Please report any abnormal behavior during launch, though, as we would like to understand what (if any) impact is seen. I couldn't see any on small jobs (the modulo functions render this number down pretty low).
This commit was SVN r10763.
Update the help text to report errors when not following that rule.
Also updated the RMAPS help text to reflect the reorganization of some of the round-robin code into the base.
The new functionality has been tested under Mac OS-X and on Odin using an MPI program. Both byslot and bynode mapping have been checked and verified. Operational support for other systems needs to be verified - I respectfully request people's help in doing so.
This commit was SVN r10708.
1. Modifies the RAS framework so it correctly stores and retrieves the actual slots in use, not just those that were allocated. Although the RAS node structure had storage for the number of slots in use, it turned out that the base function for storing and retrieving that information ignored what was in the field and simply set it equal to the number of slots allocated. This has now been fixed.
2. Modified the RMAPS framework so it updates the registry with the actual number of slots used by the mapping. Note that daemons are still NOT counted in this process as daemons are NOT mapped at this time. This will be fixed in 2.0, but will not be addressed in 1.x.
3. Added a new MCA parameter "rmaps_base_no_oversubscribe" that tells the system not to oversubscribe nodes even if the underlying environment permits it. The default is to oversubscribe if needed and the underlying environment permits it. I'm sure someone may argue "why would a user do that?", but it turns out that (looking ahead to dynamic resource reservations) sometimes users won't know how many nodes or slots they've been given in advance - this just allows them to say "hey, I'd rather not run if I didn't get enough".
4. Reorganizes the RMAPS framework to more easily support multiple components. A lot of the logic in the round_robin mapper was very valuable to any component - this has been moved to the base so others can take advantage of it.
5. Added a new test program "hello_nodename" - just does "hello_world" but also prints out the name of the node it is on.
6. Made the orte_ras_node_t object a full ORTE data type so it can more easily be copied, packed, etc. This proved helpful for the RMAPS code reorganization and might be of use elsewhere too.
This commit was SVN r10697.
using a pty for everything, which drops all buffered data on the floor when
close() is called on the daemon side, meaning EOF has some issues. Instead,
do the same thing we do for other starters that use the fork() pls -- use
a pipe/fifo for stdin and stderr and a pty for stdout. This is good enough
for what we need and avoids most of the issues with ptys.
This commit was SVN r10692.
Basically, the problem was that the allocator was grabbing everything on the cluster for which the user had access privilege. Thus, if a user had two sessions operable, each with its own allocation, mpirun in each session would grab both sets of nodes and use them. Not very polite.
This commit was SVN r10683.
* num_children should really be an int instead of size_t
since 'size_t' is not signed and num_children can (in rare cases)
drop below 0, and don't want it to roll around to MAX_INT or some
such.
* I figured out that this problem only happened to me because I use
the pls_fork_reap_timeout MCA parameter and thus the only time that
the code in pls_fork_module.c to waitpid is executed is if this is
not set to 0 (I had it set to 1 to give my procs time to exit). I
adjusted the loop from while{...} to do{...}while; so that it is
executed at least once for consistency.
* de-register the SIGCHILD callback for the pid before we attempt
to kill it, so that we don't leave the door open for both the
waitpids (the one in the callback, and the one in this function)
to race to see who can wait on the child.
* Move the 'thread release' to outside the for loop for a bit of an
optimization, and always set the value to 0 since we want to
finish after this function.
* Added a help message for the case when we can't send a kill()
signal to the process. Should never happen, but all is possible
in the wild wild west of HPC.
This commit was SVN r10666.
When we force an application to terminate (via CTRL-C to mpirun)
we send an out-of-band message to the orted to reap its children.
the fork PLS was doing an internal waitpid but never releasing or
updating the information and signaling the condition variable. So
the fork PLS callback for SIGCHLD registered with the event library
and this waitpid are in a bit of a race to 'waitpid' for the children.
