1
1

62 Коммитов

Автор SHA1 Сообщение Дата
Ralph Castain
89c80b2294 Only start a listener for processes that will actually receive connection requests. Tools such as orte-submit always initiate connections and thus do not need to start a listener. 2015-08-27 16:41:00 -07:00
Nathan Hjelm
156ce6af21 periodic whitespace purge
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-08-24 09:32:33 -06:00
Gilles Gouaillardet
429bdf1af7 oob/tcp: fix a race condition when finalizing the oob/tcp component 2015-07-28 09:16:13 +09:00
Ralph Castain
869041f770 Purge whitespace from the repo 2015-06-23 20:59:57 -07:00
Jeff Squyres
e43c8dc291 oob tcp: label a few #endif's
Only bother labeling the ones that are a little far away from their
corresponding #if statements.
2015-05-20 21:10:11 -04:00
Jeff Squyres
4b2f0d4827 oob tcp: reset MCA params from level 9
Set various MCA param levels
2015-05-20 21:10:11 -04:00
Jeff Squyres
1a4c9960e1 oob tcp: set KEEPALIVE timeout 60s, retry interval 5s
The timeout is frequency at which to send keepalive pings; the retry
interval is how often to send successive pings once a keepalive has
not replied.

Also update comments and MCA param help strings.

60 seconds -- squashme
2015-05-20 21:08:37 -04:00
Jeff Squyres
32d81af35f oob tcp: re-enable keepalive option for Mac
Plus very minor #if/#endif reduction.
2015-05-20 17:28:45 -04:00
Ralph Castain
d3d3e73099 Per request from George, use defined(__APPLE__) instead of OPAL_HAVE_MAC. Don't try to close a negative socket 2015-05-15 07:13:42 -06:00
Ralph Castain
8e30579e6e The Mac appears to have problems with the keepalive support - once keepalive starts, the memory footprint soars. So disable keepalive on the Mac 2015-05-14 18:09:13 -06:00
Ralph Castain
b5382c9bf9 Rework the OOB selection logic to allow a component (e.g., usock) to direct that it be the sole active component. Remove prior disqualifying code in the oob/tcp component as it was too restrictive - if usock wasn't able to run, it left apps with no way to communicate to their daemon. Have the local daemon check the global modex for the RML URI info of the local procs so it can route messages between them when tcp is the primary channel.
A few other minor cleanups included.
2015-05-08 11:15:21 -07:00
Ralph Castain
1f8de276de Consolidate all the QOS changes into one clean commit 2015-05-06 19:48:42 -07:00
Nathan Hjelm
45e053dbce orte: use C99 subobject naming for component initialization
This commit helps future-proof orte components by initializing each
component member by name.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-04-18 10:29:58 -06:00
Ralph Castain
0c043dbdc9 Fix typo in var name 2015-04-02 02:32:42 -07:00
Ralph Castain
a4b466efc4 Support attempts to connect async processes by allowing the oob/tcp connection to retry the attempt to connect to a peer. Off by default, operates if someone specifies how long to wait between retry attempts. 2015-04-01 20:21:23 -07:00
rhc54
2ff7575dde Merge pull request #497 from rhc54/topic/sec
Allow for different security domains.
2015-03-25 21:01:29 -07:00
Ralph Castain
10cf455080 Tools need to use the TCP OOB component 2015-03-25 19:56:49 -07:00
Ralph Castain
1b24536941 Allow for different security domains. Let the initiator of the connection determine the method to be used - if the receiver cannot support it, then that's an error that will cause the connection attempt to fail. 2015-03-25 13:22:01 -07:00
Ralph Castain
a0487e014c Further reduce the RARP load by removing getaddrinfo for IPv6 connections. Correct typo when checking return on inet_pton. Don't consider the TCP component for apps that are launched via mpirun as it will never be used. 2015-03-16 19:42:05 -07:00
Ralph Castain
64d11f170a Adjust the default keepalive interval. Refactor the code when setting keepalive options 2015-03-16 12:32:58 -07:00
Ralph Castain
4ded049cbc Modify MCA param description 2015-03-16 11:57:32 -07:00
Ralph Castain
019bba5caf Cleanup a bit - don't need to lookup the protocol number if we just use the right define 2015-03-16 11:54:51 -07:00
Ralph Castain
69ac25bf55 Add support for TCP keepalive on inter-node sockets 2015-03-16 09:59:44 -07:00
Gilles Gouaillardet
a69d935d55 oob/tcp: fix misc issues
as reported by Coverity with CIDs 70726, 710564,
1196630, 1269805, 1269803, 1269932
2015-03-10 19:32:01 +09:00
Jeff Squyres
71ae0ad5ec oob_tcp_component: add #if OPAL_ENABLE_IPV6 around IPv6-specific code
This was CID 1196629
2015-02-24 15:24:11 -05:00
Ralph Castain
d2938a144f Use the proper interface index. Thanks to Mark Kettenis for spotting the problem and providing a patch 2015-01-12 05:31:02 -08:00
Ralph Castain
4d186e6402 Properly protect the MCA parameters being registered by the OOB/TCP component when IPv6 is enabled
cmr=v1.8.3:reviewer=jsquyres

