Small optimizations

[tenderlove]

Hi,

I've added a couple small optimizations. The first one removes a runtime check of a constant. We can know whether or not Process.clock_gettime is supported at file require time, so we can eliminate a conditional in time.

The other optimization eliminates object allocations inside the Telemetry class. We can (and should) use monotonic time there in order to eliminate allocations and also not be subject to clock changes. Unfortunately that file seems to be tested with Timecop, and Timecop apparently doesn't support mocking Process.clock_gettime: travisjeffery/timecop#220

The changes I made should be equivalent to the Time.now.to_i call. I will try to get the tests passing later, but maybe someone more familiar can do it too?

I used this as a benchmark:

# frozen_string_literal: true

require 'datadog/statsd'
require 'logger'
require 'benchmark/ips'
require 'stackprof'

logger = Logger.new $stderr
logger.level = Logger::INFO

thing = Datadog::Statsd.new('localhost', 1234,
                    namespace: "sample_ns",
                    sample_rate: nil,
                    tags: %w{ abc def },
                    logger: nil,
                    telemetry_flush_interval: -1,
                   )

Benchmark.ips do |x|
  x.report("time") { thing.time("foobar") { } }
end

On master, I get:

$ ruby -I lib:spec time_speed.rb
Warming up --------------------------------------
                time     4.463k i/100ms
Calculating -------------------------------------
                time     47.775k (± 6.9%) i/s -    241.002k in   5.076456s

On this branch I get:

$ ruby -I lib:spec time_speed.rb
Warming up --------------------------------------
                time     4.700k i/100ms
Calculating -------------------------------------
                time     50.365k (± 7.7%) i/s -    253.800k in   5.078403s

I also measured allocations. Here is the benchmark for allocations:

# frozen_string_literal: true

require 'datadog/statsd'
require 'logger'
require 'benchmark/ips'
require 'stackprof'

logger = Logger.new $stderr
logger.level = Logger::INFO

thing = Datadog::Statsd.new('localhost', 1234,
                    namespace: "sample_ns",
                    sample_rate: nil,
                    tags: %w{ abc def },
                    logger: nil,
                    telemetry_flush_interval: -1,
                   )

def allocs
  x = GC.stat(:total_allocated_objects)
  yield
  GC.stat(:total_allocated_objects) - x
end

p allocs { 50.times { thing.time("foobar") { } } }

On master:

$ ruby -I lib:spec time_speed.rb
1752

On this branch:

$ ruby -I lib:spec time_speed.rb
1160

I guess that's about 11 allocations saved per call to time.

[kbogtob]

Hello @tenderlove (nice nickname btw) and thanks a lot for your contribution! I'm going to do couple tests and I will review the PR. This is actually very smart optimizations!

[kbogtob]

Great job! That's really smart and useful optimizations. But could you check the CI and fix the number of allocations comparing to each ruby versions:
https://travis-ci.org/github/DataDog/dogstatsd-ruby/builds/668978123?utm_source=github_status&utm_medium=notification

[kbogtob]

Hi @tenderlove, I have fixed the tests for you. Would you be okay for me to push to your branch so that I can update the PR? Or could you cherry-pick the commits on this branch? https://github.com/DataDog/dogstatsd-ruby/tree/tenderlove/small-optimizations
Thanks!

[tenderlove]

@kbogtob I pulled your changes in. Thanks!

[kbogtob]

Super! I'll merge the PR right after #142.