ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
27 058 коммитов 3 Ветки 0 Теги
Граф коммитов

5512 Коммитов

Автор SHA1 Сообщение Дата
Ralph Castain
29e083bffd Fix total_slots_allocated computation 
On unmanaged allocations, we need to update the total_slots_allocated once the daemons have been launched and "discovered" their topology

Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-12 08:21:52 -07:00
Ralph Castain
9164afbb08 When a daemon force-terminates, we don't get the show_help message it was trying to send because the message is at a lower priority than the termination event. Resolve this by putting the oob in its own progress thread. Also, use only that one thread by default - if someone needs more progress threads in the OOB, they can use the MCA param to get them. 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-11 06:52:55 -07:00
Ralph Castain
f47124e4d3 Finally fix the problem - the key was knowing there were more than 2 topologies involved, and that the HNP is not allocated. Give up on being cute and just search the darned list of topologies - there won't be that many, and if there are (so the scan takes awhile), then too bad. 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-10 16:44:19 -07:00
Ralph Castain
55f4b825af Add verbose output to nidmap code for debugging as this is a new, and sometimes fragile, feature 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-10 12:40:02 -07:00
Ralph Castain
911961ee21 Sigh - remove debug 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-10 11:26:42 -07:00
Ralph Castain
2d93d15aa7 Merge pull request #3502 from rhc54/topic/cisco 
Fix nidmap computation to deal with hetero nodes
 2017-05-10 11:21:12 -07:00
Ralph Castain
50646b07ce Update the RML OFI by copying the updated files from @anandhis branch 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-10 09:17:06 -07:00
Ralph Castain
442e307a6e Fix the nidmap computation to deal with hetero nodes 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-10 08:43:28 -07:00
Ralph Castain
0afcb1a448 Update to support server self-notifications 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-08 10:04:50 -07:00
Ralph Castain
ef0e0171c9 Implement the changes required to support cross-library coordination. Update PMIx to support intra-process notifications and ensure that we always notify ourselves for events. Add a new ompi/interlib directory where cross-lib coordination code can go, and put the code to declare ourselves there (called from ompi_mpi_init.c). 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-08 10:04:50 -07:00
Ralph Castain
42d31454a5 Merge pull request #3469 from rhc54/topic/nidmap 
Do not pass topologies during tree spawn of daemons as there is no wa…
 2017-05-08 06:22:50 -07:00
Gilles Gouaillardet
e101f2b3f9 orte/util: fix vpids parsing in orte_util_nidmap_parse() 
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
 2017-05-08 16:46:13 +09:00
Ralph Castain
180809f2ef Do not pass topologies during tree spawn of daemons as there is no way the HNP can know the backend topologies at that point. Any needed topologies will be sent along with the launch_apps command 
Do not pass param file MCA params if the user has requested that no param files be read - required when trying to avoid launch time penalties from large numbers of processes reading default param files. The daemon picks them up and passes them along anyway, so it isn't clear what value we gain from having them all read the defaults

Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-07 21:14:43 -07:00
Ralph Castain
a143800bce Enable full operations under SLURM on Cray systems by co-locating a daemon with mpirun when mpirun is executing on a compute node in that environment. This allows local application procs to inherit their security credential from the daemon as it will have been launched via SLURM 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-06 19:08:50 -07:00
Ralph Castain
3a434d75d6 By default, use the system default snd/recv buffer sizes 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-05-05 09:58:05 -07:00
Gilles Gouaillardet
57b4144e57 orte: use compression for ORTE_DAEMON_REPORT_TOPOLOGY_CMD answer 
Refs open-mpi/ompi#3414

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
 2017-04-27 17:21:59 +09:00
Gilles Gouaillardet
49cd40b2df compress the topology sent by the first orted 
Refs open-mpi/ompi#3414

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
 2017-04-27 16:20:11 +09:00
Gilles Gouaillardet
c38ef3d46f oob/tcp: fix short writev handling in send_msg() 
Fixes open-mpi/ompi#3414

Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
 2017-04-27 10:24:38 +09:00
Howard Pritchard
462342d148 Merge pull request #3311 from hppritcha/topic/libfabric_moves_to_ofi 
common/libfabric: move libfabric to ofi
 2017-04-21 07:50:38 -06:00
Howard Pritchard
841192645b common/libfabric: move libfabric to ofi 
This PR renames the common library for OFI libfabric from
libfabric to ofi.  There are a number of reasons this
is good to do:

1) its shorter and replaces 9 characters with three for
   function names for what may eventually be a fairly extensive interface
2) OFI is the term used for MTL and RML components that use
   the OFI libfabric interface
3) A planned OSC component will also use the OFI term.
4) Other HPC libraries that can use OFI libfabric tend to use
   the term "ofi" internally and also in their configure options
   relevant to OFI libfabric (i.e. MPICH/CH4, Intel MPI, Sandia SHMEM)

There seem to be comments in places in the Open MPI source
code that indicate that this common library will be going away.
Far from it as we will want to be able to share things like
AV objects between OMPI and possibly OSHMEM components that
use the OFI libfabric interface.

This PR also adds a synonym to the --with-libfabric(-libdir)
configury options: --with-ofi and with-ofi-libdir.

Signed-off-by: Howard Pritchard <howardp@lanl.gov>
 2017-04-20 13:07:16 -06:00
Nathaniel Graham
34b4aeb17f Merge pull request #3339 from nrgraham23/mpirun_help_improvements 
Additional mpirun --help changes
 2017-04-19 14:05:07 -06:00
Nathaniel Graham
01312b2f90 Additional mpirun --help changes 
This commit recategorizes several mpirun arguments,
and moves the information for mpirun --help arguments
to the bottom of the general help message.  I also
added the OPAL_CMD_LINE_OTYPE field to two commands
that were missed initially because they were not
in the same area as the others.

Signed-off-by: Nathaniel Graham <ngraham@lanl.gov>
 2017-04-19 11:43:45 -06:00
Howard Pritchard
3918b7a796 Merge pull request #3213 from hppritcha/topic/remove_loadleveer 
orte/ras: remove loadleveler support
 2017-04-18 09:18:54 -06:00
Ralph Castain
bb1aaa3286 Use the node index to compare to daemon vpid when identifying procs to bind 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-14 02:37:25 -07:00
Ralph Castain
67156556ce On behalf of Josh, ensure we flag that the child is no longer alive since we are killing it with SIGKILL 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-13 21:07:26 -07:00
Ralph Castain
fbf714d326 Merge pull request #3331 from artpol84/orte_cleanup/master 
orte/pmix: Do not set orted exit status to one from proc abort
 2017-04-13 15:14:14 -07:00
Ralph Castain
1585854335 Minor coverity cleanups 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-12 19:31:35 -07:00
Ralph Castain
0500cc1c66 Update the debugger launch code to reflect the new backend mapping method. 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-12 13:31:18 -07:00
Artem Polyakov
4af7a0827f orte/pmix: Do not set orted exit status to one from proc abort 
The fact that application proc called Abort (read failed) doesn't
mean that ORTE subsystem has failed - vice versa it does it's work
to gracefuly exit the whole application.

orted exiting with non-zero status creates a problem for at least
plm/slurm environments where orteds are launched via `srun` with
"--kill-on-bad-exit" flag. If one of orteds has exited with non-
zero status slurm will immediately kill all other orteds. As the
result we see a lot of leftover in the `/tmp` directory.

