Hi,

thank you very much for the analysis.

here's the knldiag and rtedump from the latest crash (last night :-( ) as you 
requested.

http://www.signal7.de/latestcrash.tgz

Best regards,

Robert


On Tuesday 08 October 2002 10:06, Mensing, Joerg wrote:
> Hallo Robert,
> your vserver reports a disappeared COMMUNIC semid. What is that for? For
> each connection a vserver is started. The vserver uses two semaphores for
> communication. For each direction a semaphore is needed to synchronize with
> the SAP DB kernel shared communication memory segment usage. In an
> established connection, the vserver gets packets from socket and puts the
> content into a shared memory. Then it triggers the semaphore which is used
> by the corresponding user kernel thread (UKT) inside SAP DB. Now it will
> wait on the communication channel specific semaphore until this is
> triggered by the SAP DB kernel.
>
> The SAP DB kernel will read the shared memory content request section and
> after its work is done, it will put its reply into the shared memory reply
> section. After the reply is available, the kernel uses the communication
> channel specific semaphore to wakeup vserver.
>
> The vserver triggered by the semaphore will now transfer the reply section
> content into the socket and wait for the next request.
>
> If your link is still valid, from the knldiag i can see that the SAP DB
> kernel was kill by a 'Signal 9'. How can such a signal be recongnized? The
> SAP DB Kernel after it is started immediatly forks a child process and
> watches over its status. The 'Watchdog' reopens the knldiag after the child
> process terminated and writes 'famous last words' and flushes content of
> shared memory trace buffers. A signal '9' is the KILL signal which cannot
> be caught or ignored. Whoever was that friendly to send that signal to SAP
> DB kernel now triggered the 'cleanup' procedure in the watchdog. This
> cleanup uses the 'tag file' directories, which are found in
> /usr/spool/sql/ipc/db:dbname and /usr/spool/sql/ipc/us:dbname. In these
> directories all shared memory segments and semaphores that belong to the
> SAP DB instance 'dbname' are found. The cleanup code uses the directory
> entries to removed the semaphores and shared memory segments for all
> clients (including all vservers connected to SAP DB) and fo! r the SAP DB
> kernel itself. This cleanup code in your case removed the IPC semaphores
> with 'ipcrm()' system calls so that the vserver reported them as
> 'disappeared'. If it uses those disapperaed semaphores it ends up with
> 'invalid argument'. So what you see is a follow up of a 'Signal 9'.
>
> Best regards
> jrg
>
> P.S.: It would be nice if you could upload your knldiag of the second crash
> you recently had. The way you did with the first crash helped a lot for
> this analysis. The excerpt you send us is not enaugh, since all i know now
> is that you ran out of memory :-(
>

_______________________________________________
sapdb.general mailing list
[EMAIL PROTECTED]
http://listserv.sap.com/mailman/listinfo/sapdb.general

Reply via email to