ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
25 992 коммитов 3 Ветки 0 Теги 130 MiB
c3614d30fa
Граф коммитов

3940 Коммитов

Автор SHA1 Сообщение Дата
Nathan Hjelm
c3614d30fa ras/alps: use cpuCnt if using hwthreads as cores 
This commit updates the alps ras component to allow the use of
hyperthreads on compute nodes. In this case we need to use the cpuCnt
value from the node structure instead of numPEs.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
 2016-10-27 09:51:17 -06:00
Ralph Castain
d031946c46 When mpirun operates in --continuous mode, we won't terminate the job when a remote process dies. In that case, we have to activate both the waitpid _and_ the IOF complete states to ensure we properly mark the proc as dead and perform any required notifications 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
 2016-10-25 12:18:14 -07:00
Ralph Castain
227d4d9609 Open the conduits for application procs - we probably can remove all the 
RML-related frameworks from MPI applications now, but let's wait a bit
to ensure we have cleaned up all the points where messaging might occur.
 2016-10-24 16:53:19 -07:00
Ralph Castain
649301a3a2 Revise the routed framework to be multi-select so it can support the new conduit system. Update all calls to rml.send* to the new syntax. Define an orte_mgmt_conduit for admin and IOF messages, and an orte_coll_conduit for all collective operations (e.g., xcast, modex, and barrier). 
Still not completely done as we need a better way of tracking the routed module being used down in the OOB - e.g., when a peer drops connection, we want to remove that route from all conduits that (a) use the OOB and (b) are routed, but we don't want to remove it from an OFI conduit.
 2016-10-23 21:52:39 -07:00
Ralph Castain
df8ac7b747 Properly mark a node as down and decrease the number of daemons so any 
subsequent grpcomm collectives can correctly operate. Note that only the
direct grpcomm component knows how to deal with down nodes.
 2016-10-21 09:53:37 -07:00
Gilles Gouaillardet
1846c2d8ad plm/rsh: use an alternate port if the ORTE_NODE_PORT attribute is set 2016-10-19 16:18:52 +09:00
Ralph Castain
16540c7422 Properly report failure to launch when someone mis-types the name of the application 
Fixes #2233
 2016-10-18 10:09:30 -07:00
Ralph Castain
7be607582e ORTE applications need to commit any modex send's prior to calling fence 2016-10-18 09:22:56 -07:00
Ralph Castain
57114a09ae Pickup the npernode and npersocket options and include them in the job object 2016-10-17 12:26:21 -07:00
Gilles Gouaillardet
bd1b6fe661 rml/oob: add a missing include file 2016-10-16 10:25:00 +09:00
Gilles Gouaillardet
451b9dc467 ess: tear down pmix (if any) before oob 2016-10-13 14:08:02 +09:00
Ralph Castain
fca1556787 Some compilers apparently complain about this, so modify the typedef statements 2016-10-12 08:44:03 -07:00
Ralph Castain
a2919174d0 Bring the RML modifications across. This is the first step in a revamp of the ORTE messaging subsystem to support fabric-based communications during launch and wireup phases. When completed, the grpcomm and plm frameworks will each have their own "conduit" for communication - each conduit corresponds to a particular RML messaging transport. This can be the active OOB-based component, or a provider from within the RML/OFI component. Messages sent down the conduit will flow across the associated transport. 
Multiple conduits can exist at the same time, and can even point to the same base transport. Each conduit can have its own characteristics (e.g., flow control) based on the info keys provided to the "open_conduit" call. For ease during the transition period, the "legacy" RML interfaces remain as wrappers over the new conduit-based APIs using a default conduit opened during orte_init - this default conduit is tied to the OOB framework so that current behaviors are preserved. Once the transition has been completed, a one-time cleanup will be done to update all RML calls to the new APIs and the "legacy" interfaces will be deleted.

