Bug #19237
closedHash default_proc is not thread-safe to lazy-initialize value for a given key
Description
1000.times do
h = Hash.new do |hash, key|
# Note, with [] instead of Array.new below it seems to work
# but then e.g. anything else in this block like any call also breaks it
hash[key] = Array.new
end
go = false
threads = 100.times.map do
Thread.new do
Thread.pass until go
h[:na] << true
end
end
go = true
threads.each(&:join)
raise "Expected 100 elements but was #{h[:na].size}" if h[:na].size != 100
end
gives (in 3 runs):
concurrent_hash_default_proc.rb:17:in `block in <main>': Expected 100 elements but was 1 (RuntimeError)
concurrent_hash_default_proc.rb:17:in `block in <main>': Expected 100 elements but was 3 (RuntimeError)
concurrent_hash_default_proc.rb:17:in `block in <main>': Expected 100 elements but was 2 (RuntimeError)
So what happens is the same Hash entry gets assigned multiple times, which feels quite unexpected and cause fairly surprising bugs (e.g. elements "disappearing" etc).
Any thoughts on how to solve that in CRuby?
In my PhD thesis one way I found is to actually pass a different object than the Hash itself as the first argument to the block.
See https://eregon.me/blog/assets/research/thesis-thread-safe-data-representations-in-dynamic-languages.pdf page 83 Idiomatic Concurrent Hash Operations
. In short, it replaces []=
calls in the initializer block with put_if_absent
by passing a different object than the Concurrent::Hash itself, which overrides []=
and delegates the rest.
#19069 could be another way but there are compatibility issues (e.g. storing when it previously would not for Hash.new { |h,k| k * 3 }
.
There is also the question whether the block should be allowed to execute more than once for a given key.
I think that is difficult to solve (probably the only way is via some lock, but then that can lead to deadlocks), and less important than ensuring the value is only assigned once for a given key.
Note that concurrent-ruby has the same issue because it uses ::Array/::Hash on CRuby: https://github.com/ruby-concurrency/concurrent-ruby/issues/970#issuecomment-1346338557 There is also more discussions about trade-offs there which apply to Hash
too.
Updated by mame (Yusuke Endoh) almost 2 years ago
Discussed at the dev meeting. @matz (Yukihiro Matsumoto) said "The performance penalty is not acceptable. Addition to the documentation looks good enough."
Updated by jeremyevans (Jeremy Evans) almost 2 years ago
- Status changed from Open to Closed
Applied in changeset git|ffd52412ab813854d134dbbc2b60f698fe536487.
Document thread-safety issues with Hash default_proc
Fixes [Bug #19237]