1
1
openmpi/orte/mca/odls/odls.h

127 строки
4.1 KiB
C
Исходник Обычный вид История

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2008 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2011-2015 Los Alamos National Security, LLC. All rights
* reserved.
Refactor the ORTE DVM code so that external codes can submit multiple jobs using only a single connection to the HNP. * Clean up the DVM so it continues to run even when applications error out and we would ordinarily abort the daemons. * Create a new errmgr component for the DVM to handle the differences. * Cleanup the DVM state component. * Add ORTE bindings directory and brief README * Pass a local tool index around to match jobs. * Pass the jobid on job completion. * Fix initialization logic. * Add framework for python wrapper. * Fix terminate-with-non-zero-exit behavior so it properly terminates only the indicated procs, notifies orte-submit, and orte-dvm continues executing. * Add some missing options to orte-dvm * Fix a bug in -host processing that caused us to ignore the #slots designator. Add a new attribute to indicate "do not expand the DVM" when submitting job spawn requests. * It actually makes no sense that we treat the termination of all children differently than terminating the children of a specific job - it only creates confusion over the difference in behavior. So terminate children the same way regardless. Extend the cmd_line utility to easily allow layering of command line definitions Catch up with ORTE interface change and make build more generic. Disable "fixed dvm" logic for now. Add another cmd_line function to merge a table of cmd line options with another one, reporting as errors any duplicate entries. Use this to allow orterun to reuse the orted_submit code Fix the "fixed_dvm" logic by ensuring we reset num_new_daemons to zero. Also ensure that the nidmap is sent with the first job so the downstream daemons get the node info. Remove a duplicate cmd line entry in orterun. Revise the DVM startup procedure to pass the nidmap only once, at the startup of the DVM. This reduces the overhead on each job launch and ensures that the nidmap doesn't get overwritten. Add new commands to get_orted_comm_cmd_str(). Move ORTE command line options to orte_globals.[ch]. Catch up with extra orte_submit_init parameter. Add example code. Add documentation. Bump version. The nidmap and routing data must be updated prior to propagating the xcast or else the xcast will fail. Fix the return code so it is something more expected when an error occurs. Ensure we get an error returned to us when we fail to launch for some reason. In this case, we will always get a launch_cb as we did indeed attempt to spawn it. The error code will be returned in the complete_cb. Fix the return code from orte_submit_job - it was returning the tracker index instead of "success". Take advantage of ORTE's pretty-print capabilities to provide a nice error output explaining why we failed to launch. Ensure we always get a launch_cb when we fail to launch, but no complete_cb as the job never launched. Extend the error reporting capability to job completion as well. Add index parameter to orte_submit_job(). Add orte_job_cancel and implement ORTE_DAEMON_TERMINATE_JOB_CMD. Factor out dvm termination. Parse the terminate option at tool level. Add error string for ORTE_ERR_JOB_CANCELLED. Add some safeguards. Cleanup and/of comments. Enable the return. Properly ORTE_DECLSPEC orte_submit_halt. Add orte_submit_halt and orte_submit_cancel to interface. Use the plm interface to terminate the job
2015-12-17 02:30:40 +03:00
* Copyright (c) 2016 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* The OpenRTE Daemon's Local Launch Subsystem
*
*/
#ifndef ORTE_MCA_ODLS_H
#define ORTE_MCA_ODLS_H
#include "orte_config.h"
#include "orte/types.h"
#include "orte/mca/mca.h"
#include "opal/class/opal_pointer_array.h"
#include "opal/dss/dss_types.h"
#include "orte/mca/rml/rml_types.h"
#include "orte/runtime/orte_globals.h"
#include "orte/mca/odls/odls_types.h"
BEGIN_C_DECLS
/*
* odls module functions
*/
2015-06-24 06:59:57 +03:00
/*
* Construct a buffer for use in adding local processes
* In order to reuse daemons, we need a way for the HNP to construct a buffer that
* contains the data needed by the active ODLS component to launch a local process. Since the
* only one that knows what a particular ODLS component needs is that component, we require an
* entry point that the HNP can call to get the required buffer. This is constructed
