ssg.c 47.6 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>
17
#include <linux/limits.h>
Jonathan Jenkins's avatar
Jonathan Jenkins committed
18
#include <assert.h>
Shane Snyder's avatar
Shane Snyder committed
19
#ifdef SSG_HAVE_MPI
20 21
#include <mpi.h>
#endif
Shane Snyder's avatar
Shane Snyder committed
22 23 24
#ifdef SSG_HAVE_PMIX
#include <pmix.h>
#endif
Jonathan Jenkins's avatar
Jonathan Jenkins committed
25

Shane Snyder's avatar
Shane Snyder committed
26
#include <mercury.h>
27
#include <abt.h>
Shane Snyder's avatar
Shane Snyder committed
28
#include <margo.h>
29

30
#include "ssg.h"
Shane Snyder's avatar
Shane Snyder committed
31 32 33
#ifdef SSG_HAVE_MPI
#include "ssg-mpi.h"
#endif
Shane Snyder's avatar
Shane Snyder committed
34 35 36
#ifdef SSG_HAVE_PMIX
#include "ssg-pmix.h"
#endif
37
#include "ssg-internal.h"
Shane Snyder's avatar
Shane Snyder committed
38
#include "swim-fd/swim-fd.h"
39

40 41 42
/* arguments for group lookup ULTs */
struct ssg_group_lookup_ult_args
{
43
    const char *addr_str;
44
    ssg_group_view_t *view;
Shane Snyder's avatar
Shane Snyder committed
45
    ABT_rwlock lock;
46
    int out;
47 48 49
};
static void ssg_group_lookup_ult(void * arg);

50
/* SSG helper routine prototypes */
Shane Snyder's avatar
Shane Snyder committed
51 52 53 54 55 56 57
static ssg_group_t * ssg_group_create_internal(
    const char * group_name, const char * const group_addr_strs[],
    int group_size, ssg_membership_update_cb update_cb, void *update_cb_dat);
static int ssg_group_view_create(
    const char * const group_addr_strs[], int group_size,
    const char * self_addr_str, ABT_rwlock view_lock,
    ssg_group_view_t * view, ssg_member_id_t * self_id);
58
static ssg_member_state_t * ssg_group_view_add_member(
59 60
    const char * addr_str, hg_addr_t addr, ssg_member_id_t member_id,
    ssg_group_view_t * view);
61
static ssg_group_descriptor_t * ssg_group_descriptor_create(
62
    uint64_t name_hash, const char * leader_addr_str, int owner_status);
63 64 65 66 67 68
static ssg_group_descriptor_t * ssg_group_descriptor_dup(
    ssg_group_descriptor_t * descriptor);
static void ssg_group_destroy_internal(
    ssg_group_t * g);
static void ssg_attached_group_destroy(
    ssg_attached_group_t * ag);
Shane Snyder's avatar
Shane Snyder committed
69 70 71 72 73 74
static void ssg_group_view_destroy(
    ssg_group_view_t * view);
static void ssg_group_descriptor_free(
    ssg_group_descriptor_t * descriptor);
static ssg_member_id_t ssg_gen_member_id(
    const char * addr_str);
75 76
static const char ** ssg_addr_str_buf_to_list(
    const char * buf, int num_addrs);
77

78 79
/* XXX: i think we ultimately need per-mid ssg instances rather than 1 global? */
ssg_instance_t *ssg_inst = NULL;
80

81 82 83
/***************************************************
 *** SSG runtime intialization/shutdown routines ***
 ***************************************************/
84

85 86
int ssg_init(
    margo_instance_id mid)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
87
{
88 89
    struct timespec ts;

90 91 92 93 94 95 96 97 98
    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;
99
    ABT_rwlock_create(&ssg_inst->lock);
100

101
    ssg_register_rpcs();
102

103 104 105 106
    /* seed RNG */
    clock_gettime(CLOCK_MONOTONIC, &ts);
    srand(ts.tv_nsec + getpid());

107 108
    return SSG_SUCCESS;
}
Jonathan Jenkins's avatar
Jonathan Jenkins committed
109

110
int ssg_finalize()
111
{
112 113
    ssg_group_t *g, *g_tmp;
    ssg_attached_group_t *ag, *ag_tmp;
114 115 116 117

    if (!ssg_inst)
        return SSG_FAILURE;

