ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #921 from rhc54/topic/pmix

Просмотр исходного кода 
Sync to PMIx master
Этот коммит содержится в:
rhc54 2015-09-22 05:52:47 -07:00
родитель 6b2307c88a 4c654ffd94
Коммит 2b45969d16
10 изменённых файлов: 277 добавлений и 29 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

3
opal/mca/pmix/pmix1xx/pmix/Makefile.am
Просмотреть файл

@ -37,7 +37,8 @@ endif
man_MANS = \
man/man3/pmix_init.3 \
man/man7/pmix.7
man/man7/pmix.7 \
man/man7/pmix_constants.7
include config/Makefile.am
include include/Makefile.am

4
opal/mca/pmix/pmix1xx/pmix/VERSION
Просмотреть файл

@ -30,7 +30,7 @@ greek=a1
# command, or with the date (if "git describe" fails) in the form of
# "date<date>".
repo_rev=git89680d6
repo_rev=git4dec948
# If tarball_version is not empty, it is used as the version string in
# the tarball filename, regardless of all other versions listed in
@ -44,7 +44,7 @@ tarball_version=
# The date when this release was created
date="Sep 10, 2015"
date="Sep 21, 2015"
# The shared library version of each of PMIx's public libraries.
# These versions are maintained in accordance with the "Library

11
opal/mca/pmix/pmix1xx/pmix/include/pmix/pmix_common.h.in
Просмотреть файл

@ -80,7 +80,6 @@ BEGIN_C_DECLS
/* define maximum value and key sizes */
#define PMIX_MAX_NSLEN 255
#define PMIX_MAX_VALLEN 1023
#define PMIX_MAX_KEYLEN 511
/* define a *wildcard* value for requests involving rank */
@ -175,6 +174,7 @@ BEGIN_C_DECLS
#define PMIX_PROC_DATA "pmix.pdata" // (pmix_value_array_t) starts with rank, then contains more data
#define PMIX_NODE_MAP "pmix.nmap" // (char*) regex of nodes containing procs for this job
#define PMIX_PROC_MAP "pmix.pmap" // (char*) regex describing procs on each node within this job
#define PMIX_ANL_MAP "pmix.anlmap" // (char*) process mapping in ANL notation (used in PMI-1/PMI-2)
/* attributes used internally to communicate data from the server to the client */
#define PMIX_PROC_BLOB "pmix.pblob" // (pmix_byte_object_t) packed blob of process data
@ -307,6 +307,7 @@ typedef enum {
/* define a "persistence" policy for data published by clients */
typedef enum {
PMIX_PERSIST_INDEF = 0, // retain until specifically deleted
PMIX_PERSIST_FIRST_READ, // delete upon first access
PMIX_PERSIST_PROC, // retain until publishing process terminates
PMIX_PERSIST_APP, // retain until application terminates
PMIX_PERSIST_SESSION // retain until session/allocation terminates
@ -468,13 +469,11 @@ typedef struct {
} \
} while(0);
#define PMIX_INFO_LOAD(m, d, k, v, t) \
do { \
#define PMIX_INFO_LOAD(m, k, v, t) \
if (NULL != (m)) { \
(m)->mandatory = (d); \
(void)strncpy((m)->key, (k), PMIX_MAX_KEYLEN); \
pmix_value_load(&((m)->value), (v), (t)); \
} while(0);
}
/**** PMIX LOOKUP RETURN STRUCT ****/
@ -827,7 +826,7 @@ const char* PMIx_Get_version(void);
PMIX_VAL_SET_ ## _field(_v, _field, _val)
#define PMIX_VAL_cmp_val(_val1, _val2) ((_val1) != (_val2))
#define PMIX_VAL_cmp_float(_val1, _val2) (abs((_val1) - (_val2)) > 0.000001)
#define PMIX_VAL_cmp_float(_val1, _val2) (((_val1)>(_val2))?(((_val1)-(_val2))>0.000001):(((_val2)-(_val1))>0.000001))
#define PMIX_VAL_cmp_ptr(_val1, _val2) strncmp(_val1, _val2, strlen(_val1)+1)
#define PMIX_VAL_CMP_int PMIX_VAL_cmp_val

