* Ensure to destroy the correct QP (local->id[num]->qp will always
have a valid pointer in it, even if we setup a dummy qp)
* Note two notable places where we need to figure out how to
propagate errors up from the CPC to the main BTL / PML when errors
occur. Probably have the same issue in IBCM, too.
This commit was SVN r18700.
two processes on the same server (!). So for today, we'll simply mark
all local processes that use iWARP adapters as "unreachable".
More details in #1352.
This commit was SVN r18699.
(e.g., if you're excluding some devices, their destructors will be
invoked before the async event thread was setup for them).
This commit was SVN r18698.
multiple adapters (eg., Chelsio T3).
But we need to figure out how to determine a good value for the
resident adapter(s) at runtime. It's problematic because, for
example, Mellanox ConnectX and Chelsio T3 report max_inline values
differently at run-time. If you ibv_create_qp with a max_inline value
of 0, ConnectX reports back a value that is a formular based on a few
other values (e.g., max_send_sge and max_recv_sge). But T3 always
reports back "64".
We're looking into this to figure out the best way -- reducing the
default right now should allow other adapters to run while we figure
it out.
This commit was SVN r18697.
Some minor changes to help facilitate debugger support so that both mpirun and yod can operate with it. Still to be completed.
This commit was SVN r18664.
a standalone library named libopenmpi-malloc. Users wanting to
use leave_pinned with ptmalloc2 will now need to link the library
into their application explicitly. All other users will use the
libc-provided allocator instead of Open MPI's ptmalloc2. This change
may be overriden with the configure option enable-ptmalloc2-internal
- The leave_pinned options will now default to using mallopt on
Linux in the cases where ptmalloc2 was not linked in. mallopt
will also only be available if munmap can be intercepted (the
default whenever Open MPI is not compiled with --without-memory-
manager.
- Open MPI will now complain and refuse to use leave_pinned if
no memory intercept / mallopt option is available.
This commit was SVN r18654.
After much work by Jeff and myself, and quite a lot of discussion, it has become clear that we simply cannot resolve the infinite loops caused by RML-involved subsystems calling orte_output. The original rationale for the change to orte_output has also been reduced by shifting the output of XML-formatted vs human readable messages to an alternative approach.
I have globally replaced the orte_output/ORTE_OUTPUT calls in the code base, as well as the corresponding .h file name. I have test compiled and run this on the various environments within my reach, so hopefully this will prove minimally disruptive.
This commit was SVN r18619.
We already show_help when we fail to create queues, so I just made the
message a little more verbose such that it may be that OMPI is trying
to use a feature that is not supported on the hardware.
This commit was SVN r18553.
The following Trac tickets were found above:
Ticket 1121 --> https://svn.open-mpi.org/trac/ompi/ticket/1121
1. The send path get shorter. The BTL is allowed to return > 0 to specify that the
descriptor was pushed to the networks, and that the memory attached to it is
available again for the upper layer. The MCA_BTL_DES_SEND_ALWAYS_CALLBACK flag
can be used by the PML to force the BTL to always trigger the callback.
Unmodified BTL will continue to work as expected, as they will return OMPI_SUCCESS
which force the PML to have exactly the same behavior as before. Some BTLs have
been modified: self, sm, tcp, mx.
2. Add send immediate interface to BTL.
The idea is to have a mechanism of allowing the BTL to take advantage of
send optimizations such as the ability to deliver data "inline". Some
network APIs such as Portals allow data to be sent using a "thin" event
without packing data into a memory descriptor. This interface change
allows the BTL to use such capabilities and allows for other optimizations
in the future. All existing BTLs except for Portals and sm have this interface
set to NULL.
This commit was SVN r18551.
than using a single receive callback followed by a switch on the header.
Also fast pathed the matching for small fragments.
This commit was SVN r18549.
that it's a directory. That's good enough to know that the
OpenFabrics kernel drivers have been loaded. If you have no RDMA
devices and don't want to see the OMPI warning about not finding any
devices, then don't start the OpenFabrics kernel drivers.