118 119
    ABT_rwlock_wrlock(ssg_inst->lock);

120
    /* destroy all active groups */
121
    HASH_ITER(hh, ssg_inst->group_table, g, g_tmp)
122 123
    {
        HASH_DELETE(hh, ssg_inst->group_table, g);
124
        ABT_rwlock_unlock(ssg_inst->lock);
125
        ssg_group_destroy_internal(g);
126
        ABT_rwlock_wrlock(ssg_inst->lock);
127 128
    }

129 130 131 132 133 134
    /* detach from all attached groups */
    HASH_ITER(hh, ssg_inst->attached_group_table, ag, ag_tmp)
    {
        ssg_attached_group_destroy(ag);
    }

135 136 137
    ABT_rwlock_unlock(ssg_inst->lock);
    ABT_rwlock_free(&ssg_inst->lock);

138 139 140
    free(ssg_inst);
    ssg_inst = NULL;

141
    return SSG_SUCCESS;
142
}
Jonathan Jenkins's avatar
Jonathan Jenkins committed
143

144 145 146
/*************************************
 *** SSG group management routines ***
 *************************************/
Jonathan Jenkins's avatar
Jonathan Jenkins committed
147

Shane Snyder's avatar
Shane Snyder committed
148 149 150 151 152 153
ssg_group_id_t ssg_group_create(
    const char * group_name,
    const char * const group_addr_strs[],
    int group_size,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
154
{
Shane Snyder's avatar
Shane Snyder committed
155 156
    ssg_group_t *g;
    ssg_group_id_t g_id = SSG_GROUP_ID_NULL;
157

Shane Snyder's avatar
Shane Snyder committed
158 159 160
    g = ssg_group_create_internal(group_name, group_addr_strs,
            group_size, update_cb, update_cb_dat);
    if (g)
Shane Snyder's avatar
Shane Snyder committed
161
        g_id = (ssg_group_id_t)g->descriptor;
Shane Snyder's avatar
Shane Snyder committed
162 163

    return g_id;
164
}
Shane Snyder's avatar
Shane Snyder committed
165

Shane Snyder's avatar
Shane Snyder committed
166 167 168 169 170
ssg_group_id_t ssg_group_create_config(
    const char * group_name,
    const char * file_name,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
171
{
Shane Snyder's avatar
Shane Snyder committed
172 173 174 175 176 177 178 179
    int fd;
    struct stat st;
    char *rd_buf = NULL;
    ssize_t rd_buf_size;
    char *tok;
    void *addr_str_buf = NULL;
    int addr_str_buf_len = 0, num_addrs = 0;
    const char **addr_strs = NULL;
Shane Snyder's avatar
Shane Snyder committed
180
    int ret;
Shane Snyder's avatar
Shane Snyder committed
181
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
182

Shane Snyder's avatar
Shane Snyder committed
183 184 185 186 187 188 189 190
    /* 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;
    }
191

Shane Snyder's avatar
Shane Snyder committed
192 193 194
    /* get file size and allocate a buffer to store it */
    ret = fstat(fd, &st);
    if (ret == -1)
Shane Snyder's avatar
Shane Snyder committed
195
    {
Shane Snyder's avatar
Shane Snyder committed
196 197 198
        fprintf(stderr, "Error: SSG unable to stat config file %s for group %s\n",
            file_name, group_name);
        goto fini;
Shane Snyder's avatar
Shane Snyder committed
199
    }
Shane Snyder's avatar
Shane Snyder committed
200 201
    rd_buf = malloc(st.st_size+1);
    if (rd_buf == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
202

Shane Snyder's avatar
Shane Snyder committed
203 204 205
    /* load it all in one fell swoop */
    rd_buf_size = read(fd, rd_buf, st.st_size);
    if (rd_buf_size != st.st_size)
Shane Snyder's avatar
Shane Snyder committed
206
    {
Shane Snyder's avatar
Shane Snyder committed
207 208 209
        fprintf(stderr, "Error: SSG unable to read config file %s for group %s\n",
            file_name, group_name);
        goto fini;
Shane Snyder's avatar
Shane Snyder committed
210
    }
Shane Snyder's avatar
Shane Snyder committed
211 212 213 214 215 216 217
    rd_buf[rd_buf_size]='\0';

    /* strtok the result - each space-delimited address is assumed to be
     * a unique mercury address
     */
    tok = strtok(rd_buf, "\r\n\t ");
    if (tok == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
218

