will be the final solution. But I'm committing it now so that
Oracle's Solaris Studio builds can resume.
The issue is that the C++ bindings are now (eventually) including
<hwloc.h>. We use !__hwloc_inline__ and #define it to an appropriate
value at compile-time. The issue is that when we're compiling C++
code, we should just set !__hwloc_inline__ to "inline", because that's
a keyword in the C++ language (as opposed to !__inline__, or
somesuch).
This commit was SVN r26418.
helper file, even if we find that the system has <infiniband/verbs.h>.
The reason is because there are some inline functions in that verbs
helper file that invoke ibv_* functions. Some linkers (e.g., Solaris
Studio Compilers) will instantiate those static inline functions --
even if we don't use them -- and therefore we need to be able to
resolve the ibv_* symbols at link time.
But since -libverbs is only specified in places where we use other
ibv_* functions (e.g., the OpenFabrics-based BTLs), that means that
linking random executables can/will fail (e.g., orterun).
So instead, introduce a new #define: OPAL_HWLOC_WANT_VERBS_HELPER. If
this macro is set to 1 before including opal/mca/hwloc/hwloc.h, then
you'll also get the hwloc OpenFabrics verbs helper header file (*if*
hwloc found <infiniband/verbs.h> -- otherwise, it'll #error).
This commit was SVN r26417.
* Remove paffinity, maffinity, and carto frameworks -- they've been
wholly replaced by hwloc.
* Move ompi_mpi_init() affinity-setting/checking code down to ORTE.
* Update sm, smcuda, wv, and openib components to no longer use carto.
Instead, use hwloc data. There are still optimizations possible in
the sm/smcuda BTLs (i.e., making multiple mpools). Also, the old
carto-based code found out how many NUMA nodes were ''available''
-- not how many were used ''in this job''. The new hwloc-using
code computes the same value -- it was not updated to calculate how
many NUMA nodes are used ''by this job.''
* Note that I cannot compile the smcuda and wv BTLs -- I ''think''
they're right, but they need to be verified by their owners.
* The openib component now does a bunch of stuff to figure out where
"near" OpenFabrics devices are. '''THIS IS A CHANGE IN DEFAULT
BEHAVIOR!!''' and still needs to be verified by OpenFabrics vendors
(I do not have a NUMA machine with an OpenFabrics device that is a
non-uniform distance from multiple different NUMA nodes).
* Completely rewrite the OMPI_Affinity_str() routine from the
"affinity" mpiext extension. This extension now understands
hyperthreads; the output format of it has changed a bit to reflect
this new information.
* Bunches of minor changes around the code base to update names/types
from maffinity/paffinity-based names to hwloc-based names.
* Add some helper functions into the hwloc base, mainly having to do
with the fact that we have the hwloc data reporting ''all''
topology information, but sometimes you really only want the
(online | available) data.
This commit was SVN r26391.
(http://www.open-mpi.org/community/lists/devel/2012/04/10905.php), set
opal_cache_line_size via hwloc data, if we have it.
opal_cache_line_size will be set to an hwloc-inspired value by the end
of orte_init(), but will always have a safe value to use (i.e., a
default value 128) -- even before opal_init() has completed.
Default to the same value of 128 that Open MPI has used for several
years if a) we have no hwloc data, or b) we weren't able to find L2
objects in the hwloc data.
This commit was SVN r26322.
Roll in the ORTE state machine. Remove last traces of opal_sos. Remove UTK epoch code.
Please see the various emails about the state machine change for details. I'll send something out later with more info on the new arch.
This commit was SVN r26242.
fakeroot. If so, exit out of the pre-main hook immediately (without
calling functions such as stat, which will be replaced by fakeroot to
things that are not safe to call in a pre-main environment).
This commit was SVN r26203.
The following Trac tickets were found above:
Ticket 2812 --> https://svn.open-mpi.org/trac/ompi/ticket/2812
Worked with Oracle to verify that hwloc PCI detection is correctly
disabled on the Suse 10/64 bit platform and is enabled by default on
all other platforms. The --[en|dis]able-hwloc-pci switch is also
available for manual override of the configure decision about hwloc
PCI support.
This commit was SVN r26175.
This commit was SVN r26165.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4402
hwloc component weren't reverse applied to the external hwloc
component. Additionaly, if we add stuff to LDFLAGS/LIBS, we also may
need to append (DY)LD_LIBRARY_PATH (here in this configure process
only), otherwise future configure tests may fail because they can't
find libhwloc.so (e.g., if you --with-hwloc=/some/path, we need to add
/some/path/lib to (DY)LD_LIBRARY_PATH).
This commit was SVN r26082.
The following Trac tickets were found above:
Ticket 3043 --> https://svn.open-mpi.org/trac/ompi/ticket/3043
This commit was SVN r26037.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4340
This commit was SVN r25987.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4319
This commit was SVN r25986.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4314
Easiest solution is to set the include in a POST_CONFIG macro based on
whether the configure system says the component was selected or not.