94
opal/mca/pmix/pmix1xx/pmix/man/man7/pmix_constants.7 Обычный файл
Просмотреть файл

@ -0,0 +1,94 @@
.TH "pmix_constants" "7" "2015\-09\-16" "PMIx Programmer\[aq]s Manual" "\@VERSION\@"
.SH NAME
.PP
PMIx Constants
.SH SYNOPSIS
.IP
.nf
\f[C]
#include\ <pmix_common.h>
\f[]
.fi
.SH OVERVIEW
.PP
PMIx relies on the following types of constants:
.PP
\f[I]Maximum Sizes\f[] : In order to minimize malloc performance
penalties, PMIx utilizes constant\-sized arrays wherever possible.
These constants provide the user with the maximum size of the various
array types.
.PP
\f[I]Attributes\f[] : .
.PP
\f[I]Errors\f[] : PMIx uses negative error constants, with 0 indicating
"success".
.SH MAXIMUM SIZES
.PP
The .
.PP
\f[I]PMIX_MAX_NSLEN\f[] : The maximum length of a namespace.
Note that any declaration of an array to hold a key string must include
one extra space for the terminating \f[I]NULL\f[].
.PP
\f[I]PMIX_MAX_KEYLEN\f[] : Maximum length of the key string used in
structures such as the \f[I]pmix\f[]info_t_.
Note that any declaration of an array to hold a key string must include
one extra space for the terminating \f[I]NULL\f[].
.SH ATTRIBUTES
.PP
Define a set of "standard" PMIx attributes that can be queried using the
PMIx_Get function.
Implementations (and users) are free to extend as desired \- thus,
functions calling PMIx_Get must be capable of handling the "not found"
condition.
Note that these are attributes of the system and the job as opposed to
values the application (or underlying programming library) might choose
to expose \- i.e., they are values provided by the resource manager as
opposed to the application.
Thus, these keys are RESERVED for use by PMIx, and users should avoid
defining any attribute starting with the keyword \f[I]PMIX\f[].
.PP
A list of the current PMIx attributes, and the type of their associated
data value, is provided here.
.PP
\f[I]PMIX_ATTR_UNDEF (NULL)\f[] : Used to initialize an attribute field,
indicating that the attribute has not yet been assigned.
.PP
\f[I]PMIX_USERID (uint32_t)\f[] : .
.PP
\f[I]PMIX_GRPID (uint32_t)\f[] : An access domain represents a single
logical connection into a fabric.
It may map to a single physical or virtual NIC or a port.
An access domain defines the boundary across which fabric resources may
be associated.
Each access domain belongs to a single fabric domain.
.PP
\f[I]PMIX_CPUSET (char*)\f[] : .
.SH ERROR CONSTANTS
.PP
\&.
.PP
\f[I]PMIX_SUCCESS\f[] : Indicates that the operation was successful.
.PP
\f[I]PMIX_ERROR\f[] : A general error code \- an error occurred, but no
specific reason can be provided.
.PP
\f[I]fi_rma \- Remote Memory Access\f[] : RMA transfers are one\-sided
operations that read or write data directly to a remote memory region.
Other than defining the appropriate memory region, RMA operations do not
require interaction at the target side for the data transfer to
complete.
.PP
\f[I]fi_atomic \- Atomic\f[] : Atomic operations can perform one of
several operations on a remote memory region.
Atomic operations include well\-known functionality, such as atomic\-add
and compare\-and\-swap, plus several other pre\-defined calls.
Unlike other data transfer interfaces, atomic operations are aware of
the data formatting at the target memory region.
.SH SEE ALSO
.PP
\f[C]pmix\f[](7), \f[C]fi_getinfo\f[](3), \f[C]fi_endpoint\f[](3),
\f[C]fi_domain\f[](3), \f[C]fi_av\f[](3), \f[C]fi_eq\f[](3),
\f[C]fi_cq\f[](3), \f[C]fi_cntr\f[](3), \f[C]fi_mr\f[](3)
.SH AUTHORS
PMIx.

