To suppress Valgrind warnings, opal_tsd_keys_destruct() needs to explicitly
release TSD values of the main thread. However, they were not freed if keys are
created by non-main threads. This patch fixes it.
This patch also optimizes allocation of opal_tsd_key_values by doubling its size
when count >= length instead of increasing the size by one.
Signed-off-by: Shintaro Iwasaki <siwasaki@anl.gov>
Add a framework to support different types of threading models including
user space thread packages such as Qthreads and argobot:
https://github.com/pmodels/argobotshttps://github.com/Qthreads/qthreads
The default threading model is pthreads. Alternate thread models are
specificed at configure time using the --with-threads=X option.
The framework is static. The theading model to use is selected at
Open MPI configure/build time.
mca/threads: implement Argobots threading layer
config: fix thread configury
- Add double quotations
- Change Argobot to Argobots
config: implement Argobots check
If the poll time is too long, MPI hangs.
This quick fix just sets it to 0, but it is not good for the
Pthreads version. Need to find a good way to abstract it.
Note that even 1 (= 1 millisecond) causes disastrous performance
degradation.
rework threads MCA framework configury
It now works more like the ompi/mca/rte configury,
modulo some edge items that are special for threading package
linking, etc.
qthreads module
some argobots cleanup
Signed-off-by: Noah Evans <noah.evans@gmail.com>
Signed-off-by: Shintaro Iwasaki <siwasaki@anl.gov>
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
Deprecate --am and --amca options
Avoid default param files on backend nodes
Any parameters in the PRRTE default or user param files will have been
picked up by prte and included in the environment sent to the prted, so
don't open those files on the backend.
Avoid picking up MCA param file info on backend
Avoid the scaling problem at PRRTE startup by only reading the system
and user param files on the frontend.
Complete revisions to cmd line parser for OMPI
Per specification, enforce following precedence order:
1. system-level default parameter file
1. user-level default parameter file
1. Anything found in the environment
1. "--tune" files. Note that "--amca" goes away and becomes equivalent to "--tune". Okay if it is provided more than once on a cmd line (we will aggregate the list of files, retaining order), but an error if a parameter is referenced in more than one file with a different value
1. "--mca" options. Again, error if the same option appears more than once with a different value. Allowed to override a parameter referenced in a "tune" file
1. "-x" options. Allowed to overwrite options given in a "tune" file, but cannot conflict with an explicit "--mca" option
1. all other options
Fix special handling of "-np"
Get agreement on jobid across the layers
Need all three pieces (PRRTE, PMIx, and OPAL) to agree on the nspace
conversion to jobid method
Ensure prte show_help messages get output
Print abnormal termination messages
Cleanup error reporting in persistent operations
Signed-off-by: Ralph Castain <rhc@pmix.org>
dd
Signed-off-by: Ralph Castain <rhc@pmix.org>
We currently save the hostname of a proc when we create the ompi_proc_t for it. This was originally done because the only method we had for discovering the host of a proc was to include that info in the modex, and we had to therefore store it somewhere proc-local. Obviously, this ccarried a memory penalty for storing all those strings, and so we added a "cutoff" parameter so that we wouldn't collect hostnames above a certain number of procs.
Unfortunately, this still results in an 8-byte/proc memory cost as we have a char* pointer in the opal_proc_t that is contained in the ompi_proc_t so that we can store the hostname of the other procs if we fall below the cutoff. At scale, this can consume a fair amount of memory.
With the switch to relying on PMIx, there is no longer a need to cache the proc hostnames. Using the "optional" feature of PMIx_Get, we restrict the retrieval to be purely proc-local - i.e., we retrieve the info either via shared memory or from within the proc-internal hash storage (depending upon the active PMIx components). Thus, the retrieval of a hostname is purely a local operation involving no communication.
All RM's are required to provide a complete hostname map of all procs at startup. Thus, we have full access to all hostnames without including them in a modex or having to cache them on each proc. This allows us to remove the char* pointer from the opal_proc_t, saving us 8-bytes/proc.
