ssg.c 33.4 KB
Newer Older
1 2 3 4 5 6
/*
 * Copyright (c) 2016 UChicago Argonne, LLC
 *
 * See COPYRIGHT in top-level directory.
 */

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

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

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

26
#include "ssg.h"
Shane Snyder's avatar
Shane Snyder committed
27 28 29
#ifdef SSG_HAVE_MPI
#include "ssg-mpi.h"
#endif
30
#include "ssg-internal.h"
31
#ifdef SSG_USE_SWIM_FD
32 33
#include "swim-fd/swim-fd.h"
#endif
34
#include "uthash.h"
35

36 37 38 39 40 41 42 43
/* arguments for group lookup ULTs */
struct ssg_group_lookup_ult_args
{
    ssg_member_state_t *member_state;
    hg_return_t out;
};
static void ssg_group_lookup_ult(void * arg);

44
/* SSG helper routine prototypes */
45
static ssg_group_descriptor_t * ssg_group_descriptor_create(
46
    uint64_t name_hash, const char * leader_addr_str, int owner_status);
47 48 49 50 51 52 53 54 55 56 57 58 59 60 61
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(
    const char * const group_addr_strs[], const char * self_addr_str,
    int group_size, ssg_group_view_t * view, ssg_member_id_t * self_id);
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

Shane Snyder's avatar
Shane Snyder committed
126
ssg_group_id_t ssg_group_create(
127 128
    const char * group_name,
    const char * const group_addr_strs[],
129 130 131
    int group_size,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
132
{
133 134
    uint32_t upper, lower;
    uint64_t name_hash;
Shane Snyder's avatar
Shane Snyder committed
135
    hg_addr_t self_addr = HG_ADDR_NULL;
136
    char *self_addr_str = NULL;
137
    hg_size_t self_addr_str_size = 0;
138
    ssg_group_descriptor_t *tmp_descriptor;
139
    ssg_group_t *g = NULL;
140
    hg_return_t hret;
141
    int sret;
Shane Snyder's avatar
Shane Snyder committed
142
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
143

144
    if (!ssg_inst) return group_id;
145

146 147 148
    ssg_hashlittle2(group_name, strlen(group_name), &lower, &upper);
    name_hash = lower + (((uint64_t)upper)<<32);

149
    /* generate a unique ID for this group  */
150 151
    tmp_descriptor = ssg_group_descriptor_create(name_hash, group_addr_strs[0],
        SSG_OWNER_IS_MEMBER);
152
    if (tmp_descriptor == NULL) return group_id;
153

Shane Snyder's avatar
Shane Snyder committed
154
    /* make sure we aren't re-creating an existing group */
155
    HASH_FIND(hh, ssg_inst->group_table, &tmp_descriptor->name_hash,
Shane Snyder's avatar
Shane Snyder committed
156
        sizeof(uint64_t), g);
157 158 159 160 161
    if (g)
    {
        g = NULL;
        goto fini;
    }
162

163
    /* get my address */
164
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
165
    if (hret != HG_SUCCESS) goto fini;
166
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
167
    if (hret != HG_SUCCESS) goto fini;
168
    self_addr_str = malloc(self_addr_str_size);
169
    if (self_addr_str == NULL) goto fini;
170
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
171 172
    if (hret != HG_SUCCESS) goto fini;

173 174 175 176 177 178 179
    /* 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;
180 181
    g->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
182 183 184 185 186

    /* initialize the group view */
    sret = ssg_group_view_create(group_addr_strs, self_addr_str, group_size,
        &g->view, &g->self_id);
    if (sret != SSG_SUCCESS) goto fini;
187
    if (g->self_id == SSG_MEMBER_ID_INVALID)
188
    {
189
        /* if unable to resolve my rank within the group, error out */
190 191 192 193
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
    }
194
    g->view.member_states[g->self_id].addr = self_addr;
195

196
#ifdef SSG_USE_SWIM_FD
197
    /* initialize swim failure detector */
198 199
    // TODO: we should probably barrier or sync somehow to avoid rpc failures
    // due to timing skew of different ranks initializing swim
200
    g->fd_ctx = (void *)swim_init(g, 1);
201
    if (g->fd_ctx == NULL) goto fini;
202 203
#endif