Shane Snyder's avatar
Shane Snyder committed
219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
    /* build up the address buffer */
    addr_str_buf = malloc(rd_buf_size);
    if (addr_str_buf == NULL) goto fini;
    do
    {
        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;
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);
    if (addr_str_buf_len != rd_buf_size)
    {
        /* adjust buffer size if our initial guess was wrong */
        void *tmp = realloc(addr_str_buf, addr_str_buf_len);
        if (tmp == NULL) goto fini;
        addr_str_buf = tmp;
    }
Shane Snyder's avatar
Shane Snyder committed
237

Shane Snyder's avatar
Shane Snyder committed
238 239 240
    /* set up address string array for group members */
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, num_addrs);
    if (!addr_strs) goto fini;
Shane Snyder's avatar
Shane Snyder committed
241

Shane Snyder's avatar
Shane Snyder committed
242 243 244
    /* invoke the generic group create routine using our list of addrs */
    group_id = ssg_group_create(group_name, addr_strs, num_addrs,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
245

Shane Snyder's avatar
Shane Snyder committed
246 247 248 249 250 251
fini:
    /* cleanup before returning */
    if (fd != -1) close(fd);
    free(rd_buf);
    free(addr_str_buf);
    free(addr_strs);
Shane Snyder's avatar
Shane Snyder committed
252

Shane Snyder's avatar
Shane Snyder committed
253
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
254 255
}

Shane Snyder's avatar
Shane Snyder committed
256 257 258 259 260 261
#ifdef SSG_HAVE_MPI
ssg_group_id_t ssg_group_create_mpi(
    const char * group_name,
    MPI_Comm comm,
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
Shane Snyder's avatar
Shane Snyder committed
262
{
Shane Snyder's avatar
Shane Snyder committed
263 264
    int i;
    char *self_addr_str = NULL;
265
    int self_addr_str_size = 0;
Shane Snyder's avatar
Shane Snyder committed
266 267 268 269 270 271
    char *addr_str_buf = NULL;
    int *sizes = NULL;
    int *sizes_psum = NULL;
    int comm_size = 0, comm_rank = 0;
    const char **addr_strs = NULL;
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;
Shane Snyder's avatar
Shane Snyder committed
272

Shane Snyder's avatar
Shane Snyder committed
273
    if (!ssg_inst) goto fini;
Shane Snyder's avatar
Shane Snyder committed
274

Shane Snyder's avatar
Shane Snyder committed
275
    /* get my address */
276
    SSG_GET_SELF_ADDR_STR(ssg_inst->mid, self_addr_str);
Shane Snyder's avatar
Shane Snyder committed
277
    if (self_addr_str == NULL) goto fini;
278
    self_addr_str_size = (int)strlen(self_addr_str) + 1;
279

Shane Snyder's avatar
Shane Snyder committed
280 281 282 283 284
    /* gather the buffer sizes */
    MPI_Comm_size(comm, &comm_size);
    MPI_Comm_rank(comm, &comm_rank);
    sizes = malloc(comm_size * sizeof(*sizes));
    if (sizes == NULL) goto fini;
285
    sizes[comm_rank] = self_addr_str_size;
Shane Snyder's avatar
Shane Snyder committed
286
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);
287

Shane Snyder's avatar
Shane Snyder committed
288 289
    /* compute a exclusive prefix sum of the data sizes, including the
     * total at the end
290
     */
Shane Snyder's avatar
Shane Snyder committed
291 292 293 294 295
    sizes_psum = malloc((comm_size+1) * sizeof(*sizes_psum));
    if (sizes_psum == NULL) goto fini;
    sizes_psum[0] = 0;
    for (i = 1; i < comm_size+1; i++)
        sizes_psum[i] = sizes_psum[i-1] + sizes[i-1];
296

Shane Snyder's avatar
Shane Snyder committed
297 298 299
    /* allgather the addresses */
    addr_str_buf = malloc(sizes_psum[comm_size]);
    if (addr_str_buf == NULL) goto fini;
300
    MPI_Allgatherv(self_addr_str, self_addr_str_size, MPI_BYTE,
Shane Snyder's avatar
Shane Snyder committed
301
            addr_str_buf, sizes, sizes_psum, MPI_BYTE, comm);
Shane Snyder's avatar
Shane Snyder committed
302