This commit was SVN r25968.
1. no binding support - indicated by a negative return code from get_cpubind
2. binding supported, but not bound - the bitset returned by get_cpubind is the same as the available cpuset
3. binding supported and bound - bitset from get_cpubind is a subset of available cpuset
4. only one cpu is available - in this case, get_cpubind matches the available cpuset, but we are effectively bound
This commit was SVN r25957.
in the tarball. Thanks to Paul Hargrove for the fix.
This commit was SVN r25952.
The following Trac tickets were found above:
Ticket 2951 --> https://svn.open-mpi.org/trac/ompi/ticket/2951
paffinity hwloc was returning "NOT_SUPPORTED" when the real problem
was that the underlying hwloc simply hadn't been initialized yet. So
let's clearly delineate this case: return OPAL_ERR_NOT_INITIALIZED if
the underlying hwloc is not initialized.
This commit was SVN r25902.
into account when configuring the components of the faramework. If
--without-memory-manager was given, then we really don't want any
memory managers to be used.
This commit was SVN r25807.
The following Trac tickets were found above:
Ticket 2844 --> https://svn.open-mpi.org/trac/ompi/ticket/2844
problem on SuSE 10 (which might be related to Oracle's dual-bitness
builds, but we aren't completely sure yet).
So just turn it off for now, and bring this over to v1.5. Find a
proper fix (that enables pci support properly) for trunk/v1.7 later.
This commit was SVN r25769.
The following Trac tickets were found above:
Ticket 2952 --> https://svn.open-mpi.org/trac/ompi/ticket/2952
This commit was SVN r25759.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4094
This commit was SVN r25707.
The following SVN revision numbers were found above:
r4102 --> open-mpi/ompi@8961ca568d
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4104
.ompi_ignored to allow other developers to test with it. It is
expected that we'll remove the .ompi_ignore here soon, and
simultaneously remove the hwloc 1.2.2ompi component.
There was one very minor patch added to stock hwloc 1.3.1 in
hwloc/config/hwloc.m4:
{{{
--- hwloc-1.3.1/config/hwloc.m4 2011-12-14
+++ ompi3/opal/mca/hwloc/hwloc131/hwloc/config/hwloc.m4
@@ -583,6 +583,7 @@
])
fi
AC_SUBST(HWLOC_PCI_LIBS)
+ HWLOC_LIBS="$HWLOC_LIBS $HWLOC_PCI_LIBS"
# If we asked for pci support but couldn't deliver, fail
AS_IF([test "$enable_pci" = "yes" -a "$hwloc_pci_happy" = "no"],
[AC_MSG_WARN([Specified --enable-pci switch, but could
not])
}}}
This will be pushed upstream to hwloc.
This commit was SVN r25706.
No need for it to be in the base (we mistakenly thought it was used in
multiple shmem components).
This commit was SVN r25662.
The following SVN revision numbers were found above:
r25652 --> open-mpi/ompi@7e223b5799
* Fixes trac:2329 : Improves the error message, and ensures opal-restart will not segv in opal_finalize.
This commit was SVN r25586.
The following Trac tickets were found above:
Ticket 2329 --> https://svn.open-mpi.org/trac/ompi/ticket/2329
to -I the opal/config directory so that autoregen can find our .m4
files.
This commit was SVN r25564.
The following SVN revision numbers were found above:
r25511 --> open-mpi/ompi@5751c45916
supposed to. I.e., half-baked/not complete stuff.
This commit backs out all of r25545. Sorry folks!
This commit was SVN r25546.
The following SVN revision numbers were found above:
r25545 --> open-mpi/ompi@7f9ae11faf
to make MPI_IN_PLACE (and other sentinel Fortran constants) work on OS
X, we need to use the following compiler (linker) flag:
-Wl,-commons,use_dylibs
So if we're compiling on OS X, test to see if that flag works with the
compiler. If so, add it to the wrapper FFLAGS and FCFLAGS (note that
per a future update, we'll only have one Fortran compiler anyway).
Fixes trac:1982.
This commit was SVN r25545.
The following Trac tickets were found above:
Ticket 1982 --> https://svn.open-mpi.org/trac/ompi/ticket/1982
https://svn.open-mpi.org/trac/ompi/wiki/ProcessPlacement
The wiki page is incomplete at the moment, but I hope to complete it over the next few days. I will provide updates on the devel list. As the wiki page states, the default and most commonly used options remain unchanged (except as noted below). New, esoteric and complex options have been added, but unless you are a true masochist, you are unlikely to use many of them beyond perhaps an initial curiosity-motivated experimentation.
