ssg.c 28.7 KB
Newer Older
1 2 3 4 5 6
/*
 * Copyright (c) 2016 UChicago Argonne, LLC
 *
 * See COPYRIGHT in top-level directory.
 */

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

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

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

26
#include "ssg.h"
Shane Snyder's avatar
Shane Snyder committed
27 28 29
#ifdef SSG_HAVE_MPI
#include "ssg-mpi.h"
#endif
30
#include "ssg-internal.h"
31
#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(
Shane Snyder's avatar
Shane Snyder committed
46
    const char * name, const char * leader_addr_str);
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
{
Shane Snyder's avatar
Shane Snyder committed
133
    hg_addr_t self_addr = HG_ADDR_NULL;
134
    char *self_addr_str = NULL;
135
    hg_size_t self_addr_str_size = 0;
136
    ssg_group_descriptor_t *tmp_descriptor;
137
    ssg_group_t *g = NULL;
138
    hg_return_t hret;
139
    int sret;
Shane Snyder's avatar
Shane Snyder committed
140
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
141

142
    if (!ssg_inst) return group_id;
143 144

    /* generate a unique ID for this group  */
145 146
    tmp_descriptor = ssg_group_descriptor_create(group_name, group_addr_strs[0]);
    if (tmp_descriptor == NULL) return group_id;
147

Shane Snyder's avatar
Shane Snyder committed
148
    /* make sure we aren't re-creating an existing group */
149
    HASH_FIND(hh, ssg_inst->group_table, &tmp_descriptor->name_hash,
Shane Snyder's avatar
Shane Snyder committed
150
        sizeof(uint64_t), g);
151 152 153 154 155
    if (g)
    {
        g = NULL;
        goto fini;
    }
156

157
    /* get my address */
158
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
159
    if (hret != HG_SUCCESS) goto fini;
160
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
161
    if (hret != HG_SUCCESS) goto fini;
162
    self_addr_str = malloc(self_addr_str_size);
163
    if (self_addr_str == NULL) goto fini;
164
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
165 166
    if (hret != HG_SUCCESS) goto fini;

167 168 169 170 171 172 173
    /* 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;
174 175
    g->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
176 177 178 179 180

    /* 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;
181
    if (g->self_id == SSG_MEMBER_ID_INVALID)
182
    {
183
        /* if unable to resolve my rank within the group, error out */
184 185 186 187
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
    }
188
    g->view.member_states[g->self_id].addr = self_addr;
189

190
#ifdef SSG_USE_SWIM_FD
191
    /* initialize swim failure detector */
192 193
    // TODO: we should probably barrier or sync somehow to avoid rpc failures
    // due to timing skew of different ranks initializing swim
194
    g->fd_ctx = (void *)swim_init(g, 1);
195
    if (g->fd_ctx == NULL) goto fini;
196 197
#endif

Shane Snyder's avatar
Shane Snyder committed
198 199 200
    /* everything successful -- set the output group identifier, which is just
     * an opaque pointer to the group descriptor structure
     */
201 202
    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
203

204 205 206 207
    /* 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
208
    SSG_DEBUG(g, "group create successful (size=%d)\n", group_size);
209

210
    /* don't free these pointers on success */
211
    self_addr = HG_ADDR_NULL;
212 213
    g = NULL;
fini:
214
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
Shane Snyder's avatar
Shane Snyder committed
215
    free(self_addr_str);
216 217
    if (g)
    {
218
        ssg_group_view_destroy(&g->view);
219
        free(g->name);
220 221
        free(g);
    }
222 223
    if (group_id == SSG_GROUP_ID_NULL)
        ssg_group_descriptor_free(tmp_descriptor);
224

Shane Snyder's avatar
Shane Snyder committed
225
    return group_id;
226 227
}

Shane Snyder's avatar
Shane Snyder committed
228
ssg_group_id_t ssg_group_create_config(
229
    const char * group_name,
230 231 232
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
233 234 235 236
{
    int fd;
    struct stat st;
    char *rd_buf = NULL;
237
    ssize_t rd_buf_size;
238
    char *tok;
239 240
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
241 242
    int ret;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
243
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265

