ssg.c 39.4 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(
Shane Snyder's avatar
Shane Snyder committed
150 151 152 153
    ssg_group_t *g);

static int ssg_get_swim_dping_target(
    void *group_data,
Shane Snyder's avatar
Shane Snyder committed
154 155 156
    swim_member_id_t *target_id,
    swim_member_inc_nr_t *target_inc_nr,
    hg_addr_t *target_addr)
157 158
{
    ssg_group_t *g = (ssg_group_t *)group_data;
159 160
    ssg_member_state_t *dping_target_ms;
    unsigned int nondead_list_len;
161 162 163

    assert(g != NULL);

164 165
    /* XXX MUTEX */

166 167
    nondead_list_len = utarray_len(g->nondead_member_list);
    if (nondead_list_len == 0)
Shane Snyder's avatar
Shane Snyder committed
168 169
        return -1; /* no targets */

170 171
    /* reshuffle member list after a complete traversal */
    if (g->dping_target_ndx == nondead_list_len)
Shane Snyder's avatar
Shane Snyder committed
172
    {
173
        ssg_shuffle_member_list(g);
174
        g->dping_target_ndx = 0;
Shane Snyder's avatar
Shane Snyder committed
175
    }
Shane Snyder's avatar
Shane Snyder committed
176

177 178
    /* pull next dping target using saved state */
    dping_target_ms = *(ssg_member_state_t **)utarray_eltptr(
179
        g->nondead_member_list, g->dping_target_ndx);
Shane Snyder's avatar
Shane Snyder committed
180

181 182 183
    *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
184

185
    /* increment dping target index for next iteration */
186
    g->dping_target_ndx++;
Shane Snyder's avatar
Shane Snyder committed
187

188
    return 0;
Shane Snyder's avatar
Shane Snyder committed
189 190
}

Shane Snyder's avatar
Shane Snyder committed
191 192
static int ssg_get_swim_iping_targets(
    void *group_data,
193
    swim_member_id_t dping_target_id,
Shane Snyder's avatar
Shane Snyder committed
194
    int *num_targets,
195
    swim_member_id_t *target_ids,
Shane Snyder's avatar
Shane Snyder committed
196 197 198 199
    hg_addr_t *target_addrs)
{
    ssg_group_t *g = (ssg_group_t *)group_data;
    int max_targets = *num_targets;
200 201 202 203
    int iping_target_count = 0;
    int i = 0;
    unsigned int r_start, r_ndx;
    ssg_member_state_t **r_ms_p;
Shane Snyder's avatar
Shane Snyder committed
204 205 206 207 208

    assert(g != NULL);

    *num_targets = 0;

209
#if 0
210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
    /* XXX MUTEX */

    /* pick random index in the nondead list, and pull out a set of iping
     * targets starting from that index
     */
    r_start = rand() % g->nondead_member_count;
    while (iping_target_count < max_targets)
    {
        r_ndx = (r_start + i) % g->nondead_member_count;
        /* if we've iterated through the entire nondead list, stop */
        if ((i > 0 ) && (r_ndx == r_start)) break;

        r_ms_p = (ssg_member_state_t **)utarray_eltptr(g->nondead_member_list, r_ndx);

        /* do not select the dping target as an iping target */
        if ((swim_member_id_t)(*r_ms_p)->id == dping_target_id)
        {
            i++;
            continue;
        }

        target_ids[iping_target_count] = (swim_member_id_t)(*r_ms_p)->id;
        target_addrs[iping_target_count] = (*r_ms_p)->addr;
        iping_target_count++;
        i++;
    }

    *num_targets = iping_target_count;
238
#endif
239

Shane Snyder's avatar
Shane Snyder committed
240 241 242
    return 0;
}

Shane Snyder's avatar
Shane Snyder committed
243 244 245 246 247 248 249 250 251 252 253 254
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
255
    assert(ms != NULL);
Shane Snyder's avatar
Shane Snyder committed
256 257 258 259 260 261 262 263 264

    *addr = ms->addr;

    return;
}

static void ssg_get_swim_member_state(
    void *group_data,
    swim_member_id_t id,
265
    swim_member_state_t **state)
Shane Snyder's avatar
Shane Snyder committed
266 267 268 269 270 271 272 273
{
    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
274
    assert(ms != NULL);
Shane Snyder's avatar
Shane Snyder committed
275

276
    *state = &ms->swim_state;
Shane Snyder's avatar
Shane Snyder committed
277 278 279 280

    return;
}

281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314
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;
}

