Bug #14943: ruby 2.4 doesn't work with jemalloc 5.1.0 properly.

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: 2.4.4
• Backport: 2.3: UNKNOWN, 2.4: REQUIRED, 2.5: DONTNEED

[repeatedly (Masahiro Nakagawa)]

With jemalloc 5.1.0, ruby 2.4.4 failed to initalize timer thread and it causes broken runtime, e.g. signals are ignored.

• ruby 2.4.4

$ LD_PRELOAD=~/jemalloc_lib/5.1.0/lib/libjemalloc.so.2 ruby print_version.rb 
<main>: warning: pthread_create failed for timer: Invalid argument, scheduling broken
2.4.4

• ruby 2.5.1

$ LD_PRELOAD=~/jemalloc_lib/5.1.0/lib/libjemalloc.so.2 ruby print_version.rb 
2.5.1

This problem has been reported in jemalloc issue. See https://github.com/jemalloc/jemalloc/issues/1006

[naruse (Yui NARUSE)]

• Backport changed from 2.3: UNKNOWN, 2.4: UNKNOWN, 2.5: UNKNOWN to 2.3: UNKNOWN, 2.4: REQUIRED, 2.5: DONTNEED

[normalperson (Eric Wong)]

[e-mail] wrote:

With jemalloc 5.1.0, ruby 2.4.4 failed to initalize timer thread and it causes broken runtime, e.g. signals are ignored.

This problem has been reported in jemalloc issue. See
https://github.com/jemalloc/jemalloc/issues/1006

I think we need to backport r58874 to 2.4. Can you file a
backport ticket? (too tired to open browser, even w3m :P)

Increased TSD (aka TLS/thread-local-storage/thread-specific-data)
usage in jemalloc will be a problem with tiny stack size used by
timer thread. I made 58874 to 2.5 as a workaround for high TSD
use in libkqueue.

Also, I may have an idea to remove timer-thread completely, even
under contention...

[jeremyevans0 (Jeremy Evans)]

• Status changed from Open to Closed