Since the PLS callback was the only one that handled the signal properly
when it 'won' then things were great -- as in the normal termination case.
But when it 'lost' -- as in the abnormal termination case -- the orted
never received the proper signal that its children had gone away.
We want to preserve the internal fork PLS callback since it allows
for a timeout while waiting for the child, which the event library
won't do.
This allows both to exist, and behave properly.
This was introduced in r9068.
The ticket is still open since the orted's hang in other situations
still. This is a fix for one of the causes.
This commit was SVN r10662.
The following SVN revision numbers were found above:
r9068 --> open-mpi/ompi@c2c2daa966
/tmp/tm-merge). Validated by RHC. Summary:
- Add --nolocal (and -nolocal) options to orterun
- Make some scalability improvements to the tm pls
This commit was SVN r10651.
with an error status (< 0) then the req buffer is NULL. Put checks around the
OBJ_RELEASE(req) calls so that we don't try to release NULL :/
This commit was SVN r10641.
After seeing the uglyness that is removing directories in the
codebase I decided to push down this to the OPAL by extending the
opal/os_create_dirpath.(c|h) to contain some more functionality.
In this process I renamed 'os_create_dirpath' to 'os_dirpath' since it
is a bit more general now.
Added a few functions to:
- check if an directory is empty
- check to see if the access permissions are set correctly
- destroy the directory at the end of the dirpath
- By using a caller callback function (a la Perl, I believe)
for every file, the caller can have fine grained control over
whether a specific file is deleted or not.
This simplifies things a bit for orte_session_dir_(finalize|cleanup)
as it should no longer contain any of this functionality, but uses
these functions to do the work.
From the external perspective nothing has changed, from the
developer point of view we have some cleaner, more generic code.
This commit was SVN r10640.
from the tmp/jjhursey-ft-cr branch.
In this commit we change the way universe names are created.
Before we by default first created "default-universe" then
if there was a conflict we created "default-universe-PID"
where PID is the PID of the HNP.
Now we create "default-universe-PID" all the time (when
a default universe name is used). This makes it much
easier when trying to find a HNP from an outside app
(e.g. orte-ps, orteconsole, ...)
This also adds a "search" function to find all of the
universes on the machine. This is useful in many contexts
when trying to find a persistent daemon or when trying to
connect to a HNP.
This commit also makes orte_universe_t an opal_object_t,
which is something that needed to happen, and only effected
the SDS in one of it's base functions.
I was asked to bring this over to aid in fixing orteconsole
and orteprobe. Due to the change of orte_universe_t to
an object orteprobe may need to be updated to reflect this
change. Since orteprobe needs to be looked at anyway I'll
leave this to Ralph to take care of.
*Note*:
These changes do not depend upon any of the FT work (but
the FT work does depend upon them). These were brought over
to help in fixing some of the ORTE tool set that require
the functionality layed out in this patch.
Testing:
Ran the 'ibm' tests before and after this change, and all was
as well as before the change. If anyone notices additional
irregularities in the system let me know. But none are expected.
This commit was SVN r10550.
This commit will apply cleanly to the v1.1 branch, and should
be moved over once I get someone to verify it.
The problem is outlined in the bug. The fix was to move the
setting of the app context index (idx) before we put it in the
GPR so that it is propogated to the gpr.
The reason this hasn't bitten us before is because we init
app->idx to 0, which is true most of the time. Except that is
when MPI_Comm_spawn_multiple in which we put in more than
one app context, thus care about correct indexing.
This was causing down the line memory corruption by overrunning
the mapping array. This commit also puts in a check to make
sure that we error out if we ever try to do that again.
This commit was SVN r10380.
1. Changed the RMGR and PLS APIs to add "signal_job" and "signal_proc" entry points. Only the "signal_job" entries are implemented - none of the components have implementations for "signal_proc" at this time. Thus, you can signal all of the procs in a job, but cannot currently signal only one specific proc.
