[GHC] #9239: Program loops in syscall retries cancelled by rt_sigreturn signal

[GHC]

#9239: Program loops in syscall retries cancelled by rt_sigreturn signal
--------------------------+------------------------------------------------
       Reporter:  k-bx    |             Owner:
           Type:  bug     |            Status:  new
       Priority:  normal  |         Milestone:
      Component:          |           Version:  7.8.2
  Compiler                |  Operating System:  Unknown/Multiple
       Keywords:          |   Type of failure:  Incorrect result at runtime
   Architecture:          |         Test Case:
  Unknown/Multiple        |          Blocking:
     Difficulty:          |
  Unknown                 |
     Blocked By:          |
Related Tickets:          |
--------------------------+------------------------------------------------
 Sorry if this bug is not related to ghc, but it seems to me that it might
 be.

 Originally I reported it against hedis package, you can find description
 in it's bug tracker https://github.com/informatikr/hedis/issues/15

 Short recap: one specific operation against redis database
 (zrangebyscoreLimit against big ZSET), program behaves differently. For
 me, most of the time it would hang, sometimes it succeeds, sometimes
 outputs "ConnectionLost" twice and dies. My environments are: Ubuntu
 12.04/14.04, redis versions tested are 2.2 through 2.8 (latest in each
 even branch).

 Under GHC 7.6 and profiling enabled, strace would end up showing an
 infinite loop of these operations:

 {{{
 --- SIGVTALRM {si_signo=SIGVTALRM, si_code=SI_TIMER, si_pid=0, si_uid=0,
 si_value=0} ---
 rt_sigreturn() = -1 EINTR (Interrupted system call)
 select(4, [3], [], NULL, NULL) = ? ERESTARTNOHAND (To be restarted if no
 handler)
 }}}

 Under 7.8.2 profiling enabled, it would also loop, but in few seconds it
 would hang, produring these last lines of output (plus result of Ctrl+C):

 {{{
 select(4, [3], [], NULL, NULL)          = ? ERESTARTNOHAND (To be
 restarted if no handler)
 --- SIGVTALRM {si_signo=SIGVTALRM, si_code=SI_TIMER, si_pid=0, si_uid=0,
 si_value=0} ---
 timer_settime(0, 0, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
 rt_sigreturn()                          = -1 EINTR (Interrupted system
 call)
 select(4, [3], [], NULL, NULL^CProcess 16483 detached
  <detached ...>
 }}}

 With profiling disabled on 7.8.2 it would sometimes hang same way as with
 profiling, but sometimes it would finish with outputting "ConnectionLost"
 errors (you can find links to full strace output in github hedis issue
 page linked above).

 Sorry again if it's not related to GHC, meanwhile I'll try to investigate
 further.

 Thank you.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9239>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler