Charley Kamolpornwijit created THRIFT-3912:
----------------------------------------------

             Summary: TNonblockingServer crashes when file descriptor numbers > 
FD_SETSIZE
                 Key: THRIFT-3912
                 URL: https://issues.apache.org/jira/browse/THRIFT-3912
             Project: Thrift
          Issue Type: Bug
          Components: C++ - Library
    Affects Versions: 0.9.3
         Environment: Ubuntu 14.04 LTS
            Reporter: Charley Kamolpornwijit


We experienced crashes by TNonblockingServer when we used it in our system with 
high numbers of file descriptors. The stacktrace output were similar to the 
following:
{quote}
Program terminated with signal SIGABRT, Aborted.
#0  0x00007f536e0f9c37 in __GI_raise (sig=sig@entry=6) at 
../nptl/sysdeps/unix/sysv/linux/raise.c:56
#0  0x00007f536e0f9c37 in __GI_raise (sig=sig@entry=6) at 
../nptl/sysdeps/unix/sysv/linux/raise.c:56
#1  0x00007f536e0fd028 in __GI_abort () at abort.c:89
#2  0x00007f536e1362a4 in __libc_message (do_abort=do_abort@entry=2, 
fmt=fmt@entry=0x7f536e242113 "*** %s ***: %s terminated\n") at 
../sysdeps/posix/libc_fatal.c:175
#3  0x00007f536e1cdbbc in __GI___fortify_fail (msg=<optimized out>, 
msg@entry=0x7f536e2420aa "buffer overflow detected") at fortify_fail.c:38
#4  0x00007f536e1cca90 in __GI___chk_fail () at chk_fail.c:28
#5  0x00007f536e1cdb07 in __fdelt_chk (d=d@entry=1024) at fdelt_chk.c:25
#6  0x0000000000a8a745 in apache::thrift::server::TNonblockingIOThread::notify 
(this=<optimized out>, conn=0x7f536d432780) at 
src/thrift/server/TNonblockingServer.cpp:1408
{quote}

By investigating the problem, we found that this cause by TNonblcokingServer 
using {{FD_SET()}} with a file descriptor number that greater than the 
`FD_SETSIZE` constant. We also found that the patch in THRIFT-3080 was the 
cause of this problem, as {{select()}}, in contrast to `poll()`, has such limit.

Currently, we revert the patch in THRIFT-3080 
(https://github.com/apache/thrift/commit/b5ebcd199c1b603cea652847bfc9177c60fb8e28)
 to make it works in our environment. However, in the long term, I believe it 
is important that TNonblockingSystem uses {{poll()}} instead of {{select()}} 
whenever {{poll()}} exists in the build system to avoid this problem.

Thank you.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to