ssg.c 39.7 KB
Newer Older
1 2 3 4 5 6
/*
 * Copyright (c) 2016 UChicago Argonne, LLC
 *
 * See COPYRIGHT in top-level directory.
 */

7
#include "ssg-config.h"
Shane Snyder's avatar
Shane Snyder committed
8

Jonathan Jenkins's avatar
Jonathan Jenkins committed
9 10 11 12
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
13
#include <errno.h>
Jonathan Jenkins's avatar
Jonathan Jenkins committed
14 15
#include <stdlib.h>
#include <string.h>
16
#include <time.h>
Jonathan Jenkins's avatar
Jonathan Jenkins committed
17
#include <assert.h>
Shane Snyder's avatar
Shane Snyder committed
18
#ifdef SSG_HAVE_MPI
19 20
#include <mpi.h>
#endif
Jonathan Jenkins's avatar
Jonathan Jenkins committed
21

Shane Snyder's avatar
Shane Snyder committed
22
#include <mercury.h>
23
#include <abt.h>
Shane Snyder's avatar
Shane Snyder committed
24
#include <margo.h>
25

26
#include "ssg.h"
Shane Snyder's avatar
Shane Snyder committed
27 28 29
#ifdef SSG_HAVE_MPI
#include "ssg-mpi.h"
#endif
30
#include "ssg-internal.h"
31

32 33 34
/* arguments for group lookup ULTs */
struct ssg_group_lookup_ult_args
{
35
    ssg_group_view_t *view;
36 37 38 39 40
    ssg_member_state_t *member_state;
    hg_return_t out;
};
static void ssg_group_lookup_ult(void * arg);

41
/* SSG helper routine prototypes */
42
static ssg_group_descriptor_t * ssg_group_descriptor_create(
43
    uint64_t name_hash, const char * leader_addr_str, int owner_status);
44 45 46 47 48
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor);
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor);
static int ssg_group_view_create(
49 50 51
    const char * const group_addr_strs[], int group_size,
    hg_addr_t self_addr, ssg_member_id_t * self_id,
    ssg_group_view_t * view);
52 53
static ssg_member_id_t ssg_gen_member_id(
    const char * addr_str);
54 55 56 57 58 59 60 61
static void ssg_group_view_destroy(
    ssg_group_view_t * view);
static void ssg_group_destroy_internal(
    ssg_group_t * g);
static void ssg_attached_group_destroy(
    ssg_attached_group_t * ag);
static const char ** ssg_addr_str_buf_to_list(
    const char * buf, int num_addrs);
62

63 64
/* XXX: i think we ultimately need per-mid ssg instances rather than 1 global? */
ssg_instance_t *ssg_inst = NULL;
65

66 67 68
/***************************************************
 *** SSG runtime intialization/shutdown routines ***
 ***************************************************/
69

70 71
int ssg_init(
    margo_instance_id mid)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
72
{
73 74
    struct timespec ts;

75 76 77 78 79 80 81 82 83 84
    if (ssg_inst)
        return SSG_FAILURE;

    /* initialize an SSG instance for this margo instance */
    ssg_inst = malloc(sizeof(*ssg_inst));
    if (!ssg_inst)
        return SSG_FAILURE;
    memset(ssg_inst, 0, sizeof(*ssg_inst));
    ssg_inst->mid = mid;

85
    ssg_register_rpcs();
86

87 88 89 90
    /* seed RNG */
    clock_gettime(CLOCK_MONOTONIC, &ts);
    srand(ts.tv_nsec + getpid());

91 92
    return SSG_SUCCESS;
}
Jonathan Jenkins's avatar
Jonathan Jenkins committed
93

94
int ssg_finalize()
95
{
96 97
    ssg_group_t *g, *g_tmp;
    ssg_attached_group_t *ag, *ag_tmp;
98 99 100 101 102

    if (!ssg_inst)
        return SSG_FAILURE;

    /* destroy all active groups */
103
    HASH_ITER(hh, ssg_inst->group_table, g, g_tmp)
104 105 106 107 108
    {
        HASH_DELETE(hh, ssg_inst->group_table, g);
        ssg_group_destroy_internal(g);
    }

109 110 111 112 113 114 115
    /* detach from all attached groups */
    HASH_ITER(hh, ssg_inst->attached_group_table, ag, ag_tmp)
    {
        HASH_DELETE(hh, ssg_inst->attached_group_table, ag);
        ssg_attached_group_destroy(ag);
    }

116 117 118
    free(ssg_inst);
    ssg_inst = NULL;

119
    return SSG_SUCCESS;
120
}
Jonathan Jenkins's avatar
Jonathan Jenkins committed
121

122 123 124
/*************************************
 *** SSG group management routines ***
 *************************************/
Jonathan Jenkins's avatar
Jonathan Jenkins committed
125

126 127
static int ssg_get_swim_dping_target(
    void *group_data,
Shane Snyder's avatar
Shane Snyder committed
128 129 130 131 132
    swim_member_id_t *target_id,
    swim_member_inc_nr_t *target_inc_nr,
    hg_addr_t *target_addr);
static int ssg_get_swim_iping_targets(
    void *group_data,
133
    swim_member_id_t dping_target_id,
Shane Snyder's avatar
Shane Snyder committed
134
    int *num_targets,
Shane Snyder's avatar
Shane Snyder committed
135 136 137 138 139 140 141 142 143
    swim_member_id_t *target_ids,
    hg_addr_t *target_addrs);
static void ssg_get_swim_member_addr(
    void *group_data,
    swim_member_id_t id,
    hg_addr_t *target_addr);
static void ssg_get_swim_member_state(
    void *group_data,
    swim_member_id_t id,
144
    swim_member_state_t **state);
145 146 147
static void ssg_apply_swim_member_update(
    void *group_data,
    swim_member_update_t update);
Shane Snyder's avatar
Shane Snyder committed
148