In a nutshell, this commit revamps the map/rank/bind procedure to take into account topology info on the compute nodes. I have, for the most part, preserved the default behaviors, with three notable exceptions:
1. I have at long last bowed my head in submission to the system admin's of managed clusters. For years, they have complained about our default of allowing users to oversubscribe nodes - i.e., to run more processes on a node than allocated slots. Accordingly, I have modified the default behavior: if you are running off of hostfile/dash-host allocated nodes, then the default is to allow oversubscription. If you are running off of RM-allocated nodes, then the default is to NOT allow oversubscription. Flags to override these behaviors are provided, so this only affects the default behavior.
2. both cpus/rank and stride have been removed. The latter was demanded by those who didn't understand the purpose behind it - and I agreed as the users who requested it are no longer using it. The former was removed temporarily pending implementation.
3. vm launch is now the sole method for starting OMPI. It was just too darned hard to maintain multiple launch procedures - maybe someday, provided someone can demonstrate a reason to do so.
As Jeff stated, it is impossible to fully test a change of this size. I have tested it on Linux and Mac, covering all the default and simple options, singletons, and comm_spawn. That said, I'm sure others will find problems, so I'll be watching MTT results until this stabilizes.
This commit was SVN r25476.
http://www.open-mpi.org/community/lists/devel/2011/10/9878.php
I am making a final decision to decide the behavior of what happens
when an MCA parameter is re-registered and changes types. In
developer builds (i.e., OPAL_ENABLE_DEBUG==1), a show_help message
will be displayed. In all builds, an error status will be returned.
Specifically, the logic looks like this:
{{{
if (detect_re-registration_with_type_change) {
#if OPAL_ENABLE_DEBUG
opal_show_help(...);
#endif
return OPAL_ERR_VALUE_OUT_OF_BOUNDS;
}
}}}
If someone would like to change this behavior, they are welcome to do
so. :-) I am committing this so that ''some'' action occurs (rather
than talking about the issue and then nothing happens).
This commit was SVN r25432.
handled properly when MCA parameters are re-registered and their types
change. Specifically, this case was broken:
1. Register an int MCA param with a non-zero default value
1. Re-register the same MCA param as a string with a NULL default value
The 2nd step would cause a segv because the first int default value
wasn't being reset properly. Here's sample code that shows the issue:
{{{
{
int ibogus;
char *sbogus;
opal_init(&argc, &argv);
mca_base_param_reg_int_name("type", "name", "help", false, false, 3, &ibogus);
printf("Ibogus: %d\n", ibogus);
mca_base_param_reg_string_name("type", "name", "help", false, false, NULL, &sbogus);
printf("Sbogus: %s\n", (NULL == sbogus) ? "NULL" : sbogus);
exit(0);
}
}}}
This commit fixes the problem from the sample code above as well as
the a similar issue for file-set MCA params and override values. It
also resets default values for MCA params initially registered as a
string but then re-registered as an int.
This commit was SVN r25392.
Use hwloc to obtain the cpuset for each process during mpi_init, and share that info in the modex. As it arrives, use a new opal_hwloc_base utility function to parse the value against the local proc's cpuset and determine where they overlap. Cache the value in the pmap object as it may be referenced multiple times.
Thus, the return value from orte_ess.proc_get_locality is a 16-bit bitmask that describes the resources being shared with you. This bitmask can be tested using the macros in opal/mca/paffinity/paffinity.h
Locality is available for all procs, whether launched via mpirun or directly with an external launcher such as slurm or aprun.
This commit was SVN r25331.
calling main():
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
xxx:~/openmpi-1.5.4/COMPILE-intel-12.1.0> which ompi_info
~/openmpi-1.5.4/COMPILE-intel-12.1.0/usr/bin/ompi_info
xxx:~/openmpi-1.5.4/COMPILE-intel-12.1.0> ompi_info
Segmentation fault
xxx:~/openmpi-1.5.4/COMPILE-intel-12.1.0> gdb usr/bin/ompi_info
...
(gdb) run
Starting program:
...
Program received signal SIGSEGV, Segmentation fault.
opal_memory_ptmalloc2_int_malloc (av=0x7ffff7fe83d8, bytes=4102) at
../../../../../opal/mca/memory/linux/malloc.c:4080
4080 /* remove from unsorted list */
(gdb) where
#0 opal_memory_ptmalloc2_int_malloc (av=0x7ffff7fe83d8, bytes=4102) at
../../../../../opal/mca/memory/linux/malloc.c:4080
#1 0x00007ffff7c232b9 in opal_memory_linux_malloc_hook
(sz=140737354040280, caller=0x1006) at
../../../../../opal/mca/memory/linux/hooks.c:687
#2 0x0000003dd96a6871 in __alloc_dir () from /lib64/libc.so.6
#3 0x0000003ddfa053cd in ?? () from /usr/lib64/libnuma.so.1
#4 0x0000003dd8e0e445 in _dl_init_internal () from /lib64/ld-linux-x86-64.so.2
...
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
A lot of combinations and trials have been done, yet to no avail.
Intel v11.0 worked...