117
opal/mca/pmix/pmix1xx/pmix/src/client/pmi1.c
Просмотреть файл

@ -30,6 +30,8 @@
#endif
#include PMIX_EVENT_HEADER
#define ANL_MAPPING "PMI_process_mapping"
#include "src/buffer_ops/buffer_ops.h"
#include "src/util/argv.h"
#include "src/util/error.h"
@ -39,6 +41,7 @@
static pmix_status_t convert_int(int *value, pmix_value_t *kv);
static int convert_err(pmix_status_t rc);
static pmix_proc_t myproc;
static bool data_commited = false;
int PMI_Init( int *spawned )
{
@ -93,6 +96,10 @@ int PMI_KVS_Put(const char kvsname[], const char key[], const char value[])
pmix_status_t rc;
pmix_value_t val;
pmix_output_verbose(2, pmix_globals.debug_output,
"PMI_KVS_Put: KVS=%s, key=%s value=%s",
kvsname, key, value);
val.type = PMIX_STRING;
val.data.string = (char*)value;
rc = PMIx_Put(PMIX_GLOBAL, key, &val);
@ -104,15 +111,106 @@ int PMI_KVS_Commit(const char kvsname[])
{
pmix_status_t rc;
pmix_output_verbose(2, pmix_globals.debug_output,
"PMI_KVS_Commit: KVS=%s", kvsname);
rc = PMIx_Commit();
/* PMIx permits only one data commit! */
data_commited = true;
return convert_err(rc);
}
int PMI_KVS_Get( const char kvsname[], const char key[], char value[], int length)
{
pmix_value_t *val;
uint32_t i;
static pmix_proc_t proc;
uint32_t procnum;
proc = myproc;
int rc;
pmix_output_verbose(2, pmix_globals.debug_output,
"PMI_KVS_Get: KVS=%s, key=%s value=%s",
kvsname, key, value);
/* PMI-1 expects resource manager to set
* process mapping in ANL notation. */
if( !strcmp(key, ANL_MAPPING) ) {
/* we are looking in the job-data. If there is nothing there
* we don't want to look in rank's data, thus set rank to widcard */
proc.rank = PMIX_RANK_WILDCARD;
if( PMIX_SUCCESS == PMIx_Get(&proc, PMIX_ANL_MAP, NULL, 0, &val) &&
(NULL != val) && (PMIX_STRING == val->type) ){
strncpy(value,val->data.string,length);
PMIX_VALUE_FREE(val,1);
return PMI_SUCCESS;
} else {
/* artpol:
* Some RM's (i.e. SLURM) already have ANL precomputed. The export it
* through PMIX_ANL_MAP variable.
* If we haven't found it we want to have our own packing functionality
* since it's common.
* Somebody else has to write it since I've already done that for
* GPL'ed SLURM :) */
return PMI_FAIL;
}
}
/* We don't know what process keeps this data. So it looks like we need to
* check each process.
* TODO: Is there any beter way?
* WARNING: this may lead to the VERY long HANG's if we ask for the unknown key
* before we've done Commit on all nodes. We need a workaround for that.
*
* SOLUTION: perhaps rovide "OK if nothing" info flag to tell PMIx that
* the key supposed to already be there and if nothing there - gave up with
* an error and don't try to use direct modex.
*/
if (PMIX_SUCCESS != (rc = PMIx_Get(&myproc,PMIX_JOB_SIZE, NULL, 0,&val))) {
pmix_output_verbose(2, pmix_globals.debug_output,
"pmi1: executing put for KVS %s, key %s value %s",
kvsname, key, value);
return convert_err(rc);
}
procnum = val->data.uint32;
PMIX_VALUE_FREE(val,1);
for( i=0; i < procnum; i++){
proc.rank = i;
if( PMIX_SUCCESS == PMIx_Get(&proc, key, NULL, 0, &val) &&
(NULL != val) && (PMIX_STRING == val->type) ){
strncpy(value,val->data.string,length);
PMIX_VALUE_FREE(val,1);
return PMI_SUCCESS;
}
PMIX_VALUE_FREE(val,1);
}
return PMI_FAIL;
}
/* Barrier only applies to our own nspace, and we want all
* data to be collected upon completion */
int PMI_Barrier(void)
{
return PMIx_Fence(NULL, 0, NULL, 0);
pmix_info_t buf;
int rc, ninfo = 0;
pmix_info_t *info = NULL;
if( data_commited ){
bool val = 1;
info = &buf;
PMIX_INFO_CONSTRUCT(info);
PMIX_INFO_LOAD(info, PMIX_COLLECT_DATA, &val, PMIX_BOOL );
ninfo = 1;
}
rc = PMIx_Fence(NULL, 0, info, ninfo);
if( NULL != info ){
PMIX_INFO_DESTRUCT(info);
}
return rc;
}
int PMI_Get_size(int *size)
@ -229,8 +327,12 @@ int PMI_Lookup_name(const char service_name[], char port[])
return convert_err(PMIX_ERR_NOT_FOUND);
}
/* return the port */
(void)strncpy(port, pdata.value.data.string, PMIX_MAX_VALLEN);
/* return the port - sadly, this API doesn't tell us
* the size of the port array, and so there is a
* potential we could overrun it. As this feature
* isn't widely supported in PMI-1, try being
* conservative */
(void)strncpy(port, pdata.value.data.string, PMIX_MAX_KEYLEN);
PMIX_PDATA_DESTRUCT(&pdata);
return PMIX_SUCCESS;
@ -260,7 +362,7 @@ int PMI_Get_id_length_max(int *length)
if (NULL == length) {
return PMI_ERR_INVALID_VAL_LENGTH;
}
*length = PMIX_MAX_VALLEN;
*length = PMIX_MAX_KEYLEN;
return PMI_SUCCESS;
}
@ -327,7 +429,9 @@ int PMI_KVS_Get_value_length_max(int *length)
if (NULL == length) {
return PMI_ERR_INVALID_VAL_LENGTH;
}
*length = PMIX_MAX_VALLEN;
/* don't give them an enormous size of some implementations
* immediately malloc a data block for their use */
*length = 4096;
return PMI_SUCCESS;
}
@ -484,6 +588,9 @@ static pmix_status_t convert_int(int *value, pmix_value_t *kv)
case PMIX_SIZE:
*value = kv->data.size;
break;
case PMIX_BOOL:
*value = kv->data.flag;
break;
default:
/* not an integer type */
return PMIX_ERR_BAD_PARAM;

