ssg.c 37.3 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,
Shane Snyder's avatar
Shane Snyder committed
133
    int *num_targets,
Shane Snyder's avatar
Shane Snyder committed
134 135 136 137 138 139 140 141 142
    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,
143
    swim_member_state_t **state);
144 145 146
static void ssg_apply_swim_member_update(
    void *group_data,
    swim_member_update_t update);
Shane Snyder's avatar
Shane Snyder committed
147

Shane Snyder's avatar
Shane Snyder committed
148 149 150 151 152
static void ssg_gen_rand_member_list(
    ssg_group_t *g);

static int ssg_get_swim_dping_target(
    void *group_data,
Shane Snyder's avatar
Shane Snyder committed
153 154 155
    swim_member_id_t *target_id,
    swim_member_inc_nr_t *target_inc_nr,
    hg_addr_t *target_addr)
156 157
{
    ssg_group_t *g = (ssg_group_t *)group_data;
Shane Snyder's avatar
Shane Snyder committed
158
    ssg_member_state_t *target_ms;
159 160 161

    assert(g != NULL);

Shane Snyder's avatar
Shane Snyder committed
162 163
    /* if current member list is empty, generate a new random one */
    if (g->member_list == NULL)
Shane Snyder's avatar
Shane Snyder committed
164
    {
Shane Snyder's avatar
Shane Snyder committed
165
        ssg_gen_rand_member_list(g);
Shane Snyder's avatar
Shane Snyder committed
166 167 168
        if(g->member_list == NULL)
            return -1;
    }
Shane Snyder's avatar
Shane Snyder committed
169 170 171 172

    /* pull random member off head of list and return addr */
    target_ms = g->member_list;
    LL_DELETE(g->member_list, target_ms);
Shane Snyder's avatar
Shane Snyder committed
173 174 175 176 177 178
    *target_id = (swim_member_id_t)target_ms->id;
    *target_inc_nr = target_ms->swim_state.inc_nr;
    *target_addr = target_ms->addr;

    return 0;
}
Shane Snyder's avatar
Shane Snyder committed
179 180 181 182 183 184 185 186 187 188 189 190 191
static void ssg_gen_rand_member_list(ssg_group_t *g)
{
    ssg_member_state_t *ms, *tmp;

    /* XXX generate random LL permutation from membership map */
    HASH_ITER(hh, g->view.member_map, ms, tmp)
    {
        LL_APPEND(g->member_list, ms);
    }

    return;
}

Shane Snyder's avatar
Shane Snyder committed
192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
static int ssg_get_swim_iping_targets(
    void *group_data,
    int *num_targets,
    swim_member_id_t *target_id,
    hg_addr_t *target_addrs)
{
    ssg_group_t *g = (ssg_group_t *)group_data;
    int max_targets = *num_targets;

    assert(g != NULL);

    *num_targets = 0;

    return 0;
}

Shane Snyder's avatar
Shane Snyder committed
208 209 210 211 212 213 214 215 216 217 218 219
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
220
    assert(ms != NULL);
Shane Snyder's avatar
Shane Snyder committed
221 222 223 224 225 226 227 228 229

    *addr = ms->addr;

    return;
}

static void ssg_get_swim_member_state(
    void *group_data,
    swim_member_id_t id,
230
    swim_member_state_t **state)
Shane Snyder's avatar
Shane Snyder committed
231 232 233 234 235 236 237 238
{
    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
239
    assert(ms != NULL);
Shane Snyder's avatar
Shane Snyder committed
240

241
    *state = &ms->swim_state;
Shane Snyder's avatar
Shane Snyder committed
242 243 244 245

    return;
}

246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281
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 */
            LL_DELETE(g->member_list, ms);
            margo_addr_free(ssg_inst->mid, ms->addr);
            g->view.size--;
            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;
}

Shane Snyder's avatar
Shane Snyder committed
282
ssg_group_id_t ssg_group_create(
283 284
    const char * group_name,
    const char * const group_addr_strs[],
285 286 287
    int group_size,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
288
{
289
    uint64_t name_hash;
Shane Snyder's avatar
Shane Snyder committed
290
    hg_addr_t self_addr = HG_ADDR_NULL;
291
    ssg_group_descriptor_t *tmp_descriptor;
292
    ssg_group_t *g = NULL;
293
    hg_return_t hret;
294
    int sret;
Shane Snyder's avatar
Shane Snyder committed
295
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
296