315 316 317 318 319 320
static void ssg_shuffle_member_list(
    ssg_group_t *g)
{
    unsigned int i, r;
    ssg_member_state_t **tmp_ms;
    UT_array *list = g->nondead_member_list;
321 322 323
    unsigned int len = utarray_len(g->nondead_member_list);

    SSG_DEBUG(g, "...SHUFFLING\n");
324

Shane Snyder's avatar
Shane Snyder committed
325 326
    if (len <= 1) return;

327 328 329 330 331 332 333 334 335 336 337 338 339 340
    /* run fisher-yates shuffle over list of nondead members */
    for (i = len - 1; i > 0; i--)
    {
        r = rand() % (i + 1);
        tmp_ms = (ssg_member_state_t **)utarray_eltptr(list, r);
        utarray_erase(list, r, 1);
        utarray_insert(list, *(ssg_member_state_t **)utarray_eltptr(list, i), r);
        utarray_erase(list, i, 1);
        utarray_insert(list, *tmp_ms, i);
    }

    return;
}

Shane Snyder's avatar
Shane Snyder committed
341
ssg_group_id_t ssg_group_create(
342 343
    const char * group_name,
    const char * const group_addr_strs[],
344 345 346
    int group_size,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
347
{
348
    uint64_t name_hash;
Shane Snyder's avatar
Shane Snyder committed
349
    hg_addr_t self_addr = HG_ADDR_NULL;
350
    ssg_group_descriptor_t *tmp_descriptor;
351
    ssg_group_t *g = NULL;
352
    ssg_member_state_t *ms, *tmp_ms;
353
    hg_return_t hret;
354
    int sret;
Shane Snyder's avatar
Shane Snyder committed
355
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
356

357
    if (!ssg_inst) return group_id;
358

359
    name_hash = ssg_hash64_str(group_name);
360

361
    /* generate a unique ID for this group  */
362 363
    tmp_descriptor = ssg_group_descriptor_create(name_hash, group_addr_strs[0],
        SSG_OWNER_IS_MEMBER);
364
    if (tmp_descriptor == NULL) return group_id;
365

Shane Snyder's avatar
Shane Snyder committed
366
    /* make sure we aren't re-creating an existing group */
367
    HASH_FIND(hh, ssg_inst->group_table, &tmp_descriptor->name_hash,
Shane Snyder's avatar
Shane Snyder committed
368
        sizeof(uint64_t), g);
369 370 371 372 373
    if (g)
    {
        g = NULL;
        goto fini;
    }
374

375 376 377 378 379 380 381
    /* 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;
382 383
    g->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
384

385 386 387 388
    /* get my address */
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
    if (hret != HG_SUCCESS) goto fini;

389
    /* initialize the group view */
390 391
    sret = ssg_group_view_create(group_addr_strs, group_size, self_addr,
        &g->self_id, &g->view);
392
    if (sret != SSG_SUCCESS) goto fini;
393
    if (g->self_id == SSG_MEMBER_ID_INVALID)
394
    {
395
        /* if unable to resolve my rank within the group, error out */
396 397 398 399 400
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
    }

401
    /* create a list of all nondead member states and shuffle it */
402 403 404 405 406
    UT_icd ms_icd = {sizeof(ssg_member_state_t *), NULL, NULL, NULL};
    utarray_new(g->nondead_member_list, &ms_icd);
    utarray_reserve(g->nondead_member_list, g->view.size);
    HASH_ITER(hh, g->view.member_map, ms, tmp_ms)
    {
Shane Snyder's avatar
Shane Snyder committed
407
        utarray_push_back(g->nondead_member_list, &ms);
408
    }
409
    ssg_shuffle_member_list(g);
Shane Snyder's avatar
Shane Snyder committed
410

411
    /* initialize swim failure detector */
412 413
    // TODO: we should probably barrier or sync somehow to avoid rpc failures
    // due to timing skew of different ranks initializing swim
414 415
    swim_group_mgmt_callbacks_t swim_callbacks = {
        .get_dping_target = &ssg_get_swim_dping_target,
Shane Snyder's avatar
Shane Snyder committed
416 417 418
        .get_iping_targets = &ssg_get_swim_iping_targets,
        .get_member_addr = ssg_get_swim_member_addr,
        .get_member_state = ssg_get_swim_member_state,
419
        .apply_member_update = ssg_apply_swim_member_update,
420
    };
421 422
    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
423
    if (g->swim_ctx == NULL) goto fini;
424