149
static void ssg_shuffle_member_list(
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
    ssg_member_state_t **list,
    unsigned int len);

void print_nondead_list(ssg_group_t *g, char *tag)
{
    unsigned int i = 0;

    printf("***SDS %s nondead_member_list [%lu]: ", tag, g->self_id);

    for (i = 0; i < g->view.size; i++)
    {
        printf("%p\t", g->nondead_member_list[i]);
    }
    printf("\n");
}
Shane Snyder's avatar
Shane Snyder committed
165 166 167

static int ssg_get_swim_dping_target(
    void *group_data,
Shane Snyder's avatar
Shane Snyder committed
168 169 170
    swim_member_id_t *target_id,
    swim_member_inc_nr_t *target_inc_nr,
    hg_addr_t *target_addr)
171 172
{
    ssg_group_t *g = (ssg_group_t *)group_data;
173 174
    ssg_member_state_t *dping_target_ms;
    unsigned int nondead_list_len;
175 176 177

    assert(g != NULL);

178 179
    /* XXX MUTEX */

180
    nondead_list_len = g->view.size;
181
    if (nondead_list_len == 0)
Shane Snyder's avatar
Shane Snyder committed
182 183
        return -1; /* no targets */

184 185
    /* reshuffle member list after a complete traversal */
    if (g->dping_target_ndx == nondead_list_len)
Shane Snyder's avatar
Shane Snyder committed
186
    {
187 188 189
        SSG_DEBUG(g, "...SHUFFLING...\n");
        ssg_shuffle_member_list(g->nondead_member_list, g->view.size);
        print_nondead_list(g, "post-shuffle");
190
        g->dping_target_ndx = 0;
Shane Snyder's avatar
Shane Snyder committed
191
    }
Shane Snyder's avatar
Shane Snyder committed
192

193
    /* pull next dping target using saved state */
194
    dping_target_ms = g->nondead_member_list[g->dping_target_ndx];
Shane Snyder's avatar
Shane Snyder committed
195

196 197 198
    *target_id = (swim_member_id_t)dping_target_ms->id;
    *target_inc_nr = dping_target_ms->swim_state.inc_nr;
    *target_addr = dping_target_ms->addr;
Shane Snyder's avatar
Shane Snyder committed
199

200
    /* increment dping target index for next iteration */
201
    g->dping_target_ndx++;
Shane Snyder's avatar
Shane Snyder committed
202

203
    return 0;
Shane Snyder's avatar
Shane Snyder committed
204 205
}

Shane Snyder's avatar
Shane Snyder committed
206 207
static int ssg_get_swim_iping_targets(
    void *group_data,
208
    swim_member_id_t dping_target_id,
Shane Snyder's avatar
Shane Snyder committed
209
    int *num_targets,
210
    swim_member_id_t *target_ids,
Shane Snyder's avatar
Shane Snyder committed
211 212 213
    hg_addr_t *target_addrs)
{
    ssg_group_t *g = (ssg_group_t *)group_data;
214
    unsigned int nondead_list_len;
Shane Snyder's avatar
Shane Snyder committed
215
    int max_targets = *num_targets;
216 217
    int iping_target_count = 0;
    int i = 0;
218 219
    int r_start, r_ndx;
    ssg_member_state_t *tmp_ms;
Shane Snyder's avatar
Shane Snyder committed
220 221 222 223 224

    assert(g != NULL);

    *num_targets = 0;

225 226
    /* XXX MUTEX */

227 228 229 230
    nondead_list_len = g->view.size;
    if (nondead_list_len == 0)
        return -1; /* no targets */

231 232 233
    /* pick random index in the nondead list, and pull out a set of iping
     * targets starting from that index
     */
234
    r_start = rand() % nondead_list_len;
235 236
    while (iping_target_count < max_targets)
    {
237
        r_ndx = (r_start + i) % nondead_list_len;
238 239 240
        /* if we've iterated through the entire nondead list, stop */
        if ((i > 0 ) && (r_ndx == r_start)) break;

241
        tmp_ms = g->nondead_member_list[r_ndx];
242 243

        /* do not select the dping target as an iping target */
244
        if ((swim_member_id_t)tmp_ms->id == dping_target_id)
245 246 247 248 249
        {
            i++;
            continue;
        }

250 251
        target_ids[iping_target_count] = (swim_member_id_t)tmp_ms->id;
        target_addrs[iping_target_count] = tmp_ms->addr;
252 253 254 255 256 257
        iping_target_count++;
        i++;
    }

    *num_targets = iping_target_count;

Shane Snyder's avatar
Shane Snyder committed
258 259 260
    return 0;
}

Shane Snyder's avatar
Shane Snyder committed
261 262 263 264 265 266 267 268 269 270 271 272
static void ssg_get_swim_member_addr(
    void *group_data,
    swim_member_id_t id,
    hg_addr_t *addr)
{
    ssg_group_t *g = (ssg_group_t *)group_data;
    ssg_member_id_t ssg_id = (ssg_member_id_t)id;
    ssg_member_state_t *ms;

    assert(g != NULL);