297
    if (!ssg_inst) return group_id;
298

299
    name_hash = ssg_hash64_str(group_name);
300

301
    /* generate a unique ID for this group  */
302 303
    tmp_descriptor = ssg_group_descriptor_create(name_hash, group_addr_strs[0],
        SSG_OWNER_IS_MEMBER);
304
    if (tmp_descriptor == NULL) return group_id;
305

Shane Snyder's avatar
Shane Snyder committed
306
    /* make sure we aren't re-creating an existing group */
307
    HASH_FIND(hh, ssg_inst->group_table, &tmp_descriptor->name_hash,
Shane Snyder's avatar
Shane Snyder committed
308
        sizeof(uint64_t), g);
309 310 311 312 313
    if (g)
    {
        g = NULL;
        goto fini;
    }
314

315 316 317 318 319 320 321
    /* 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;
322 323
    g->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
324

325 326 327 328
    /* get my address */
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
    if (hret != HG_SUCCESS) goto fini;

329
    /* initialize the group view */
330 331
    sret = ssg_group_view_create(group_addr_strs, group_size, self_addr,
        &g->self_id, &g->view);
332
    if (sret != SSG_SUCCESS) goto fini;
333
    if (g->self_id == SSG_MEMBER_ID_INVALID)
334
    {
335
        /* if unable to resolve my rank within the group, error out */
336 337 338 339 340
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
    }

Shane Snyder's avatar
Shane Snyder committed
341 342 343
    /* generate random list permutation of the member map */
    ssg_gen_rand_member_list(g);

344
    /* initialize swim failure detector */
345 346
    // TODO: we should probably barrier or sync somehow to avoid rpc failures
    // due to timing skew of different ranks initializing swim
347 348
    swim_group_mgmt_callbacks_t swim_callbacks = {
        .get_dping_target = &ssg_get_swim_dping_target,
Shane Snyder's avatar
Shane Snyder committed
349 350 351
        .get_iping_targets = &ssg_get_swim_iping_targets,
        .get_member_addr = ssg_get_swim_member_addr,
        .get_member_state = ssg_get_swim_member_state,
352
        .apply_member_update = ssg_apply_swim_member_update,
353
    };
354 355
    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
356
    if (g->swim_ctx == NULL) goto fini;
357

Shane Snyder's avatar
Shane Snyder committed
358 359 360
    /* everything successful -- set the output group identifier, which is just
     * an opaque pointer to the group descriptor structure
     */
361 362
    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
363

364 365 366 367
    /* 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
368
    SSG_DEBUG(g, "group create successful (size=%d)\n", group_size);
369

370
    /* don't free group pointer on success */
371 372
    g = NULL;
fini:
373
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
374 375
    if (g)
    {
376
        ssg_group_view_destroy(&g->view);
377
        free(g->name);
378 379
        free(g);
    }
380 381
    if (group_id == SSG_GROUP_ID_NULL)
        ssg_group_descriptor_free(tmp_descriptor);
382

Shane Snyder's avatar
Shane Snyder committed
383
    return group_id;
384 385
}

Shane Snyder's avatar
Shane Snyder committed
386
ssg_group_id_t ssg_group_create_config(
387
    const char * group_name,
388 389 390
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
391 392 393 394
{
    int fd;
    struct stat st;
    char *rd_buf = NULL;
395
    ssize_t rd_buf_size;
396
    char *tok;
397 398
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
399 400
    int ret;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
401
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423

    /* 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 */
424 425
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
426 427 428 429 430
    {
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
431
    rd_buf[rd_buf_size]='\0';
432 433 434 435 436

    /* 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
437 438
    if (tok == NULL) goto fini;

439
    /* build up the address buffer */
440 441
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
442 443
    do
    {
444 445 446
        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
447 448 449
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
450
    if (addr_str_buf_len != rd_buf_size)
451 452
    {
        /* adjust buffer size if our initial guess was wrong */
453
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
454
        if (tmp == NULL) goto fini;
455
        addr_str_buf = tmp;
456
    }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
457

458
    /* set up address string array for group members */
459
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
460
    if (!addr_strs) goto fini;
461

462
    /* invoke the generic group create routine using our list of addrs */
463 464
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
465 466

fini:
467
    /* cleanup before returning */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
468
    if (fd != -1) close(fd);
469
    free(rd_buf);
470
    free(addr_str_buf);
471 472
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
473
    return group_id;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
474 475
}