Unfortunately, PMIx_Get does not currently support the return of a static pointer to memory. Thus, even though PMIx has the hostname in its memory, it can only return a malloc'd version of it. I have therefore ensured that the return from opal_get_proc_hostname is consistently malloc'd and free'd wherever used. This shouldn't be a burden as the hostname is only used in one of two circumstances:
(a) in an error message
(b) in a verbose output for debugging purposes
Thus, there should be no performance penalty associated with the malloc/free requirement. PMIx will eventually be returning static pointers, and so we can eventually simplify this method and return a "const char*" - but as noted, this really isn't an issue even today.
Signed-off-by: Ralph Castain <rhc@pmix.org>
Use "prte" instead of "prun" for proxy execution of cmds like mpirun.
This avoids the fork/exec-rendezvous complexities and should result in
more reliable operation.
Signed-off-by: Ralph Castain <rhc@pmix.org>
Extend the PMIx modex recv macros to cover the full set of
immediate/optional combinations. If PMIx_Init cannot reach a server,
then declare the MPI proc to be a singleton.
Provide full support for info values via PMIx
Catch all the values used in the "info" area of OMPI using data
available from PMIx instead of via envars. Update PMIx and PRRTE to sync
with their capabilities.
PMIx
- ensure cleanup of fork/exec children
- fix bug in gds/hash that left app info off of list
PRRTE
- fix multi-app bugs
- port setup_child logic from orte
- OMPI env changes
- set app->first_rank
- ensure common hostname across prun, prte, and pmix
- Fix "nolocal" support
Silence a warning from btl/vader
Signed-off-by: Ralph Castain <rhc@pmix.org>
- Avoid modifying single-dash options of applications
- Fix fetch of node/app-level info
- Return correct status code
Signed-off-by: Ralph Castain <rhc@pmix.org>
This commit fixes an issue with the include usage in some
ompi source files. These source files are using the <> form
of include when the "" form is correct (as these are internal,
**not** system headers).
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Signed-off-by: Nathan Hjelm <hjelmn@google.com>
Per suggestion of @awlauria, added some comments about
the need to free ep before resources it points to.
Signed-off-by: Harumi Kuno <harumi.kuno@hpe.com>
PMIx:
- Ensure that launchers open all required frameworks
- Pass back the tool's ID
- Fix race condition in IOF
PRRTE:
- Begin conversion to use of nspace in place of numeric jobid
- Restore support:
--report-bindings
--display-map
--display-devel-map
--display-topo
--do-not-launch
--xml-output
--display-allocation
Signed-off-by: Ralph Castain <rhc@pmix.org>
- Port memchecker call from a1d502c.
- Remove unused memcheck macro variables.
- Some code readability improvements.
- Remove some stray +1's in dynamic comm cleanup.
- Re-add OPAL_ENABLE_DEBUG macro to osc header.
- Cleanup some printf's, and includes.
- Refactor cleanup of dpm_disconnect_objs.
Signed-off-by: Austen Lauria <awlauria@us.ibm.com>
This fix is from John L. Byrne (john.l.byrne@hpe.com).
When OFI Libfabric binds objects to endpoints, before the object can
be successfully closed, the endpoint must first be freed. For scalable
endpoints, objects can also be bound to transmit and receive contexts,
and for objects that are bound to contexts, we need to first free the
contexts before freeing the endpoint. We also need to clear the memory
registration cache.
If we don't clean up properly, then fi\_close may not be able to close
the domain because the dom will have a non-zero ref count.
Signed-off-by: harumi kuno <harumi.kuno@hpe.com>
Keep track of the connected procs in vader_add_procs().
Otherwise, the same rank will reconnect the same shmem
segment (rank 0+...) multiple times instead of the next
one as intended.
Signed-off-by: Austen Lauria <awlauria@us.ibm.com>
Temporary solution for the PML inconsistency issue discussed in #7475.
This patch address 2 things: first it make the PMIx key optional so that
if we are not in a full modex mode we don't do a direct modex, and
second it get the PML info from the vpid 0 instead of from the local
rank.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Restrict the search to the "immediate" range so at worst we check with
our local server and don't go up to the host daemon.