    HASH_FIND(hh, g->view.member_map, &ssg_id, sizeof(ssg_member_id_t), ms);
Shane Snyder's avatar
Shane Snyder committed
273
    assert(ms != NULL);
Shane Snyder's avatar
Shane Snyder committed
274 275 276 277 278 279 280 281 282

    *addr = ms->addr;

    return;
}

static void ssg_get_swim_member_state(
    void *group_data,
    swim_member_id_t id,
283
    swim_member_state_t **state)
Shane Snyder's avatar
Shane Snyder committed
284 285 286 287 288 289 290 291
{
    ssg_group_t *g = (ssg_group_t *)group_data;
    ssg_member_id_t ssg_id = (ssg_member_id_t)id;
    ssg_member_state_t *ms;

    assert(g != NULL);

    HASH_FIND(hh, g->view.member_map, &ssg_id, sizeof(ssg_member_id_t), ms);
Shane Snyder's avatar
Shane Snyder committed
292
    assert(ms != NULL);
Shane Snyder's avatar
Shane Snyder committed
293

294
    *state = &ms->swim_state;
Shane Snyder's avatar
Shane Snyder committed
295 296 297 298

    return;
}

299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
static void ssg_apply_swim_member_update(
    void *group_data,
    swim_member_update_t update)
{
    ssg_group_t *g = (ssg_group_t *)group_data;
    ssg_member_id_t ssg_id = (ssg_member_id_t)update.id;
    ssg_member_state_t *ms;
    ssg_member_update_t ssg_update;

    assert(g != NULL);

    if (update.state.status == SWIM_MEMBER_DEAD)
    {
        HASH_FIND(hh, g->view.member_map, &ssg_id, sizeof(ssg_member_id_t), ms);
        if (ms)
        {
            /* update group, but don't completely remove state */
            margo_addr_free(ssg_inst->mid, ms->addr);
            ssg_update.id = ssg_id;
            ssg_update.type = SSG_MEMBER_REMOVE;
        }
    }
    else
    {
        assert(0); /* XXX: dynamic group joins aren't possible yet */
    }

    /* execute user-supplied membership update callback, if given */
    if (g->update_cb)
        g->update_cb(ssg_update, g->update_cb_dat);

    return;
}

333
static void ssg_shuffle_member_list(
334 335
    ssg_member_state_t **list,
    unsigned int len)
336 337
{
    unsigned int i, r;
338
    ssg_member_state_t *tmp_ms;
339

Shane Snyder's avatar
Shane Snyder committed
340 341
    if (len <= 1) return;

342 343 344 345
    /* run fisher-yates shuffle over list of nondead members */
    for (i = len - 1; i > 0; i--)
    {
        r = rand() % (i + 1);
346 347 348
        tmp_ms = list[r];
        list[r] = list[i];
        list[i] = tmp_ms;
349 350 351 352 353
    }

    return;
}

Shane Snyder's avatar
Shane Snyder committed
354
ssg_group_id_t ssg_group_create(
355 356
    const char * group_name,
    const char * const group_addr_strs[],
357 358 359
    int group_size,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
360
{
361
    uint64_t name_hash;
Shane Snyder's avatar
Shane Snyder committed
362
    hg_addr_t self_addr = HG_ADDR_NULL;
363
    ssg_group_descriptor_t *tmp_descriptor;
364
    ssg_group_t *g = NULL;
365
    ssg_member_state_t *ms, *tmp_ms;
366
    unsigned int i = 0;
367
    hg_return_t hret;
368
    int sret;
Shane Snyder's avatar
Shane Snyder committed
369
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
370

371
    if (!ssg_inst) return group_id;
372

373
    name_hash = ssg_hash64_str(group_name);
374

375
    /* generate a unique ID for this group  */
376 377
    tmp_descriptor = ssg_group_descriptor_create(name_hash, group_addr_strs[0],
        SSG_OWNER_IS_MEMBER);
378
    if (tmp_descriptor == NULL) return group_id;
379

Shane Snyder's avatar
Shane Snyder committed
380
    /* make sure we aren't re-creating an existing group */
381
    HASH_FIND(hh, ssg_inst->group_table, &tmp_descriptor->name_hash,
Shane Snyder's avatar
Shane Snyder committed
382
        sizeof(uint64_t), g);
383 384 385 386 387
    if (g)
    {
        g = NULL;
        goto fini;
    }
388

389 390 391 392 393 394 395
    /* allocate an SSG group data structure and initialize some of it */
    g = malloc(sizeof(*g));
    if (!g) goto fini;
    memset(g, 0, sizeof(*g));
    g->name = strdup(group_name);
    if (!g->name) goto fini;
    g->descriptor = tmp_descriptor;
396 397
    g->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
398

399 400 401 402
    /* get my address */
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
    if (hret != HG_SUCCESS) goto fini;

403
    /* initialize the group view */
404 405
    sret = ssg_group_view_create(group_addr_strs, group_size, self_addr,
        &g->self_id, &g->view);
406
    if (sret != SSG_SUCCESS) goto fini;
407
    if (g->self_id == SSG_MEMBER_ID_INVALID)
408
    {
409
        /* if unable to resolve my rank within the group, error out */
410 411 412 413 414
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
    }

