Keenan Brock kbrock

Looking at the difference between using ActiveSupport::Notification, and rolling our own metrics counters.

When logging, Notifications are 66% slower than manually adding the metrics code. The metrics code is actually 50% slower when not logging the results (no subscriber). Having no subscriber does speed up performance for all the cases.

The difference for timing for payload is probably due to allocations. Having a fixed payload ended up being equivalent to the no payload case. This is a reminder for us to pay attention to what we pass in the payload.

gem
Json oj	40.4 i/s
MessagePack	38.3 i/s	same
binary json	22.0 i/s	1.84x slower
Json yajl	19.0 i/s	2.13x slower
ruby json	15.6 i/s	2.59x slower
ruby marshal	13.9 i/s	2.90x slower
ruby yaml	0.4 i/s	111.99x slower
Yaml Psych	0.4 i/s	112.09x slower

ms	bytes	objects	queries	query (ms)	rows	`comments`
290.5	21,380,122*	225,801	2	28.0	1,598	before
244.9	18,410,393*	188,824	1	4.6		after

* Memory usage does not reflect 262 freed objects. (for both of them)

My comments on 7388

- s = MiqSearch.find(@sb[:planning][:options][:filter_value])
- s.options ||= {}
- s.options[:userid] = session[:userid]
- s.options[:results_format] = :objects
-vms, attrs = s.search
+ vms = MiqSearch.find(@sb[:planning][:options][:filter_value]).filtered(nil, :userid => current_user)

	def reload_models
	[Vm, Host, Storage].each { \|m\| m.update_all(:last_perf_capture_on => 50.minutes.ago) } # <== still tweaking
	[MiqTask, MiqQueue, Metric, MetricRollup, VimPerformanceState].each { \|m\| m.truncate}
	ActiveRecord::Base.connection.execute("vacuum full")
	Module.clear_all_cache_with_timeout
	[[MiqProductFeature, :@feature_cache], [MiqProductFeature, :@obj_cache], [BottleneckEvent, :@event_definitions],
	[Tenant, :@root_tenant]].each { \|i, ivar\| i.instance_variable_set(ivar, nil) }
	MiqServer.my_server ; MiqServer.my_zone
	[ExtManagementSystem, Host, EmsCluster, Tagging, Tag, HostStorage, Storage, Vm, MiqRegion, MiqTask, Relationship, Disk,
	Endpoint, Authentication, ServerRole, MiqWorker, MiqQueue, MetricRollup, BottleneckEvent].each { \|m\| m.first }

	<html>
	<head><style>
	table {border-collapse:collapse;}
	table, td, th { border:1px solid black; }
	.f0, a.f0 { color: #999; }
	.f1, a.f1 { color: #ccc; }
	</style>
	</head>
	<body>
	<h2>!</h2>

	#!/usr/bin/env ruby
	require 'thread'

	COLLECTOR_COUNT = 2 # define to run in separate threads. comment to run inline
	class Db
	def initialize
	@mutex = Mutex.new
	@data = {}
	end
	def []=(n, v) ; @mutex.synchronize { @data[n] = v } ; end

	#!/usr/bin/env ruby

	begin
	require "bundler/inline"
	rescue LoadError => e
	$stderr.puts "Bundler version 1.10 or later is required. Please update your Bundler"
	raise e
	end

	gemfile(true) do

	require 'benchmark/ips'

	def eval_eval_true
	name = "Benchmark"
	eval("defined?(#{name}) && #{name}.class").to_s.eql?('Module')
	end

	def eval_eval_false
	name = "XABCDE"
	eval("defined?(#{name}) && #{name}.class").to_s.eql?('Module')

	create index i_hosts_i_h on hosts(name, id);
	create index i_hosts_h_i on hosts(id, name);

	explain analyze verbose
	SELECT "vms".id FROM "vms" left join "hosts" on "hosts"."id" = "vms"."host_id" ORDER BY "hosts"."name"
	;
	QUERY PLAN
	-----------------------------------------------------------------------------------------------------------------------------
	Sort (cost=440.76..448.26 rows=3000 width=36) (actual time=30.724..30.884 rows=3000 loops=1)
	Output: vms.id, hosts.name