Signed-off-by: Ralph Castain <rhc@pmix.org>
- Deal with deprecated cmd line options and rndz files
- Protect against DVM collisions
- Update atomics
- Fix fork/exec and provide better tool support (PMIx)
- Ensure we cleanup completely upon terminating a tool/server that has
dropped rendezvous files
- Fix multithreaded launch race on remote node
- Fix bug where multiple threads can be modifying app->env
- Add a --no-ready-msg option to prte
- Utilize the PMIx_Spawn ability to do a fork/exec on our behalf to better
setup the "prte" DVM when running in proxy mode
- Provide better isolation between DVM instances
- Fix race condition in shutdown of PMIx fork/exec framework
- Ensure ompi personality gets added in proxy scenarios
Signed-off-by: Ralph Castain <rhc@pmix.org>
The SM BTL was effectively removed a long time ago. All that was left
was a shell that warned people if they tried to use the SM BTL. For
v5.0, we plan to finally remove this ancient shell (and possibly
replace it with vader).
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
External pmix installs are frequently in non-standard locations and
the path to their shared libraries are not ldconfig'd in because
there may be multiple pmix installs.
The way the configury was set up prior to this patch, the configuration
would fail soon after the PMIX config stuff was called because it
added some pmix lib stuff to the LDFLAGS, resulting in configury tests
for things that require running the configure test to fail.
This patch avoids this problem by resetting the LDFLAGS and LIBS back
to what they were prior to the run of the external PMIX detection.
The CFLAGS setting is left because there are many places in the ompi
and opal source code where pmix_common.h needs to be included.
Signed-off-by: Howard Pritchard <hppritcha@gmail.com>
Hwloc upstream has fixed a problem with embedded "make distcheck" that
was breaking that surfaced when you ran autogen in an Open MPI
tarball.
This submodule update takes in the upstream hwloc fixes for this
issue.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
If someone gives us a namespace that doesn't easily translate to an
integer, we have to create a mechanism for working around the
disconnect. PRRTE has been updated to give us a flag so we know we were
"natively" launched. If we don't see it, then fall back to generating a
hash of the nspace as our jobid. We then have to translate back/forth
between nspace and jobid using a lookup table.
Probably not the right long-term solution, but hopefully helps get us
thru for a bit.
Includes update of PRRTE pointer
Signed-off-by: Ralph Castain <rhc@pmix.org>
- remove stale s390 and MIPS atomics
- ensure envars from spawn are propagated
- fix make tarball
- ensure cleanup of default hostfile
Signed-off-by: Ralph Castain <rhc@pmix.org>
This commit fixes an issue in the MCA base variable system. The
code was retrieving the user home directory (from HOME) and
attempting to use it to build a search path for config files.
In this case user-level configuration directories have been
enabled so the appropriate thing to do is to print an error
message and return. This commit makes that change. It does not
ensure that HOME is set correctly.
Signed-off-by: Nathan Hjelm <hjelmn@google.com>
If you autogen.pl --without-prrte, we wouldn't configure or build PRRTE
support. However, configuring with --disable-internal-rte wasn't working
as it was being ignored. This led to some false errors when compiling
with an earlier PMIx v2.2 release.
That said, there were a couple of places that needed protection against
PMIx v2.2.
Signed-off-by: Ralph Castain <rhc@pmix.org>
Fix a couple of spots in OMPI to resolve warnings. The one in comm_cid
in particular may be responsible for some/all of the comm_spawn issues
as it was passing an incorrect pointer to a macro, thus causing memory
corruption.
Update PRRTE and PMIx to deal with v3/v4 differences.
Signed-off-by: Ralph Castain <rhc@pmix.org>
When a job fails very quickly, it is possible that the spawning tool
won't get notified that the spawn completed prior to be told that the
job terminated. This can cause the tool to "hang" in PMIx_Spawn. Ensure
that PRRTE handles this case by guaranteeing we notify the spawner.
Track both PRRTE and PMIx masters as both have changed, though only the
PRRTE one is involved in this particular fix.
