESULT=$?
+
+if [ $RESULT -gt 0 ]; then
+	echo -n $OUTPUT
+	exit 1
+fi
+echo $OUTPUT | grep -q "No database devices with low disk space."
+
+RESULT=$?
+
+if [ $RESULT -gt 0 ]; then
+	echo "Warning - Some dataprotector tablespaces are running low"
+	exit 1
+else
+	echo "OK - No database devices with low disk space."
+	exit 0
+fi
+
diff --git a/check_dataprotector/trunk/nrpe_local/dataprotector.cfg b/check_dataprotector/trunk/nrpe_local/dataprotector.cfg
new file mode 100644
index 0000000..b227e9a
--- /dev/null
+++ b/check_dataprotector/trunk/nrpe_local/dataprotector.cfg
@@ -0,0 +1,7 @@
+
+
+command[check_dp_pool]=/usr/lib/nagios/plugins/contrib/check_dp_pool "free"
+command[check_dp_tablespace]=/usr/lib/nagios/plugins/contrib/check_dp_tablespace
+command[check_dp_services]=/usr/lib/nagios/plugins/contrib/check_dp_services
+command[check_dp_mountrequest]=/usr/lib/nagios/plugins/contrib/check_dp_mountrequest
+command[check_dp_idb]=/usr/lib/nagios/plugins/contrib/check_dp_idb