415
    /* create a list of all nondead member states and shuffle it */
416 417 418
    g->nondead_member_list = malloc(g->view.size * sizeof(*g->nondead_member_list));
    if (g->nondead_member_list == NULL) goto fini;
    g->nondead_member_list_nslots = g->view.size;
419 420
    HASH_ITER(hh, g->view.member_map, ms, tmp_ms)
    {
421 422
        g->nondead_member_list[i] = ms;
        i++;
423
    }
424 425 426
    print_nondead_list(g, "init");
    ssg_shuffle_member_list(g->nondead_member_list, g->view.size);
    print_nondead_list(g, "init_shuffle");
Shane Snyder's avatar
Shane Snyder committed
427

428
    /* initialize swim failure detector */
429 430
    // TODO: we should probably barrier or sync somehow to avoid rpc failures
    // due to timing skew of different ranks initializing swim
431 432
    swim_group_mgmt_callbacks_t swim_callbacks = {
        .get_dping_target = &ssg_get_swim_dping_target,
Shane Snyder's avatar
Shane Snyder committed
433 434 435
        .get_iping_targets = &ssg_get_swim_iping_targets,
        .get_member_addr = ssg_get_swim_member_addr,
        .get_member_state = ssg_get_swim_member_state,
436
        .apply_member_update = ssg_apply_swim_member_update,
437
    };
438 439
    g->swim_ctx = swim_init(ssg_inst->mid, g, (swim_member_id_t)g->self_id,
        swim_callbacks, 1);
Shane Snyder's avatar
Shane Snyder committed
440
    if (g->swim_ctx == NULL) goto fini;
441

Shane Snyder's avatar
Shane Snyder committed
442 443 444
    /* everything successful -- set the output group identifier, which is just
     * an opaque pointer to the group descriptor structure
     */
445 446
    group_id = (ssg_group_id_t)ssg_group_descriptor_dup(g->descriptor);
    if (group_id == SSG_GROUP_ID_NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
447

448 449 450 451
    /* add this group reference to our group table */
    HASH_ADD(hh, ssg_inst->group_table, descriptor->name_hash,
        sizeof(uint64_t), g);

Shane Snyder's avatar
Shane Snyder committed
452
    SSG_DEBUG(g, "group create successful (size=%d)\n", group_size);
453

454
    /* don't free group pointer on success */
455 456
    g = NULL;
fini:
457
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
458 459
    if (g)
    {
460
        ssg_group_view_destroy(&g->view);
461
        free(g->name);
462 463
        free(g);
    }
464 465
    if (group_id == SSG_GROUP_ID_NULL)
        ssg_group_descriptor_free(tmp_descriptor);
466

Shane Snyder's avatar
Shane Snyder committed
467
    return group_id;
468 469
}

Shane Snyder's avatar
Shane Snyder committed
470
ssg_group_id_t ssg_group_create_config(
471
    const char * group_name,
472 473 474
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
475 476 477 478
{
    int fd;
    struct stat st;
    char *rd_buf = NULL;
479
    ssize_t rd_buf_size;
480
    char *tok;
481 482
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
483 484
    int ret;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
485
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507

    /* open config file for reading */
    fd = open(file_name, O_RDONLY);
    if (fd == -1)
    {
        fprintf(stderr, "Error: SSG unable to open config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }

    /* get file size and allocate a buffer to store it */
    ret = fstat(fd, &st);
    if (ret == -1)
    {
        fprintf(stderr, "Error: SSG unable to stat config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
    rd_buf = malloc(st.st_size+1);
    if (rd_buf == NULL) goto fini;

    /* load it all in one fell swoop */
508 509
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
510 511 512 513 514
    {
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
515
    rd_buf[rd_buf_size]='\0';
516 517 518 519 520

    /* strtok the result - each space-delimited address is assumed to be
     * a unique mercury address
     */
    tok = strtok(rd_buf, "\r\n\t ");
Jonathan Jenkins's avatar
Jonathan Jenkins committed
521 522
    if (tok == NULL) goto fini;

523
    /* build up the address buffer */
524 525
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
526 527
    do
    {
528 529 530
        int tok_size = strlen(tok);
        memcpy((char*)addr_str_buf + addr_str_buf_len, tok, tok_size+1);
        addr_str_buf_len += tok_size+1;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
531 532 533
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
534
    if (addr_str_buf_len != rd_buf_size)
535 536
    {
        /* adjust buffer size if our initial guess was wrong */
537
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
538
        if (tmp == NULL) goto fini;
539
        addr_str_buf = tmp;
540
    }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
541

542
    /* set up address string array for group members */
543
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
544
    if (!addr_strs) goto fini;
545

546
    /* invoke the generic group create routine using our list of addrs */
547 548
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
549 550

fini:
551
    /* cleanup before returning */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
552
    if (fd != -1) close(fd);
553
    free(rd_buf);
554
    free(addr_str_buf);
555 556
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
557
    return group_id;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
558 559
}

Shane Snyder's avatar
Shane Snyder committed
560
#ifdef SSG_HAVE_MPI
Shane Snyder's avatar
Shane Snyder committed
561
ssg_group_id_t ssg_group_create_mpi(
562
    const char * group_name,
563 564 565
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
566
{
567
    int i;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
568
    hg_addr_t self_addr = HG_ADDR_NULL;
569
    char *self_addr_str = NULL;
570 571
    hg_size_t self_addr_str_size = 0;
    int self_addr_str_size_int = 0; /* for mpi-friendly conversion */
572
    char *addr_str_buf = NULL;
573 574 575 576
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
577
    hg_return_t hret;
Shane Snyder's avatar
Shane Snyder committed
578
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
579