This commit was SVN r32662.
2014-09-02 14:53:00 +00:00
Ralph Castain
aec5cd08bd Per the PMIx RFC:
WHAT:    Merge the PMIx branch into the devel repo, creating a new
               OPAL “lmix” framework to abstract PMI support for all RTEs.
               Replace the ORTE daemon-level collectives with a new PMIx
               server and update the ORTE grpcomm framework to support
               server-to-server collectives

WHY:      We’ve had problems dealing with variations in PMI implementations,
               and need to extend the existing PMI definitions to meet exascale
               requirements.

WHEN:   Mon, Aug 25

WHERE:  https://github.com/rhc54/ompi-svn-mirror.git

Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding.

All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level.

Accordingly, we have:

* created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations.

* Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported.

* Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint

* removed the prior OMPI/OPAL modex code

* added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform.

* retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand

This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
Ralph Castain
5db717f090 Some small leak cleanups
cmr=v1.8.3:reviewer=artpol

This commit was SVN r32358.
2014-07-30 15:46:02 +00:00
Adrian Reber
4aca7095dc fix a syntax error in the FT code
This commit was SVN r32087.
2014-06-25 20:35:50 +00:00
Ralph Castain
e21bfeadcd Now that the BTLs are moving down to OPAL and becoming available to ORTE, there no longer is a need/desire to push performance in the OOB/TCP component. So we don't need multiple modules driving NICs in parallel, and can drop all the complicated distribution logic. Fall back to the simplified single module model, but retain the ability to run that module in its own progress thread if so directed.
This should eliminate the connectivity issues that have been reported, and will make maintenance of this component much easier.

cmr=v1.8.2:reviewer=jsquyres:subject=simplify the OOB/TCP component

This commit was SVN r31956.
2014-06-06 02:24:17 +00:00
Ralph Castain
7df500ecf5 Break the loop caused by retrying to send a message to a hop that is unknown by the TCP oob component. We attempt to provide a way for other components to try, but need to mark that the TCP component is not able to reach that process so the OOB base will know to give up.
This commit was SVN r31928.
2014-06-02 15:00:33 +00:00
Nathan Hjelm
59d09ad9de orte: fix several small memory leaks
grpcomm: fix memory leaks

We were leaking the caddy object used to pass data to the callback
function. This commit fixes these leaks.

oob,rml: fix memory leaks

This commit fixes several leaks:

 - Both the oob/base and oob/tcp were leaking objects on their peer
   hash tables. Iterate on the hash tables and free any objects.

 - Leaked sent messages because of missing OBJ_RELEASE. I placed the
   release in ORTE_RML_SEND_COMPLETE to catch all the possible
   paths.

ess/base: close the state framework

cmr=v1.8.2:reviewer=rhc

This commit was SVN r31776.
2014-05-15 15:06:27 +00:00
Ralph Castain
445b552d3a Try again to get an error message printed when a daemon fails to successfully report back to mpirun. In this case, there is no guaranteed way for the daemon to output the error report itself - we don't have a connection back to the HNP, and we have tied stderr off to /dev/null (for good reasons). So the HNP has to detect the failure itself and report it.
The HNP can't know the precise reason, of course - all it knows is that the daemon failed. So output a generic error message that provides guidance on probable causes.

Refs trac:4571

This commit was SVN r31589.

The following Trac tickets were found above:
  Ticket 4571 --> https://svn.open-mpi.org/trac/ompi/ticket/4571
2014-05-01 19:48:21 +00:00
Ralph Castain
3723b39f30 Ensure we don't silently fail when unable to make a connection - bark pleasantly first.
Refs trac:4571

This commit was SVN r31537.

The following Trac tickets were found above:
  Ticket 4571 --> https://svn.open-mpi.org/trac/ompi/ticket/4571
2014-04-28 19:16:32 +00:00
Ralph Castain
d642babff6 Derived from patch provided by Artem, cleanup the "abnormal" code path for selecting TCP OOB modules to connect to a remote process. If we can't find a direct interface-to-address match, then assign all the provided addresses to the first available TCP module and let the normal failure process determine if the remote proc is truly reachable.
cmr=v1.8.2:reviewer=artpol:subject=fix abnormal code connection path in tcp oob

This commit was SVN r31536.
2014-04-28 19:05:14 +00:00
Ralph Castain
bbdbc5f8a8 Per suggestion from George, use a pipe for terminating the thread.
Refs trac:4510

This commit was SVN r31381.

