pg 1.1.0.pre20180730144600 → 1.1.0.pre20180730171000

data/sample/replication_monitor.rb

@@ -0,0 +1,222 @@
+# -*- ruby -*-
+# vim: set noet nosta sw=4 ts=4 :
+#
+# Get the current WAL segment and offset from a master postgresql
+# server, and compare slave servers to see how far behind they
+# are in MB.  This script should be easily modified for use with
+# Nagios/Mon/Monit/Zabbix/whatever, or wrapping it in a display loop,
+# and is suitable for both WAL shipping or streaming forms of replication.
+#
+# Mahlon E. Smith <mahlon@martini.nu>
+#
+# First argument is the master server, all other arguments are treated
+# as slave machines.
+#
+#	db_replication.monitor db-master.example.com ...
+#
+
+require 'ostruct'
+require 'optparse'
+require 'pathname'
+require 'etc'
+require 'pg'
+require 'pp'
+
+
+### A class to encapsulate the PG handles.
+###
+class PGMonitor
+
+	VERSION = %q$Id$
+
+	# When to consider a slave as 'behind', measured in WAL segments.
+	# The default WAL segment size is 16, so we'll alert after
+	# missing two WAL files worth of data.
+	#
+	LAG_ALERT = 32
+
+	### Create a new PGMonitor object.
+	###
+	def initialize( opts, hosts )
+		@opts        = opts
+		@master      = hosts.shift
+		@slaves      = hosts
+		@current_wal = {}
+		@failures    = []
+	end
+
+	attr_reader :opts, :current_wal, :master, :slaves, :failures
+
+
+	### Perform the connections and check the lag.
+	###
+	def check
+		# clear prior failures, get current xlog info
+		@failures = []
+		return unless self.get_current_wal
+
+		# check all slaves
+		self.slaves.each do |slave|
+			begin
+				slave_db = PG.connect(
+					:dbname   => self.opts.database,
+					:host     => slave,
+					:port     => self.opts.port,
+					:user     => self.opts.user,
+					:password => self.opts.pass,
+					:sslmode  => 'prefer'
+				)
+
+				xlog = slave_db.exec( 'SELECT pg_last_xlog_receive_location()' ).getvalue( 0, 0 )
+				slave_db.close
+
+				lag_in_megs = ( self.find_lag( xlog ).to_f / 1024 / 1024 ).abs
+				if lag_in_megs >= LAG_ALERT
+					failures << { :host => slave,
+						:error => "%0.2fMB behind the master." % [ lag_in_megs ] }
+				end
+			rescue => err
+				failures << { :host => slave, :error => err.message }
+			end
+		end
+	end
+
+
+	#########
+	protected
+	#########
+
+	### Ask the master for the current xlog information, to compare
+	### to slaves.  Returns true on succcess.  On failure, populates
+	### the failures array and returns false.
+	###
+	def get_current_wal
+		master_db = PG.connect(
+			:dbname   => self.opts.database,
+			:host     => self.master,
+			:port     => self.opts.port,
+			:user     => self.opts.user,
+			:password => self.opts.pass,
+			:sslmode  => 'prefer'
+		)
+
+		self.current_wal[ :segbytes ] = master_db.exec( 'SHOW wal_segment_size' ).
+			getvalue( 0, 0 ).sub( /\D+/, '' ).to_i << 20
+
+		current = master_db.exec( 'SELECT pg_current_xlog_location()' ).getvalue( 0, 0 )
+		self.current_wal[ :segment ], self.current_wal[ :offset ] = current.split( /\// )
+
+		master_db.close
+		return true
+
+	# If we can't get any of the info from the master, then there is no
+	# point in a comparison with slaves.
+	#
+	rescue => err
+		self.failures << { :host => self.master,
+			:error => 'Unable to retrieve required info from the master (%s)' % [ err.message ] }
+
+		return false
+	end
+
+
+	### Given an +xlog+ position from a slave server, return
+	### the number of bytes the slave needs to replay before it
+	### is caught up to the master.
+	###
+	def find_lag( xlog )
+		s_segment, s_offset = xlog.split( /\// )
+		m_segment  = self.current_wal[ :segment ]
+		m_offset   = self.current_wal[ :offset ]
+		m_segbytes = self.current_wal[ :segbytes ]
+
+		return (( m_segment.hex - s_segment.hex ) * m_segbytes) + ( m_offset.hex - s_offset.hex )
+	end
+
+end
+
+
+### Parse command line arguments.  Return a struct of global options.