Shane Snyder's avatar
Shane Snyder committed
476
#ifdef SSG_HAVE_MPI
Shane Snyder's avatar
Shane Snyder committed
477
ssg_group_id_t ssg_group_create_mpi(
478
    const char * group_name,
479 480 481
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
482
{
483
    int i;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
484
    hg_addr_t self_addr = HG_ADDR_NULL;
485
    char *self_addr_str = NULL;
486 487
    hg_size_t self_addr_str_size = 0;
    int self_addr_str_size_int = 0; /* for mpi-friendly conversion */
488
    char *addr_str_buf = NULL;
489 490 491 492
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
493
    hg_return_t hret;
Shane Snyder's avatar
Shane Snyder committed
494
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
495

496 497
    if (!ssg_inst) goto fini;

498
    /* get my address */
499
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
500
    if (hret != HG_SUCCESS) goto fini;
501
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
Shane Snyder's avatar
Shane Snyder committed
502
    if (hret != HG_SUCCESS) goto fini;
503
    self_addr_str = malloc(self_addr_str_size);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
504
    if (self_addr_str == NULL) goto fini;
505
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
506
    if (hret != HG_SUCCESS) goto fini;
507
    self_addr_str_size_int = (int)self_addr_str_size; /* null char included in call */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
508

509
    /* gather the buffer sizes */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
510 511 512 513
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
514
    sizes[comm_rank] = self_addr_str_size_int;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
515 516
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

517 518 519
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
     */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
520 521 522
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
523
    for (i = 1; i < comm_size+1; i++)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
524 525
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];

526
    /* allgather the addresses */
527 528
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
529
    MPI_Allgatherv(self_addr_str, self_addr_str_size_int, MPI_BYTE,
530
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
531

532
    /* set up address string array for group members */
533
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
534 535 536
    if (!addr_strs) goto fini;

    /* invoke the generic group create routine using our list of addrs */
537 538
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
539 540

fini:
541
    /* cleanup before returning */
542
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
543
    free(self_addr_str);
544 545 546
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
547 548
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
549
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
550 551 552
}
#endif

553 554
int ssg_group_destroy(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
555
{
Shane Snyder's avatar
Shane Snyder committed
556
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
557
    ssg_group_t *g;
558

Shane Snyder's avatar
Shane Snyder committed
559
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
560
        return SSG_FAILURE;
561

562 563 564 565 566 567
    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;
    }

568
    /* find the group structure and destroy it */
Shane Snyder's avatar
Shane Snyder committed
569 570
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
571 572 573 574 575
    if (!g)
    {
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
    }
576
    HASH_DELETE(hh, ssg_inst->group_table, g);
577
    ssg_group_destroy_internal(g);
578

579
    return SSG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
580 581
}

582 583 584
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
585
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
586 587 588 589 590 591
    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;
592

593
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
594

595 596 597 598 599 600 601 602 603 604 605
    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;
    }
606

607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627
    /* 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 */
628 629
    sret = ssg_group_view_create(addr_strs, group_size, HG_ADDR_NULL,
        NULL, &ag->view);
630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
    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)
    {
645
        ssg_group_view_destroy(&ag->view);
646 647
        free(ag->name);
        free(ag);
648
        ssg_group_descriptor_free(ag->descriptor);
649 650 651
    }

    return sret;
652 653 654 655 656
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677
    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);

678 679 680
    return SSG_SUCCESS;
}

681 682 683
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
684

685
ssg_member_id_t ssg_get_group_self_id(
686
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
687
{
Shane Snyder's avatar
Shane Snyder committed
688
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
689 690
    ssg_group_t *g;

Shane Snyder's avatar
Shane Snyder committed
691
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
692
        return SSG_MEMBER_ID_INVALID;
693

694 695 696 697
    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");
698
        return SSG_MEMBER_ID_INVALID;
699 700
    }

Shane Snyder's avatar
Shane Snyder committed
701 702
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
703
    if (!g)
704
        return SSG_MEMBER_ID_INVALID;
705 706

    return g->self_id;
Shane Snyder's avatar
Shane Snyder committed
707 708
}

709 710
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
711
{
Shane Snyder's avatar
Shane Snyder committed
712
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
713
    ssg_group_view_t *group_view = NULL;
714

Shane Snyder's avatar
Shane Snyder committed
715
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
716 717
        return 0;

718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739
    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");
740
        return 0;
741
    }
742

743 744 745 746
    if (group_view)
        return group_view->size;
    else
        return 0;
Shane Snyder's avatar
Shane Snyder committed
747 748
}

749 750 751
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
752
{
Shane Snyder's avatar
Shane Snyder committed
753
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
754
    ssg_group_view_t *group_view = NULL;
755
    ssg_member_state_t *member_state;
756

757 758
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL ||
            member_id == SSG_MEMBER_ID_INVALID)
759 760
        return HG_ADDR_NULL;