580 581
    if (!ssg_inst) goto fini;

582
    /* get my address */
583
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
584
    if (hret != HG_SUCCESS) goto fini;
585
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
Shane Snyder's avatar
Shane Snyder committed
586
    if (hret != HG_SUCCESS) goto fini;
587
    self_addr_str = malloc(self_addr_str_size);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
588
    if (self_addr_str == NULL) goto fini;
589
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
590
    if (hret != HG_SUCCESS) goto fini;
591
    self_addr_str_size_int = (int)self_addr_str_size; /* null char included in call */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
592

593
    /* gather the buffer sizes */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
594 595 596 597
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
598
    sizes[comm_rank] = self_addr_str_size_int;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
599 600
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

601 602 603
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
     */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
604 605 606
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
607
    for (i = 1; i < comm_size+1; i++)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
608 609
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];

610
    /* allgather the addresses */
611 612
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
613
    MPI_Allgatherv(self_addr_str, self_addr_str_size_int, MPI_BYTE,
614
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
615

616
    /* set up address string array for group members */
617
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
618 619 620
    if (!addr_strs) goto fini;

    /* invoke the generic group create routine using our list of addrs */
621 622
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
623 624

fini:
625
    /* cleanup before returning */
626
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
627
    free(self_addr_str);
628 629 630
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
631 632
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
633
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
634 635 636
}
#endif

637 638
int ssg_group_destroy(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
639
{
Shane Snyder's avatar
Shane Snyder committed
640
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
641
    ssg_group_t *g;
642

Shane Snyder's avatar
Shane Snyder committed
643
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
644
        return SSG_FAILURE;
645

646 647 648 649 650 651
    if (group_descriptor->owner_status != SSG_OWNER_IS_MEMBER)
    {
        fprintf(stderr, "Error: SSG unable to destroy a group it is not a member of\n");
        return SSG_FAILURE;
    }

652
    /* find the group structure and destroy it */
Shane Snyder's avatar
Shane Snyder committed
653 654
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
655 656 657 658 659
    if (!g)
    {
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
    }
660
    HASH_DELETE(hh, ssg_inst->group_table, g);
661
    ssg_group_destroy_internal(g);
662

663
    return SSG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
664 665
}

666 667 668
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
669
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
670 671 672 673 674 675
    ssg_attached_group_t *ag = NULL;
    char *group_name = NULL;
    int group_size;
    void *view_buf = NULL;
    const char **addr_strs = NULL;
    int sret = SSG_FAILURE;
676

677
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
678

679 680 681 682 683 684 685 686 687 688 689
    if (group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
    {
        fprintf(stderr, "Error: SSG unable to attach a group it is a member of\n");
        goto fini;
    }
    else if (group_descriptor->owner_status == SSG_OWNER_IS_ATTACHER)
    {
        fprintf(stderr, "Error: SSG unable to attach a group it is" \
            " already attached to\n");
        goto fini;
    }
690

691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
    /* send the attach request to a group member to initiate a bulk transfer
     * of the group's membership view
     */
    sret = ssg_group_attach_send(group_descriptor, &group_name,
        &group_size, &view_buf);
    if (sret != SSG_SUCCESS || !group_name || !view_buf) goto fini;

    /* set up address string array for group members */
    addr_strs = ssg_addr_str_buf_to_list(view_buf, group_size);
    if (!addr_strs) goto fini;

    /* allocate an SSG attached group data structure and initialize some of it */
    ag = malloc(sizeof(*ag));
    if (!ag) goto fini;
    memset(ag, 0, sizeof(*ag));
    ag->name = group_name;
    ag->descriptor = ssg_group_descriptor_dup(group_descriptor);
    if (!ag->descriptor) goto fini;
    ag->descriptor->owner_status = SSG_OWNER_IS_ATTACHER;

    /* create the view for the group */
712 713
    sret = ssg_group_view_create(addr_strs, group_size, HG_ADDR_NULL,
        NULL, &ag->view);
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728
    if (sret != SSG_SUCCESS) goto fini;

    /* add this group reference to our group table */
    HASH_ADD(hh, ssg_inst->attached_group_table, descriptor->name_hash,
        sizeof(uint64_t), ag);

    sret = SSG_SUCCESS;

    /* don't free on success */
    ag = NULL;
fini:
    free(view_buf);
    free(addr_strs);
    if (ag)
    {
729
        ssg_group_view_destroy(&ag->view);
730 731
        free(ag->name);
        free(ag);
732
        ssg_group_descriptor_free(ag->descriptor);
733 734 735
    }

    return sret;
736 737 738 739 740
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
    ssg_attached_group_t *ag;

    if (group_descriptor->owner_status != SSG_OWNER_IS_ATTACHER)
    {
        fprintf(stderr, "Error: SSG unable to detach from group that" \
            " was never attached\n");
        return SSG_FAILURE;
    }