This commit was SVN r18540.
non-empty. If not, then exit the openib btl silently. This addresses
the case where libibverbs is installed (which is getting more common)
and therefore the openib BTL was built/installed, but the kernel
drivers are not loaded (assumedly because there is no RDMA hardware
present). In this case, "mpirun a.out" will not issue a warning.
There appears to be no good way to definitely tell if there are no
RDMA hardware devices present. For example, if libibverbs/the openib
BTL is installed, there are no RDMA devices present, but the RDMA
hardware kernel drivers ''are'' loaded, OMPI will warn that it was
unable to find suitable devices. This warning is easily eliminated by
unloading the kernel drivers.
This commit was SVN r18530.
The following Trac tickets were found above:
Ticket 1305 --> https://svn.open-mpi.org/trac/ompi/ticket/1305
* s/port/tcp_port/g where relevant to disambiguate TCP port from
device port
* Rework ipaddrcheck to make it work in the LMC>0 case
This commit was SVN r18482.
The following Trac tickets were found above:
Ticket 1281 --> https://svn.open-mpi.org/trac/ompi/ticket/1281
* Ensure _iwarp.h is always included, or you'll get warnings on
platforms that don't have the RDMACM
* Add skeleton for function descriptions in comments in iwarp.h
This commit was SVN r18477.
opal_ifnext() return -1 upon completion); don't check it against
opal_ifcount() -- the interface indexes aren't necessarily related to
how many interfaces were found.
This commit was SVN r18476.
This commit has the same commit message as r18450, but without the
extra bonus memory corruption that was introduced.
This commit was SVN r18467.
The following SVN revision numbers were found above:
r18450 --> open-mpi/ompi@5295902ebe
The following Trac tickets were found above:
Ticket 1285 --> https://svn.open-mpi.org/trac/ompi/ticket/1285
However, no decision logic is changed by this commit so default behavior has not changed. This
is only selectable by runtime parameters.
This commit was SVN r18464.
Need to release the items and the item list after selecting the collective
modules that are being used. Reviewed by Jeff Squyres.
This commit was SVN r18457.
1. We can't use orte_output in the CPC service thread because orte is
not thread safe
1. Use the macro version sso that they're compiled out of production
builds
This commit was SVN r18455.
* allow receive_queues to be specified in the INI file
* detect when multiple different receive_queues are specified and
gracefully abort
However, accomplishing these goals ran into multiple difficulties. By
putting receive_queues in the INI file:
1. we may not find the value until we've already traversed multiple HCAs
1. we may find multiple different receive_queues values
But since the openib btl initializes as it discovers each HCA/port/LID
(including the BSRQ data), if we find a new receive_queues value late
in the discovery process, then all the BSRQ data that was previously
initialized will likely be invalid. So I had to pull all the BSRQ
initialization out until after the rest of the discovery /
initialization process.
Additionally, note that if the user specifies the MCA parameter
btl_openib_receive_queues, it trumps whatever was in the INI file. So
in this case, there can never be a receive_queues conflict. This
commit does the following (Jon wrote part of this, too):
* adapt _ini.c to accept the "receive_queues" field in the file
* move 90% of _setup_qps() from _ini.c to _component.c
* move what was left of _setup_qps() into the main
_register_mca_params() function
* adapt init_one_hca() to detect conflicting receive_queues values
from the INI file
* after the _component.c loop calling init_one_hca():
* call setup_qps() to parse the final receive_queues string value
* traverse all resulting btls and initialize their HCAs (if they
weren't already): setup some lists and call prepare_hca_for_use()
I tested this code on a dual-HCA system where I artificially put in
differing receive_queues values in the INI file for the two different
types of HCAs that I have and it all seemed to work.
This commit was SVN r18450.
The following Trac tickets were found above:
Ticket 1285 --> https://svn.open-mpi.org/trac/ompi/ticket/1285
There are more details in the code regarding how to use this feature.
