Fixes #20776 - Do not re-render template if it can be loaded

[adamruzicka]

When the user visits the job invocation details page, we rendered the
templates for first 20 hosts, which made the page load slowly. With
this change we first try to look into a task for this host, which may
contain the already rendered template.

[houndci-bot]

Lint/AssignmentInCondition: Use == if you meant to do a comparison or wrap the expression in parentheses to indicate you meant to assign in a condition.

[ares]

one question, otherwise looks good so far, I'll test it in a bit

[ares]

what are the conditions for input not yet have rendered template? e.g. future execution?

[adamruzicka]

For future execution the task object would not exist at all (only the parent task exists at that time). This is to prevent a race condition where the RunHostJob already exists, but its ProxyAction was not planned yet

[ares]

Sorry I failed to look at your benchmarks, but you convinced me already :-) I'll do a quick test, but 👍 so far

[adamruzicka]

I guess I should have attached the benchmarks to the PR

[adamruzicka]

Not probably the best of benchmarks but should give a rough idea about this affects performance

def benchmark(count, template_invocation, target)
  require 'benchmark'
  Benchmark.benchmark do |bm|
    bm.report('db load') do
      count.times { load_template_from_task(template_invocation, target) }
    end

    bm.report('render') do
      count.times do
        renderer = InputTemplateRenderer.new(template_invocation.template, target, template_invocation)
        renderer.preview
      end
    end
  end
end

pry> template_invocation.template.template
"<%= input(\"command\") %>"

pry> benchmark 100, template_invocation, target
db load  0.061721   0.000000   0.061721 (  0.062268)
render  0.310252   0.021698   0.331950 (  0.329095)

pry> template_invocation.template.name
"Package Action - SSH Default"

pry> benchmark 100, template_invocation, target
db load  0.207657   0.027209   0.234866 (  0.337088)
render  3.192303   0.043801   3.236104 (  3.247061)

[ares]

Thanks @adamruzicka, works well! Failures are unrelated, merging.