[ruby-core:82185] [Ruby trunk Bug#13772] Memory leak recycling stacks for threads in 2.4.1

Issue #13772 has been updated by sam.saffron (Sam Saffron).


valgrind reports tons of these:

==18152== 463,665 (86,200 direct, 377,465 indirect) bytes in 431 blocks are definitely lost in loss record 16,407 of 16,429
==18152==    at 0x4C2FB55: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==18152==    by 0x14BECE: objspace_xcalloc (gc.c:7944)
==18152==    by 0x170E12: iseq_alloc (iseq.c:215)
==18152==    by 0x170E12: rb_iseq_new_with_opt (iseq.c:478)
==18152==    by 0x2E3B2F: new_child_iseq (compile.c:1119)
==18152==    by 0x2E3B2F: iseq_compile_each (compile.c:5835)
==18152==    by 0x2E4499: iseq_compile_each (compile.c:3996)
==18152==    by 0x2EE93B: rb_iseq_compile_node (compile.c:608)
==18152==    by 0x170E53: rb_iseq_new_with_opt (iseq.c:483)
==18152==    by 0x2E232C: new_child_iseq (compile.c:1119)
==18152==    by 0x2E232C: iseq_compile_each (compile.c:5905)
==18152==    by 0x2E4499: iseq_compile_each (compile.c:3996)
==18152==    by 0x2E2CCE: iseq_compile_each (compile.c:6280)
==18152==    by 0x2EDAA7: rb_iseq_compile_node (compile.c:651)
==18152==    by 0x170E53: rb_iseq_new_with_opt (iseq.c:483)


----------------------------------------
Bug #13772: Memory leak recycling stacks for threads in 2.4.1
https://bugs.ruby-lang.org/issues/13772#change-65948

* Author: sam.saffron (Sam Saffron)
* Status: Open
* Priority: Normal
* Assignee: 
* Target version: 
* ruby -v: 2.4.1
* Backport: 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN
----------------------------------------
Per: 

https://github.com/rest-client/rest-client/issues/611

gem install rest-client

```
100000.times.each_slice(32) do |slice|
    slice.map { Thread.new { RestClient.get('echo.jsontest.com/key/value/one/two') } }.each(&:join)
    GC.start(full_mark: true, immediate_sweep: true)
    slots = GC.stat[:heap_live_slots]
    puts "slots #{slots} #{i+=1} threads #{Thread.list.count}"

end
```

Causes unbound memory growth on 2.4.1, this was not the case in earlier versions of Ruby or Ruby master. 

When running through heaptrack and massif-visualizer it looks like all unreclaimed allocations come from thread_recycle_stack

Ruby heaps do not grow during tests, all the growth is around unmanaged memory. 

I feel getting this sorted is urgent, we have seen similar leaks at Discourse. 



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

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>