Shane Snyder's avatar
Shane Snyder committed
425 426 427
    /* everything successful -- set the output group identifier, which is just
     * an opaque pointer to the group descriptor structure
     */
428 429
    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
430

431 432 433 434
    /* 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
435
    SSG_DEBUG(g, "group create successful (size=%d)\n", group_size);
436

437
    /* don't free group pointer on success */
438 439
    g = NULL;
fini:
440
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
441 442
    if (g)
    {
443
        ssg_group_view_destroy(&g->view);
444
        free(g->name);
445 446
        free(g);
    }
447 448
    if (group_id == SSG_GROUP_ID_NULL)
        ssg_group_descriptor_free(tmp_descriptor);
449

Shane Snyder's avatar
Shane Snyder committed
450
    return group_id;
451 452
}

Shane Snyder's avatar
Shane Snyder committed
453
ssg_group_id_t ssg_group_create_config(
454
    const char * group_name,
455 456 457
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
458 459 460 461
{
    int fd;
    struct stat st;
    char *rd_buf = NULL;
462
    ssize_t rd_buf_size;
463
    char *tok;
464 465
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
466 467
    int ret;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
468
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490

    /* 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 */
491 492
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
493 494 495 496 497
    {
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
498
    rd_buf[rd_buf_size]='\0';
499 500 501 502 503

    /* 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
504 505
    if (tok == NULL) goto fini;

506
    /* build up the address buffer */
507 508
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
509 510
    do
    {
511 512 513
        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
514 515 516
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
517
    if (addr_str_buf_len != rd_buf_size)
518 519
    {
        /* adjust buffer size if our initial guess was wrong */
520
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
521
        if (tmp == NULL) goto fini;
522
        addr_str_buf = tmp;
523
    }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
524

525
    /* set up address string array for group members */
526
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
527
    if (!addr_strs) goto fini;
528

529
    /* invoke the generic group create routine using our list of addrs */
530 531
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
532 533

fini:
534
    /* cleanup before returning */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
535
    if (fd != -1) close(fd);
536
    free(rd_buf);
537
    free(addr_str_buf);
538 539
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
540
    return group_id;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
541 542
}

Shane Snyder's avatar
Shane Snyder committed
543
#ifdef SSG_HAVE_MPI
Shane Snyder's avatar
Shane Snyder committed
544
ssg_group_id_t ssg_group_create_mpi(
545
    const char * group_name,
546 547 548
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
549
{
550
    int i;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
551
    hg_addr_t self_addr = HG_ADDR_NULL;
552
    char *self_addr_str = NULL;
553 554
    hg_size_t self_addr_str_size = 0;
    int self_addr_str_size_int = 0; /* for mpi-friendly conversion */
555
    char *addr_str_buf = NULL;
556 557 558 559
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
560
    hg_return_t hret;
Shane Snyder's avatar
Shane Snyder committed
561
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
562

563 564
    if (!ssg_inst) goto fini;

565
    /* get my address */
566
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
567
    if (hret != HG_SUCCESS) goto fini;
568
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
Shane Snyder's avatar
Shane Snyder committed
569
    if (hret != HG_SUCCESS) goto fini;
570
    self_addr_str = malloc(self_addr_str_size);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
571
    if (self_addr_str == NULL) goto fini;
572
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
573
    if (hret != HG_SUCCESS) goto fini;
574
    self_addr_str_size_int = (int)self_addr_str_size; /* null char included in call */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
575

576
    /* gather the buffer sizes */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
577 578 579 580
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
581
    sizes[comm_rank] = self_addr_str_size_int;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
582 583
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

584 585 586
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
     */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
587 588 589
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
590
    for (i = 1; i < comm_size+1; i++)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
591 592
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];

593
    /* allgather the addresses */
594 595
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
596
    MPI_Allgatherv(self_addr_str, self_addr_str_size_int, MPI_BYTE,
597
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
598

599
    /* set up address string array for group members */
600
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
601 602 603
    if (!addr_strs) goto fini;

    /* invoke the generic group create routine using our list of addrs */
604 605
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
606 607

fini:
608
    /* cleanup before returning */
609
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
610
    free(self_addr_str);
611 612 613
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
614 615
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
616
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
617 618 619
}
#endif

620 621
int ssg_group_destroy(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
622
{
Shane Snyder's avatar
Shane Snyder committed
623
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
624
    ssg_group_t *g;
625

Shane Snyder's avatar
Shane Snyder committed
626
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
627
        return SSG_FAILURE;
628

629 630 631 632 633 634
    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;
    }

