Hi. We have a product that is composed of many individual executables.
Several of these are full-blown resource managers. Over the course of
normal operation, there are a fair number of opens and closes (and
corresponding _IO_OPEN, _IO_DUP, and _IO_CLOSE messages). We also use
_IO_CLOSE to detect crashes in subordinate processes.
Since moving to 4.25E, we have started to notice that we do not always
receive _IO_CLOSE message to notify us of crashes. _IO_CLOSE always
arrives when an explicit close() is done, but if an executable is killed
in a non-clean fashion (either intentionally or due to a bug) the
resource manager does not receive an _IO_CLOSE. This always worked
flawlessly under 4.25D.
I’m in the process of attempting to duplicate this with a test case;
since this is rather important part of the OS, however, I was wondering
if anyone else has seen similar missing _IO_CLOSE messages.
Josh Hamacher
FAAC Incorporated
“Josh Hamacher” <hamacher@faac.com> wrote in message
news:3BF2A377.5070208@faac.com…
Hi. We have a product that is composed of many individual executables.
Several of these are full-blown resource managers. Over the course of
normal operation, there are a fair number of opens and closes (and
corresponding _IO_OPEN, _IO_DUP, and _IO_CLOSE messages). We also use
_IO_CLOSE to detect crashes in subordinate processes.
Since moving to 4.25E, we have started to notice that we do not always
receive _IO_CLOSE message to notify us of crashes. _IO_CLOSE always
arrives when an explicit close() is done, but if an executable is killed
in a non-clean fashion (either intentionally or due to a bug) the
resource manager does not receive an _IO_CLOSE. This always worked
flawlessly under 4.25D.
I’m in the process of attempting to duplicate this with a test case;
since this is rather important part of the OS, however, I was wondering
if anyone else has seen similar missing _IO_CLOSE messages.
we also use custom written resource managers and i havn’t saw reported
behaviour under QNX 4.25D and prior. loosing close notofication messages
from client processes can be a really problem. sorry at this moment i have
no information about patch E due to we have no reasons to use it yet but
please let me know if your appregensions will be confirmed or not.
thanks.
Josh Hamacher
FAAC Incorporated
// wbr