ssg.c 37.2 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
128 129 130 131 132 133 134 135 136 137 138 139 140 141
    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,
    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,
142
    swim_member_state_t **state);
Shane Snyder's avatar
Shane Snyder committed
143

Shane Snyder's avatar
Shane Snyder committed
144 145 146 147 148
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
149 150 151
    swim_member_id_t *target_id,
    swim_member_inc_nr_t *target_inc_nr,
    hg_addr_t *target_addr)
152 153
{
    ssg_group_t *g = (ssg_group_t *)group_data;
Shane Snyder's avatar
Shane Snyder committed
154
    ssg_member_state_t *target_ms;
155 156 157

    assert(g != NULL);

Shane Snyder's avatar
Shane Snyder committed
158 159 160 161 162 163 164
    /* if current member list is empty, generate a new random one */
    if (g->member_list == NULL)
        ssg_gen_rand_member_list(g);

    /* 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
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
    *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;
}

static int ssg_get_swim_iping_targets(
    void *group_data,
    swim_member_id_t *target_id,
    hg_addr_t *target_addrs)
{
    ssg_group_t *g = (ssg_group_t *)group_data;

    assert(g != NULL);
180 181 182 183

    return 0;
}

Shane Snyder's avatar
Shane Snyder committed
184 185 186 187 188 189 190 191 192 193 194 195 196
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
197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218
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);
    /* XXX ASSERT */

    *addr = ms->addr;

    return;
}

static void ssg_get_swim_member_state(
    void *group_data,
    swim_member_id_t id,
219
    swim_member_state_t **state)
Shane Snyder's avatar
Shane Snyder committed
220 221 222 223 224 225 226 227 228 229
{
    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);
    /* XXX ASSERT */

230
    *state = &ms->swim_state;
Shane Snyder's avatar
Shane Snyder committed
231 232 233 234

    return;
}

Shane Snyder's avatar
Shane Snyder committed
235
ssg_group_id_t ssg_group_create(
236 237
    const char * group_name,
    const char * const group_addr_strs[],
238 239 240
    int group_size,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
241
{
242
    uint64_t name_hash;
Shane Snyder's avatar
Shane Snyder committed
243
    hg_addr_t self_addr = HG_ADDR_NULL;
244
    ssg_group_descriptor_t *tmp_descriptor;
245
    ssg_group_t *g = NULL;
246
    hg_return_t hret;
247
    int sret;
Shane Snyder's avatar
Shane Snyder committed
248
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
249

250
    if (!ssg_inst) return group_id;
251

252
    name_hash = ssg_hash64_str(group_name);
253

254
    /* generate a unique ID for this group  */
255 256
    tmp_descriptor = ssg_group_descriptor_create(name_hash, group_addr_strs[0],
        SSG_OWNER_IS_MEMBER);
257
    if (tmp_descriptor == NULL) return group_id;
258

Shane Snyder's avatar
Shane Snyder committed
259
    /* make sure we aren't re-creating an existing group */
260
    HASH_FIND(hh, ssg_inst->group_table, &tmp_descriptor->name_hash,
Shane Snyder's avatar
Shane Snyder committed
261
        sizeof(uint64_t), g);
262 263 264 265 266
    if (g)
    {
        g = NULL;
        goto fini;
    }
267

268 269 270 271 272 273 274
    /* 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;
275 276
    g->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
277

278 279 280 281
    /* get my address */
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
    if (hret != HG_SUCCESS) goto fini;

282
    /* initialize the group view */
283 284
    sret = ssg_group_view_create(group_addr_strs, group_size, self_addr,
        &g->self_id, &g->view);
285
    if (sret != SSG_SUCCESS) goto fini;
286
    if (g->self_id == SSG_MEMBER_ID_INVALID)
287
    {
288
        /* if unable to resolve my rank within the group, error out */
289 290 291 292 293
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
    }

Shane Snyder's avatar
Shane Snyder committed
294 295 296
    /* generate random list permutation of the member map */
    ssg_gen_rand_member_list(g);

