ruby-core@ruby-lang.org archive (unofficial mirror)
 help / color / mirror / Atom feed
From: "byroot (Jean Boussier)" <noreply@ruby-lang.org>
To: ruby-core@ruby-lang.org
Subject: [ruby-core:107733] [Ruby master Bug#18599] `vm->loaded_features_snapshot` became very inefficient in Ruby 3.1
Date: Wed, 23 Feb 2022 16:09:33 +0000 (UTC)	[thread overview]
Message-ID: <redmine.journal-96656.20220223160932.7941@ruby-lang.org> (raw)
In-Reply-To: redmine.issue-18599.20220222173705.7941@ruby-lang.org

Issue #18599 has been updated by byroot (Jean Boussier).

Backport changed from 2.6: DONTNEED, 2.7: DONTNEED, 3.0: DONTNEED, 3.1: UNKNOWN to 2.6: DONTNEED, 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN

Ok, so my assessment about shared arrays wasn't quite correct. What caused the difference between 3.0 and 3.1 here is http://github.com/ruby/ruby/commit/aeae6e2842e1702dfb89b8ae69b48c4f5f64c662

Previously the shared array would be force recycled immediately, which was hiding the problem, but it was already there.

With the help of @peterzhu2118, we figured a fairly simple patch for this: https://github.com/ruby/ruby/pull/5593


----------------------------------------
Bug #18599: `vm->loaded_features_snapshot` became very inefficient in Ruby 3.1
https://bugs.ruby-lang.org/issues/18599#change-96656

* Author: byroot (Jean Boussier)
* Status: Open
* Priority: Normal
* ruby -v: ruby 3.1.1p18 (2022-02-18 revision 53f5fc4236) [x86_64-darwin20]
* Backport: 2.6: DONTNEED, 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN
----------------------------------------
A change in how shared arrays behave has caused a fairly major performance regression in `Kernel.require`. In short on every single call, a full copy of `$LOADED_FEATURES` is made.

In our app this causes over 6GiB worth of useless allocations.

### Reproduction

```ruby
# frozen_string_literal: true
require "objspace"
files = 5.times.map { |i| "/tmp/foo-#{i}.rb" }
files.each { |f| File.write(f, "") }

require files.shift
4.times { GC.start }
generation = GC.count

ObjectSpace.trace_object_allocations do
  files.each { |f| require f }
end

ObjectSpace.dump_all(output: File.open("/tmp/allocated.heap", "w+"), since: generation)
ObjectSpace.dump_all(output: File.open("/tmp/full.heap", "w+"))
```

If you inspect the `allocated.heap`, you'll notice 4 array, each increasingly bigger.

```
$ grep ARRAY /tmp/allocated.heap 
{"address":"0x10fe2b3b8", "type":"ARRAY", "frozen":true, "length":92, "references":["0x10fcd3da8", "0x10fcca2a8", "0x10fcc21c0", ...], "file":"<internal:/opt/rubies/3.1.0/lib/ruby/3.1.0/rubygems/core_ext/kernel_require.rb>", "line":85, "method":"require", "generation":13, "memsize":776, "flags":{"wb_protected":true}}
...
{"address":"0x10fe2b688", "type":"ARRAY", "frozen":true, "length":91, "references":["0x10fcd3da8", "0x10fcca2a8", "0x10fcc21c0", ...], "file":"<internal:/opt/rubies/3.1.0/lib/ruby/3.1.0/rubygems/core_ext/kernel_require.rb>", "line":85, "method":"require", "generation":13, "memsize":768, "flags":{"wb_protected":true}}
...
{"address":"0x10fe2b958", "type":"ARRAY", "frozen":true, "length":89, "references":["0x10fcd3da8", "0x10fcca2a8", "0x10fcc21c0", ...], "file":"<internal:/opt/rubies/3.1.0/lib/ruby/3.1.0/rubygems/core_ext/kernel_require.rb>", "line":85, "method":"require", "generation":13, "memsize":752, "flags":{"wb_protected":true}}
...
{"address":"0x10fe2bc28", "type":"ARRAY", "frozen":true, "length":88, "references":["0x10fcd3da8", "0x10fcca2a8", "0x10fcc21c0", ...], "file":"<internal:/opt/rubies/3.1.0/lib/ruby/3.1.0/rubygems/core_ext/kernel_require.rb>", "line":85, "method":"require", "generation":13, "memsize":744, "flags":{"wb_protected":true}}
...
```

And if you lookup the addresses in `full.heap`, it makes no doubt that it's `$LOADED_FEATURES`.

Note however how the array is `"frozen":true`, so it's not `$LOADED_FEATURES` directly, but the shared array that backs `vm->loaded_features` and `vm->loaded_features_snapshot`.


### More scoped reproduction

If we scope on shared arrays behavior specifically, we can get a much smaller reproduction:

```ruby
array = 10.times.to_a
copy = array.dup
before = GC.stat(:total_allocated_objects)
copy << 1
p GC.stat(:total_allocated_objects) - before # 1 on Ruby 3.1, 0 on 3.0 and older
```

In short, back in 3.0 and older, appending to a shared array wouldn't "unshare" the array, but starting in 3.1 it now does.
The problem being that `Kernel.require` heavily assumed this implementation detail:

```c
static void
reset_loaded_features_snapshot(rb_vm_t *vm)
{
    rb_ary_replace(vm->loaded_features_snapshot, vm->loaded_features);
}

rb_provide_feature(rb_vm_t *vm, VALUE feature)
{
    //...
    get_loaded_features_index(vm);
    rb_ary_push(features, rb_fstring(feature));
    features_index_add(vm, feature, INT2FIX(RARRAY_LEN(features)-1));
    reset_loaded_features_snapshot(vm);
}
```

`rb_provide_feature` first append to `$LOADED_FEATURES` and then resets `vm->loaded_features_snapshot` to be a shared copy of `$LOADED_FEATURES`.





-- 
https://bugs.ruby-lang.org/

  reply	other threads:[~2022-02-23 16:09 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-02-22 17:37 [ruby-core:107715] [Ruby master Bug#18599] `vm->loaded_features_snapshot` became very inefficient in Ruby 3.1 byroot (Jean Boussier)
2022-02-23 16:09 ` byroot (Jean Boussier) [this message]
2022-02-23 17:05 ` [ruby-core:107734] " byroot (Jean Boussier)
2022-11-11 15:32 ` [ruby-core:110706] " thomthom (Thomas Thomassen)
2022-11-11 23:38 ` [ruby-core:110713] " byroot (Jean Boussier)
2022-11-13  8:08 ` [ruby-core:110735] " nagachika (Tomoyuki Chikanaga)

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-list from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: https://www.ruby-lang.org/en/community/mailing-lists/

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=redmine.journal-96656.20220223160932.7941@ruby-lang.org \
    --to=ruby-core@ruby-lang.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).