761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787
    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)
    {
788 789 790 791 792
        HASH_FIND(hh, group_view->member_map, &member_id, sizeof(ssg_member_id_t),
            member_state);
        if (member_state)
            return member_state->addr;
        else
793 794 795
            return HG_ADDR_NULL;
    }
    else
796
        return HG_ADDR_NULL;
797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813
}

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
814
    descriptor = SSG_GROUP_ID_NULL;
815 816 817
    return;
}

818 819 820 821 822 823 824 825 826
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);
}


827 828 829 830 831 832 833
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;
834
    char *gid_buf, *p; 
835 836 837 838 839 840 841 842 843 844 845 846 847

    *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 */
848 849 850 851 852 853
    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);
854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
    /* 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;
}

1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
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");
1030
            sprintf(group_self_id, "%lu", g->self_id);
1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
        }
    }
    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)
    {
1055
        ssg_member_state_t *member_state, *tmp_ms;
1056 1057 1058 1059 1060

        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
1061
        printf("\tsize: %d\n", group_view->size);
1062
        printf("\tview:\n");
1063
        HASH_ITER(hh, group_view->member_map, member_state, tmp_ms)
1064
        {
Shane Snyder's avatar
Shane Snyder committed
1065
            printf("\t\tid: %20lu\taddr: %s\n", member_state->id,
1066
                member_state->addr_str);
1067
        }
1068 1069 1070 1071
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
1072

1073
    return;
Shane Snyder's avatar
Shane Snyder committed
1074 1075
}

1076 1077 1078
/************************************
 *** SSG internal helper routines ***
 ************************************/
1079

1080
static ssg_group_descriptor_t * ssg_group_descriptor_create(
1081
    uint64_t name_hash, const char * leader_addr_str, int owner_status)
1082
{
1083
    ssg_group_descriptor_t *descriptor;
1084

1085 1086
    descriptor = malloc(sizeof(*descriptor));
    if (!descriptor) return NULL;
Shane Snyder's avatar
Shane Snyder committed
1087

1088
    /* hash the group name to obtain an 64-bit unique ID */
1089
    descriptor->magic_nr = SSG_MAGIC_NR;
1090
    descriptor->name_hash = name_hash;
1091 1092
    descriptor->addr_str = strdup(leader_addr_str);
    if (!descriptor->addr_str)
Shane Snyder's avatar
Shane Snyder committed
1093
    {
1094
        free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
1095 1096
        return NULL;
    }
1097
    descriptor->owner_status = owner_status;
1098
    descriptor->ref_count = 1;
1099 1100
    return descriptor;
}
1101

1102 1103 1104
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor)
{
Shane Snyder's avatar
Shane Snyder committed
1105 1106
    descriptor->ref_count++;
    return descriptor;
1107 1108
}

1109 1110
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor)
1111
{
1112 1113
    if (descriptor)
    {
Shane Snyder's avatar
Shane Snyder committed
1114 1115 1116 1117 1118 1119 1120 1121 1122
        if(descriptor->ref_count == 1)
        {
            free(descriptor->addr_str);
            free(descriptor);
        }
        else
        {
            descriptor->ref_count--;
        }
1123 1124 1125
    }
    return;
}
1126

1127
static int ssg_group_view_create(
1128 1129 1130
    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)
1131 1132
{
    int i, j, r;
1133 1134 1135 1136
    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;
1137 1138
    const char *self_addr_substr = NULL;
    const char *addr_substr = NULL;
1139 1140
    ssg_member_state_t *tmp_ms;
    hg_return_t hret;
1141
    int aret;
1142
    int sret = SSG_FAILURE;
1143

1144 1145 1146 1147
    if (self_id)
        *self_id = SSG_MEMBER_ID_INVALID;
        
    if ((self_id != NULL && self_addr == HG_ADDR_NULL) || !view) goto fini;
1148 1149 1150

    /* allocate lookup ULTs */
    lookup_ults = malloc(group_size * sizeof(*lookup_ults));
1151 1152
    if (lookup_ults == NULL) goto fini;
    for (i = 0; i < group_size; i++) lookup_ults[i] = ABT_THREAD_NULL;
1153
    lookup_ult_args = malloc(group_size * sizeof(*lookup_ult_args));
1154
    if (lookup_ult_args == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
1155

1156
    if(self_addr)
1157
    {
1158 1159 1160 1161 1162 1163
        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;
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175

        /* 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;
    }

1176 1177
    /* construct view using ULTs to lookup the address of each group member */
    view->size = group_size;
1178 1179 1180 1181 1182 1183 1184 1185
    r = rand() % view->size;
    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
         */
        j = (r + i) % view->size;