    /* 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 */
266 267
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
268 269 270 271 272
    {
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
    }
273
    rd_buf[rd_buf_size]='\0';
274 275 276 277 278

    /* 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
279 280
    if (tok == NULL) goto fini;

281
    /* build up the address buffer */
282 283
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
284 285
    do
    {
286 287 288
        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
289 290 291
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
292
    if (addr_str_buf_len != rd_buf_size)
293 294
    {
        /* adjust buffer size if our initial guess was wrong */
295
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
296
        if (tmp == NULL) goto fini;
297
        addr_str_buf = tmp;
298
    }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
299

300
    /* set up address string array for group members */
301
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
302
    if (!addr_strs) goto fini;
303

304
    /* invoke the generic group create routine using our list of addrs */
305 306
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
307 308

fini:
309
    /* cleanup before returning */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
310
    if (fd != -1) close(fd);
311
    free(rd_buf);
312
    free(addr_str_buf);
313 314
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
315
    return group_id;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
316 317
}

Shane Snyder's avatar
Shane Snyder committed
318
#ifdef SSG_HAVE_MPI
Shane Snyder's avatar
Shane Snyder committed
319
ssg_group_id_t ssg_group_create_mpi(
320
    const char * group_name,
321 322 323
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
324
{
Jonathan Jenkins's avatar
Jonathan Jenkins committed
325
    hg_addr_t self_addr = HG_ADDR_NULL;
326
    char *self_addr_str = NULL;
327 328
    hg_size_t self_addr_str_size = 0;
    int self_addr_str_size_int = 0; /* for mpi-friendly conversion */
329
    char *addr_str_buf = NULL;
330 331 332 333
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
334
    hg_return_t hret;
Shane Snyder's avatar
Shane Snyder committed
335
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
336

337 338
    if (!ssg_inst) goto fini;

339
    /* get my address */
340
    hret = margo_addr_self(ssg_inst->mid, &self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
341
    if (hret != HG_SUCCESS) goto fini;
342
    hret = margo_addr_to_string(ssg_inst->mid, NULL, &self_addr_str_size, self_addr);
Shane Snyder's avatar
Shane Snyder committed
343
    if (hret != HG_SUCCESS) goto fini;
344
    self_addr_str = malloc(self_addr_str_size);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
345
    if (self_addr_str == NULL) goto fini;
346
    hret = margo_addr_to_string(ssg_inst->mid, self_addr_str, &self_addr_str_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
347
    if (hret != HG_SUCCESS) goto fini;
348
    self_addr_str_size_int = (int)self_addr_str_size; /* null char included in call */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
349

350
    /* gather the buffer sizes */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
351 352 353 354
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
355
    sizes[comm_rank] = self_addr_str_size_int;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
356 357
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

358 359 360
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
     */
Jonathan Jenkins's avatar
Jonathan Jenkins committed
361 362 363 364 365 366
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
    for (int i = 1; i < comm_size+1; i++)
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];

367
    /* allgather the addresses */
368 369
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
370
    MPI_Allgatherv(self_addr_str, self_addr_str_size_int, MPI_BYTE,
371
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
372

373
    /* set up address string array for group members */
374
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
375 376 377
    if (!addr_strs) goto fini;

    /* invoke the generic group create routine using our list of addrs */
378 379
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
380 381

fini:
382
    /* cleanup before returning */
383
    if (self_addr != HG_ADDR_NULL) margo_addr_free(ssg_inst->mid, self_addr);
384
    free(self_addr_str);
385 386 387
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
388 389
    free(addr_strs);

Shane Snyder's avatar
Shane Snyder committed
390
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
391 392 393
}
#endif

394 395
int ssg_group_destroy(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
396
{
Shane Snyder's avatar
Shane Snyder committed
397
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
398
    ssg_group_t *g;
399

Shane Snyder's avatar
Shane Snyder committed
400
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
401
        return SSG_FAILURE;
402

403 404 405 406 407 408
    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;
    }