While we are at it: Remove oob/usock component to eliminate the TMPDIR length problem - get all working, including oob_stress
 2016-10-11 16:01:02 -07:00
Gilles Gouaillardet
c92e9a5406 use the new OPAL_HASH_TABLE_FOREACH convenience macro 2016-10-08 16:58:20 +09:00
Gilles Gouaillardet
0931d09afa ess/singleton: silence a valgrind warning 
initialize a pointer and keep valgrind happy about it
 2016-09-27 15:22:39 +09:00
Gilles Gouaillardet
f9ebba4668 ess/singleton: only realloc() when required in fork_hnp() 2016-09-23 16:35:59 +09:00
Gilles Gouaillardet
c7bf9a0ec9 ess/singleton: fix read on the pipe to spawn'ed orted 
and close the pipe on both ends when it is no more needed
 2016-09-22 14:21:52 +09:00
Ralph Castain
de7b1494d9 Clean out old cruft from the ORCM project 2016-09-21 00:13:30 -07:00
Gilles Gouaillardet
83399adb3f singleton: "safe" read/write to the pipe between (spawn'ed) orted and singleton 2016-09-20 14:56:58 +09:00
Gilles Gouaillardet
e7ae6975d0 orted: fix spawn in singleton mode 
in singleton mode, have the spawn'ed orted invoke orte_pre_condition_transports()
and send the transport key back to the singleton
 2016-09-20 14:39:22 +09:00
Ralph Castain
a16b3cc33d Fix some minor complaints - missing "void" in function parameters 2016-09-15 15:18:42 -07:00
Ralph Castain
6f086189e6 Fix trivial typo 2016-09-15 13:10:55 -07:00
Gregory M. Kurtzer
16794cc260 Updates to support Singularity containers v2.2 2016-09-15 09:52:06 -07:00
Gilles Gouaillardet
11ebf3ab23 ess/singleton: when forking hnp, use the PMIX_NAMESPACE sent by the hnp 
as the jobid
 2016-09-15 13:57:23 +09:00
Gilles Gouaillardet
e84b35217f oob/tcp: plug a memory leak 
as reported by Coverity with CID 1196711
 2016-09-08 18:50:18 +09:00
Gilles Gouaillardet
b2a2be0e5a odls: fix memory leak plug 
This fixes commit open-mpi/ompi@e2c343cdfc.
 2016-09-08 10:02:52 +09:00
Artem Polyakov
9eba1b0b75 Merge pull request #2042 from artpol84/pmix_sdirs 
Several fixes related to session directories:
 2016-09-07 14:15:47 +07:00
Artem Polyakov
a9a7f39773 ess/pmi: fix the comments about MCA/PMIx setting conflict resolution. 2016-09-07 07:47:35 +03:00
Gilles Gouaillardet
e2c343cdfc odls: plus memory leak 
as reported by Coverity with CID 710645
 2016-09-07 10:08:44 +09:00
Gilles Gouaillardet
c09899f6af plm: plus resource leaks 
as reported by Coverity with CIDs 72274 and 1196733
 2016-09-07 10:08:44 +09:00
Josh Hursey
f6337f9eae Merge pull request #2047 from jjhursey/topic/mixed-host2 
orte: !FQDN implementation to use opal_net_isaddr
 2016-09-06 13:08:54 -05:00
Ralph Castain
f85dcaee2a Fixes CID 1369067 and CID 1196684 
Fixes CID 1369648

    Fixes CID 1372409
 2016-09-06 08:43:15 -07:00
Artem Polyakov
74a11d7832 Fix session dir cleanup code. 2016-09-05 07:53:55 +03:00
Artem Polyakov
dc0ab674de Add PMIx key to provide RM with ability to indicate that it will cleanup 
session directories provided at through OPAL_PMIX_TMPDIR,
OPAL_PMIX_NSDIR, OPAL_PMIX_PROCDIR
 2016-09-05 07:48:44 +03:00
Artem Polyakov
81195ab724 Several fixes related to session directories: 
* enable OMPI to retrieve paths from RM through PMIx
* cleanups related to tempdirs.
 2016-09-05 07:48:44 +03:00
Ralph Castain
fb51d65049 Minor change: check for NULL before using the job map to avoid segfault when erroring out prior to creating the map 2016-09-04 07:53:12 -07:00
Joshua Hursey
fe937d1e82 orte: !FQDN implementation to use opal_net_isaddr 
* Switch to use opal_net_isaddr() for checking if a name is an IP
   address - as it is a bit cleaner, and uses common functionality.
 2016-09-02 13:31:49 -05:00
Ralph Castain
4e0788e9ad Enable PSM to support dynamic processes 
Fix comm_spawn to correctly reference the actual parent process that requested the spawn when looking for the parent job object
 2016-09-02 10:22:04 -07:00
Ralph Castain
0ea1cff733 Implement notification of completion on comm_spawn'd child jobs. Add a configure flag to enable PMIx 3's shared memory datastore, and set it disable by default so that comm_spawn functions again. Will reverse the default once that feature is fully functional 2016-09-01 13:10:10 -07:00
Gilles Gouaillardet
0b8c58298d oob/usock: fix handling of orte_process_name_t * 
orte_process_name_t is aligned on 32 bits, so it cannot simply be casted
into an int64_t. use memcpy() instead