Thanks to Hubert Haberstock (Intel) providing the hint in:
http://software.intel.com/en-us/forums/showthread.php?t=87132
This was tested on openmpi-1.5.4 and therefore should
cmr:v1.5
This commit was SVN r25290.
match similar stuff in the event framework; only add CPPFLAGS /
LDFLAGS / LIBS / and WRAPPER_EXTRA_* of the same for the one, single,
winning component (because this framework is compile-time,
one-of-many).
This commit was SVN r25199.
only became evident when there was more than one event component.
The libevent2013 component is still ompi_ignore'd for most developers.
This commit was SVN r25198.
Since hwloc has a dynamic bitmap size, it could actually have bits set
that will not fit in the paffinity mask. We already made sure that we
didn't overrun the paffinity mask; now also set the return value to
OPAL_ERR_VALUE_OUT_OF_BOUNDS (wow, we really thought of everything
with those error codes, eh?) if the hwloc bitmap has bits set higher
than what will fit into the paffinity bitmask.
This commit was SVN r25179.
The following Trac tickets were found above:
Ticket 2854 --> https://svn.open-mpi.org/trac/ompi/ticket/2854
* change components from setting <framework>_base_include to
opal_<framework>_<component>_include; the framework m4 will figure
out the winning component and pick the right "include" shell
variable. Ditto for the other shell variables (cppflags, ldflags,
etc.).
* misc fixes to hwloc/external
* add a bunch of missing "opal_" prefixes to shell variables
* add a few more / update a few comments in framework m4's
This commit was SVN r25174.
maffinity_base_bind_failure_action MCA params to the hwloc base
(hwloc_base_alloc_polocy and hwloc_base_bind_failure_action). Since
these MCA parameters were never on a release branch, I'm just
moving/renaming them outright and not leaving aliases to the old
names.
Note that some upper layer needs to call
opal_hwloc_base_set_process_membind_policy() to set the
set-by-MCA-param process-wide memory affinity policy. We can't do
this automatically during hwloc_base_open() because, for reasons
described elsewhere, opal_hwloc_topology is not automatically filled
during hwloc_base_open() (in short: potential scalability issues when
launching many MPI processes simultaneously on a single machine, for
example).
This commit was SVN r25156.
paffinity/hwloc components were still calling hwloc_topology_init/load
themselves, and not using the opal_hwloc_topology. Doh!
This commit fixes that -- these 2 components no longer have their own
copy of the topology tree; they just use opal_hwloc_topology.
This commit was SVN r25151.
the command line, hwloc is just like any other external dependency
in OMPI: if we find it, we'll use it. If we don't find it, we'll
ignore it. See comments in opal/mca/hwloc/configure.m4 for an
explanation.
* Fix some copy-n-paste errors in opal/mca/hwloc/configure.m4
w.r.t. flags coming in from the winning component.
* Add another line in ompi_info's output about whether hwloc support
is included or not.
This commit was SVN r25134.
specify btl_tcp_if_include because btl_tcp_if_exclude is defaulted to
the loopback devices.
This commit does a few things:
* Introduce a new OPAL MCA base function:
mca_base_param_check_exclusive_string(). It checks to see that the
''user'' does not set two MCA parameters that are mutually
exclusive by checking the source of those MCS param values.
* Use the above function in many BTLs (and the OOB TCP) to ensure
that <foo>_if_include and <foo>_if_exclude are not both specified
''by the user''.
* Re-arrange many of these BTLs to move their MCA registration code
into a separate component_register() function (vs. the
component_open() function).
This code has been nominally reviewed and checked by Ralph, George,
Terry, and Shiqing.
This commit was SVN r25043.
The following SVN revision numbers were found above:
r24976 --> open-mpi/ompi@8f4ac54336
that enabling "local_only" by default could cause excessive
by-NUMA-node paging and/or OOMs (rather than allowing memory
allocations to spill over to other NUMA nodes).
This brought home the very real-world example of people buying servers
with more processors/cores than they need, just to get more memory.
We wouldn't want Badness to occur in such scenarios by default.
Instead, let people turn on "only allow memory allocations on my local
NUMA node" if their application would benefit from it.
This commit was SVN r24648.
After a long period of development with many starts and stops, we
finally got this where we wanted it.
This commit introduces 2 new MCA params (note that the
"maffinity_libnuma_policy" MCA param introduced by r24290 was removed
when libnuma support was removed). Remember that maffinity policies
are only in effect when paffinity is enaabled -- i.e., when processes
are bound to processors!
* '''maffinity_base_alloc_policy:''' Policy that determines how
general memory allocations are bound after MPI_INIT. A value of
"none" means that no memory policy is applied. A value of
"local_only" means that all memory allocations will be restricted
to the local NUMA node where each process is placed. Note that
operating system paging policies are unaffected by this setting.
For example, if "local_only" is used and local NUMA node memory is
exhausted, a new memory allocation may cause paging.