Shane Snyder's avatar
Shane Snyder committed
204 205 206
    /* everything successful -- set the output group identifier, which is just
     * an opaque pointer to the group descriptor structure
     */
207 208
    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
209

210 211 212 213
    /* 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
214
    SSG_DEBUG(g, "group create successful (size=%d)\n", group_size);
215

216
    /* don't free these pointers on success */
217
    self_addr = HG_ADDR_NULL;
218 219
    g = NULL;
fini:
220
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
Shane Snyder's avatar
Shane Snyder committed
221
    free(self_addr_str);
222 223
    if (g)
    {
224
        ssg_group_view_destroy(&g->view);
225
        free(g->name);
226 227
        free(g);
    }
228 229
    if (group_id == SSG_GROUP_ID_NULL)
        ssg_group_descriptor_free(tmp_descriptor);
230

Shane Snyder's avatar
Shane Snyder committed
231
    return group_id;
232 233
}

Shane Snyder's avatar
Shane Snyder committed
234
ssg_group_id_t ssg_group_create_config(
235
    const char * group_name,
236 237 238
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
239 240 241 242
{
    int fd;
    struct stat st;
    char *rd_buf = NULL;
243
    ssize_t rd_buf_size;
244
    char *tok;
245 246
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
247 248
    int ret;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
249
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271

    /* 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 */
272 273
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
274 275 276 277 278
    {
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
279
    rd_buf[rd_buf_size]='\0';
280 281 282 283 284

    /* 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
285 286
    if (tok == NULL) goto fini;

287
    /* build up the address buffer */
288 289
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
290 291
    do
    {
292 293 294
        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
295 296 297
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
298
    if (addr_str_buf_len != rd_buf_size)
299 300
    {
        /* adjust buffer size if our initial guess was wrong */
301
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
302
        if (tmp == NULL) goto fini;
303
        addr_str_buf = tmp;
304
    }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
305

306
    /* set up address string array for group members */
307
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
308
    if (!addr_strs) goto fini;
309

310
    /* invoke the generic group create routine using our list of addrs */
311 312
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
313 314

fini:
315
    /* cleanup before returning */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
316
    if (fd != -1) close(fd);
317
    free(rd_buf);
318
    free(addr_str_buf);
319 320
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
321
    return group_id;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
322 323
}

Shane Snyder's avatar
Shane Snyder committed
324
#ifdef SSG_HAVE_MPI
Shane Snyder's avatar
Shane Snyder committed
325
ssg_group_id_t ssg_group_create_mpi(
326
    const char * group_name,
327 328 329
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
330
{
331
    int i;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
332
    hg_addr_t self_addr = HG_ADDR_NULL;
333
    char *self_addr_str = NULL;
334 335
    hg_size_t self_addr_str_size = 0;
    int self_addr_str_size_int = 0; /* for mpi-friendly conversion */
336
    char *addr_str_buf = NULL;
337 338 339 340
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
341
    hg_return_t hret;
Shane Snyder's avatar
Shane Snyder committed
342
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
343

344 345
    if (!ssg_inst) goto fini;

346
    /* get my address */
347
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
348
    if (hret != HG_SUCCESS) goto fini;
349
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
Shane Snyder's avatar
Shane Snyder committed
350
    if (hret != HG_SUCCESS) goto fini;
351
    self_addr_str = malloc(self_addr_str_size);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
352
    if (self_addr_str == NULL) goto fini;
353
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
354
    if (hret != HG_SUCCESS) goto fini;
355
    self_addr_str_size_int = (int)self_addr_str_size; /* null char included in call */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
356

357
    /* gather the buffer sizes */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
358 359 360 361
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
362
    sizes[comm_rank] = self_addr_str_size_int;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
363 364
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

365 366 367
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
     */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
368 369 370
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
371
    for (i = 1; i < comm_size+1; i++)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
372 373
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];

374
    /* allgather the addresses */
375 376
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
377
    MPI_Allgatherv(self_addr_str, self_addr_str_size_int, MPI_BYTE,
378
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
379