297
    /* initialize swim failure detector */
298 299
    // TODO: we should probably barrier or sync somehow to avoid rpc failures
    // due to timing skew of different ranks initializing swim
300 301
    swim_group_mgmt_callbacks_t swim_callbacks = {
        .get_dping_target = &ssg_get_swim_dping_target,
Shane Snyder's avatar
Shane Snyder committed
302 303 304
        .get_iping_targets = &ssg_get_swim_iping_targets,
        .get_member_addr = ssg_get_swim_member_addr,
        .get_member_state = ssg_get_swim_member_state,
305
    };
306 307
    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
308
    if (g->swim_ctx == NULL) goto fini;
309

Shane Snyder's avatar
Shane Snyder committed
310 311 312
    /* everything successful -- set the output group identifier, which is just
     * an opaque pointer to the group descriptor structure
     */
313 314
    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
315

316 317 318 319
    /* 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
320
    SSG_DEBUG(g, "group create successful (size=%d)\n", group_size);
321

322
    /* don't free group pointer on success */
323 324
    g = NULL;
fini:
325
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
326 327
    if (g)
    {
328
        ssg_group_view_destroy(&g->view);
329
        free(g->name);
330 331
        free(g);
    }
332 333
    if (group_id == SSG_GROUP_ID_NULL)
        ssg_group_descriptor_free(tmp_descriptor);
334

Shane Snyder's avatar
Shane Snyder committed
335
    return group_id;
336 337
}

Shane Snyder's avatar
Shane Snyder committed
338
ssg_group_id_t ssg_group_create_config(
339
    const char * group_name,
340 341 342
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
343 344 345 346
{
    int fd;
    struct stat st;
    char *rd_buf = NULL;
347
    ssize_t rd_buf_size;
348
    char *tok;
349 350
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
351 352
    int ret;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
353
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375

    /* 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 */
376 377
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
378 379 380 381 382
    {
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
383
    rd_buf[rd_buf_size]='\0';
384 385 386 387 388

    /* 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
389 390
    if (tok == NULL) goto fini;

391
    /* build up the address buffer */
392 393
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
394 395
    do
    {
396 397 398
        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
399 400 401
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
402
    if (addr_str_buf_len != rd_buf_size)
403 404
    {
        /* adjust buffer size if our initial guess was wrong */
405
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
406
        if (tmp == NULL) goto fini;
407
        addr_str_buf = tmp;
408
    }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
409

410
    /* set up address string array for group members */
411
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
412
    if (!addr_strs) goto fini;
413

414
    /* invoke the generic group create routine using our list of addrs */
415 416
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
417 418

fini:
419
    /* cleanup before returning */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
420
    if (fd != -1) close(fd);
421
    free(rd_buf);
422
    free(addr_str_buf);
423 424
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
425
    return group_id;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
426 427
}

Shane Snyder's avatar
Shane Snyder committed
428
#ifdef SSG_HAVE_MPI
Shane Snyder's avatar
Shane Snyder committed
429
ssg_group_id_t ssg_group_create_mpi(
430
    const char * group_name,
431 432 433
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
434
{
435
    int i;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
436
    hg_addr_t self_addr = HG_ADDR_NULL;
437
    char *self_addr_str = NULL;
438 439
    hg_size_t self_addr_str_size = 0;
    int self_addr_str_size_int = 0; /* for mpi-friendly conversion */
440
    char *addr_str_buf = NULL;
441 442 443 444
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
445
    hg_return_t hret;
Shane Snyder's avatar
Shane Snyder committed
446
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
447

448 449
    if (!ssg_inst) goto fini;

450
    /* get my address */
451
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
452
    if (hret != HG_SUCCESS) goto fini;
453
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
Shane Snyder's avatar
Shane Snyder committed
454
    if (hret != HG_SUCCESS) goto fini;
455
    self_addr_str = malloc(self_addr_str_size);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
456
    if (self_addr_str == NULL) goto fini;
457
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
458
    if (hret != HG_SUCCESS) goto fini;
459
    self_addr_str_size_int = (int)self_addr_str_size; /* null char included in call */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
460

461
    /* gather the buffer sizes */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
462 463 464 465
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
466
    sizes[comm_rank] = self_addr_str_size_int;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
467 468
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

469 470 471
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
     */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
472 473 474
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
475
    for (i = 1; i < comm_size+1; i++)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
476 477
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];

478
    /* allgather the addresses */
479 480
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
481
    MPI_Allgatherv(self_addr_str, self_addr_str_size_int, MPI_BYTE,
482
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
483