    /* find the attached group structure and destroy it */
    HASH_FIND(hh, ssg_inst->attached_group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), ag);
    if (!ag)
    {
        fprintf(stderr, "Error: SSG unable to find expected attached group reference\n");
        return SSG_FAILURE;
    }
    HASH_DELETE(hh, ssg_inst->attached_group_table, ag);
    ssg_attached_group_destroy(ag);

762 763 764
    return SSG_SUCCESS;
}

765 766 767
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
768

769
ssg_member_id_t ssg_get_group_self_id(
770
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
771
{
Shane Snyder's avatar
Shane Snyder committed
772
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
773 774
    ssg_group_t *g;

Shane Snyder's avatar
Shane Snyder committed
775
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
776
        return SSG_MEMBER_ID_INVALID;
777

778 779 780 781
    if (group_descriptor->owner_status != SSG_OWNER_IS_MEMBER)
    {
        fprintf(stderr, "Error: SSG can only obtain a self ID from a group the" \
            " caller is a member of\n");
782
        return SSG_MEMBER_ID_INVALID;
783 784
    }

Shane Snyder's avatar
Shane Snyder committed
785 786
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
787
    if (!g)
788
        return SSG_MEMBER_ID_INVALID;
789 790

    return g->self_id;
Shane Snyder's avatar
Shane Snyder committed
791 792
}

793 794
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
795
{
Shane Snyder's avatar
Shane Snyder committed
796
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
797
    ssg_group_view_t *group_view = NULL;
798

Shane Snyder's avatar
Shane Snyder committed
799
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
800 801
        return 0;

802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823
    if (group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
    {
        ssg_group_t *g;

        HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), g);
        if (g)
            group_view = &g->view;
    }
    else if (group_descriptor->owner_status == SSG_OWNER_IS_ATTACHER)
    {
        ssg_attached_group_t *ag;

        HASH_FIND(hh, ssg_inst->attached_group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), ag);
        if (ag)
            group_view = &ag->view;
    }
    else
    {
        fprintf(stderr, "Error: SSG can only obtain size of groups that the caller" \
            " is a member of or an attacher of\n");
824
        return 0;
825
    }
826

827 828 829 830
    if (group_view)
        return group_view->size;
    else
        return 0;
Shane Snyder's avatar
Shane Snyder committed
831 832
}

833 834 835
hg_addr_t ssg_get_addr(
    ssg_group_id_t group_id,
    ssg_member_id_t member_id)
Shane Snyder's avatar
Shane Snyder committed
836
{
Shane Snyder's avatar
Shane Snyder committed
837
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
838
    ssg_group_view_t *group_view = NULL;
839
    ssg_member_state_t *member_state;
840

841 842
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL ||
            member_id == SSG_MEMBER_ID_INVALID)
843 844
        return HG_ADDR_NULL;

845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871
    if (group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
    {
        ssg_group_t *g;

        HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), g);
        if (g)
            group_view = &g->view;
    }
    else if (group_descriptor->owner_status == SSG_OWNER_IS_ATTACHER)
    {
        ssg_attached_group_t *ag;

        HASH_FIND(hh, ssg_inst->attached_group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), ag);
        if (ag)
            group_view = &ag->view;
    }
    else
    {
        fprintf(stderr, "Error: SSG can only obtain member addresses of groups" \
            " that the caller is a member of or an attacher of\n");
        return HG_ADDR_NULL;
    }

    if (group_view)
    {
872 873 874 875 876
        HASH_FIND(hh, group_view->member_map, &member_id, sizeof(ssg_member_id_t),
            member_state);
        if (member_state)
            return member_state->addr;
        else
877 878 879
            return HG_ADDR_NULL;
    }
    else
880
        return HG_ADDR_NULL;
881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897
}

ssg_group_id_t ssg_group_id_dup(
    ssg_group_id_t group_id)
{
    ssg_group_descriptor_t *dup;

    dup = ssg_group_descriptor_dup((ssg_group_descriptor_t *)group_id);
    return (ssg_group_id_t)dup;
}

void ssg_group_id_free(
    ssg_group_id_t group_id)
{
    ssg_group_descriptor_t *descriptor = (ssg_group_descriptor_t *)group_id;

    ssg_group_descriptor_free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
898
    descriptor = SSG_GROUP_ID_NULL;
899 900 901
    return;
}

902 903 904 905 906 907 908 909 910
char *ssg_group_id_get_addr_str(
    ssg_group_id_t group_id)
{
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;

    return strdup(group_descriptor->addr_str);
}


911 912 913 914 915 916 917
void ssg_group_id_serialize(
    ssg_group_id_t group_id,
    char ** buf_p,
    size_t * buf_size_p)
{
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
    size_t alloc_size;
918
    char *gid_buf, *p; 
919 920 921 922 923 924 925 926 927 928 929 930 931

    *buf_p = NULL;
    *buf_size_p = 0;

    /* determine needed buffer size */
    alloc_size = (sizeof(group_descriptor->magic_nr) + sizeof(group_descriptor->name_hash) +
        strlen(group_descriptor->addr_str) + 1);

    gid_buf = malloc(alloc_size);
    if (!gid_buf)
        return;

    /* serialize */
932 933 934 935 936 937
    p = gid_buf;
    *(uint64_t *)p = group_descriptor->magic_nr;
    p += sizeof(uint64_t);
    *(uint64_t *)p = group_descriptor->name_hash;
    p += sizeof(uint64_t);
    strcpy(p, group_descriptor->addr_str);
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093
    /* the rest of the descriptor is stateful and not appropriate for serializing... */