380
    /* set up address string array for group members */
381
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
382 383 384
    if (!addr_strs) goto fini;

    /* invoke the generic group create routine using our list of addrs */
385 386
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
387 388

fini:
389
    /* cleanup before returning */
390
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
391
    free(self_addr_str);
392 393 394
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
395 396
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
397
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
398 399 400
}
#endif

401 402
int ssg_group_destroy(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
403
{
Shane Snyder's avatar
Shane Snyder committed
404
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
405
    ssg_group_t *g;
406

Shane Snyder's avatar
Shane Snyder committed
407
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
408
        return SSG_FAILURE;
409

410 411 412 413 414 415
    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;
    }

416
    /* find the group structure and destroy it */
Shane Snyder's avatar
Shane Snyder committed
417 418
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
419 420 421 422 423
    if (!g)
    {
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
    }
424
    HASH_DELETE(hh, ssg_inst->group_table, g);
425
    ssg_group_destroy_internal(g);
426

427
    return SSG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
428 429
}

430 431 432
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
433
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
434 435 436 437 438 439
    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;
440

441
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
442

443 444 445 446 447 448 449 450 451 452 453
    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;
    }
454

455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
    /* 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 */
    sret = ssg_group_view_create(addr_strs, NULL, group_size, &ag->view, NULL);
    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)
    {
492
        ssg_group_view_destroy(&ag->view);
493 494
        free(ag->name);
        free(ag);
495
        ssg_group_descriptor_free(ag->descriptor);
496 497 498
    }

    return sret;
499 500 501 502 503
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
    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);

525 526 527
    return SSG_SUCCESS;
}

528 529 530
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
531

532
ssg_member_id_t ssg_get_group_self_id(
533
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
534
{
Shane Snyder's avatar
Shane Snyder committed
535
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
536 537
    ssg_group_t *g;

Shane Snyder's avatar
Shane Snyder committed
538
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
539
        return SSG_MEMBER_ID_INVALID;
540

541 542 543 544
    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");
545
        return SSG_MEMBER_ID_INVALID;
546 547
    }

Shane Snyder's avatar
Shane Snyder committed
548 549
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
550
    if (!g)
551
        return SSG_MEMBER_ID_INVALID;
552 553

    return g->self_id;
Shane Snyder's avatar
Shane Snyder committed
554 555
}

556 557
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
558
{
Shane Snyder's avatar
Shane Snyder committed
559
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
560
    ssg_group_view_t *group_view = NULL;
561

Shane Snyder's avatar
Shane Snyder committed
562
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
563 564
        return 0;

565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586
    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");
587
        return 0;
588
    }
589

590 591 592 593
    if (group_view)
        return group_view->size;
    else
        return 0;
Shane Snyder's avatar
Shane Snyder committed
594 595
}

596 597 598
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
599
{
Shane Snyder's avatar
Shane Snyder committed
600
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
601
    ssg_group_view_t *group_view = NULL;
602

Shane Snyder's avatar
Shane Snyder committed
603
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
604 605
        return HG_ADDR_NULL;

606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641
    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)
    {
        /* XXX for now we assume member ids are dense ranks and error out
         * if they are not within allowable range for the group view size
         */
        if (member_id >= group_view->size)
            return HG_ADDR_NULL;

        return group_view->member_states[member_id].addr;
    }
    else
642
        return HG_ADDR_NULL;
643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659
}

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
660
    descriptor = SSG_GROUP_ID_NULL;
661 662 663
    return;
}

664 665 666 667 668 669 670 671 672
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);
}


673 674 675 676 677 678 679
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;
680
    char *gid_buf, *p; 
681 682 683 684 685 686 687 688 689 690 691 692 693

    *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 */
694 695 696 697 698 699
    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);
700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
    /* 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;
}

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
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");
            sprintf(group_self_id, "%"PRIu64, g->self_id);
        }
    }
    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)
    {
        unsigned int i;

        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
907
        printf("\tsize: %d\n", group_view->size);
908 909
        printf("\tview:\n");
        for (i = 0; i < group_view->size; i++)
910 911 912 913 914
        {
            if (group_view->member_states[i].is_member)
                printf("\t\tid: %d\taddr: %s\n", i,
                    group_view->member_states[i].addr_str);
        }
915 916 917 918
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
919