* for *all* nodes - the individual orteds then parse that data to find the specific launch info
* for procs on their node
*/
typedef int (*orte_odls_base_module_get_add_procs_data_fn_t)(opal_buffer_t *data,
orte_jobid_t job);
/**
* Locally launch the provided processes
*/
typedef int (*orte_odls_base_module_launch_local_processes_fn_t)(opal_buffer_t *data);
/**
* Kill the local processes on this node
*/
typedef int (*orte_odls_base_module_kill_local_processes_fn_t)(opal_pointer_array_t *procs);
/**
* Signal local processes
*/
typedef int (*orte_odls_base_module_signal_local_process_fn_t)(const orte_process_name_t *proc,
int32_t signal);
/**
* Restart a local process
*/
typedef int (*orte_odls_base_module_restart_proc_fn_t)(orte_proc_t *child);
These changes were mostly captured in a prior RFC (except for #2 below) and are aimed specifically at improving startup performance and setting up the remaining modifications described in that RFC. The commit has been tested for C/R and Cray operations, and on Odin (SLURM, rsh) and RoadRunner (TM). I tried to update all environments, but obviously could not test them. I know that Windows needs some work, and have highlighted what is know to be needed in the odls process component. This represents a lot of work by Brian, Tim P, Josh, and myself, with much advice from Jeff and others. For posterity, I have appended a copy of the email describing the work that was done: As we have repeatedly noted, the modex operation in MPI_Init is the single greatest consumer of time during startup. To-date, we have executed that operation as an ORTE stage gate that held the process until a startup message containing all required modex (and OOB contact info - see #3 below) info could be sent to it. Each process would send its data to the HNP's registry, which assembled and sent the message when all processes had reported in. In addition, ORTE had taken responsibility for monitoring process status as it progressed through a series of "stage gates". The process reported its status at each gate, and ORTE would then send a "release" message once all procs had reported in. The incoming changes revamp these procedures in three ways: 1. eliminating the ORTE stage gate system and cleanly delineating responsibility between the OMPI and ORTE layers for MPI init/finalize. The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info. The allgather is implemented using the orte_grpcomm framework since the BTL's are not active at that point. At the moment, the grpcomm framework only has a "basic" component analogous to OMPI's "basic" coll framework - I would recommend that the MPI team create additional, more advanced components to improve performance of this step. The other stage gates have been replaced by orte_grpcomm barrier functions. We tried to use MPI barriers instead (since the BTL's are active at that point), but - as we discussed on the telecon - these are not currently true barriers so the job would hang when we fell through while messages were still in process. Note that the grpcomm barrier doesn't actually resolve that problem, but Brian has pointed out that we are unlikely to ever see it violated. Again, you might want to spend a little time on an advanced barrier algorithm as the one in "basic" is very simplistic. Summarizing this change: ORTE no longer tracks process state nor has direct responsibility for synchronizing jobs. This is now done via collective operations within the MPI layer, albeit using ORTE collective communication services. I -strongly- urge the MPI team to implement advanced collective algorithms to improve the performance of this critical procedure. 2. reducing the volume of data exchanged during modex. Data in the modex consisted of the process name, the name of the node where that process is located (expressed as a string), plus a string representation of all contact info. The nodename was required in order for the modex to determine if the process was local or not - in addition, some people like to have it to print pretty error messages when a connection failed. The size of this data has been reduced in three ways: (a) reducing the size of the process name itself. The process name consisted of two 32-bit fields for the jobid and vpid. This is far larger than any current system, or system likely to exist in the near future, can support. Accordingly, the default size of these fields has been reduced to 16-bits, which means you can have 32k procs in each of 32k jobs. Since the daemons must have a