Project

General

Profile

Actions

Feature #18589

closed

Finer-grained constant invalidation

Added by kddnewton (Kevin Newton) about 2 years ago. Updated over 1 year ago.

Status:
Closed
Assignee:
-
Target version:
-
[ruby-core:107603]

Description

This is related to https://github.com/ruby/ruby/pull/5433.

Current behavior

Caches depend on a global counter. All constant mutations cause all caches to be invalidated.

class A
  B = 1
end

def foo
  A::B # inline cache depends on global counter
end

foo # populate inline cache
foo # hit inline cache

C = 1 # global counter increments, all caches are invalidated

foo # misses inline cache due to `C = 1`

Proposed behavior

Caches depend on name components. Only constant mutations with corresponding names will invalidate the cache.

class A
  B = 1
end

def foo
  A::B # inline cache depends constants named "A" and "B"
end

foo # populate inline cache
foo # hit inline cache

C = 1 # caches that depend on the name "C" are invalidated

foo # hits inline cache because IC only depends on "A" and "B"

Examples of breaking the new cache:

module C
  # Breaks `foo` cache because "A" constant is set and the cache in foo depends
  # on "A" and "B"
  class A; end
end

B = 1

We expect the new cache scheme to be invalidated less often because names aren't frequently reused. With the cache being invalidated less, we can rely on its stability more to keep our constant references fast and reduce the need to throw away generated code in YJIT.

Performance benchmarks

The following benchmark (included in this pull request) performs about 2x faster than master.

CONSTANT1 = 1
CONSTANT2 = 1
CONSTANT3 = 1
CONSTANT4 = 1
CONSTANT5 = 1

def constants
  [CONSTANT1, CONSTANT2, CONSTANT3, CONSTANT4, CONSTANT5]
end

500_000.times do
  constants
  INVALIDATE = true
end

In terms of macro benchmarks, I ran with this code on railsbench and there was not a statistically significant different in startup time or overall runtime performance.

@byroot (Jean Boussier) also ran performance benchmarks on our production application. He noticed that there were several cache busts related to Object#extend (from core libraries), ActiveRecord::Relation#extending (from Rails), and autoload (from various gems, both internal and external). After a lot of work, the cache busts went down:

Cache bust changes

but they're still frequent enough that it's a problem. These changes had a measurable performance difference in request speed:

Request speed changes

Memory benchmarks

In terms of memory, this includes an increase in VM size by about 500KiB when running on railsbench. This is because we're now tracking cache associations ({ ID => IC[] }) on the VM to know how to invalidate specific caches when constants change.

I booted Shopify's core monolith with this branch as well. It increased proportional to the number of constant caches found in the application. For each constant cache 1 level deep (e.g., Foo) the increase is about 33 bytes. For a constant cache 2 levels deep (e.g., Foo::Bar) the increase is about 67 bytes. The overall increase was around 16Mb or about 1% of the total retained memory.

Actions

Also available in: Atom PDF

Like0
Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0