2
opal/mca/pmix/pmix1xx/pmix/src/client/pmi2.c
Просмотреть файл

@ -304,7 +304,7 @@ int PMI2_Nameserv_lookup(const char service_name[], const PMI_keyval_t *info_ptr
}
/* return the port */
(void)strncpy(port, pdata[0].value.data.string, PMIX_MAX_VALLEN);
(void)strncpy(port, pdata[0].value.data.string, portLen);
PMIX_PDATA_DESTRUCT(&pdata[0]);
if (NULL != info_ptr) {

61
opal/mca/pmix/pmix1xx/pmix/src/client/pmix_client.c
Просмотреть файл

@ -266,13 +266,12 @@ int PMIx_Init(pmix_proc_t *proc)
pmix_output_verbose(2, pmix_globals.debug_output,
"pmix: init called");
pmix_bfrop_open();
pmix_usock_init(pmix_client_notify_recv);
pmix_sec_init();
/* we require the nspace */
if (NULL == (evar = getenv("PMIX_NAMESPACE"))) {
/* let the caller know that the server isn't available yet */
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return PMIX_ERR_INVALID_NAMESPACE;
}
(void)strncpy(proc->nspace, evar, PMIX_MAX_NSLEN);
@ -285,11 +284,17 @@ int PMIx_Init(pmix_proc_t *proc)
* then we need to return an error */
if (NULL == (evar = getenv("PMIX_SERVER_URI"))) {
/* let the caller know that the server isn't available */
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return PMIX_ERR_SERVER_NOT_AVAIL;
}
uri = pmix_argv_split(evar, ':');
if (3 != pmix_argv_count(uri)) {
pmix_argv_free(uri);
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return PMIX_ERROR;
}
@ -308,6 +313,9 @@ int PMIx_Init(pmix_proc_t *proc)
/* if the rendezvous file doesn't exist, that's an error */
if (0 != access(uri[2], R_OK)) {
pmix_argv_free(uri);
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return PMIX_ERR_NOT_FOUND;
}
pmix_argv_free(uri);
@ -315,14 +323,28 @@ int PMIx_Init(pmix_proc_t *proc)
/* we also require our rank */
if (NULL == (evar = getenv("PMIX_RANK"))) {
/* let the caller know that the server isn't available yet */
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return PMIX_ERR_DATA_VALUE_NOT_FOUND;
}
pmix_globals.myid.rank = strtol(evar, NULL, 10);
proc->rank = pmix_globals.myid.rank;
pmix_globals.pindex = -1;
/* setup the support */
pmix_bfrop_open();
pmix_usock_init(pmix_client_notify_recv);
pmix_sec_init();
/* create an event base and progress thread for us */
if (NULL == (pmix_globals.evbase = pmix_start_progress_thread())) {
pmix_sec_finalize();
pmix_usock_finalize();
pmix_bfrop_close();
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return -1;
}
@ -332,6 +354,13 @@ int PMIx_Init(pmix_proc_t *proc)
/* connect to the server - returns job info if successful */
if (PMIX_SUCCESS != (rc = connect_to_server(&address, &cb))){
PMIX_DESTRUCT(&cb);
pmix_stop_progress_thread(pmix_globals.evbase);