409
    /* find the group structure and destroy it */
Shane Snyder's avatar
Shane Snyder committed
410 411
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
412 413 414 415 416
    if (!g)
    {
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
    }
417
    HASH_DELETE(hh, ssg_inst->group_table, g);
418
    ssg_group_destroy_internal(g);
419

420
    return SSG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
421 422
}

423 424 425
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
426
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
427 428 429 430 431 432
    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;
433

434
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
435

436 437 438 439 440 441 442 443 444 445 446
    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;
    }
447

448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484
    /* 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)
    {
485
        ssg_group_view_destroy(&ag->view);
486 487
        free(ag->name);
        free(ag);
488
        ssg_group_descriptor_free(ag->descriptor);
489 490 491
    }

    return sret;
492 493 494 495 496
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
    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);

518 519 520
    return SSG_SUCCESS;
}

521 522 523
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
524

525
ssg_member_id_t ssg_get_group_self_id(
526
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
527
{
Shane Snyder's avatar
Shane Snyder committed
528
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
529 530
    ssg_group_t *g;

Shane Snyder's avatar
Shane Snyder committed
531
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
532
        return SSG_MEMBER_ID_INVALID;
533

534 535 536 537
    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");
538
        return SSG_MEMBER_ID_INVALID;
539 540
    }

Shane Snyder's avatar
Shane Snyder committed
541 542
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
543
    if (!g)
544
        return SSG_MEMBER_ID_INVALID;
545 546

    return g->self_id;
Shane Snyder's avatar
Shane Snyder committed
547 548
}

549 550
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
551
{
Shane Snyder's avatar
Shane Snyder committed
552
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
553
    ssg_group_view_t *group_view = NULL;
554

Shane Snyder's avatar
Shane Snyder committed
555
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
556 557
        return 0;

558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579
    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");
580
        return 0;
581
    }
582

583 584 585 586
    if (group_view)
        return group_view->size;
    else
        return 0;
Shane Snyder's avatar
Shane Snyder committed
587 588
}

589 590 591
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
592
{
Shane Snyder's avatar
Shane Snyder committed
593
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
594
    ssg_group_view_t *group_view = NULL;
595

Shane Snyder's avatar
Shane Snyder committed
596
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL)
597 598
        return HG_ADDR_NULL;

599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
    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
635
        return HG_ADDR_NULL;
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
}

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
653
    descriptor = SSG_GROUP_ID_NULL;
654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707
    return;
}

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
708
        printf("\tsize: %d\n", group_view->size);
709 710
        printf("\tview:\n");
        for (i = 0; i < group_view->size; i++)
711 712 713 714 715
        {
            if (group_view->member_states[i].is_member)
                printf("\t\tid: %d\taddr: %s\n", i,
                    group_view->member_states[i].addr_str);
        }
716 717 718 719
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
720

721
    return;
Shane Snyder's avatar
Shane Snyder committed
722 723
}

724 725 726
/************************************
 *** SSG internal helper routines ***
 ************************************/
727

728 729 730 731
void ssg_apply_membership_update(
    ssg_group_t *g,
    ssg_membership_update_t update)
{
732 733
    if(!ssg_inst || !g) return;

734 735
    if (update.type == SSG_MEMBER_REMOVE)
    {
736
        margo_addr_free(ssg_inst->mid, g->view.member_states[update.member].addr);
737 738
        free(g->view.member_states[update.member].addr_str);
        g->view.member_states[update.member].addr_str = NULL;
739
        g->view.member_states[update.member].is_member = 0;
740
        /* XXX: need to update size ... g->view.size--; */
741 742 743 744 745 746 747 748 749 750 751 752 753
    }
    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;
}

754
static ssg_group_descriptor_t * ssg_group_descriptor_create(
Shane Snyder's avatar
Shane Snyder committed
755
    const char * name, const char * leader_addr_str)
756
{
757
    ssg_group_descriptor_t *descriptor;
758 759
    uint32_t upper, lower;