1186 1187
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

1188 1189 1190 1191 1192 1193 1194
        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)
1195
        {
1196 1197
            free(tmp_ms);
            goto fini;
1198
        }
1199
        SWIM_MEMBER_STATE_INIT(tmp_ms->swim_state);
1200
        
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211
        /* 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)
            {
1212 1213 1214 1215
                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
1216 1217
                free(tmp_ms->addr_str);
                free(tmp_ms);
1218
                continue;
1219 1220 1221 1222
            }
        }

        /* XXX limit outstanding lookups to some max */
1223 1224
        lookup_ult_args[j].view = view;
        lookup_ult_args[j].member_state = tmp_ms;
1225 1226
        ABT_pool pool;
        margo_get_handler_pool(ssg_inst->mid, &pool);
1227 1228
        aret = ABT_thread_create(pool, &ssg_group_lookup_ult,
            &lookup_ult_args[j], ABT_THREAD_ATTR_NULL,
1229 1230 1231
            &lookup_ults[j]);
        if (aret != ABT_SUCCESS)
        {
1232 1233
            free(tmp_ms->addr_str);
            free(tmp_ms);
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
            goto fini;
        }
    }

    /* wait on all lookup ULTs to terminate */
    for (i = 0; i < group_size; i++)
    {
        if (lookup_ults[i] == ABT_THREAD_NULL) continue;

        aret = ABT_thread_join(lookup_ults[i]);
        ABT_thread_free(&lookup_ults[i]);
        lookup_ults[i] = ABT_THREAD_NULL;
1246
        if (aret != ABT_SUCCESS) goto fini;
1247 1248
        else if (lookup_ult_args[i].out != HG_SUCCESS)
        {
1249
            fprintf(stderr, "Error: SSG unable to lookup HG address for member %lu"
1250 1251 1252
                "(err=%d)\n", lookup_ult_args[i].member_state->id,
                lookup_ult_args[i].out);
            goto fini;
1253 1254 1255
        }
    }

1256 1257 1258
    /* clean exit */
    sret = SSG_SUCCESS;

1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273
fini:
    if (sret != SSG_SUCCESS)
    {
        for (i = 0; i < group_size; i++)
        {
            if (lookup_ults[i] != ABT_THREAD_NULL)
            {
                ABT_thread_cancel(lookup_ults[i]);
                ABT_thread_free(&lookup_ults[i]);
            }
        }
        ssg_group_view_destroy(view);
    }
    free(lookup_ults);
    free(lookup_ult_args);
1274
    free(self_addr_str);
1275 1276

    return sret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
1277
}
1278

1279 1280 1281
static ssg_member_id_t ssg_gen_member_id(
    const char * addr_str)
{
1282 1283 1284 1285 1286 1287 1288 1289 1290
    char tmp[64] = {0};
    ssg_member_id_t id = (ssg_member_id_t)ssg_hash64_str(addr_str);
    while (id == SSG_MEMBER_ID_INVALID)
    {
        if (tmp[0] == 0) strncpy(tmp, addr_str, 63);
        tmp[0]++;
        id = (ssg_member_id_t)ssg_hash64_str(tmp);
    }
    return id;
1291 1292 1293
}


1294 1295 1296 1297 1298
static void ssg_group_lookup_ult(
    void * arg)
{
    struct ssg_group_lookup_ult_args *l = arg;

1299
    /* XXX: should be a timeout here? */
1300 1301
    l->out = margo_addr_lookup(ssg_inst->mid, l->member_state->addr_str,
        &l->member_state->addr);
1302 1303 1304 1305 1306
    if (l->out == HG_SUCCESS)
    {
        HASH_ADD(hh, l->view->member_map, id, sizeof(ssg_member_id_t),
            l->member_state);
    }
1307 1308 1309 1310
    else
    {
        /* XXX */
    }
1311

1312 1313 1314 1315 1316
    return;
}

static void ssg_group_view_destroy(
    ssg_group_view_t * view)
1317
{
1318
    ssg_member_state_t *state, *tmp;
1319

1320
    /* destroy state for all group members */
Shane Snyder's avatar
Shane Snyder committed
1321
    HASH_ITER(hh, view->member_map, state, tmp)
1322
    {
1323 1324 1325 1326
        HASH_DEL(view->member_map, state);
        free(state->addr_str);
        margo_addr_free(ssg_inst->mid,  state->addr);
        free(state);
1327
    }
1328
    view->member_map = NULL;
1329 1330 1331 1332 1333 1334 1335

    return<