Also shift a few of the orte_output back to opal_output. I'm experiencing an odd problem with locks in the oob/tcp when using orte_output. I haven't had time to track it down yet.
This commit was SVN r18439.
such, the commit message back to the master SVN repository is fairly
long.
= ORTE Job-Level Output Messages =
Add two new interfaces that should be used for all new code throughout
the ORTE and OMPI layers (we already make the search-and-replace on
the existing ORTE / OMPI layers):
* orte_output(): (and corresponding friends ORTE_OUTPUT,
orte_output_verbose, etc.) This function sends the output directly
to the HNP for processing as part of a job-specific output
channel. It supports all the same outputs as opal_output()
(syslog, file, stdout, stderr), but for stdout/stderr, the output
is sent to the HNP for processing and output. More on this below.
* orte_show_help(): This function is a drop-in-replacement for
opal_show_help(), with two differences in functionality:
1. the rendered text help message output is sent to the HNP for
display (rather than outputting directly into the process' stderr
stream)
1. the HNP detects duplicate help messages and does not display them
(so that you don't see the same error message N times, once from
each of your N MPI processes); instead, it counts "new" instances
of the help message and displays a message every ~5 seconds when
there are new ones ("I got X new copies of the help message...")
opal_show_help and opal_output still exist, but they only output in
the current process. The intent for the new orte_* functions is that
they can apply job-level intelligence to the output. As such, we
recommend that all new ORTE and OMPI code use the new orte_*
functions, not thei opal_* functions.
=== New code ===
For ORTE and OMPI programmers, here's what you need to do differently
in new code:
* Do not include opal/util/show_help.h or opal/util/output.h.
Instead, include orte/util/output.h (this one header file has
declarations for both the orte_output() series of functions and
orte_show_help()).
* Effectively s/opal_output/orte_output/gi throughout your code.
Note that orte_output_open() takes a slightly different argument
list (as a way to pass data to the filtering stream -- see below),
so you if explicitly call opal_output_open(), you'll need to
slightly adapt to the new signature of orte_output_open().
* Literally s/opal_show_help/orte_show_help/. The function signature
is identical.
=== Notes ===
* orte_output'ing to stream 0 will do similar to what
opal_output'ing did, so leaving a hard-coded "0" as the first
argument is safe.
* For systems that do not use ORTE's RML or the HNP, the effect of
orte_output_* and orte_show_help will be identical to their opal
counterparts (the additional information passed to
orte_output_open() will be lost!). Indeed, the orte_* functions
simply become trivial wrappers to their opal_* counterparts. Note
that we have not tested this; the code is simple but it is quite
possible that we mucked something up.
= Filter Framework =
Messages sent view the new orte_* functions described above and
messages output via the IOF on the HNP will now optionally be passed
through a new "filter" framework before being output to
stdout/stderr. The "filter" OPAL MCA framework is intended to allow
preprocessing to messages before they are sent to their final
destinations. The first component that was written in the filter
framework was to create an XML stream, segregating all the messages
into different XML tags, etc. This will allow 3rd party tools to read
the stdout/stderr from the HNP and be able to know exactly what each
text message is (e.g., a help message, another OMPI infrastructure
message, stdout from the user process, stderr from the user process,
etc.).
Filtering is not active by default. Filter components must be
specifically requested, such as:
{{{
$ mpirun --mca filter xml ...
}}}
There can only be one filter component active.
= New MCA Parameters =
The new functionality described above introduces two new MCA
parameters:
* '''orte_base_help_aggregate''': Defaults to 1 (true), meaning that
help messages will be aggregated, as described above. If set to 0,
all help messages will be displayed, even if they are duplicates
(i.e., the original behavior).
* '''orte_base_show_output_recursions''': An MCA parameter to help
debug one of the known issues, described below. It is likely that
this MCA parameter will disappear before v1.3 final.
= Known Issues =
* The XML filter component is not complete. The current output from
this component is preliminary and not real XML. A bit more work
needs to be done to configure.m4 search for an appropriate XML
library/link it in/use it at run time.
