check_slony 0.0.6 → 0.0.7

data/Manifest.txt

@@ -1,4 +1,3 @@
-.autotest
 History.txt
 Manifest.txt
 README.txt

data/README.txt

@@ -14,6 +14,7 @@ This is an check plugin for Nagios.
  % check_slony.rb -d slonydb -c replication -e 10 -l 10
  POSTGRES_REPLICATION_LAG OK: SUBSCRIBER 1 ON ORIGIN 2 : EVENT LAG=0 TIME LAG=8s || 
 
+
 == REQUIREMENTS:
 
 pg

data/bin/check_slony

@@ -5,7 +5,6 @@ require 'pg'
 require 'optparse'
 require 'timeout'
 
-
 pgoptions=''
 pgtty=''
 @result=''
@@ -25,6 +24,8 @@ OptionParser.new do |opt|
   opt.on("-e","--events=EVENTS", Integer, "Specify the Event Counts"){|e| @event = e }
   opt.on("-l","--lagtime=LAGTIME", Integer, "Specify the LAGTIME"){|l| @lagtime = l }
   opt.on("-t","--timeout=TIMEOUT", Integer, "Specify the TIMEOUT"){|t| @timeout = t }
+  opt.on("-c","--countall", "All Tables count check(must be against master node)"){ @countall = true }
+  opt.on("-v","--verbose", "verbose output(when specify the '-c')"){ @verbose = true }
   
   begin
     opt.parse!(ARGV)
@@ -34,46 +35,141 @@ OptionParser.new do |opt|
   end
 end
 
-begin
-  timeout(@timeout){
-    # sleep 3 # for debug
-    con = PGconn.new(@hostname, @port, pgoptions, pgtty, @dbname, @user, @password) 
-    query = 'SELECT st_origin, st_received, st_lag_num_events, round(extract(epoch from st_lag_time)) from "_' + @cluster + '".sl_status'
+def countcheck_all_tables(options, tty)
+  result_hash = {}
+  result_hash["status"] = "OK"
+  error_str = ""
+  con = PGconn.new(@hostname, @port, options, tty, @dbname, @user, @password)
+  # 
+  query = 'SELECT tab_relname from "_' + @cluster + '".sl_table'
+  res = con.exec(query)
+
+  sl_table = []
+  res.each do |r|
+    sl_table << r["tab_relname"]
+  end
+  # 
+  query = 'SELECT pa_server, pa_conninfo from "_' + @cluster + '".sl_path order by pa_server'
+  res = con.exec(query)
+  @pg_hash = {}
+  res.each do |r|
+    info_hash = {}
+    r["pa_conninfo"].split(" ").each do |info|
+#       puts "******"
+#       puts info.split("=")[0] + '=' + info.split("=")[1] 
+      info_hash[info.split("=")[0]] = info.split("=")[1] 
+    end
+    @pg_hash[r["pa_server"]] = info_hash
+  end
+  con.close
 
+  @master = @pg_hash.delete("1")
+  # save master node data
+  con = PGconn.new(@master["host"], @master["port"], options, tty, @master["dbname"], @master["user"], @master["password"])
+  sl_table.each do |t|
+    query = 'SELECT count(*) as cnt from ' + t 
     res = con.exec(query)
-    unless res
-      print "POSTGRES_REPLICATION_LAG CRITICAL: Cannot prepare $DBI::errstr\n";
-      exit 2
+    @master[t] = res[0]["cnt"]
+  end
+  con.close
+
+  # save slave node data
+  @pg_hash.each {|node, info|
+    con = PGconn.new(info["host"], info["port"], options, tty, info["dbname"], info["user"], info["password"])
+    sl_table.each do |t|
+      query = 'SELECT count(*) as cnt from ' + t 
+      res = con.exec(query)
+      info[t] = [res[0]["cnt"], false]
+      if @master[t] != info[t][0]
+        info[t][1] = true
+        result_hash["status"] = "NG"
+        error_str += "node #{node} #{t}"
+      end
+    end
+    con.close
+  }
+  result_hash["error"] = error_str
+
+  # only '-v'
+  if @verbose
+    fmt_str = '| %10s'
+    cols = fmt_str *  @pg_hash.count
+    state = '| %2s'
+    fmt = "%-35s | %10s" + cols + state + "\n"
+    @val_ary = ["tablename", "master"]
+    @pg_hash.each{ |node,val|
+      @val_ary << node
+    }
+    @val_ary << "status"
+    puts fmt % @val_ary 
+    puts '-' * ((@pg_hash.count * 15) + 60)
+    sl_table.each do |t|
+      temp_array = [t, @master[t]]
+      @pg_hash.each{ |k,v|
+        temp_array << v[t][0]
+        temp_array << (v[t][1] ? "NG" : "OK")
+      }
+      puts fmt % temp_array 
     end
+  end
 
-    res.each do |r|
+  return result_hash
+end
 
-      node = r["st_received"]
-      master = r["st_origin"]
-      lag = r["st_lag_num_events"]
-      round = r["round"]
 