pmix_sec_finalize();
pmix_usock_finalize();
pmix_bfrop_close();
pmix_output_close(pmix_globals.debug_output);
pmix_output_finalize();
pmix_class_finalize();
return rc;
}
PMIX_WAIT_FOR_COMPLETION(cb.active);
@ -399,10 +428,6 @@ pmix_status_t PMIx_Finalize(void)
}
pmix_stop_progress_thread(pmix_globals.evbase);
event_base_free(pmix_globals.evbase);
#ifdef HAVE_LIBEVENT_GLOBAL_SHUTDOWN
libevent_global_shutdown();
#endif
pmix_usock_finalize();
PMIX_DESTRUCT(&pmix_client_globals.myserver);
@ -411,6 +436,11 @@ pmix_status_t PMIx_Finalize(void)
if (0 <= pmix_client_globals.myserver.sd) {
CLOSE_THE_SOCKET(pmix_client_globals.myserver.sd);
}
event_base_free(pmix_globals.evbase);
#ifdef HAVE_LIBEVENT_GLOBAL_SHUTDOWN
libevent_global_shutdown();
#endif
pmix_bfrop_close();
pmix_sec_finalize();
@ -874,6 +904,9 @@ void pmix_client_process_nspace_blob(const char *nspace, pmix_buffer_t *bptr)
PMIX_CONSTRUCT(&buf2, pmix_buffer_t);
PMIX_LOAD_BUFFER(&buf2, bo->bytes, bo->size);
/* protect the data */
kptr->value->data.bo.bytes = NULL;
kptr->value->data.bo.size = 0;
PMIX_RELEASE(kptr);
bo->bytes = NULL;
bo->size = 0;
/* start by unpacking the rank */
@ -904,7 +937,7 @@ void pmix_client_process_nspace_blob(const char *nspace, pmix_buffer_t *bptr)
kp2 = PMIX_NEW(pmix_kval_t);
}
/* cleanup */
PMIX_DESTRUCT(&buf2);
PMIX_DESTRUCT(&buf2); // releases the original kptr data
PMIX_RELEASE(kp2);
} else if (0 == strcmp(kptr->key, PMIX_MAP_BLOB)) {
/* transfer the byte object for unpacking */
@ -912,6 +945,9 @@ void pmix_client_process_nspace_blob(const char *nspace, pmix_buffer_t *bptr)
PMIX_CONSTRUCT(&buf2, pmix_buffer_t);
PMIX_LOAD_BUFFER(&buf2, bo->bytes, bo->size);
/* protect the data */
kptr->value->data.bo.bytes = NULL;
kptr->value->data.bo.size = 0;
PMIX_RELEASE(kptr);
bo->bytes = NULL;
bo->size = 0;
/* start by unpacking the number of nodes */
@ -974,17 +1010,24 @@ void pmix_client_process_nspace_blob(const char *nspace, pmix_buffer_t *bptr)
pmix_argv_free(procs);
PMIX_DESTRUCT(&kv);
}
/* cleanup */
PMIX_DESTRUCT(&buf2); // releases the original kptr data
} else {
/* this is job-level data, so just add it to that hash_table
* with the wildcard rank */
if (PMIX_SUCCESS != (rc = pmix_hash_store(&nsptr->internal, PMIX_RANK_WILDCARD, kptr))) {
PMIX_ERROR_LOG(rc);
}
/* maintain accounting - but note that the kptr remains
* alive and stored in the hash table! So we cannot reuse
* it for some other purpose */
PMIX_RELEASE(kptr);
}
kptr = PMIX_NEW(pmix_kval_t);
cnt = 1;
}
/* need to release the leftover kptr */
PMIX_RELEASE(kptr);
}
static int usock_connect(struct sockaddr *addr)