484
    /* set up address string array for group members */
485
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
486 487 488
    if (!addr_strs) goto fini;

    /* invoke the generic group create routine using our list of addrs */
489 490
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
491 492

fini:
493
    /* cleanup before returning */
494
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
495
    free(self_addr_str);
496 497 498
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
499 500
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
501
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
502 503 504
}
#endif

505 506
int ssg_group_destroy(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
507
{
Shane Snyder's avatar
Shane Snyder committed
508
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
509
    ssg_group_t *g;
510

Shane Snyder's avatar
Shane Snyder committed
511
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
512
        return SSG_FAILURE;
513

514 515 516 517 518 519
    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;
    }

520
    /* find the group structure and destroy it */
Shane Snyder's avatar
Shane Snyder committed
521 522
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
523 524 525 526 527
    if (!g)
    {
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
    }
528
    HASH_DELETE(hh, ssg_inst->group_table, g);
529
    ssg_group_destroy_internal(g);
530

531
    return SSG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
532 533
}

534 535 536
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
537
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
538 539 540 541 542 543
    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;
544

545
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
546

547 548 549 550 551 552 553 554 555 556 557
    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;
    }
558

559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579
    /* 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 */
580 581
    sret = ssg_group_view_create(addr_strs, group_size, HG_ADDR_NULL,
        NULL, &ag->view);
582 583 584 585 586 587 588 589 590 591 592 593 594 595 596
    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)
    {
597
        ssg_group_view_destroy(&ag->view);
598 599
        free(ag->name);
        free(ag);
600
        ssg_group_descriptor_free(ag->descriptor);
601 602 603
    }

    return sret;
604 605 606 607 608
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629
    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);

630 631 632
    return SSG_SUCCESS;
}

633 634 635
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
636

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

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

646 647 648 649
    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");
650
        return SSG_MEMBER_ID_INVALID;
651 652
    }

Shane Snyder's avatar
Shane Snyder committed
653 654
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
655
    if (!g)
656
        return SSG_MEMBER_ID_INVALID;
657 658

    return g->self_id;
Shane Snyder's avatar
Shane Snyder committed
659 660
}

661 662
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
663
{
Shane Snyder's avatar
Shane Snyder committed
664
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
665
    ssg_group_view_t *group_view = NULL;
666

Shane Snyder's avatar
Shane Snyder committed
667
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
668 669
        return 0;

670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691
    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");
692
        return 0;
693
    }
694

695 696 697 698
    if (group_view)
        return group_view->size;
    else
        return 0;
Shane Snyder's avatar
Shane Snyder committed
699 700
}

701 702 703
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
704
{
Shane Snyder's avatar
Shane Snyder committed
705
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
706
    ssg_group_view_t *group_view = NULL;
707
    ssg_member_state_t *member_state;
708

709 710
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL ||
            member_id == SSG_MEMBER_ID_INVALID)
711 712
        return HG_ADDR_NULL;

713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739
    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 member addresses of groups" \
            " that the caller is a member of or an attacher of\n");
        return HG_ADDR_NULL;
    }

    if (group_view)
    {
740 741 742 743 744
        HASH_FIND(hh, group_view->member_map, &member_id, sizeof(ssg_member_id_t),
            member_state);
        if (member_state)
            return member_state->addr;
        else
745 746 747
            return HG_ADDR_NULL;
    }
    else
748
        return HG_ADDR_NULL;
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
}

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
766
    descriptor = SSG_GROUP_ID_NULL;
767 768 769
    return;
}

770 771 772 773 774 775 776 777 778
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);
}


779 780 781 782 783 784 785
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;
786
    char *gid_buf, *p; 
787 788 789 790 791 792 793 794 795 796 797 798 799

    *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 */
800 801 802 803 804 805
    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);
806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
    /* 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;
}

962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981
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");
982
            sprintf(group_self_id, "%lu", g->self_id);
983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006
        }
    }
    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)
    {
1007
        ssg_member_state_t *member_state, *tmp_ms;
1008 1009 1010 1011 1012

        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
1013
        printf("\tsize: %d\n", group_view->size);
1014
        printf("\tview:\n");
1015
        HASH_ITER(hh, group_view->member_map, member_state, tmp_ms)
1016
        {
Shane Snyder's avatar
Shane Snyder committed
1017
            printf("\t\tid: %20lu\taddr: %s\n", member_state->id,
1018
                member_state->addr_str);
1019
        }
1020 1021 1022 1023
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
1024