* There are possible recursion loops in the orte_output() and
orte_show_help() functions -- e.g., if RML send calls orte_output()
or orte_show_help(). We have some ideas how to fix these, but
figured that it was ok to commit before feature freeze with known
issues. The code currently contains sub-optimal workarounds so
that this will not be a problem, but it would be good to actually
solve the problem rather than have hackish workarounds before v1.3 final.
This commit was SVN r18434.
start sending fragment by copy in/out before ACK is received as we don't
know pointer to receive request yet.
Pipeline protocol sometimes doesn't send ACK though, so this case is still
broken.
This commit was SVN r18423.
* Remove the opal_only option. This was suffering from bit rot, and no one uses it. It can be added back fairly easily if wanted.
* Cleanup metadata interactions at the local level.
* Touch up some of the INC funcitonality (fix typos and a minor ordering issue)
This commit was SVN r18416.
The iWARP subnet ID determination should not be in the RDMACM cpc, as
it was in the preversion, as this violates the cpc abstract that is
present throughout the code. Also, this patch uses the opal_list_t
data struct instead of using its own linked lists.
This attempt includes *iwarp.c and *iwarp.h
This commit was SVN r18414.
the btl_openib_iwarp.c and btl_openib_iwarp.h files.
This commit was SVN r18410.
The following SVN revision numbers were found above:
r18409 --> open-mpi/ompi@056bbb68c8
The iWARP subnet ID determination should not be in the RDMACM cpc, as
it was in the preversion, as this violates the cpc abstract that is
present throughout the code. Also, this patch uses the opal_list_t
data struct instead of using its own linked lists.
This commit was SVN r18409.
All spawned procs must decode the port of the spawning process so they can communicate in direct routed mode.
This fixes comm_spawn for all routing modes.
This commit was SVN r18395.
This enables subnet differientation for iWARP devices, and rearrange
initilization so that the services are available when they are needed.
This commit was SVN r18393.
If there is no IP Address, have rdmacm log the correct error and let
another cpc have a go at it. This is being done by splitting off the
IP address checking logic for the modex message creation, and having
it log the correct error in the error case.
This commit was SVN r18392.
For iWARP, the TCP connection is tied to the QP once the QP is in RTS.
And destroying the QP is thus tied to connection teardown for iWARP.
This is a key distinction from IB, I think. Anyway, to destroy the
connection in iWARP you must move the QP out of RTS, either into CLOSING
for a nice graceful close, or to ERROR if you want to be rude. In both
cases, all pending non-completed SQ and RQ WRs must be flushed.
This patch ignores all flush errors reaped by the cq and removes an
earlier attempt to work around this in the rdmacm cpc.
This commit was SVN r18388.
http://www.open-mpi.org/community/lists/devel/2008/04/3779.php
{{{
svn merge -r 18276:18380 https://svn.open-mpi.org/svn/ompi/tmp-public/jjh-mca-play .
}}}
Any components not in the trunk, but in one of the effected frameworks *must* be
updated. Contact the list, look at the RFC, or look at the diff for how to do this.
Sorry for the early commit of this, but I wanted to get it in today (per RFC) and
didn't know if I would have a chance later today.
This commit was SVN r18381.
If there are multiple QP's, RDMACM will not send a message if the
qpnum != 0. In doing so, it will log an error unecessarily. This
removes that.
This commit was SVN r18363.
Add the logic to support using port numbers, instead of simply using
the IP address of the sending node to determine which endpoint to
connect. Since each process calls the cpc query function, it will
generate its own port to listen on thus enablign this to work.
This commit was SVN r18362.
Update the rsh tree spawn capability so we spawn the next wave of daemons before launching our own local procs.
Add an ability to encode nodenames for large clusters with contiguous node name numbering schemes - this allows communication of all node names in a few bytes instead of tens-of-bytes/node.
This commit was SVN r18338.
The endpoint may be appended to list during XOOB connection bring up.