635
    /* find the group structure and destroy it */
Shane Snyder's avatar
Shane Snyder committed
636 637
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
638 639 640 641 642
    if (!g)
    {
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
    }
643
    HASH_DELETE(hh, ssg_inst->group_table, g);
644
    ssg_group_destroy_internal(g);
645

646
    return SSG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
647 648
}

649 650 651
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
652
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
653 654 655 656 657 658
    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;
659

660
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
661

662 663 664 665 666 667 668 669 670 671 672
    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;
    }
673

674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694
    /* 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 */
695 696
    sret = ssg_group_view_create(addr_strs, group_size, HG_ADDR_NULL,
        NULL, &ag->view);
697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
    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)
    {
712
        ssg_group_view_destroy(&ag->view);
713 714
        free(ag->name);
        free(ag);
715
        ssg_group_descriptor_free(ag->descriptor);
716 717 718
    }

    return sret;
719 720 721 722 723
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
    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);

745 746 747
    return SSG_SUCCESS;
}

748 749 750
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
751

752
ssg_member_id_t ssg_get_group_self_id(
753
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
754
{
Shane Snyder's avatar
Shane Snyder committed
755
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
756 757
    ssg_group_t *g;

Shane Snyder's avatar
Shane Snyder committed
758
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
759
        return SSG_MEMBER_ID_INVALID;
760

761 762 763 764
    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");
765
        return SSG_MEMBER_ID_INVALID;
766 767
    }

Shane Snyder's avatar
Shane Snyder committed
768 769
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
770
    if (!g)
771
        return SSG_MEMBER_ID_INVALID;
772 773

    return g->self_id;
Shane Snyder's avatar
Shane Snyder committed
774 775
}

776 777
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
778
{
Shane Snyder's avatar
Shane Snyder committed
779
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
780
    ssg_group_view_t *group_view = NULL;
781

Shane Snyder's avatar
Shane Snyder committed
782
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
783 784
        return 0;

785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806
    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");
807
        return 0;
808
    }
809

810 811 812 813
    if (group_view)
        return group_view->size;
    else
        return 0;
Shane Snyder's avatar
Shane Snyder committed
814 815
}

816 817 818
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
819
{
Shane Snyder's avatar
Shane Snyder committed
820
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
821
    ssg_group_view_t *group_view = NULL;
822
    ssg_member_state_t *member_state;
823

824 825
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL ||
            member_id == SSG_MEMBER_ID_INVALID)
826 827
        return HG_ADDR_NULL;

828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854
    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)
    {
855 856 857 858 859
        HASH_FIND(hh, group_view->member_map, &member_id, sizeof(ssg_member_id_t),
            member_state);
        if (member_state)
            return member_state->addr;
        else
860 861 862
            return HG_ADDR_NULL;
    }
    else
863
        return HG_ADDR_NULL;
864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880
}

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
881
    descriptor = SSG_GROUP_ID_NULL;
882 883 884
    return;
}

885 886 887 888 889 890 891 892 893
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);
}


894 895 896 897 898 899 900
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;
901
    char *gid_buf, *p; 
902 903 904 905 906 907 908 909 910 911 912 913 914

    *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 */
915 916 917 918 919 920
    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);
921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 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
    /* 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;
}

1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096
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");
1097
            sprintf(group_self_id, "%lu", g->self_id);
1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
        }
    }
    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)
    {
1122
        ssg_member_state_t *member_state, *tmp_ms;
1123 1124 1125 1126 1127

        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
1128
        printf("\tsize: %d\n", group_view->size);
1129
        printf("\tview:\n");
1130
        HASH_ITER(hh, group_view->member_map, member_state, tmp_ms)
1131
        {
Shane Snyder's avatar
Shane Snyder committed
1132
            printf("\t\tid: %20lu\taddr: %s\n", member_state->id,
1133
                member_state->addr_str);
1134
        }
1135 1136 1137 1138
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
1139

1140
    return;
Shane Snyder's avatar
Shane Snyder committed
1141 1142
}

1143 1144 1145
/************************************
 *** SSG internal helper routines ***
 ************************************/
1146

1147
static ssg_group_descriptor_t * ssg_group_descriptor_create(
1148
    uint64_t name_hash, const char * leader_addr_str, int owner_status)
1149
{
1150
    ssg_group_descriptor_t *descriptor;
1151

1152 1153
    descriptor = malloc(sizeof(*descriptor));
    if (!descriptor) return NULL;
Shane Snyder's avatar
Shane Snyder committed
1154