* '''maffinity_base_bind_failure_action:''' What Open MPI will do if
it explicitly tries to bind memory to a specific NUMA location, and
fails. Note that this is a different case than the general
allocation policy described by maffinity_base_alloc_policy. A
value of "warn" means that Open MPI will warn the first time this
happens, but allow the job to continue (possibly with degraded
performance). A value of "error" means that Open MPI will abort
the job if this happens.
This needs at least a little soak time on the trunk before going to
v1.5.
This commit was SVN r24639.
The following SVN revision numbers were found above:
r24290 --> open-mpi/ompi@afa654746c
The following Trac tickets were found above:
Ticket 2698 --> https://svn.open-mpi.org/trac/ompi/ticket/2698
Upgrade to hwloc 1.2 (from hwloc 1.1.2). This should fix the problems
Nathan's seeing in #2778.
Let's let this soak on the trunk for a little while and see how LANL's
MTT's work out. If that works, then we can CMR this to v1.5.
This commit was SVN r24635.
The following Trac tickets were found above:
Ticket 2778 --> https://svn.open-mpi.org/trac/ompi/ticket/2778
Nth core, so it fell over to try to find the Nth PU.
-----
hwloc isn't able to find cores on all platforms. Example: PPC64
running RHEL 5.4 (linux kernel 2.6.18) only reports NUMA nodes and
PU's. Fine.
However, note that hwloc_get_obj_by_type() will return NULL in 2
(effectively) different cases:
- no objects of the requested type were found
- the Nth object of the requested type was not found
So first we have to see if we can find *any* cores by looking for the
0th core. If we find it, then try to find the Nth core. Otherwise,
try to find the Nth PU.
This commit was SVN r24632.
Rename the memusage sensor plugin to "resusage" as it will soon be updated to include full process stat monitoring.
Extend the heartbeat sensor to report node and process stats in the heartbeat.
Store the process and node stats in their respective orte_xxx_t object.
This commit was SVN r24629.
released):
backport hwloc r 3416 from trunk: Add cache info entry _after_ checking
that we need one, thanks Andriy Gapon for the fix
This commit was SVN r24612.
The following SVN revision numbers were found above:
r3418 --> open-mpi/ompi@9972663a12
appeared multiple times in ompi_info output (so did others, but this
is the one that was noticed). Ensure that we don't repeat
opal_paffinity_base_register_params() multiple times.
This commit was SVN r24569.
fix will be included in hwloc 1.1.2.
Brad -- can you verify that this fixes the issue for you?
Fixes trac:2732.
This commit was SVN r24450.
The following Trac tickets were found above:
Ticket 2732 --> https://svn.open-mpi.org/trac/ompi/ticket/2732
filenames -- don't include the project name ("opal")
* Don't link maffinity/hwloc and paffinity/hwloc against the common
hwloc in the static build case (because this will result in
duplicate symbols)
This commit was SVN r24447.
hopefully, this now compiles for libnuma 0.9.x and libnuma 2.0.x.
Fixes for the strategy discussed in the commit message for r24442
(i.e., check against numa_get_mems_allowed(), which only exists in
libnuma 2.0.x) and the new new new plan on #2698 coming in a separate
commit.
This commit was SVN r24443.
The following SVN revision numbers were found above:
r24442 --> open-mpi/ompi@90a8fe4aad
(with libnuma-2.0.4 / LIBNUMA_API_VERSION 2): numa_get_run_node_mask
returns a struct bitmask *.
Whether it's a good idea to blindly pass that on to
numa_set_membind() is another matter: one might want to match against
the list returned by numa_get_mems_allowed(), which may be set by the
outside environment.
Refs trac:2698.
This commit was SVN r24442.
The following SVN revision numbers were found above:
r24421 --> open-mpi/ompi@31510e683b
The following Trac tickets were found above:
Ticket 2698 --> https://svn.open-mpi.org/trac/ompi/ticket/2698
what memory node the process is running on (which is guaranteed to be
a good answer because maffinity won't be invoked unless the process is
already bound to a specific processor), and then bind our memory to
that.
Refs trac:2698.
This commit was SVN r24421.
The following SVN revision numbers were found above:
r24290 --> open-mpi/ompi@afa654746c
The following Trac tickets were found above:
Ticket 2698 --> https://svn.open-mpi.org/trac/ompi/ticket/2698
Temporarily remove hwloc's internal version of myriexpress.h. It is
causing a problem when compiling Open MPI with MX support because
hwloc uses AC_CONFIG_HEADER in hwloc's hwloc.m4 to generate
opal/mca/paffinity/hwloc/hwloc/include/hwloc/config.h.
AC_CONFIG_HEADER apparently has the (undocumented) side effect of
adding -I$(top_builddir)/opal/mca/paffinity/hwloc/hwloc/include/hwloc
to OMPI's compilation flags. Hence, when the OMPI MX components are
compiled and #include "myriexpress.h" (or <myriexpress.h>) they see
hwloc's myriexpress.h before the system one. Badness ensures.