Thanks Paul Hargrove for the report
 2016-09-01 13:18:02 +09:00
Ralph Castain
c1050bc01e Provide a mechanism for obtaining memory profiles of daemons and application profiles for use in studying our memory footprint. Setting OMPI_MEMPROFILE=N causes mpirun to set a timer for N seconds. When the timer fires, mpirun will query each daemon in the job to report its own memory usage plus the average memory usage of its child processes. The Proportional Set Size (PSS) is used for this purpose. 2016-08-31 09:32:07 -07:00
Ralph Castain
9b991bd1f5 Ensure that the "running" state is correctly updated 
It is possible that one or more procs could get thru PMIx_Init, and thus be marked as in state "registered", before all local procs have been started. If that happens, then we would report some of the procs in state "running", and the others in state "registered" - which means that the HNP would miss the "running" stage of the state machine.

Thanks to Jingchao Zhang for his patience in tracking this down on the 2.0 branch
 2016-08-30 19:24:39 -07:00
Josh Hursey
b0d8638824 Merge pull request #2015 from jjhursey/topic/mixed-hostnames 
orte: Expand use of !orte_keep_fqdn_hostnames MCA parameter
 2016-08-29 09:14:54 -05:00
Ralph Castain
2f6e0fec90 Provide the number of nodes in the job 2016-08-26 14:50:41 -07:00
Joshua Hursey
d26dd2c20e orte: Expand the application of !orte_keep_fqdn_hostnames 
* Expand the use of the `orte_keep_fqdn_hostnames` MCA parameter when
   it is set to false.
 * If that parameter is set to false (default) then short hostnames
   (e.g., `node01`) will match with the long hostnames (e.g.,
   `node01.mycluster.org`). This allows a user (or resource manager)
    to mix the use of short and long hostnames.
  - Note that this mechanism does _not_ perform a DNS lookup, but
    instead strips off the FQDN by truncating the hostname string at
    the first `.` character (when not an IP address).
     - By default (`false`) the following is true:
       `node01 == node01.mycluster.org == node01.bogus.com`
       since we use `node01` as the hostname.
 2016-08-26 16:09:04 -05:00
Artem Polyakov
55ac3b0be3 orte/schizo: fix binding detection in slurm component 
in SLURM 16.05 the SLURM_CPU_BIND_TYPE is equal to "mask_cpu:"
instead of "mask_cpu". Account for that.
 2016-08-26 09:55:52 +03:00
rhc54
19b0f4db9f Merge pull request #1995 from rhc54/topic/pe-per-rank 
Change the behavior of cpus-per-rank.
 2016-08-25 14:38:12 -05:00
Ralph Castain
440eae90ec Correct the binding algorithm to decouple it from oversubscribe. 
Oversubscribe stipulates that we allow more procs on the node than assigned slots - it has nothing to do with the number of available pe's. Let overload directives handle the pe situation.
 2016-08-24 21:17:22 -07:00
Gilles Gouaillardet
93e73841f9 ess/singleton: push all PMIX_* environment variables, regardless how many there are 2016-08-23 09:46:55 +09:00
Gilles Gouaillardet
a1e8e58a8a ess/singleton: expects 4 PMIX_* environment variables or more 2016-08-23 09:34:03 +09:00