Signed-off-by: Ralph Castain <rhc@pmix.org>
- Ensure we accurately handle node name aliases
- Apply the local launch environ to apps prior to spawn
- Add error check if PMIx_Spawn fails
- Fix compiler warning
- Fix PGI vendor check
- Prevent mpirun from hanging if prte segfaults
- Fix absolute/relative path names to ensure "prte" and
"prted" are taken from same distribution as "mpirun"
Signed-off-by: Ralph Castain <rhc@pmix.org>
This commit fixes a potential deadlock that can occur between the
memory hooks and region registation. This deadlock occurs because
of a hold and wait error between two mutexes. The first mutex is
the VMA lock used to protect internal rcache/grdma structures and
the reader/writer lock in the interval tree.
In the case of the memory hooks a reader lock is obtained on the
interval tree then the VMA lock is obtained to remove the
registration from the LRU. In the case of LRU evictions the VMA
lock is obtained then the writer lock on the interval tree is
obtained. This leads to the deadlock.
To fix the issue the code that evicts from the LRU has been
updated to only invalidate the registration while the VMA lock
is held then remove the registration from the VMA after the
lock is released. This should completely eliminate the above
deadlock.
Signed-off-by: Nathan Hjelm <hjelmn@google.com>
Will be replaced by PRRTE. Ensure that OMPI and OPAL layers build
without reference to ORTE. Setup opal/pmix framework to be static.
Remove support for all PMI-1 and PMI-2 libraries. Add support for
"external" pmix component as well as internal v4 one.
remove orte: misc fixes
- UCX fixes
- VPATH issue
- oshmem fixes
- remove useless definition
- Add PRRTE submodule
- Get autogen.pl to traverse PRRTE submodule
- Remove stale orcm reference
- Configure embedded PRRTE
- Correctly pass the prefix to PRRTE
- Correctly set the OMPI_WANT_PRRTE am_conditional
- Move prrte configuration to the end of OMPI's configure.ac
- Make mpirun a symlink to prun, when available
- Fix makedist with --no-orte/--no-prrte option
- Add a `--no-prrte` option which is the same as the legacy
`--no-orte` option.
- Remove embedded PMIx tarball. Replace it with new submodule
pointing to OpenPMIx master repo's master branch
- Some cleanup in PRRTE integration and add config summary entry
- Correctly set the hostname
- Fix locality
- Fix singleton operations
- Fix support for "tune" and "am" options
Signed-off-by: Ralph Castain <rhc@pmix.org>
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
Previously we used a fairly simple algorithm in
mca_btl_tcp_proc_insert() to pair local and remote modules. This was a
point in time solution rather than a global optimization problem (where
global means all modules between two peers). The selection logic would
often fail due to pairing interfaces that are not routable for traffic.
The complexity of the selection logic was Θ(n^n), which was expensive.
Due to poor scalability, this logic was only used when the number of
interfaces was less than MAX_PERMUTATION_INTERFACES (default 8). More
details can be found in this ticket:
https://svn.open-mpi.org/trac/ompi/ticket/2031 (The complexity estimates
in the ticket do not match what I calculated from the function)
As a fallback, when interfaces surpassed this threshold, a brute force
O(n^2) double for loop was used to match interfaces.
This commit solves two problems. First, the point-in-time solution is
turned into a global optimization solution. Second, the reachability
framework was used to create a more realistic reachability map. We
switched from using IP/netmask to using the reachability framework,
which supports route lookup. This will help many corner cases as well as
utilize any future development of the reachability framework.
The solution implemented in this commit has a complexity mainly derived
from the bipartite assignment solver. If the local and remote peer both
have the same number of interfaces (n), the complexity of matching will
be O(n^5).
With the decrease in complexity to O(n^5), I calculated and tested
that initialization costs would be 5000 microseconds with 30 interfaces
per node (Likely close to the maximum realistic number of interfaces we
will encounter). For additional datapoints, data up to 300 (a very
unrealistic number) of interfaces was simulated. Up until 150
interfaces, the matching costs will be less than 1 second, climbing to
10 seconds with 300 interfaces. Reflecting on these results, I removed
the suboptimal O(n^2) fallback logic, as it no longer seems necessary.