1155
    /* hash the group name to obtain an 64-bit unique ID */
1156
    descriptor->magic_nr = SSG_MAGIC_NR;
1157
    descriptor->name_hash = name_hash;
1158 1159
    descriptor->addr_str = strdup(leader_addr_str);
    if (!descriptor->addr_str)
Shane Snyder's avatar
Shane Snyder committed
1160
    {
1161
        free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
1162 1163
        return NULL;
    }
1164
    descriptor->owner_status = owner_status;
1165
    descriptor->ref_count = 1;
1166 1167
    return descriptor;
}
1168

1169 1170 1171
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor)
{
Shane Snyder's avatar
Shane Snyder committed
1172 1173
    descriptor->ref_count++;
    return descriptor;
1174 1175
}

1176 1177
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor)
1178
{
1179 1180
    if (descriptor)
    {
Shane Snyder's avatar
Shane Snyder committed
1181 1182 1183 1184 1185 1186 1187 1188 1189
        if(descriptor->ref_count == 1)
        {
            free(descriptor->addr_str);
            free(descriptor);
        }
        else
        {
            descriptor->ref_count--;
        }
1190 1191 1192
    }
    return;
}
1193

1194
static int ssg_group_view_create(
1195 1196 1197
    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)
1198 1199
{
    int i, j, r;
1200 1201 1202 1203
    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;
1204 1205
    const char *self_addr_substr = NULL;
    const char *addr_substr = NULL;
1206 1207
    ssg_member_state_t *tmp_ms;
    hg_return_t hret;
1208
    int aret;
1209
    int sret = SSG_FAILURE;
1210

1211 1212 1213 1214
    if (self_id)
        *self_id = SSG_MEMBER_ID_INVALID;
        
    if ((self_id != NULL && self_addr == HG_ADDR_NULL) || !view) goto fini;
1215 1216 1217

    /* allocate lookup ULTs */
    lookup_ults = malloc(group_size * sizeof(*lookup_ults));
1218 1219
    if (lookup_ults == NULL) goto fini;
    for (i = 0; i < group_size; i++) lookup_ults[i] = ABT_THREAD_NULL;
1220
    lookup_ult_args = malloc(group_size * sizeof(*lookup_ult_args));
1221
    if (lookup_ult_args == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
1222

1223
    if(self_addr)
1224
    {
1225 1226 1227 1228 1229 1230
        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;
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242

        /* 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;
    }

1243 1244
    /* construct view using ULTs to lookup the address of each group member */
    view->size = group_size;
Shane Snyder's avatar
Shane Snyder committed
1245
    r = rand() % group_size;
1246 1247 1248 1249 1250
    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
1251
        j = (r + i) % group_size;
1252

1253 1254
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

1255 1256 1257 1258 1259 1260 1261
        tmp_ms = malloc(sizeof(*tmp_ms));
        if (!tmp_ms) goto fini;

        /* generate a unique member ID for this address */
        tmp_ms->id = ssg_gen_member_id(group_addr_strs[j]);
        tmp_ms->addr_str = strdup(group_addr_strs[j]);
        if (!tmp_ms->addr_str)
1262
        {
1263 1264
            free(tmp_ms);
            goto fini;
1265
        }
1266
        SWIM_MEMBER_STATE_INIT(tmp_ms->swim_state);
1267
        
1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
        /* resolve self id in group if caller asked for it */
        if (self_addr_substr)
        {
            addr_substr = strstr(group_addr_strs[j], "://");
            if (addr_substr == NULL)
                addr_substr = group_addr_strs[j];
            else
                addr_substr += 3;

            if (strcmp(self_addr_substr, addr_substr) == 0)
            {
1279 1280 1281 1282
                if (self_id)               
                    *self_id = tmp_ms->id;

                /* don't look up our own address, we already know it */
Shane Snyder's avatar
Shane Snyder committed
1283
                view->size--;
Shane Snyder's avatar
Shane Snyder committed
1284 1285
                free(tmp_ms->addr_str);
                free(tmp_ms);
1286
                continue;
1287 1288 1289 1290
            }
        }

        /* XXX limit outstanding lookups to some max */
1291 1292
        lookup_ult_args[j].view = view;
        lookup_ult_args[j].member_state = tmp_ms;
1293 1294
        ABT_pool pool;
        margo_get_handler_pool(ssg_inst->mid, &pool);
1295 1296
        aret = ABT_thread_create(pool, &ssg_group_lookup_ult,
            &lookup_ult_args[j], ABT_THREAD_ATTR_NULL,
1297 1298 1299
            &lookup_ults[j]);
        if (aret != ABT_SUCCESS)
        {