    *buf_p = gid_buf;
    *buf_size_p = alloc_size;

    return;
}

void ssg_group_id_deserialize(
    const char * buf,
    size_t buf_size,
    ssg_group_id_t * group_id_p)
{
    size_t min_buf_size;
    uint64_t magic_nr;
    uint64_t name_hash;
    const char *addr_str;
    ssg_group_descriptor_t *group_descriptor;

    *group_id_p = SSG_GROUP_ID_NULL;

    /* check to ensure the buffer contains enough data to make a group ID */
    min_buf_size = (sizeof(group_descriptor->magic_nr) +
        sizeof(group_descriptor->name_hash) + 1);
    if (buf_size < min_buf_size)
    {
        fprintf(stderr, "Error: Serialized buffer does not contain a valid SSG group ID\n");
        return;
    }

    /* deserialize */
    magic_nr = *(uint64_t *)buf;
    if (magic_nr != SSG_MAGIC_NR)
    {
        fprintf(stderr, "Error: Magic number mismatch when deserializing SSG group ID\n");
        return;
    }
    buf += sizeof(uint64_t);
    name_hash = *(uint64_t *)buf;
    buf += sizeof(uint64_t);
    addr_str = buf;

    group_descriptor = ssg_group_descriptor_create(name_hash, addr_str,
        SSG_OWNER_IS_UNASSOCIATED);
    if (!group_descriptor)
        return;

    *group_id_p = (ssg_group_id_t)group_descriptor;

    return;
}

int ssg_group_id_store(
    const char * file_name,
    ssg_group_id_t group_id)
{
    int fd;
    char *buf;
    size_t buf_size;
    ssize_t bytes_written;

    fd = open(file_name, O_WRONLY | O_CREAT | O_TRUNC, 0644);
    if (fd < 0)
    {
        fprintf(stderr, "Error: Unable to open file %s for storing SSG group ID\n",
            file_name);
        return SSG_FAILURE;
    }

    ssg_group_id_serialize(group_id, &buf, &buf_size);
    if (buf == NULL)
    {
        fprintf(stderr, "Error: Unable to serialize SSG group ID.\n");
        close(fd);
        return SSG_FAILURE;
    }

    bytes_written = write(fd, buf, buf_size);
    if (bytes_written != (ssize_t)buf_size)
    {
        fprintf(stderr, "Error: Unable to write SSG group ID to file %s\n", file_name);
        close(fd);
        free(buf);
        return SSG_FAILURE;
    }

    close(fd);
    free(buf);
    return SSG_SUCCESS;
}


int ssg_group_id_load(
    const char * file_name,
    ssg_group_id_t * group_id_p)
{
    int fd;
    struct stat fstats;
    char *buf;
    ssize_t bytes_read;
    int ret;

    *group_id_p = SSG_GROUP_ID_NULL;

    fd = open(file_name, O_RDONLY);
    if (fd < 0)
    {
        fprintf(stderr, "Error: Unable to open file %s for loading SSG group ID\n",
            file_name);
        return SSG_FAILURE;
    }

    ret = fstat(fd, &fstats);
    if (ret != 0)
    {
        fprintf(stderr, "Error: Unable to stat file %s\n", file_name);
        close(fd);
        return SSG_FAILURE;
    }
    if (fstats.st_size == 0)
    {
        fprintf(stderr, "Error: SSG group ID file %s is empty\n", file_name);
        close(fd);
        return SSG_FAILURE;
    }

    buf = malloc(fstats.st_size);
    if (buf == NULL)
    {
        close(fd);
        return SSG_FAILURE;
    }

    bytes_read = read(fd, buf, fstats.st_size);
    if (bytes_read != (ssize_t)fstats.st_size)
    {
        fprintf(stderr, "Error: Unable to read SSG group ID from file %s\n", file_name);
        close(fd);
        free(buf);
        return SSG_FAILURE;
    }

    ssg_group_id_deserialize(buf, (size_t)bytes_read, group_id_p);
    if (*group_id_p == SSG_GROUP_ID_NULL)
    {
        fprintf(stderr, "Error: Unable to deserialize SSG group ID\n");
        close(fd);
        free(buf);
        return SSG_FAILURE;
    }

    close(fd);
    free(buf);
    return SSG_SUCCESS;
}

1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113
void ssg_group_dump(
    ssg_group_id_t group_id)
{
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
    ssg_group_view_t *group_view = NULL;
    char *group_name = NULL;
    char group_role[32];
    char group_self_id[32];

    if (group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
    {
        ssg_group_t *g;

        HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), g);
        if (g)
        {
            group_view = &g->view;
            group_name = g->name;
            strcpy(group_role, "member");
1114
            sprintf(group_self_id, "%lu", g->self_id);
1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138
        }
    }
    else if (group_descriptor->owner_status == SSG_OWNER_IS_ATTACHER)
    {
        ssg_attached_group_t *ag;