Data was gathered comparing the scaling of initialization costs with
ranks. For low number of interfaces, the impact of initialization is
negligible. At an interface count of 7-8, the new code has slightly
faster initialization costs. At an interface count of 15, the new code
has slower initialization costs. However, all initialization costs
scale linearly with the number of ranks.
In order to use the reachable function, we populate local and remote
lists of interfaces. We then convert the interface matching problem
into a graph problem. We create a bipartite graph with the local and
remote interfaces as vertices and use negative reachability weights as
costs. Using the bipartite assignment solver, we generate the matches
for the graph. To ensure that both the local and remote process have
the same output, we ensure we mirror their respective inputs for the
graphs. Finally, we store the endpoint matches that we created earlier
in a hash table. This is stored with the btl_index as the key and a
struct mca_btl_tcp_addr_t* as the value. This is then retrieved during
insertion time to set the endpoint address.
Signed-off-by: William Zhang <wilzhang@amazon.com>
Advance to hwloc-2.1.0rc2-33-g38433c0f, which includes a .gitignore
update that we want here in Open MPI.
Be warned; this is actually 33 commits beyond the hwloc v2.1.0 tag.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
This PR removes the constant defining the max attachment address and
replaces it with the largest address that shows up in /proc/self/maps.
This should address issues found on AARCH64 where the max address
may differ based on the configuration.
Since the calculated max address may differ between processes the
max address is sent as part of the modex and stored in the endpoint
data.
Signed-off-by: Nathan Hjelm <hjelmn@google.com>
The opal_gethostname() function provides a more robust mechanism
to retrieve the hostname than gethostname(), which can return
results that are not null-terminated, and which can vary in its
behavior from system to system.
opal_gethostname() just returns the value in opal_process_info.nodename;
this is populated in opal_init_gethostname() inside opal_init.c.
-Changed all gethostname calls in opal subtree to opal_gethostname
-Changed all gethostname calls in orte subtree to opal_gethostname
-Changed all gethostname calls in ompi subdir to opal_gethostname
-Changed all gethostname calls in oshmem subdir to opal_gethostname
-Changed opal_if.c in test subdir to use opal_gethostname
-Changed opal_init.c to include opal_init_gethostname. This function
returns an int and directly sets opal_process_info.nodename per
jsquyres' modifications.
Relates to open-mpi#6801
Signed-off-by: Charles Shereda <cpshereda@lanl.gov>
Make sure to get an RDM provider that can provide both local and
remote communication. We need this check because some providers could
be selected via RXD or RXM, but can't provide local communication, for
example.
Add OPAL_CHECK_OFI_VERSION_GE() m4 macro to check that the Libfabric
we're building against is >= a target version. Use this check in two
places:
1. MTL/OFI: Make sure it is >= v1.5, because the FI_LOCAL_COMM /
FI_REMOTE_COMM constants were introduced in Libfabric API v1.5.
2. BTL/usnic: It already had similar configury to check for Libfabric
>= v1.1, but the usnic component was checking for >= v1.3. So
update the btl/usnic configury to use the new macro and check for
>= v1.3.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
This commit fixes an issue discovered in the XPMEM registration cache. It
was possible for a registration to be invalidated by multiple threads
leading to a double-free situation or re-use of an invalidated registration.
This commit fixes the issue by setting the INVALID flag on a registation
when it will be deleted. The flag is set while iterating over the tree
to take advantage of the fact that a registration can not be removed
from the VMA tree by a thread while another thread is traversing the VMA
tree.
References #6524
References #7030Closes#6534
Signed-off-by: Nathan Hjelm <hjelmn@google.com>
Rename the component to be "hwloc2" (since it can now be any v2.x.y
version of hwloc), and make the embedded copy of hwloc be a git
submodule.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Clarify in README what --with-hwloc does in its different use cases.
Also, ensure that the behavior when specifying `--with-hwloc` is the
same as if that option is not specified at all. This is what we did
in Open MPI <= v3.x; looks like we inadvertantly caused `--with-hwloc`
to be synonymous with `--with-hwloc=external` in v4.0.0.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