760 761
    descriptor = malloc(sizeof(*descriptor));
    if (!descriptor) return NULL;
Shane Snyder's avatar
Shane Snyder committed
762

763 764
    /* hash the group name to obtain an 64-bit unique ID */
    ssg_hashlittle2(name, strlen(name), &lower, &upper);
765 766 767 768
    descriptor->magic_nr = SSG_MAGIC_NR;
    descriptor->name_hash = lower + (((uint64_t)upper)<<32);
    descriptor->addr_str = strdup(leader_addr_str);
    if (!descriptor->addr_str)
Shane Snyder's avatar
Shane Snyder committed
769
    {
770
        free(descriptor);
Shane Snyder's avatar
Shane Snyder committed
771 772
        return NULL;
    }
773
    descriptor->owner_status = SSG_OWNER_IS_MEMBER;
774
    descriptor->ref_count = 1;
775 776
    return descriptor;
}
777

778 779 780
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor)
{
Shane Snyder's avatar
Shane Snyder committed
781 782
    descriptor->ref_count++;
    return descriptor;
783 784
}

785 786
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor)
787
{
788 789
    if (descriptor)
    {
Shane Snyder's avatar
Shane Snyder committed
790 791 792 793 794 795 796 797 798
        if(descriptor->ref_count == 1)
        {
            free(descriptor->addr_str);
            free(descriptor);
        }
        else
        {
            descriptor->ref_count--;
        }
799 800 801
    }
    return;
}
802

803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821
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)
822
    {
823 824
        free(lookup_ults);
        return SSG_FAILURE;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
825
    }
826 827 828 829 830 831 832 833 834 835

    /* 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
836

837 838 839 840
    for (i = 0; i < group_size; i++)
    {
        view->member_states[i].addr_str = NULL;
        view->member_states[i].addr = HG_ADDR_NULL;
841
        view->member_states[i].is_member = 0;
842 843 844 845 846 847 848 849 850 851 852 853 854 855 856
        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;

857
        *self_id = SSG_MEMBER_ID_INVALID;
858 859 860 861 862 863 864 865 866 867 868
    }

    /* 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;

869 870
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

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
        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];
        aret = ABT_thread_create(*margo_get_handler_pool(ssg_inst->mid),
            &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;
        }
927
        view->member_states[i].is_member = 1;
928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946
    }

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
947
}
948

949 950 951 952 953
static void ssg_group_lookup_ult(
    void * arg)
{
    struct ssg_group_lookup_ult_args *l = arg;

954
    /* XXX: should be a timeout here? */
955 956 957 958 959 960 961
    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)
962
{
Shane Snyder's avatar
Shane Snyder committed
963
    unsigned int i;
964

965 966 967
    for (i = 0; i < view->size; i++)
    {
        free(view->member_states[i].addr_str);
968
        margo_addr_free(ssg_inst->mid, view->member_states[i].addr);
969 970 971 972 973 974 975 976 977
    }
    free(view->member_states);

    return;
}

static void ssg_group_destroy_internal(
    ssg_group_t * g)
{
978 979 980 981 982 983 984 985 986
    /* 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 */
987
    ssg_group_view_destroy(&g->view);
Shane Snyder's avatar
Shane Snyder committed
988
    g->descriptor->owner_status = SSG_OWNER_IS_UNASSOCIATED;
989 990
    ssg_group_descriptor_free(g->descriptor);
    free(g->name);
991 992
    free(g);

993 994 995 996 997 998
    return;
}

static void ssg_attached_group_destroy(
    ssg_attached_group_t * ag)
{
Shane Snyder's avatar
Shane Snyder committed
999 1000
    ssg_group_view_destroy(&ag->view);
    ag->descriptor->owner_status = SSG_OWNER_IS_UNASSOCIATED;
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019
    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)
{
    const char **ret = malloc(num_addrs * sizeof(*ret));
    if (ret == NULL) return NULL;

    ret[0] = buf;
    for (int i = 1; i < num_addrs; i++)
    {
        const char * a = ret[i-1];
        ret[i] = a + strlen(a) + 1;
    }
    return ret;
1020
}