        HASH_FIND(hh, ssg_inst->attached_group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), ag);
        if (ag)
        {
            group_view = &ag->view;
            group_name = ag->name;
            strcpy(group_role, "attacher");
        }
    }
    else
    {
        fprintf(stderr, "Error: SSG can only dump membership information for" \
            " groups that the caller is a member of or an attacher of\n");
        return;
    }

    if (group_view)
    {
1139
        ssg_member_state_t *member_state, *tmp_ms;
1140 1141 1142 1143 1144

        printf("SSG membership information for group '%s':\n", group_name);
        printf("\trole: '%s'\n", group_role);
        if (strcmp(group_role, "member") == 0)
            printf("\tself_id: %s\n", group_self_id);
Shane Snyder's avatar
Shane Snyder committed
1145
        printf("\tsize: %d\n", group_view->size);
1146
        printf("\tview:\n");
1147
        HASH_ITER(hh, group_view->member_map, member_state, tmp_ms)
1148
        {
Shane Snyder's avatar
Shane Snyder committed
1149
            printf("\t\tid: %20lu\taddr: %s\n", member_state->id,
1150
                member_state->addr_str);
1151
        }
1152 1153 1154 1155
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
1156

1157
    return;
Shane Snyder's avatar
Shane Snyder committed
1158 1159
}

1160 1161 1162
/************************************
 *** SSG internal helper routines ***
 ************************************/
1163

1164
static ssg_group_descriptor_t * ssg_group_descriptor_create(
1165
    uint64_t name_hash, const char * leader_addr_str, int owner_status)
1166
{
1167
    ssg_group_descriptor_t *descriptor;
1168

1169 1170
    descriptor = malloc(sizeof(*descriptor));
    if (!descriptor) return NULL;
Shane Snyder's avatar
Shane Snyder committed
1171

1172
    /* hash the group name to obtain an 64-bit unique ID */
1173
    descriptor->magic_nr = SSG_MAGIC_NR;
1174
    descriptor->name_hash = name_hash;
1175 1176
    descriptor->addr_str = strdup(leader_addr_str);
    if (!descriptor->addr_str)
Shane Snyder's avatar
Shane Snyder committed
1177
    {
1178
        free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
1179 1180
        return NULL;
    }
1181
    descriptor->owner_status = owner_status;
1182
    descriptor->ref_count = 1;
1183 1184
    return descriptor;
}
1185

1186 1187 1188
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor)
{
Shane Snyder's avatar
Shane Snyder committed
1189 1190
    descriptor->ref_count++;
    return descriptor;
1191 1192
}

1193 1194
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor)
1195
{
1196 1197
    if (descriptor)
    {
Shane Snyder's avatar
Shane Snyder committed
1198 1199 1200 1201 1202 1203 1204 1205 1206
        if(descriptor->ref_count == 1)
        {
            free(descriptor->addr_str);
            free(descriptor);
        }
        else
        {
            descriptor->ref_count--;
        }
1207 1208 1209
    }
    return;
}
1210

1211
static int ssg_group_view_create(
1212 1213 1214
    const char * const group_addr_strs[], int group_size,
    hg_addr_t self_addr, ssg_member_id_t * self_id,
    ssg_group_view_t * view)
1215 1216
{
    int i, j, r;
1217 1218 1219 1220
    ABT_thread *lookup_ults = NULL;
    struct ssg_group_lookup_ult_args *lookup_ult_args = NULL;
    char *self_addr_str = NULL;
    hg_size_t self_addr_str_size = 0;
1221 1222
    const char *self_addr_substr = NULL;
    const char *addr_substr = NULL;
1223 1224
    ssg_member_state_t *tmp_ms;
    hg_return_t hret;
1225
    int aret;
1226
    int sret = SSG_FAILURE;
1227

1228 1229 1230 1231
    if (self_id)
        *self_id = SSG_MEMBER_ID_INVALID;
        
    if ((self_id != NULL && self_addr == HG_ADDR_NULL) || !view) goto fini;
1232 1233 1234

    /* allocate lookup ULTs */
    lookup_ults = malloc(group_size * sizeof(*lookup_ults));
1235 1236
    if (lookup_ults == NULL) goto fini;
    for (i = 0; i < group_size; i++) lookup_ults[i] = ABT_THREAD_NULL;
1237
    lookup_ult_args = malloc(group_size * sizeof(*lookup_ult_args));
1238
    if (lookup_ult_args == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
1239

1240
    if(self_addr)
1241
    {
1242 1243 1244 1245 1246 1247
        hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
        if (hret != HG_SUCCESS) goto fini;
        self_addr_str = malloc(self_addr_str_size);
        if (self_addr_str == NULL) goto fini;
        hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
        if (hret != HG_SUCCESS) goto fini;
1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259

        /* strstr is used here b/c there may be inconsistencies in whether the class
         * is included in the address or not (it should not be in HG_Addr_to_string,
         * but it's possible that it is in the list of group address strings)
         */
        self_addr_substr = strstr(self_addr_str, "://");
        if (self_addr_substr == NULL)
            self_addr_substr = self_addr_str;
        else
            self_addr_substr += 3;
    }

1260 1261
    /* construct view using ULTs to lookup the address of each group member */
    view->size = group_size;
Shane Snyder's avatar
Shane Snyder committed
1262
    r = rand() % group_size;
1263 1264 1265 1266 1267
    for (i = 0; i < group_size; i++)
    {
        /* randomize our starting index so all group members aren't looking
         * up other group members in the same order
         */
Shane Snyder's avatar
Shane Snyder committed
1268
        j = (r + i) % group_size;
1269

1270 1271
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

1272 1273 1274 1275 1276 1277 1278
        tmp_ms = malloc(sizeof(*tmp_ms));
        if (!tmp_ms) goto fini;

        /* generate a unique member ID for this address */