vpid, and we require one daemon/node, this also restricts the default configuration to 32k nodes. To support any future "mega-clusters", a configuration option --enable-jumbo-apps has been added. This option increases the jobid and vpid field sizes to 32-bits. Someday, if necessary, someone can add yet another option to increase them to 64-bits, I suppose. (b) replacing the string nodename with an integer nodeid. Since we have one daemon/node, the nodeid corresponds to the local daemon's vpid. This replaces an often lengthy string with only 2 (or at most 4) bytes, a substantial reduction. (c) when the mca param requesting that nodenames be sent to support pretty error messages, a second mca param is now used to request FQDN - otherwise, the domain name is stripped (by default) from the message to save space. If someone wants to combine those into a single param somehow (perhaps with an argument?), they are welcome to do so - I didn't want to alter what people are already using. While these may seem like small savings, they actually amount to a significant impact when aggregated across the entire modex operation. Since every proc must receive the modex data regardless of the collective used to send it, just reducing the size of the process name removes nearly 400MBytes of communication from a 32k proc job (admittedly, much of this comm may occur in parallel). So it does add up pretty quickly. 3. routing RML messages to reduce connections. The default messaging system remains point-to-point - i.e., each proc opens a socket to every proc it communicates with and sends its messages directly. A new option uses the orteds as routers - i.e., each proc only opens a single socket to its local orted. All messages are sent from the proc to the orted, which forwards the message to the orted on the node where the intended recipient proc is located - that orted then forwards the message to its local proc (the recipient). This greatly reduces the connection storm we have encountered during startup. It also has the benefit of removing the sharing of every proc's OOB contact with every other proc. The orted routing tables are populated during launch since every orted gets a map of where every proc is being placed. Each proc, therefore, only needs to know the contact info for its local daemon, which is passed in via the environment when the proc is fork/exec'd by the daemon. This alone removes ~50 bytes/process of communication that was in the current STG1 startup message - so for our 32k proc job, this saves us roughly 32k*50 = 1.6MBytes sent to 32k procs = 51GBytes of messaging. Note that you can use the new routing method by specifying -mca routed tree - if you so desire. This mode will become the default at some point in the future. There are a few minor additional changes in the commit that I'll just note in passing: * propagation of command line mca params to the orteds - fixes ticket #1073. See note there for details. * requiring of "finalize" prior to "exit" for MPI procs - fixes ticket #1144. See note there for details. * cleanup of some stale header files This commit was SVN r16364.
2007-10-05 23:48:23 +04:00
/**
* pls module version
*/
struct orte_odls_base_module_1_3_0_t {
orte_odls_base_module_get_add_procs_data_fn_t get_add_procs_data;
orte_odls_base_module_launch_local_processes_fn_t launch_local_procs;
orte_odls_base_module_kill_local_processes_fn_t kill_local_procs;
orte_odls_base_module_signal_local_process_fn_t signal_local_procs;
orte_odls_base_module_restart_proc_fn_t restart_proc;
};
/** shorten orte_odls_base_module_1_3_0_t declaration */
typedef struct orte_odls_base_module_1_3_0_t orte_odls_base_module_1_3_0_t;
/** shorten orte_odls_base_module_t declaration */
typedef struct orte_odls_base_module_1_3_0_t orte_odls_base_module_t;
/**
* odls component
*/
struct orte_odls_base_component_2_0_0_t {
/** component version */
mca_base_component_t version;
/** component data */
mca_base_component_data_t base_data;
};
/** Convenience typedef */
typedef struct orte_odls_base_component_2_0_0_t orte_odls_base_component_2_0_0_t;
/** Convenience typedef */
typedef orte_odls_base_component_2_0_0_t orte_odls_base_component_t;
/**
* Macro for use in modules that are of type odls
*/
#define ORTE_ODLS_BASE_VERSION_2_0_0 \
ORTE_MCA_BASE_VERSION_2_1_0("odls", 2, 0, 0)
/* Global structure for accessing ODLS functions
*/
ORTE_DECLSPEC extern orte_odls_base_module_t orte_odls; /* holds selected module's function pointers */
END_C_DECLS
#endif /* MCA_ODLS_H */