+###
+def parse_args( args )
+	options          = OpenStruct.new
+	options.database = 'postgres'
+	options.port     = 5432
+	options.user     = Etc.getpwuid( Process.uid ).name
+	options.sslmode  = 'prefer'
+
+	opts = OptionParser.new do |opts|
+		opts.banner = "Usage: #{$0} [options] <master> <slave> [slave2, slave3...]"
+
+		opts.separator ''
+		opts.separator 'Connection options:'
+
+		opts.on( '-d', '--database DBNAME',
+				"specify the database to connect to (default: \"#{options.database}\")" ) do |db|
+			options.database = db
+		end
+
+		opts.on( '-h', '--host HOSTNAME', 'database server host' ) do |host|
+			options.host = host
+		end
+
+		opts.on( '-p', '--port PORT', Integer,
+				"database server port (default: \"#{options.port}\")" ) do |port|
+			options.port = port
+		end
+
+		opts.on( '-U', '--user NAME',
+				"database user name (default: \"#{options.user}\")" ) do |user|
+			options.user = user
+		end
+
+		opts.on( '-W', 'force password prompt' ) do |pw|
+			print 'Password: '
+			begin
+				system 'stty -echo'
+				options.pass = $stdin.gets.chomp
+			ensure
+				system 'stty echo'
+				puts
+			end
+		end
+
+		opts.separator ''
+		opts.separator 'Other options:'
+
+		opts.on_tail( '--help', 'show this help, then exit' ) do
+			$stderr.puts opts
+			exit
+		end
+
+		opts.on_tail( '--version', 'output version information, then exit' ) do
+			puts PGMonitor::VERSION
+			exit
+		end
+	end
+
+	opts.parse!( args )
+	return options
+end
+
+
+
+if __FILE__ == $0
+	opts = parse_args( ARGV )
+	raise ArgumentError, "At least two PostgreSQL servers are required." if ARGV.length < 2
+	mon = PGMonitor.new( opts, ARGV )
+
+	mon.check
+	if mon.failures.empty?
+		puts "All is well!"
+		exit 0
+	else
+		puts "Database replication delayed or broken."
+		mon.failures.each do |bad|
+			puts "%s: %s" % [ bad[ :host ], bad[ :error ] ]
+		end
+		exit 1
+	end
+end
+
+

data/sample/test_binary_values.rb

@@ -0,0 +1,33 @@
+# -*- ruby -*-1.9.1
+
+require 'pg'
+
+db = PG.connect( :dbname => 'test' )
+db.exec "DROP TABLE IF EXISTS test"
+db.exec "CREATE TABLE test (a INTEGER, b BYTEA)"
+
+a = 42
+b = [1, 2, 3]
+db.exec "INSERT INTO test(a, b) VALUES($1::int, $2::bytea)",
+	[a, {:value => b.pack('N*'), :format => 1}]
+
+db.exec( "SELECT a::int, b::bytea FROM test LIMIT 1", [], 1 ) do |res|
+
+	res.nfields.times do |i|
+		puts "Field %d is: %s, a %s (%s) column from table %p" % [
+			i,
+			res.fname( i ),
+			db.exec( "SELECT format_type($1,$2)", [res.ftype(i), res.fmod(1)] ).getvalue(0,0),
+			res.fformat( i ).zero? ? "string" : "binary",
+			res.ftable( i ),
+		]
+	end
+
+	res.each do |row|
+		puts "a = #{row['a'].inspect}"
+		puts "a (unpacked) = #{row['a'].unpack('N*').inspect}"
+		puts "b = #{row['b'].unpack('N*').inspect}"
+	end
+end
+
+

data/sample/wal_shipper.rb

@@ -0,0 +1,434 @@
+# -*- ruby -*-
+#
+# A script to wrap ssh and rsync for PostgreSQL WAL files shipping.
+# Mahlon E. Smith <mahlon@martini.nu>
+#
+# Based off of Joshua Drake's PITRTools concept, but with some important
+# differences:
+#
+#	- Only supports PostgreSQL >= 8.3
+#	- No support for rsync version < 3
+#	- Only shipping, no client side sync (too much opportunity for failure,
+#	  and it's easy to get a base backup manually)
+#	- WAL files are only stored once, regardless of how many
+#	  slaves are configured or not responding, and are removed from
+#	  the master when they are no longer needed.