Signed-off-by: Artem Polyakov <artpol84@gmail.com>
 2017-04-13 01:37:36 +07:00
Ralph Castain
539f71d0cc Merge pull request #3310 from marksantcroos/fix/alps_wdir 
Bring ALPS ODLS up to par regarding wdir.
 2017-04-11 17:30:04 -07:00
Mark Santcroos
27fa8aabd6 Hardcode basename to "orted" for error reporting. 
Signed-off-by: Mark Santcroos <mark.santcroos@rutgers.edu>
 2017-04-11 18:59:23 -04:00
Mark Santcroos
af3a6e1a29 Verify that the chdir(2) succeeds. 
Signed-off-by: Mark Santcroos <mark.santcroos@rutgers.edu>
 2017-04-12 00:37:37 +02:00
Ralph Castain
97e38e6d84 Move a free to a little later in case the verbose output needs it 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-11 11:21:12 -07:00
Ralph Castain
bb81f3b5db Always setup the attach fifo, even when we initially launch under a debugger so that the user can detach and reattach later 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-11 08:24:03 -07:00
Mark Santcroos
36ac54b5d8 Bring ALPS ODLS up to par regarding wdir. 
Signed-off-by: Mark Santcroos <mark.santcroos@rutgers.edu>
 2017-04-10 08:15:07 -04:00
Ralph Castain
95ae0d1df3 Cleanup timing macros for portability across compilers. Rename the --enable-timing configure option to be --enable-pmix-timing so it doesn't pickup external timing requests. Remove a stale function reference in PMIx so it can compile with timing enabled. 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-10 12:56:38 +06:00
Artem Polyakov
45898a9c65 opal/timing: add the draft of env-based timings 
This commit adds new timing feature that uses environment variables to
expose timing information. This allows easy access to this data (if
timing is enabled) from any other part of the application for the subsequent
postprocessing.
In particular this will be integrated with OMPI-level timing framework that
whill use MPI_Reduce functionality to provide more compact and easy-to use
information.

This commit also adds the example of usage of this framework by annotating
rte_init function. The result is not used anywhere for now. It will be
postprocessed in subsequent commits.

NOTE: that functionality is currently disabled untill it will be verified at runtime

Signed-off-by: Artem Polyakov <artpol84@gmail.com>
 2017-04-07 21:16:22 +06:00
Artem Polyakov
88ed79ea25 opal/timing: remove old framework 
Signed-off-by: Artem Polyakov <artpol84@gmail.com>
 2017-04-07 21:16:22 +06:00
Artem Polyakov
482d7c9322 opal/timing: remove RML timings 
Signed-off-by: Artem Polyakov <artpol84@gmail.com>
 2017-04-07 21:16:21 +06:00
Artem Polyakov
79100de014 opal/timing: Remove oob tracing 
Signed-off-by: Artem Polyakov <artpol84@gmail.com>
 2017-04-07 21:16:21 +06:00
Ralph Castain
b526bca56c Fix a potential segfault by avoiding NULL topologies prior to launching the VM. 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-06 20:51:19 -07:00
Ralph Castain
b33b4607df Correctly identify the source of the event when notifying of abnormal termination by a process 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-06 20:50:38 -07:00
Ralph Castain
a29ca2bb0d Enable slurm operations on Cray with constraints 
Cleanup some errors in the nidmap code that caused us to send unnecessary topologies

Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-06 08:58:06 -07:00
Ralph Castain
db8943cedd Provide further (hopefully) helpful messages about the hotel size 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-05 04:27:32 -07:00
Ralph Castain
b7e9711f45 Resolve the direct modex race condition. The request hotel was running out of rooms, thereby returning an error upon checkin - and we had missed error_logging a couple of those places. Hence no error message and things just hung. 
Output a (hopefully) helpful message when we timeout an operation

Thanks to Nathan for tracking it down.

Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-04 21:32:44 -07:00
Ralph Castain
9a69b20d09 Merge pull request #3282 from rhc54/topic/direct 
Set the PARENT vpid for direct routed module
 2017-04-04 20:55:12 -07:00
Ralph Castain
40ca43e157 Set the PARENT vpid for direct routed module 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-04 19:03:28 -07:00
Ralph Castain
734b90aa6b Adjust the timeout for direct modex requests to reflect the size of the job. It can take several seconds to start all the procs, and we don't want to timeout due to differences in start times of the various procs 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-04 18:20:51 -07:00
Ralph Castain
9cb18b8348 Merge pull request #3280 from rhc54/topic/dvm 
Fix the DVM by ensuring that all nodes, even those that didn't partic…
 2017-04-04 18:15:33 -07:00
Ralph Castain
74863a0ea4 Fix the DVM by ensuring that all nodes, even those that didn't participate (i.e., didn't have any local children) in a job, clean up all resources associated with that job upon its completion. With the advent of backend distributed mapping, nodes that weren't part of the job would still allocate resources on other nodes - and then start from that point when mapping the next job. This change ensures that all daemons start from the same point each time. 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2017-04-04 17:31:38 -07:00