The following Trac tickets were found above:
  Ticket 4510 --> https://svn.open-mpi.org/trac/ompi/ticket/4510
2014-04-14 01:02:46 +00:00
Ralph Castain
2d8dff837c Ensure we properly terminate the listening thread prior to exiting, but do so in a way that doesn't make us wait for select to timeout.
Refs trac:4510

This commit was SVN r31376.

The following Trac tickets were found above:
  Ticket 4510 --> https://svn.open-mpi.org/trac/ompi/ticket/4510
2014-04-12 15:01:24 +00:00
Ralph Castain
9b30b2b783 Shave some time off of mpirun's operation by not waiting for the listener thread to terminate before exiting
cmr=v1.8.1:reviewer=rhc

This commit was SVN r31368.
2014-04-11 04:16:28 +00:00
Dave Goodell
5f3b81e291 oob: delete events when destroying a peer
Without this patch running ring_c with the usnic BTL under valgrind will
cause the orteds to segfault.

Reviewed-by: Jeff Squyres <jsquyres@cisco.com>
Reviewed-by: Ralph Castain <rhc@open-mpi.org>

cmr=v1.7.5:reviewer=ompi-rm1.7

This commit was SVN r31161.
2014-03-19 22:15:49 +00:00
Ralph Castain
2abed09d7c Continue to resolve priority issues. Cleanup the case of forced termination in mpirun during launch processing by ensuring we can respond to socket closures, and ensuring that the remote daemons correctly close their sockets when terminating.
Jeff: please test a variety of conditions to ensure we get this right

cmr=v1.7.5:reviewer=jsquyres

This commit was SVN r31058.
2014-03-13 04:02:24 +00:00
Ralph Castain
a254d2db34 Silence warning when CR is not enabled
This commit was SVN r31025.
2014-03-12 13:47:03 +00:00
Adrian Reber
4512b3375e OOB/TCP: wire up the existing ft_event() function
This commit was SVN r31022.
2014-03-12 12:47:20 +00:00
Ralph Castain
da4cb39683 If we can't find a route to communicate, emit an error message rather than just exiting with a non-zero status
cmr=v1.7.5:reviewer=jsquyres:subject=print error if cannot communicate

This commit was SVN r30922.
2014-03-04 04:57:53 +00:00
Ralph Castain
14bb7a117c Fix bugs in the oob base - ensure we get the components in high-to-low priority, and that we correctly track reachability via all components. Adjust the priority of the tcp component to leave headroom for others
Refs trac:267

This commit was SVN r30740.

The following Trac tickets were found above:
  Ticket 267 --> https://svn.open-mpi.org/trac/ompi/ticket/267
2014-02-16 03:19:08 +00:00
Ralph Castain
3f9db36e0d Make Jeff smile - pretty-up the indentation
Refs trac:4267

This commit was SVN r30733.

The following Trac tickets were found above:
  Ticket 4267 --> https://svn.open-mpi.org/trac/ompi/ticket/4267
2014-02-14 23:25:48 +00:00
Ralph Castain
4e1c07cbf2 If we are given a TCP oob address that doesn't match any active module, it is still possible that we could route to the address if a router is in the system. No harm in trying, so arbitrarily pick the first connection in the active module list and assign the peer to it. If that module can't reach it, we'll follow the usual failover mechanism until finally concluding that nobody can get there.
cmr=v1.7.5:reviewer=jsquyres:subject=handle non-matching addresses

This commit was SVN r30719.
2014-02-13 23:37:22 +00:00
Ralph Castain
fc6101b508 Handle "localhost" better
Refs trac:4263

This commit was SVN r30702.

The following Trac tickets were found above:
  Ticket 4263 --> https://svn.open-mpi.org/trac/ompi/ticket/4263
2014-02-12 20:30:39 +00:00
Ralph Castain
a8a9801a0b Ensure an orted exits with non-zero status if it is unable to send a message. Add more diagnostic messages to the OOB set_addr code
cmr=v1.7.5:reviewer=jsquyres

This commit was SVN r30701.
2014-02-12 19:44:01 +00:00
Ralph Castain
fa7b686ccc Provide better messages when we don't find any included interfaces, and/or don't find any interfaces for use by OOB.
cmr=v1.7.5:reviewer=jsquyres

This commit was SVN r30675.
2014-02-11 19:29:03 +00:00