This commit was SVN r18328.
The following SVN revision numbers were found above:
r17940 --> open-mpi/ompi@ebfdd133f5
expected one. It doesn't necessarily means the message is duplicated,
it can simply signify the message is out of sequence and the counter
overflowed.
This commit was SVN r18323.
The following SVN revision numbers were found above:
r18274 --> open-mpi/ompi@73c9de3af9
The problem was caused by a bad ordering between the restart of the ORTE level tcp connections (in the OOB - out-of-band communication) and the Open MPI level tcp connections (BTLs). Before this commit ORTE would shutdown and restart the OOB completely before the OMPI level restarted its tcp connections. What would happen is that a socket descriptor used by the OMPI level on checkpoint was assigned to the ORTE level on restart. But the OMPI level had no knowledge that the socket descriptor it was previously using has been recycled so it closed it on restart. This caused the ORTE level to break as the newly created socket descriptor was closed without its knowledge.
The fix is to have the OMPI level shutdown tcp connections, allow the ORTE level to restart, and then allow the OMPi level to restart its connections. This seems obvious, and I'm surprised that this bug has not cropped up sooner. I'm confident that this specific problem has been fixed with this commit.
Thanks to Eric Roman and Tamer El Sayed for their help in identifying this problem, and patience while I was fixing it.
* Add a new state {{{OPAL_CRS_RESTART_PRE}}}. This state identifies when we are on the down slope of the INC (finalize-like) which is useful when you want to close, but not reopen a component set for fear of interfering with a lower level.
* Use this new state in OMPI level coordination. Here we want to make sure to play well with both the OMPI/BTL/TCP and ORTE/OOB/TCP components.
* Update ft_event functions in PML and BML to handle the new restart state.
* Add an additional flag to the error output in OOB/TCP so we can see what the socket descriptor was on failure as this can be helpful in debugging.
This commit was SVN r18276.
{{{
svn merge -r 18218:18240 https://svn.open-mpi.org/svn/ompi/tmp/jjh-scratch .
}}}
Contains:
* Primarily a fix for a user reported problem where a cached file descriptor is causing a SIGPIPE on restart.
* Cleanup some small memory leaks from using mca_base_param_env_var() - Thanks Jeff
* Cleanup ORTE FT tool compilation in non-FT builds - Thanks Tim P.
* Cleanup mpi interface with missplaced {{{OPAL_CR_ENTER_LIBRARY}}} - Thanks Terry
* Some other sundry cleanup items all dealing with C/R functionality in the trunk.
This commit was SVN r18241.
Fix the ompi-server -h cmd line option so it actually tells you something!
Add two new testing codes to the orte/test/mpi area: accept and connect.
This commit was SVN r18176.
Rational (taken from the code):
/* This is PITA. We never know which source address an
* incoming/outgoing packet will have, so even with
* btl_tcp_if_include/exclude on the remote end, we
* might get a different source address.
*
* If this address isn't included in btl_proc->proc_addrs,
* we would erroneously drop the connection
*/
merge -r18165:18167 to the trunk.
This commit was SVN r18169.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r18165
r18167
mca_btl_openib_endpoint_post_rr_nolock is freeing the endpoint lock on
the error case, but most/all of the functions calling this free the lock
regardless of its error case. Thus resulting is a double free of the
lock.
This commit was SVN r18131.
wring directory. The UH copyrights do belong into this file (i.e. because of
the fix which is in the 1.2 branch, the UH copyright notes are in the header
there alreary), but I want to have the proper log for that.
This commit was SVN r18124.
inter-communicator scatter, since the root (root==MPI_ROOT) might very well
have recvcount=0. The same fix has been applied to gather.c just the other way
round.
Fixes the bug reported on the mainling list by Martin Audet. If there is a
1.2.7 this fix might be worthwhile porting it over.
Please note, that while the test works now for basic and for inter, we get a
0byte malloc warning from the inter module, which we still have to fix in a
separate patch.
This commit was SVN r18122.