1025
    return;
Shane Snyder's avatar
Shane Snyder committed
1026 1027
}

1028 1029 1030
/************************************
 *** SSG internal helper routines ***
 ************************************/
1031

1032 1033 1034 1035
void ssg_apply_membership_update(
    ssg_group_t *g,
    ssg_membership_update_t update)
{
1036 1037
    ssg_member_state_t *member_state;

1038 1039
    if(!ssg_inst || !g) return;

1040 1041
    if (update.type == SSG_MEMBER_REMOVE)
    {
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051
        HASH_FIND(hh, g->view.member_map, &update.member, sizeof(ssg_member_id_t),
            member_state);
        if (member_state)
        {
            HASH_DELETE(hh, g->view.member_map, member_state);
            margo_addr_free(ssg_inst->mid, member_state->addr);
            free(member_state->addr_str);
            free(member_state);
            g->view.size--;
        }
1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
    }
    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(update, g->update_cb_dat);

    return;
}

1065
static ssg_group_descriptor_t * ssg_group_descriptor_create(
1066
    uint64_t name_hash, const char * leader_addr_str, int owner_status)
1067
{
1068
    ssg_group_descriptor_t *descriptor;
1069

1070 1071
    descriptor = malloc(sizeof(*descriptor));
    if (!descriptor) return NULL;
Shane Snyder's avatar
Shane Snyder committed
1072

1073
    /* hash the group name to obtain an 64-bit unique ID */
1074
    descriptor->magic_nr = SSG_MAGIC_NR;
1075
    descriptor->name_hash = name_hash;
1076 1077
    descriptor->addr_str = strdup(leader_addr_str);
    if (!descriptor->addr_str)
Shane Snyder's avatar
Shane Snyder committed
1078
    {
1079
        free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
1080 1081
        return NULL;
    }
1082
    descriptor->owner_status = owner_status;
1083
    descriptor->ref_count = 1;
1084 1085
    return descriptor;
}
1086

1087 1088 1089
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor)
{
Shane Snyder's avatar
Shane Snyder committed
1090 1091
    descriptor->ref_count++;
    return descriptor;
1092 1093
}

1094 1095
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor)
1096
{
1097 1098
    if (descriptor)
    {
Shane Snyder's avatar
Shane Snyder committed
1099 1100 1101 1102 1103 1104 1105 1106 1107
        if(descriptor->ref_count == 1)
        {
            free(descriptor->addr_str);
            free(descriptor);
        }
        else
        {
            descriptor->ref_count--;
        }
1108 1109 1110
    }
    return;
}
1111

1112
static int ssg_group_view_create(
1113 1114 1115
    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)
1116 1117
{
    int i, j, r;
1118 1119 1120 1121
    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;
1122 1123
    const char *self_addr_substr = NULL;
    const char *addr_substr = NULL;
1124 1125
    ssg_member_state_t *tmp_ms;
    hg_return_t hret;
1126
    int aret;
1127
    int sret = SSG_FAILURE;
1128

1129 1130 1131 1132
    if (self_id)
        *self_id = SSG_MEMBER_ID_INVALID;
        
    if ((self_id != NULL && self_addr == HG_ADDR_NULL) || !view) goto fini;
1133 1134 1135

    /* allocate lookup ULTs */
    lookup_ults = malloc(group_size * sizeof(*lookup_ults));
1136 1137
    if (lookup_ults == NULL) goto fini;
    for (i = 0; i < group_size; i++) lookup_ults[i] = ABT_THREAD_NULL;
1138
    lookup_ult_args = malloc(group_size * sizeof(*lookup_ult_args));
1139
    if (lookup_ult_args == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
1140

1141
    if(self_addr)
1142
    {
1143 1144 1145 1146 1147 1148
        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;
1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160

        /* 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;
    }

1161 1162
    /* construct view using ULTs to lookup the address of each group member */
    view->size = group_size;
1163 1164 1165 1166 1167 1168 1169 1170
    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;