+#	- Each slave can have completely distinct settings, instead
+#	  of a single set of options applied to all slaves
+#	- slave sync can be individually paused from the master
+#	- can run synchronously, or if you have a lot of slaves, threaded async mode
+#	- It's ruby, instead of python.  :)
+#
+# wal_shipper is configurable via an external YAML file, and will create
+# a template on its first run -- you'll need to modify it!  It expects
+# a directory structure like so:
+#
+#	postgres/
+#		data/...
+#		bin/wal_shipper.rb
+#		etc/wal_shipper.conf   <-- YAML settings!
+#		wal/
+#
+# It should be loaded from the PostgreSQL master's postgresql.conf
+# as such, after putting it into your postgres user homedir under 'bin':
+#
+#	archive_command = '/path/to/postgres_home/bin/wal_shipper.rb %p'
+#
+# Passwordless ssh keys need to be set up for the postgres user on all
+# participating masters and slaves.
+#
+# You can use any replay method of your choosing on the slaves.
+# Here's a nice example using pg_standby, to be put in data/recovery.conf:
+#
+#	restore_command = 'pg_standby -t /tmp/pgrecovery.done -s5 -w0 -c /path/to/postgres_home/wal_files/ %f %p %r'
+#
+# Or, here's another simple alternative data/recovery.conf, for using WAL shipping
+# alongside streaming replication:
+#
+#    standby_mode = 'on'
+#    primary_conninfo = 'host=master.example.com port=5432 user=repl password=XXXXXXX'
+#    restore_command = 'cp /usr/local/pgsql/wal/%f %p'
+#    trigger_file = '/usr/local/pgsql/pg.become_primary'
+#    archive_cleanup_command = '/usr/local/bin/pg_archivecleanup /usr/local/pgsql/wal %r'
+#
+#========================================================================================
+
+
+require 'pathname'
+require 'yaml'
+require 'fileutils'
+require 'ostruct'
+
+
+### Encapsulate WAL shipping functionality.
+###
+module WalShipper
+
+	### Send messages to the PostgreSQL log files.
+	###
+	def log( msg )
+		return unless @debug
+		puts "WAL Shipper: %s" % [ msg ]
+	end
+
+
+	### An object that represents a single destination from the
+	### configuration file.
+	###
+	class Destination < OpenStruct
+		include WalShipper
+
+		### Create a new WalShipper::Destination object.
+		def initialize( dest, debug=false )
+			@debug = debug
+			super( dest )
+			self.validate
+		end
+
+		#########
+		protected
+		#########
+
+
+		### Check for required keys and normalize various keys.
+		###
+		def validate
+			# Check for required destination keys
+			%w[ label kind ].each do |key|
+				if self.send( key.to_sym ).nil?
+					self.log "Destination %p missing required '%s' key." % [ self, key ]
+					self.invalid = true
+				end
+			end
+
+			# Ensure paths are Pathnames for the 'file' destination type.
+			self.path = Pathname.new( self.path ) if self.kind == 'file'
+
+			if self.kind == 'rsync-ssh'
+				self.port ||= 22
+				self.user = self.user ? "#{self.user}@" : ''
+			end
+		end
+	end # Class Destination
+
+
+
+	### Class for creating new Destination objects and determining how to
+	### ship WAL files to them.
+	###
+	class Dispatcher
+		include WalShipper
+
+		### Create a new Shipper object, given a +conf+ hash and a +wal+ file
+		### Pathname object.
+		###
+		def initialize( wal, conf )
+			# Make the config keys instance variables.
+			conf.each_pair {|key, val| self.instance_variable_set( "@#{key}", val ) }
+
+			# Spool directory check.
+			#
+			@spool = Pathname.new( @spool )
+			@spool.exist? or raise "The configured spool directory (%s) doesn't exist." % [ @spool ]
+
+			# Stop right away if we have disabled shipping.
+			#
+			unless @enabled
+				self.log "WAL shipping is disabled, queuing segment %s" % [ wal.basename ]
+				exit 1
+			end
+
+			# Instantiate Destination objects, creating new spool directories
+			# for each.
+			#
+			@destinations.
+				collect!{|dest| WalShipper::Destination.new( dest, @debug ) }.
+				reject  {|dest| dest.invalid }.
+				collect do |dest|
+					dest.spool = @spool + dest.label
+					dest.spool.mkdir( 0711 ) unless dest.spool.exist?
+					dest
+				end
+
+			# Put the WAL file into the spool for processing!
+			#
+			@waldir = @spool + 'wal_segments'
+			@waldir.mkdir( 0711 ) unless @waldir.exist?