4
opal/mca/pmix/pmix1xx/pmix/src/server/pmix_server.c
Просмотреть файл

@ -753,7 +753,7 @@ pmix_status_t PMIx_server_register_client(const pmix_proc_t *proc,
/* setup the envars for a child process */
pmix_status_t PMIx_server_setup_fork(const pmix_proc_t *proc, char ***env)
{
char rankstr[PMIX_MAX_VALLEN+1];
char rankstr[128];
pmix_output_verbose(2, pmix_globals.debug_output,
"pmix:server setup_fork for nspace %s rank %d",
@ -762,7 +762,7 @@ pmix_status_t PMIx_server_setup_fork(const pmix_proc_t *proc, char ***env)
/* pass the nspace */
pmix_setenv("PMIX_NAMESPACE", proc->nspace, true, env);
/* pass the rank */
(void)snprintf(rankstr, PMIX_MAX_VALLEN, "%d", proc->rank);
(void)snprintf(rankstr, 127, "%d", proc->rank);
pmix_setenv("PMIX_RANK", rankstr, true, env);
/* pass our rendezvous info */
pmix_setenv("PMIX_SERVER_URI", myuri, true, env);

3
opal/mca/pmix/pmix1xx/pmix/src/server/pmix_server_listener.c
Просмотреть файл

@ -511,6 +511,9 @@ static void connection_handler(int sd, short flags, void* cbdata)
"connection_handler: new connection: %d",
pnd->sd);
/* ensure the socket is in blocking mode */
pmix_usock_set_blocking(pnd->sd);
/* receive identifier info from the client and authenticate it - the
* function will lookup and return the peer object if the connection
* is successfully authenticated */

7
opal/mca/pmix/pmix1xx/pmix/src/usock/usock.c
Просмотреть файл

@ -135,9 +135,10 @@ pmix_status_t pmix_usock_send_blocking(int sd, char *ptr, size_t size)
continue;
}
if (pmix_socket_errno != EINTR) {
pmix_output(8, "usock_peer_send_blocking: send() to socket %d failed: %s (%d)\n",
sd, strerror(pmix_socket_errno),
pmix_socket_errno);
pmix_output_verbose(8, pmix_globals.debug_output,
"usock_peer_send_blocking: send() to socket %d failed: %s (%d)\n",
sd, strerror(pmix_socket_errno),
pmix_socket_errno);
return PMIX_ERR_UNREACH;
}
continue;