2. Implemented those new API functions in all components except xgrid (Brian will do so very soon). Only the rsh/ssh and fork modules have been tested, however, and only under OS-X.
3. Added signal traps and callback functions for SIGUSR1/2 to orterun/mpirun that catch those signals and call the appropriate commands to propagate them out to all processes in the job.
4. Added a new test directory under the orte branch to (eventually) hold unit and system level tests for just the run-time. Since our test branch of the repository is under restricted access, people working on the RTE were continually developing their own system-level tests - thus making it hard to help diagnose problems. I have moved the more commonly-used functions here, and added one specifically for testing the SIGUSR1/2 functionality.
I will be contacting people directly to seek help with testing the changes on more environments. Other than compile issues, you should see absolutely no change in behavior on any of your systems - this additional functionality is transparent to anyone who does not issue a SIGUSR1/2 to mpirun.
Ralph
This commit was SVN r10258.
of $libdir and $bindir (i.e., was correctly doing local launches, but
was still using $prefix/lib and $prefix/bin for remote launches).
[Re-]Fixes OFED bug 59.
This commit was SVN r10207.
The following SVN revision numbers were found above:
r9930 --> open-mpi/ompi@1d6902296c
We were stuck in an infinite loop inside the rmaps round_robin
component when the user specified a host, then over subscribed it.
Instead of retuning an error, we looped forever.
For example:
$ cat hostfile
A slots=2 max-slots=2
B slots=2 max-slots=2
$ mpirun -np 3 --hostfile hostfile --host B
<hang>
The loop would not terminate because both host A and B are in the
'nodes' structure as they are both allocated to the job. However,
after allocating 2 slots to host B, we remove it from the node list
leaving us with a 'nodes' structure with just A in it. Since we can't
use host A, we keep looping here until we find a node that we can use.
This patch checks to make sure that if we get into this situation where
rmaps is looping over the list a second time without finding a node
during the first pass then we know that there are no nodes left to
use, so we have a resource allocation error, and should return to the user.
This patch should be moved to all of the release branches
This commit was SVN r10131.
pipeline. See lengthy comment in iof_base_endpoint.c for the details, but
the short version is that we shouldn't set O_NONBLOCK on standard I/O
file descriptors, so we no longer do.
Closes ticket:9
This commit was SVN r9966.
option as discussed on the devel-core mailing list. The Big
Difference is that instead of hard-coding the strings "/lib" and
"/bin" in to append to the prefix, we append the basename of the local
libdir and bindir. Hence, if your libdir is $prefix/lib64, we'll
append /lib64 to construct the remote node's LD_LIBRARY_PATH (etc.).
Also appended the orterun.1 man page to include a description of
--prefix, how it is constructed, what it handles / what it does not,
etc.
This commit was SVN r9930.
exit out, rather than trying to have the pls exit. Since singletons
weren't started with a pls, there's no way the pls is going to be
able to kill the process. So just exit and save the error message.
This commit was SVN r9859.
as the smaller machine's SIZE_MAX won't be SIZE_MAX on the bigger machine, which
can lead to failures along the way -- in this case, with GPR triggers being
improperly fired.
This commit was SVN r9776.
is interpreted as a shutdown of the io channel on the next iteration.
Definitively not the good approach. The correct condition is
bigger than 0.
This commit was SVN r9770.
copy of the receive buffer based on the iovec struct that may have been updated
during partial reads to reflect the current offset. Need to make the copy using
the base address of the buffer.
Thanks to Sven Stork for finding this.
This should be backported to 1.0.X and 1.1.X branches.
This commit was SVN r9749.
- The constant 1 is a signed int by default. Explicitly say that
it is an unsigned value so we can't overflow
- Fix unreachable statement warnings in dss_arith by breaking out
of switch statements instead of returning - this should have
no impact on performance, since it's a non-conditional jump
- A couple of the GPR files had carriage returns and were in
DOS mode - put them in unix mode...
These should all probably go to the v1.1 branch...
This commit was SVN r9664.