+
+			self.log "Copying %s to %s" % [ wal.basename, @waldir ]
+			FileUtils::cp wal, @waldir
+
+			# 'wal' now references the copy.  The original is managed and auto-expired
+			# by PostgreSQL when a new checkpoint segment it reached.
+			@wal = @waldir + wal.basename
+		end
+
+
+		### Create hardlinks for the WAL file into each of the destination directories
+		### for separate queueing and recording of what was shipped successfully.
+		###
+		def link
+			@destinations.each do |dest|
+				self.log "Linking %s into %s" % [ @wal.basename, dest.spool.basename ]
+				FileUtils::ln @wal, dest.spool, :force => true
+			end
+		end
+
+
+		### Decide to be synchronous or threaded, and delegate each destination
+		### to the proper ship method.
+		###
+		def dispatch
+			# Synchronous mode.
+			#
+			unless @async
+				self.log "Performing a synchronous dispatch."
+				@destinations.each {|dest| self.dispatch_dest( dest ) }
+				return
+			end
+
+			tg = ThreadGroup.new
+
+			# Async, one thread per destination
+			#
+			if @async_max.nil? || @async_max.to_i.zero?
+				self.log "Performing an asynchronous dispatch: one thread per destination."
+				@destinations.each do |dest|
+					t = Thread.new do
+						Thread.current.abort_on_exception = true
+						self.dispatch_dest( dest )
+					end
+					tg.add( t )
+				end
+				tg.list.each {|t| t.join }
+				return
+			end
+
+			# Async, one thread per destination, in groups of asynx_max size.
+			#
+			self.log "Performing an asynchronous dispatch: one thread per destination, %d at a time." % [ @async_max ]
+			all_dests = @destinations.dup
+			dest_chunks = []
+			until all_dests.empty? do
+				dest_chunks << all_dests.slice!( 0, @async_max )
+			end
+
+			dest_chunks.each do |chunk|
+				chunk.each do |dest|
+					t = Thread.new do
+						Thread.current.abort_on_exception = true
+						self.dispatch_dest( dest )
+					end
+					tg.add( t )
+				end
+
+				tg.list.each {|t| t.join }
+			end
+
+			return
+		end
+
+
+		### Remove any WAL segments no longer needed by slaves.
+		###
+		def clean_spool
+			total = 0
+			@waldir.children.each do |wal|
+				if wal.stat.nlink == 1
+					total += wal.unlink
+				end
+			end
+
+			self.log "Removed %d WAL segment%s." % [ total, total == 1 ? '' : 's' ]
+		end
+
+
+
+		#########
+		protected
+		#########
+
+		### Send WAL segments to remote +dest+ via rsync+ssh.
+		### Passwordless keys between the user running this script (postmaster owner)
+		### and remote user need to be set up in advance.
+		###
+		def ship_rsync_ssh( dest )
+			if dest.host.nil?
+				self.log "Destination %p missing required 'host' key.  WAL is queued." % [ dest.host ]
+				return
+			end
+
+			rsync_flags = '-zc'
+			ssh_string = "%s -o ConnectTimeout=%d -o StrictHostKeyChecking=no -p %d" %
+				[ @ssh, @ssh_timeout || 10, dest.port ]
+			src_string = ''
+			dst_string = "%s%s:%s/" % [ dest.user, dest.host, dest.path ]
+
+			# If there are numerous files in the spool dir, it means there was
+			# an error transferring to this host in the past.  Try and ship all
+			# WAL segments, instead of just the new one.  PostgreSQL on the slave
+			# side will "do the right thing" as they come in, regardless of
+			# ordering.
+			#
+			if dest.spool.children.length > 1
+				src_string = dest.spool.to_s + '/'
+				rsync_flags << 'r'
+			else
+				src_string = dest.spool + @wal.basename
+			end
+
+
+			ship_wal_cmd = [
+				@rsync,
+				@debug ? (rsync_flags << 'vh') : (rsync_flags << 'q'),
+				'--remove-source-files',
+				'-e', ssh_string,
+				src_string, dst_string
+			]
+
+			self.log "Running command '%s'" % [ ship_wal_cmd.join(' ') ]
+			system *ship_wal_cmd
+
+			# Run external notification program on error, if one is configured.
+			#
+			unless $?.success?
+				self.log "Ack!  Error while shipping to %p, WAL is queued." % [ dest.label ]
+				system @error_cmd, dest.label if @error_cmd
+			end
+		end
+
+
+		### Copy WAL segments to remote path as set in +dest+.
+		### This is useful for longer term PITR, copying to NFS shares, etc.
