From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Wed Jan 30 2008 - 11:30:58 PST
Neal,
I am not a python programmer, but I think I see the reasons for both
your problems.
1) the "unknown error": You are printing os.strerror(err) when you
probably wanted os.strerror(errno). I believe you will find that making
that one change will print "Invalid Argument" (or similar) because you
are getting the expected post-restart condition err=-1 and errno=EINVAL.
The huge value you see now is -1 (err) when printed as an unsigned 64bit
value (ulong).
2) the close(): blcr will restart with the checkpoint filedescriptor
already closed. An additional close() in your code will return a
negative value and errno=EBAFD=9 (the C code in cr_request_file() calls
close w/o checking the return value for this reason). It appears that
python is throwing an exception when that happens. The solution is to
either not close() on restart, or to close() w/o regard to restarting
but to catch/ignore this exception.
-Paul
Neal Becker wrote:
> In the ctypes wrapper I just sent, there are a couple of small issues.
> Here is the loop I used (python version of cr_checkpoint.c)
>
> with open ("checkpoint", 'w') as cp_file:
> cr_args.cr_fd = cp_file.fileno()
>
> err,cr_handle = request_checkpoint (cr_args, cr_handle)
>
> err = -1
> while (err < 0):
> err = libcr.cr_poll_checkpoint (byref(cr_handle), POINTER(timeval_t)
> ())
> print "err:", os.strerror(err)
> if (err < 0):
> if (errno == EINVAL):
> break # restarted
> elif (errno == EINTR):
> continue
> else:
> die ("cr_poll_checkpoint")
> elif (err == 0):
> die ("cr_poll_checkpoint returned unexpected 0")
>
> On restart, I get:
> cr_restart checkpoint
> err: Unknown error 18446744073709551615
> ---------------------------------------------------------------------------
> IOError Traceback (most recent call last)
>
> /home/nbecker/idma-cdma/test/<ipython console> in <module>()
>
> /usr/tmp/python-m9AicK.py in <module>()
>
> IOError: [Errno 9] Bad file descriptor
>
> Looks like 2 errors:
> 1) On restart, the "print err" statement is executed and seems to print a
> garbage value
> 2) Looks like on restart closing the fd is a bad idea
--
Paul H. Hargrove PHHargrove_at_lbl_dot_gov
Future Technologies Group
HPC Research Department Tel: +1-510-495-2352
Lawrence Berkeley National Laboratory Fax: +1-510-486-6900