-      @result += "SUBSCRIBER " + node  + " ON ORIGIN " + master + " : EVENT LAG=" +  lag
-      if (lag.to_i > 0) && (@event < lag.to_i)
-        @result = @result + " (BEHIND " + (lag.to_i - @event).to_s + ") ";
+begin
+  timeout(@timeout){
+    # sleep 3 # for debug
+    title = "POSTGRES_"
+    if @countall
+      title += "COUNTCHECK "
+      state = countcheck_all_tables(pgoptions, pgtty)
+      if state["status"] == "OK"
+        @result = "NO DIFFERENCE"
+      else
+        @result = "DIFFERENCE OCCURED ON " + state["error"]
         @problems += 1
       end
-      @result = @result + " TIME LAG=" + round + "s";
-      
-      if (@lagtime > 0) && (@lagtime < round.to_i)
-        @result = @result + " (BEHIND " + (round.to_i - @lagtime).to_s + "s) ";
-        @problems += 1
+    else
+      title += "REPLICATION_LAG "
+      con = PGconn.new(@hostname, @port, pgoptions, pgtty, @dbname, @user, @password) 
+      query = 'SELECT st_origin, st_received, st_lag_num_events, round(extract(epoch from st_lag_time)) from "_' + @cluster + '".sl_status'
+
+      res = con.exec(query)
+      unless res
+        print "#{title} CRITICAL: Cannot prepare $DBI::errstr\n";
+        exit 2
       end
-      @result = @result + " || ";
 
+      res.each do |r|
+
+        node = r["st_received"]
+        master = r["st_origin"]
+        lag = r["st_lag_num_events"]
+        round = r["round"]
+
+        @result += "SUBSCRIBER " + node  + " ON ORIGIN " + master + " : EVENT LAG=" +  lag
+        if (lag.to_i > 0) && (@event < lag.to_i)
+          @result = @result + " (BEHIND " + (lag.to_i - @event).to_s + ") ";
+          @problems += 1
+        end
+        @result = @result + " TIME LAG=" + round + "s";
+        
+        if (@lagtime > 0) && (@lagtime < round.to_i)
+          @result = @result + " (BEHIND " + (round.to_i - @lagtime).to_s + "s) ";
+          @problems += 1
+        end
+        @result = @result + " || ";
+
+      end
     end
 
     if @problems > 0
-      @result = "POSTGRES_REPLICATION_LAG CRITICAL: " + @result + "\n"
+      @result = title + "CRITICAL: " + @result + "\n"
       print @result
       exit 2
     else 
-      @result = "POSTGRES_REPLICATION_LAG OK: " + @result + "\n"
+      @result = title + "OK: " + @result + "\n"
       print @result
       exit 0
     end
@@ -92,6 +188,3 @@ end
 
 
 
-
-exit;
-abort "you need to write me"

data/lib/check_slony.rb

@@ -1,4 +1,4 @@
 class CheckSlony
-  VERSION = '0.0.6'
+  VERSION = '0.0.7'
 end
 

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification 
 name: check_slony
 version: !ruby/object:Gem::Version 
-  hash: 19
+  hash: 17
   prerelease: 
   segments: 
   - 0
   - 0
-  - 6
-  version: 0.0.6
+  - 7
+  version: 0.0.7
 platform: ruby
 authors: 
 - kakikubo
@@ -15,8 +15,7 @@ autorequire:
 bindir: bin
 cert_chain: []
 
-date: 2011-01-10 00:00:00 +09:00
-default_executable: 
+date: 2011-11-29 00:00:00 Z
 dependencies: 
 - !ruby/object:Gem::Dependency 
   name: hoe
@@ -46,7 +45,6 @@ extra_rdoc_files:
 - Manifest.txt
 - README.txt
 files: 
-- .autotest
 - History.txt
 - Manifest.txt
 - README.txt
@@ -54,7 +52,6 @@ files:
 - bin/check_slony
 - lib/check_slony.rb
 - test/test_check_slony.rb
-has_rdoc: true
 homepage: https://github.com/kakikubo/check_slony
 licenses: []
 
@@ -85,7 +82,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
 requirements: []
 
 rubyforge_project: check_slony
-rubygems_version: 1.4.2
+rubygems_version: 1.8.11
 signing_key: 
 specification_version: 3
 summary: This is an check plugin for Nagios.

data/.autotest

@@ -1,23 +0,0 @@
-# -*- ruby -*-
-
-require 'autotest/restart'
-
-# Autotest.add_hook :initialize do |at|
-#   at.extra_files << "../some/external/dependency.rb"
-#
-#   at.libs << ":../some/external"
-#
-#   at.add_exception 'vendor'
-#
-#   at.add_mapping(/dependency.rb/) do |f, _|
-#     at.files_matching(/test_.*rb$/)
-#   end
-#
-#   %w(TestA TestB).each do |klass|
-#     at.extra_class_map[klass] = "test/test_misc.rb"
-#   end
-# end
-
-# Autotest.add_hook :run_command do |at|
-#   system "rake build"
-# end