+		###
+		def ship_file( dest )
+			if dest.path.nil?
+				self.log "Destination %p missing required 'path' key.  WAL is queued." % [ dest ]
+				return
+			end
+			dest.path.mkdir( 0711 ) unless dest.path.exist?
+
+			# If there are numerous files in the spool dir, it means there was
+			# an error transferring to this host in the past.  Try and ship all
+			# WAL segments, instead of just the new one.  PostgreSQL on the slave
+			# side will "do the right thing" as they come in, regardless of
+			# ordering.
+			#
+			if dest.spool.children.length > 1
+				dest.spool.children.each do |wal|
+					wal.unlink if self.copy_file( wal, dest.path, dest.label, dest.compress )
+				end
+			else
+				wal = dest.spool + @wal.basename
+				wal.unlink if self.copy_file( wal, dest.path, dest.label, dest.compress )
+			end
+		end
+
+
+		### Given a +wal+ Pathname, a +path+ destination, and the destination
+		### label, copy and optionally compress a WAL file.
+		###
+		def copy_file( wal, path, label, compress=false )
+			dest_file = path + wal.basename
+			FileUtils::cp wal, dest_file
+			if compress
+				system *[ 'gzip', '-f', dest_file ]
+				raise "Error while compressing: %s" % [ wal.basename ] unless $?.success?
+			end
+			self.log "Copied %s%s to %s." %
+				[ wal.basename, compress ? ' (and compressed)' : '', path ]
+			return true
+		rescue => err
+			self.log "Ack!  Error while copying '%s' (%s) to %p, WAL is queued." %
+				[ wal.basename, err.message, path ]
+			system @error_cmd, label if @error_cmd
+			return false
+		end
+
+
+		### Figure out how to send the WAL file to its intended destination +dest+.
+		###
+		def dispatch_dest( dest )
+			if ! dest.enabled.nil? && ! dest.enabled
+				self.log "Skipping explicity disabled destination %p, WAL is queued." % [ dest.label ]
+				return
+			end
+
+			# Send to the appropriate method.  ( rsync-ssh --> ship_rsync_ssh )
+			#
+			meth = ( 'ship_' + dest.kind.gsub(/-/, '_') ).to_sym
+			if WalShipper::Dispatcher.method_defined?( meth )
+				self.send( meth, dest )
+			else
+				self.log "Unknown destination kind %p for %p.  WAL is queued." % [ dest.kind, dest.label ]
+			end
+		end
+	end
+end
+
+# Ship the WAL file!
+#
+if __FILE__ == $0
+	CONFIG_DIR = Pathname.new( __FILE__ ).dirname.parent + 'etc'
+	CONFIG     = CONFIG_DIR + 'wal_shipper.conf'
+
+	unless CONFIG.exist?
+		CONFIG_DIR.mkdir( 0711 ) unless CONFIG_DIR.exist?
+		CONFIG.open('w') {|conf| conf.print(DATA.read) }
+		CONFIG.chmod( 0644 )
+		puts "No WAL shipping configuration found, default file created."
+	end
+
+	wal  = ARGV[0] or raise "No WAL file was specified on the command line."
+	wal  = Pathname.new( wal )
+	conf = YAML.load( CONFIG.read )
+
+	shipper = WalShipper::Dispatcher.new( wal, conf )
+	shipper.link
+	shipper.dispatch
+	shipper.clean_spool
+end
+
+
+__END__
+---
+# Spool from pg_xlog to the working area?
+# This must be set to 'true' for wal shipping to function!
+enabled: false
+
+# Log everything to the PostgreSQL log files?
+debug: true
+
+# The working area for WAL segments.
+spool: /opt/local/var/db/postgresql84/wal
+
+# With multiple slaves, ship WAL in parallel, or be synchronous?
+async: false
+
+# Put a ceiling on the parallel threads?
+# '0' or removing this option uses a thread for each destination,
+# regardless of how many you have.  Keep in mind that's 16 * destination
+# count megs of simultaneous bandwidth.
+async_max: 5
+
+# Paths and settings for various binaries.
+rsync: /usr/bin/rsync
+ssh: /usr/bin/ssh
+ssh_timeout: 10
+
+destinations:
+
+- label: rsync-example
+  port: 2222
+  kind: rsync-ssh
+  host: localhost
+  user: postgres
+  path: wal    # relative to the user's homedir on the remote host
+  enabled: false
+
+- label: file-example
+  kind: file
+  compress: true
+  enabled: true
+  path: /tmp/someplace
+