This removal is temporary because we need to figure out a better
solution. But for now, OMPI is not using hwloc's myriexpress.h file --
so it's safe to remove. I'll push this issue upstream to hwloc to
figure out a better solution...
This commit was SVN r24354.
The following Trac tickets were found above:
Ticket 2690 --> https://svn.open-mpi.org/trac/ompi/ticket/2690
Short Version:
--------------
Event engine needs to be flushed so it does not use old/stale file descriptors.
Long Version:
-------------
The problem was that the restarted process was waiting for the socket to the local daemon to finish establishing during the 'sync' operation. The core problem was that the daemon was sending a header of 36 bytes, but the restarted process only received 35 bytes of the message. So the restarted process became stuck waiting for the last byte to arrive.
After many hours of digging, I figured out that the event engine was using the same file descriptor for its evsig_cb functionality (to signal itself when a signal arrives). So when the daemon wrote in to the new fd the event engine was stealing the first byte (*shakes fist at event engine*) before the recv() could be posted.
The solution is to use the event_reinit() function on restart to re-establish the now-stale file descriptors in the event engine. This seems to have fixed the problem.
A few other minor things:
-------------------------
* Add a check to make sure the event engine is balanced in its init/finalize
* Add the opal_event_base_close() to the BLCR restart exec function (still not 100% sure it is needed, but there it is).
This commit was SVN r24296.
last December. :-(
Add new MCA param: maffinity_libnuma_policy. Thanks to David
Singleton for the suggestion. Here's the help text about it:
{{{
MCA maffinity: parameter "maffinity_libnuma_policy" (current value:
<loose>, data source: default value)
Binding policy that determines what happens if memory
is unavailable on the local NUMA node. A value of
"strict" means that the memory allocation will fail;
a value of "loose" means that the memory allocation
will spill over to another NUMA node.
}}}
This commit was SVN r24290.
the module to use the new hwloc bitmap API (the cpuset API is both
klunkier and deprecated), which simplified a few things.
This commit was SVN r24217.
{{{
base/paffinity_base_service.c: In function ‘opal_paffinity_base_cset2mapstr’:
base/paffinity_base_service.c:623: warning: unused variable ‘range_last’
base/paffinity_base_service.c:623: warning: unused variable ‘range_first’
base/paffinity_base_service.c:622: warning: unused variable ‘count’
base/paffinity_base_service.c:622: warning: unused variable ‘m’
}}}
{{{
connect/btl_openib_connect_oob.c: In function ‘init_ud_qp’:
connect/btl_openib_connect_oob.c:1111: warning: control reaches end of non-void function
connect/btl_openib_connect_oob.c: In function ‘init_device’:
connect/btl_openib_connect_oob.c:1235: warning: unused variable ‘i’
connect/btl_openib_connect_oob.c: In function ‘get_pathrecord_sl’:
connect/btl_openib_connect_oob.c:1323: warning: unused variable ‘i’
}}}
This commit was SVN r24196.
1. Remove it from libevent207.h because it is not needed.
2. Add compat to the include list so it can use queue.h when needed.
This commit was SVN r24144.
Change OMPI code in libevent to not use bool.
Add some comments to indicate OMPI specific code.
This should fix compiles on Sun Studio Solaris.
This commit was SVN r24062.
libevent creates its event-config.h during "make all" (vs. during
configure). The prior method around this didn't work because it wrote
an event-config.h.in in the source tree -- a Bad Idea(tm). The new
way uses AC_CONFIG_COMMAND to get stuff executed at the end of
config.status to create event-config.h. This seems to work properly
during make distcheck.
This commit was SVN r23975.
Note: the ompi_check_libfca.m4 file had to be modified to avoid it stomping on global CPPFLAGS and the like. The file was also relocated to the ompi/config directory as it pertains solely to an ompi-layer component.
Forgive the mid-day configure change, but I know Shiqing is working the windows issues and don't want to cause him unnecessary redo work.
This commit was SVN r23966.
After talking with Brian, we're pretty sure that this is only because
really, really old libevent didn't allow bitwise or-ing of the other
loop types, because what we really need is (EVLOOP_ONCE |
EVLOOP_NONBLOCK). And that's what EVLOOP_ONELOOP did (i.e., we
changed the logic of libevent's event.c to let ONELOOP do both ONCE
and NONBLOCK things).
In the new libevent version, we didn't implement EVLOOP_ONELOOP
properly. As a result, and we got hangs in the SM BTL add_procs
function. Note that the SM BTL wasn't to blame -- it was purely a
side-effect of bad ONELOOP integration (i.e., if you got past the SM
BTL add_procs, you may well have hung somewhere else).
This commit removes all ONELOOP customizations from event.c and
returns it to (almost) its original state from the libevent 2.0.7-rc
distribution. Everwhere in the code base where we used ONELOOP, we
now use (ONCE | NONBLOCK).
This commit was SVN r23957.