920
    return;
Shane Snyder's avatar
Shane Snyder committed
921 922
}

923 924 925
/************************************
 *** SSG internal helper routines ***
 ************************************/
926

927 928 929 930
void ssg_apply_membership_update(
    ssg_group_t *g,
    ssg_membership_update_t update)
{
931 932
    if(!ssg_inst || !g) return;

933 934
    if (update.type == SSG_MEMBER_REMOVE)
    {
935
        margo_addr_free(ssg_inst->mid, g->view.member_states[update.member].addr);
936 937
        free(g->view.member_states[update.member].addr_str);
        g->view.member_states[update.member].addr_str = NULL;
938
        g->view.member_states[update.member].is_member = 0;
939
        /* XXX: need to update size ... g->view.size--; */
940 941 942 943 944 945 946 947 948 949 950 951 952
    }
    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;
}

953
static ssg_group_descriptor_t * ssg_group_descriptor_create(
954
    uint64_t name_hash, const char * leader_addr_str, int owner_status)
955
{
956
    ssg_group_descriptor_t *descriptor;
957

958 959
    descriptor = malloc(sizeof(*descriptor));
    if (!descriptor) return NULL;
Shane Snyder's avatar
Shane Snyder committed
960

961
    /* hash the group name to obtain an 64-bit unique ID */
962
    descriptor->magic_nr = SSG_MAGIC_NR;
963
    descriptor->name_hash = name_hash;
964 965
    descriptor->addr_str = strdup(leader_addr_str);
    if (!descriptor->addr_str)
Shane Snyder's avatar
Shane Snyder committed
966
    {
967
        free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
968 969
        return NULL;
    }
970
    descriptor->owner_status = owner_status;
971
    descriptor->ref_count = 1;
972 973
    return descriptor;
}
974

975 976 977
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor)
{
Shane Snyder's avatar
Shane Snyder committed
978 979
    descriptor->ref_count++;
    return descriptor;
980 981
}

982 983
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor)
984
{
985 986
    if (descriptor)
    {
Shane Snyder's avatar
Shane Snyder committed
987 988 989 990 991 992 993 994 995
        if(descriptor->ref_count == 1)
        {
            free(descriptor->addr_str);
            free(descriptor);
        }
        else
        {
            descriptor->ref_count--;
        }
996 997 998
    }
    return;
}
999

1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
static int ssg_group_view_create(
    const char * const group_addr_strs[], const char * self_addr_str,
    int group_size, ssg_group_view_t * view, ssg_member_id_t * self_id)
{
    int i, j, r;
    ABT_thread *lookup_ults;
    struct ssg_group_lookup_ult_args *lookup_ult_args;
    const char *self_addr_substr = NULL;
    const char *addr_substr = NULL;
    int aret;
    int sret = SSG_SUCCESS;

    if (!view || (self_id != NULL && self_addr_str == NULL)) return SSG_FAILURE;

    /* allocate lookup ULTs */
    lookup_ults = malloc(group_size * sizeof(*lookup_ults));
    if (lookup_ults == NULL) return SSG_FAILURE;
    lookup_ult_args = malloc(group_size * sizeof(*lookup_ult_args));
    if (lookup_ult_args == NULL)
1019
    {
1020 1021
        free(lookup_ults);
        return SSG_FAILURE;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
1022
    }
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032

    /* allocate and initialize the view */
    view->size = group_size;
    view->member_states = malloc(group_size * sizeof(*view->member_states));
    if (!view->member_states)
    {
        free(lookup_ults);
        free(lookup_ult_args);
        return SSG_FAILURE;
    }
Shane Snyder's avatar
Shane Snyder committed
1033

1034 1035 1036 1037
    for (i = 0; i < group_size; i++)
    {
        view->member_states[i].addr_str = NULL;
        view->member_states[i].addr = HG_ADDR_NULL;
1038
        view->member_states[i].is_member = 0;
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053
        lookup_ults[i] = ABT_THREAD_NULL;
    }

    if (self_addr_str && self_id)
    {
        /* 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;

