Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] ompi-ps broken or just changed?
From: Ralph Castain (rhc_at_[hidden])
Date: 2009-05-21 18:27:15


Hi Ashley

I dug into this deeper and fixed the problem last night. We were actually
hanging on the send operation as the recvr (a stale HNP, in this case)
wasn't there. I had to use a non-blocking send and put a timeout around it
so we could escape that situation.

I also had to change a couple of tests to ensure we knew when to print local
help messages vs sending them to an HNP.

I have tested the result in a scenario where several stale HNP contact files
existed and it worked fine. See if it works for you. I'll tinker with it to
print out a complete list of the stale HNPs later.

Ralph

On Tue, May 19, 2009 at 7:07 AM, Ashley Pittman <ashley_at_[hidden]>wrote:

>
> Ralph,
>
> At least part of them problem is to do with error reporting, orte-ps is
> hitting the error case for a stale hnp at around line 258 and is trying
> to report the error via orte_show_help() however this function is
> calling a rpc into the orted-run which is silently ignoring it for some
> reason.
>
> The failure itself seems to come from a timeout in comm.c:1114 where the
> client process isn't waiting long enough for the orted-run to reply and
> is returning ORTE_ERR_SILENT instead. I can't think what to suggest
> here other than increasing the timeout?
>
> Ashley,
>
> On Mon, 2009-05-18 at 17:06 +0100, Ashley Pittman wrote:
> > It's certainly helped and now runs for me however if I run mpirun under
> > valgrind and then opmi-ps in another window Valgrind reports errors and
> > ompi-ps doesn't list the job so there is clearly something still amiss.
> > I'm trying to do some more diagnostics now.
> >
> > ==32362== Syscall param writev(vector[...]) points to uninitialised
> > byte(s)
> > ==32362== at 0x41BF10C: writev (writev.c:46)
> > ==32362== by 0x4EAAD52: mca_oob_tcp_msg_send_handler
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/openmpi/mca_oob_tcp.so)
> > ==32362== by 0x4EAC505: mca_oob_tcp_peer_send
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/openmpi/mca_oob_tcp.so)
> > ==32362== by 0x4EAEF89: mca_oob_tcp_send_nb
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/openmpi/mca_oob_tcp.so)
> > ==32362== by 0x4EA20BE: orte_rml_oob_send
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/openmpi/mca_rml_oob.so)
> > ==32362== by 0x4EA2359: orte_rml_oob_send_buffer
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/openmpi/mca_rml_oob.so)
> > ==32362== by 0x4050738: process_commands
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-rte.so.0.0.0)
> > ==32362== by 0x405108C: orte_daemon_cmd_processor
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-rte.so.0.0.0)
> > ==32362== by 0x4260B57: opal_event_base_loop
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x4260DF6: opal_event_loop
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x4260E1D: opal_event_dispatch
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x804B15F: orterun (orterun.c:757)
> > ==32362== Address 0x448507c is 20 bytes inside a block of size 512
> > alloc'd
> > ==32362== at 0x402613C: realloc (vg_replace_malloc.c:429)
> > ==32362== by 0x42556B7: opal_dss_buffer_extend
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x4256C4F: opal_dss_pack_int32
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x42565C9: opal_dss_pack_buffer
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x403A60D: orte_dt_pack_job
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-rte.so.0.0.0)
> > ==32362== by 0x42565C9: opal_dss_pack_buffer
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x4256FFB: opal_dss_pack
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x40506F7: process_commands
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-rte.so.0.0.0)
> > ==32362== by 0x405108C: orte_daemon_cmd_processor
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-rte.so.0.0.0)
> > ==32362== by 0x4260B57: opal_event_base_loop
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x4260DF6: opal_event_loop
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> > ==32362== by 0x4260E1D: opal_event_dispatch
> > (in
> /mnt/home/debian/ashley/code/OpenMPI/install/lib/libopen-pal.so.0.0.0)
> >
> > On Mon, 2009-05-18 at 08:22 -0600, Ralph Castain wrote:
> > > Aha! Thanks for spotting the problem - I had to move that var init to
> > > cover all cases, but it should be working now with r21249
> > >
> > >
> > >
> > > On May 18, 2009, at 8:08 AM, Ashley Pittman wrote:
> > >
> > > >
> > > > Ralph,
> > > >
> > > > This patch fixed it, num_nodes was being used initialised and hence
> > > > the
> > > > client was getting a bogus value for the number of nodes.
> > > >
> > > > Ashley,
> > > >
> > > > On Mon, 2009-05-18 at 10:09 +0100, Ashley Pittman wrote:
> > > >> No joy I'm afraid, now I get errors when I run it. This is a
> single
> > > >> node job run with the command line "mpirun -n 3 ./a.out". I've
> > > >> attached
> > > >> the strace output and gzipped /tmp files from the machine.
> > > >> Valgrind on
> > > >> the opmi-ps process doesn't show anything interesting.
> > > >>
> > > >> [alpha:29942] [[35044,0],0] ORTE_ERROR_LOG: Data unpack would read
> > > >> past
> > > >> end of buffer in
> > > >> file /mnt/home/debian/ashley/code/OpenMPI/ompi-trunk-tes/trunk/orte/
> > > >> util/comm/comm.c at line 242
> > > >> [alpha:29942] [[35044,0],0] ORTE_ERROR_LOG: Data unpack would read
> > > >> past
> > > >> end of buffer in
> > > >> file /mnt/home/debian/ashley/code/OpenMPI/ompi-trunk-tes/trunk/orte/
> > > >> tools/orte-ps/orte-ps.c at line 818
> > > >>
> > > >> Ashley.
> > > >>
> > > >> On Sat, 2009-05-16 at 08:15 -0600, Ralph Castain wrote:
> > > >>> This is fixed now, Ashley - sorry for the problem.
> > > >>>
> > > >>>
> > > >>> On May 15, 2009, at 4:47 AM, Ashley Pittman wrote:
> > > >>>
> > > >>>> On Thu, 2009-05-14 at 22:49 -0600, Ralph Castain wrote:
> > > >>>>> It is definitely broken at the moment, Ashley. I have it pretty
> > > >>>>> well
> > > >>>>> fixed, but need/want to cleanup some corner cases that have
> > > >>>>> plagued
> > > >>>>> us
> > > >>>>> for a long time.
> > > >>>>>
> > > >>>>> Should have it for you sometime Friday.
> > > >>>>
> > > >>>> Ok, thanks. I might try switching to slurm in the mean-time, I
> > > >>>> know
> > > >>>> my
> > > >>>> code works with that.
> > > >>>>
> > > >>>> Can you let me know when it's fixed on or off list and I'll do an
> > > >>>> update.
> > > >>>>
> > > >>>> Ashley,
> > > >>>>
> > > >>>> _______________________________________________
> > > >>>> devel mailing list
> > > >>>> devel_at_[hidden]
> > > >>>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > > >>>
> > > >>> _______________________________________________
> > > >>> devel mailing list
> > > >>> devel_at_[hidden]
> > > >>> http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > > >> _______________________________________________
> > > >> devel mailing list
> > > >> devel_at_[hidden]
> > > >> http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > > > <ompi-ps.patch>_______________________________________________
> > > > devel mailing list
> > > > devel_at_[hidden]
> > > > http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > >
> > > _______________________________________________
> > > devel mailing list
> > > devel_at_[hidden]
> > > http://www.open-mpi.org/mailman/listinfo.cgi/devel
> >
> > _______________________________________________
> > devel mailing list
> > devel_at_[hidden]
> > http://www.open-mpi.org/mailman/listinfo.cgi/devel
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>