Setup the event API to support multiple bases in preparation for splitting the OMPI and ORTE events. Holding here pending shared memory resolution.
This commit was SVN r23943.
This is a fairly intrusive change, but outside of the moving of opal/event to opal/mca/event, the only changes involved (a) changing all calls to opal_event functions to reflect the new framework instead, and (b) ensuring that all opal_event_t objects are properly constructed since they are now true opal_objects.
Note: Shiqing has just returned from vacation and has not yet had a chance to complete the Windows integration. Thus, this commit almost certainly breaks Windows support on the trunk. However, I want this to have a chance to soak for as long as possible before I become less available a week from today (going to be at a class for 5 days, and thus will only be sparingly available) so we can find and fix any problems.
Biggest change is moving the libevent code from opal/event to a new opal/mca/event framework. This was done to make it much easier to update libevent in the future. New versions can be inserted as a new component and tested in parallel with the current version until validated, then we can remove the earlier version if we so choose. This is a statically built framework ala installdirs, so only one component will build at a time. There is no selection logic - the sole compiled component simply loads its function pointers into the opal_event struct.
I have gone thru the code base and converted all the libevent calls I could find. However, I cannot compile nor test every environment. It is therefore quite likely that errors remain in the system. Please keep an eye open for two things:
1. compile-time errors: these will be obvious as calls to the old functions (e.g., opal_evtimer_new) must be replaced by the new framework APIs (e.g., opal_event.evtimer_new)
2. run-time errors: these will likely show up as segfaults due to missing constructors on opal_event_t objects. It appears that it became a typical practice for people to "init" an opal_event_t by simply using memset to zero it out. This will no longer work - you must either OBJ_NEW or OBJ_CONSTRUCT an opal_event_t. I tried to catch these cases, but may have missed some. Believe me, you'll know when you hit it.
There is also the issue of the new libevent "no recursion" behavior. As I described on a recent email, we will have to discuss this and figure out what, if anything, we need to do.
This commit was SVN r23925.
This merges the branch containing the revamped build system based around converting autogen from a bash script to a Perl program. Jeff has provided emails explaining the features contained in the change.
Please note that configure requirements on components HAVE CHANGED. For example. a configure.params file is no longer required in each component directory. See Jeff's emails for an explanation.
This commit was SVN r23764.
All interface APIs for accessing the info remain unchanged in opal/util/if.c.
This has been tested on Mac, Linux, and NetBSD. Nobody else seemed interested in testing it, so there may be some future problems revealed as people try it on other OSs.
This commit was SVN r23743.
linking against libibverbs on Solaris.
Sorry for the mid-day configure change folks; I meant to commit this
last night and forgot. :-(
This commit was SVN r23606.
http://www.open-mpi.org/community/lists/devel/2010/07/8240.php
Documentation:
http://osl.iu.edu/research/ft/
Major Changes:
--------------
* Added C/R-enabled Debugging support.
Enabled with the --enable-crdebug flag. See the following website for more information:
http://osl.iu.edu/research/ft/crdebug/
* Added Stable Storage (SStore) framework for checkpoint storage
* 'central' component does a direct to central storage save
* 'stage' component stages checkpoints to central storage while the application continues execution.
* 'stage' supports offline compression of checkpoints before moving (sstore_stage_compress)
* 'stage' supports local caching of checkpoints to improve automatic recovery (sstore_stage_caching)
* Added Compression (compress) framework to support
* Add two new ErrMgr recovery policies
* {{{crmig}}} C/R Process Migration
* {{{autor}}} C/R Automatic Recovery
* Added the {{{ompi-migrate}}} command line tool to support the {{{crmig}}} ErrMgr component
* Added CR MPI Ext functions (enable them with {{{--enable-mpi-ext=cr}}} configure option)
* {{{OMPI_CR_Checkpoint}}} (Fixes trac:2342)
* {{{OMPI_CR_Restart}}}
* {{{OMPI_CR_Migrate}}} (may need some more work for mapping rules)
* {{{OMPI_CR_INC_register_callback}}} (Fixes trac:2192)
* {{{OMPI_CR_Quiesce_start}}}
* {{{OMPI_CR_Quiesce_checkpoint}}}
* {{{OMPI_CR_Quiesce_end}}}
* {{{OMPI_CR_self_register_checkpoint_callback}}}
* {{{OMPI_CR_self_register_restart_callback}}}
* {{{OMPI_CR_self_register_continue_callback}}}
* The ErrMgr predicted_fault() interface has been changed to take an opal_list_t of ErrMgr defined types. This will allow us to better support a wider range of fault prediction services in the future.
* Add a progress meter to:
* FileM rsh (filem_rsh_process_meter)
* SnapC full (snapc_full_progress_meter)
* SStore stage (sstore_stage_progress_meter)
* Added 2 new command line options to ompi-restart
* --showme : Display the full command line that would have been exec'ed.