1054
        *self_id = SSG_MEMBER_ID_INVALID;
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
    }

    /* kickoff ULTs to lookup the address of each group member */
    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;

1066 1067
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093
        view->member_states[j].addr_str = strdup(group_addr_strs[j]);
        if (!view->member_states[j].addr_str)
        {
            sret = SSG_FAILURE;
            goto fini;   
        }

        /* 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)
            {
                
                *self_id = j;
                continue; /* don't look up our own address, we already know it */
            }
        }

        /* XXX limit outstanding lookups to some max */
        lookup_ult_args[j].member_state = &view->member_states[j];
1094 1095 1096
        ABT_pool pool;
        margo_get_handler_pool(ssg_inst->mid, &pool);
        aret = ABT_thread_create(pool,
1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125
            &ssg_group_lookup_ult, &lookup_ult_args[j], ABT_THREAD_ATTR_NULL,
            &lookup_ults[j]);
        if (aret != ABT_SUCCESS)
        {
            sret = SSG_FAILURE;
            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;
        if (aret != ABT_SUCCESS)
        {
            sret = SSG_FAILURE;
            break;
        }
        else if (lookup_ult_args[i].out != HG_SUCCESS)
        {
            fprintf(stderr, "Error: SSG unable to lookup HG address for member %d"
                "(err=%d)\n", i, lookup_ult_args[i].out);
            sret = SSG_FAILURE;
            break;
        }
1126
        view->member_states[i].is_member = 1;
1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
    }

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);

    return sret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
1146
}
1147

1148 1149 1150 1151 1152
static void ssg_group_lookup_ult(
    void * arg)
{
    struct ssg_group_lookup_ult_args *l = arg;

1153
    /* XXX: should be a timeout here? */
1154 1155 1156 1157 1158 1159 1160
    l->out = margo_addr_lookup(ssg_inst->mid, l->member_state->addr_str,
        &l->member_state->addr);
    return;
}

static void ssg_group_view_destroy(
    ssg_group_view_t * view)
1161
{
Shane Snyder's avatar
Shane Snyder committed
1162
    unsigned int i;
1163

1164 1165 1166
    for (i = 0; i < view->size; i++)
    {
        free(view->member_states[i].addr_str);
1167
        margo_addr_free(ssg_inst->mid, view->member_states[i].addr);
1168 1169 1170 1171 1172 1173 1174 1175 1176
    }
    free(view->member_states);

    return;
}

static void ssg_group_destroy_internal(
    ssg_group_t * g)
{
1177 1178 1179 1180 1181 1182 1183 1184 1185
    /* TODO: send a leave message to the group ? */

#ifdef SSG_USE_SWIM_FD
    /* free up failure detector state */
    if(g->fd_ctx)
        swim_finalize(g->fd_ctx);
#endif

    /* destroy group state */
1186
    ssg_group_view_destroy(&g->view);
Shane Snyder's avatar
Shane Snyder committed
1187
    g->descriptor->owner_status = SSG_OWNER_IS_UNASSOCIATED;
1188 1189
    ssg_group_descriptor_free(g->descriptor);
    free(g->name);
1190 1191
    free(g);

1192 1193 1194 1195 1196 1197
    return;
}

static void ssg_attached_group_destroy(
    ssg_attached_group_t * ag)
{
Shane Snyder's avatar
Shane Snyder committed
1198 1199
    ssg_group_view_destroy(&ag->view);
    ag->descriptor->owner_status = SSG_OWNER_IS_UNASSOCIATED;
1200 1201 1202 1203 1204 1205 1206 1207 1208
    ssg_group_descriptor_free(ag->descriptor);
    free(ag->name);
    free(ag);
    return;
}

static const char ** ssg_addr_str_buf_to_list(
    const char * buf, int num_addrs)
{
1209
    int i;
1210 1211 1212 1213
    const char **ret = malloc(num_addrs * sizeof(*ret));
    if (ret == NULL) return NULL;

    ret[0] = buf;
1214
    for (i = 1; i < num_addrs; i++)
1215 1216 1217 1218 1219
    {
        const char * a = ret[i-1];
        ret[i] = a + strlen(a) + 1;
    }
    return ret;
1220
}