1171 1172
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

1173 1174 1175 1176 1177 1178 1179
        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)
1180
        {
1181 1182
            free(tmp_ms);
            goto fini;
1183
        }
1184
        SWIM_MEMBER_STATE_INIT(tmp_ms->swim_state);
1185
        
1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196
        /* 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)
            {
1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207
                hret = margo_addr_dup(ssg_inst->mid, self_addr, &tmp_ms->addr);
                if (hret != HG_SUCCESS)
                {
                    free(tmp_ms->addr_str);
                    free(tmp_ms);
                    goto fini;
                }

                if (self_id)               
                    *self_id = tmp_ms->id;

Shane Snyder's avatar
Shane Snyder committed
1208 1209
                /* XXX: add self state to membership view? or somewhere static? */
                //HASH_ADD(hh, view->member_map, id, sizeof(ssg_member_id_t), tmp_ms);
1210 1211 1212

                /* don't look up our own address, we already know it */
                continue;
1213 1214 1215 1216
            }
        }

        /* XXX limit outstanding lookups to some max */
1217 1218
        lookup_ult_args[j].view = view;
        lookup_ult_args[j].member_state = tmp_ms;
1219 1220
        ABT_pool pool;
        margo_get_handler_pool(ssg_inst->mid, &pool);
1221 1222
        aret = ABT_thread_create(pool, &ssg_group_lookup_ult,
            &lookup_ult_args[j], ABT_THREAD_ATTR_NULL,
1223 1224 1225
            &lookup_ults[j]);
        if (aret != ABT_SUCCESS)
        {
1226 1227
            free(tmp_ms->addr_str);
            free(tmp_ms);
1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
            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;
1240
        if (aret != ABT_SUCCESS) goto fini;
1241 1242
        else if (lookup_ult_args[i].out != HG_SUCCESS)
        {
1243
            fprintf(stderr, "Error: SSG unable to lookup HG address for member %lu"
1244 1245 1246
                "(err=%d)\n", lookup_ult_args[i].member_state->id,
                lookup_ult_args[i].out);
            goto fini;
1247 1248 1249
        }
    }

1250 1251 1252
    /* clean exit */
    sret = SSG_SUCCESS;

1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
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);
1268
    free(self_addr_str);
1269 1270

    return sret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
1271
}
1272

1273 1274 1275
static ssg_member_id_t ssg_gen_member_id(
    const char * addr_str)
{
1276 1277 1278 1279 1280 1281 1282 1283 1284
    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;
1285 1286 1287
}


1288 1289 1290 1291 1292
static void ssg_group_lookup_ult(
    void * arg)
{
    struct ssg_group_lookup_ult_args *l = arg;

1293
    /* XXX: should be a timeout here? */
1294 1295
    l->out = margo_addr_lookup(ssg_inst->mid, l->member_state->addr_str,
        &l->member_state->addr);
1296 1297 1298 1299 1300
    if (l->out == HG_SUCCESS)
    {
        HASH_ADD(hh, l->view->member_map, id, sizeof(ssg_member_id_t),
            l->member_state);
    }
1301 1302 1303 1304
    else
    {
        /* XXX */
    }
1305

1306 1307 1308 1309 1310
    return;
}

static void ssg_group_view_destroy(
    ssg_group_view_t * view)
1311
{
1312
    ssg_member_state_t *state, *tmp;
1313

1314
    /* destroy state for all group members */
Shane Snyder's avatar
Shane Snyder committed
1315
    HASH_ITER(hh, view->member_map, state, tmp)
1316
    {
1317 1318 1319 1320
        HASH_DEL(view->member_map, state);
        free(state->addr_str);
        margo_addr_free(ssg_inst->mid,  state->addr);
        free(state);
1321
    }
1322
    view->member_map = NULL;
1323 1324 1325 1326 1327 1328 1329

    return;
}

static void ssg_group_destroy_internal(
    ssg_group_t * g)
{
1330 1331
    /* TODO: send a leave message to the group ? */

Shane Snyder's avatar
Shane Snyder committed
1332 1333 1334
    /* free up SWIM state */
    if(g->swim_ctx)
        swim_finalize(g->swim_ctx);
1335 1336

    /* destroy group state */
1337
    ssg_group_view_destroy(&g->view);
Shane Snyder's avatar
Shane Snyder committed
1338
    g->descriptor->owner_status = SSG_OWNER_IS_UNASSOCIATED;
1