* --mpirun_opts : Command line options to pass directly to mpirun. (Fixes trac:2413)
* Deprecated some MCA params:
* crs_base_snapshot_dir deprecated, use sstore_stage_local_snapshot_dir
* snapc_base_global_snapshot_dir deprecated, use sstore_base_global_snapshot_dir
* snapc_base_global_shared deprecated, use sstore_stage_global_is_shared
* snapc_base_store_in_place deprecated, replaced with different components of SStore
* snapc_base_global_snapshot_ref deprecated, use sstore_base_global_snapshot_ref
* snapc_base_establish_global_snapshot_dir deprecated, never well supported
* snapc_full_skip_filem deprecated, use sstore_stage_skip_filem
Minor Changes:
--------------
* Fixes trac:1924 : {{{ompi-restart}}} now recognizes path prefixed checkpoint handles and does the right thing.
* Fixes trac:2097 : {{{ompi-info}}} should now report all available CRS components
* Fixes trac:2161 : Manual checkpoint movement. A user can 'mv' a checkpoint directory from the original location to another and still restart from it.
* Fixes trac:2208 : Honor various TMPDIR varaibles instead of forcing {{{/tmp}}}
* Move {{{ompi_cr_continue_like_restart}}} to {{{orte_cr_continue_like_restart}}} to be more flexible in where this should be set.
* opal_crs_base_metadata_write* functions have been moved to SStore to support a wider range of metadata handling functionality.
* Cleanup the CRS framework and components to work with the SStore framework.
* Cleanup the SnapC framework and components to work with the SStore framework (cleans up these code paths considerably).
* Add 'quiesce' hook to CRCP for a future enhancement.
* We now require a BLCR version that supports {{{cr_request_file()}}} or {{{cr_request_checkpoint()}}} in order to make the code more maintainable. Note that {{{cr_request_file}}} has been deprecated since 0.7.0, so we prefer to use {{{cr_request_checkpoint()}}}.
* Add optional application level INC callbacks (registered through the CR MPI Ext interface).
* Increase the {{{opal_cr_thread_sleep_wait}}} parameter to 1000 microseconds to make the C/R thread less aggressive.
* {{{opal-restart}}} now looks for cache directories before falling back on stable storage when asked.
* {{{opal-restart}}} also support local decompression before restarting
* {{{orte-checkpoint}}} now uses the SStore framework to work with the metadata
* {{{orte-restart}}} now uses the SStore framework to work with the metadata
* Remove the {{{orte-restart}}} preload option. This was removed since the user only needs to select the 'stage' component in order to support this functionality.
* Since the '-am' parameter is saved in the metadata, {{{ompi-restart}}} no longer hard codes {{{-am ft-enable-cr}}}.
* Fix {{{hnp}}} ErrMgr so that if a previous component in the stack has 'fixed' the problem, then it should be skipped.
* Make sure to decrement the number of 'num_local_procs' in the orted when one goes away.
* odls now checks the SStore framework to see if it needs to load any checkpoint files before launching (to support 'stage'). This separates the SStore logic from the --preload-[binary|files] options.
* Add unique IDs to the named pipes established between the orted and the app in SnapC. This is to better support migration and automatic recovery activities.
* Improve the checks for 'already checkpointing' error path.
* A a recovery output timer, to show how long it takes to restart a job
* Do a better job of cleaning up the old session directory on restart.
* Add a local module to the autor and crmig ErrMgr components. These small modules prevent the 'orted' component from attempting a local recovery (Which does not work for MPI apps at the moment)
* Add a fix for bounding the checkpointable region between MPI_Init and MPI_Finalize.
This commit was SVN r23587.
The following Trac tickets were found above:
Ticket 1924 --> https://svn.open-mpi.org/trac/ompi/ticket/1924
Ticket 2097 --> https://svn.open-mpi.org/trac/ompi/ticket/2097
Ticket 2161 --> https://svn.open-mpi.org/trac/ompi/ticket/2161
Ticket 2192 --> https://svn.open-mpi.org/trac/ompi/ticket/2192
Ticket 2208 --> https://svn.open-mpi.org/trac/ompi/ticket/2208
Ticket 2342 --> https://svn.open-mpi.org/trac/ompi/ticket/2342
Ticket 2413 --> https://svn.open-mpi.org/trac/ompi/ticket/2413
orte_local_chip_type and orte_local_chip_model in MPI processes it the
appropriate sysinfo module found the values on the machine.
This commit was SVN r23581.
logic (even though the "else" clause for handling it was there). This
commit puts back the specific check for the word "external".
Thanks to Jed Brown for noticing the issue. Fixes trac:2503.
This commit was SVN r23475.
The following Trac tickets were found above:
Ticket 2503 --> https://svn.open-mpi.org/trac/ompi/ticket/2503
not using in Open MPI (i.e., that stuff is only used in the standalone
builds of hwloc -- it's not compiled/installed/used by Open MPI).
This commit was SVN r23416.