Shane Snyder's avatar
Shane Snyder committed
303 304 305
    /* set up address string array for group members */
    addr_strs = ssg_addr_str_buf_to_list(addr_str_buf, comm_size);
    if (!addr_strs) goto fini;
306

Shane Snyder's avatar
Shane Snyder committed
307 308 309
    /* invoke the generic group create routine using our list of addrs */
    group_id = ssg_group_create(group_name, addr_strs, comm_size,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
310

Shane Snyder's avatar
Shane Snyder committed
311 312 313 314 315 316 317
fini:
    /* cleanup before returning */
    free(self_addr_str);
    free(sizes);
    free(sizes_psum);
    free(addr_str_buf);
    free(addr_strs);
Shane Snyder's avatar
Shane Snyder committed
318

Shane Snyder's avatar
Shane Snyder committed
319
    return group_id;
Shane Snyder's avatar
Shane Snyder committed
320
}
Shane Snyder's avatar
Shane Snyder committed
321
#endif
Shane Snyder's avatar
Shane Snyder committed
322

Shane Snyder's avatar
Shane Snyder committed
323 324 325
#ifdef SSG_HAVE_PMIX
ssg_group_id_t ssg_group_create_pmix(
    const char * group_name,
Shane Snyder's avatar
Shane Snyder committed
326
    const pmix_proc_t proc,
Shane Snyder's avatar
Shane Snyder committed
327 328 329
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
{
Shane Snyder's avatar
Shane Snyder committed
330 331 332 333 334 335 336 337 338 339 340 341
    char *self_addr_str = NULL;
    pmix_proc_t tmp_proc;
    pmix_value_t value;
    pmix_value_t *val_p = &value;
    pmix_value_t *addr_vals = NULL;
    unsigned int nprocs;
    char key[128];
    pmix_info_t *info;
    bool flag;
    const char **addr_strs = NULL;
    unsigned int n;
    pmix_status_t ret;
Shane Snyder's avatar
Shane Snyder committed
342 343
    ssg_group_id_t group_id = SSG_GROUP_ID_NULL;

Shane Snyder's avatar
Shane Snyder committed
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
    if (!ssg_inst || !PMIx_Initialized()) goto fini;

    /* get my address */
    SSG_GET_SELF_ADDR_STR(ssg_inst->mid, self_addr_str);
    if (self_addr_str == NULL) goto fini;

    /* XXX default handler? */

    /* XXX note we are assuming every process in the job wants to join this group... */
    /* get the total nprocs in the job */
    PMIX_PROC_LOAD(&tmp_proc, proc.nspace, PMIX_RANK_WILDCARD);
    ret = PMIx_Get(&tmp_proc, PMIX_JOB_SIZE, NULL, 0, &val_p);
    if (ret != PMIX_SUCCESS) goto fini;
    nprocs = (int)val_p->data.uint32;
    PMIX_VALUE_RELEASE(val_p);

    /* put my address string using a well-known key */
    if (snprintf(key, 128, "%s-%d-hg-addr", proc.nspace, proc.rank) >= 128) goto fini;
    PMIX_VALUE_LOAD(val_p, self_addr_str, PMIX_STRING);
    ret = PMIx_Put(PMIX_GLOBAL, key, val_p);
    if (ret != PMIX_SUCCESS) goto fini;

    /* commit the put data to the local pmix server */
    ret = PMIx_Commit();
    if (ret != PMIX_SUCCESS) goto fini;

    /* barrier, additionally requesting to collect relevant process data */
    PMIX_INFO_CREATE(info, 1);
    flag = true;
    PMIX_INFO_LOAD(info, PMIX_COLLECT_DATA, &flag, PMIX_BOOL);
    ret = PMIx_Fence(&proc, 1, info, 1);
    PMIX_INFO_FREE(info, 1);

    addr_strs = malloc(nprocs * sizeof(*addr_strs));
    if (addr_strs == NULL) goto fini;

    /* finalize exchange by getting each member's address */
    PMIX_VALUE_CREATE(addr_vals, nprocs);
    for (n = 0; n < nprocs; n++)
    {
        /* skip ourselves */
        if(n == proc.rank) continue;

        if (snprintf(key, 128, "%s-%d-hg-addr", proc.nspace, n) >= 128) goto fini;

        tmp_proc.rank = n;
        val_p = &addr_vals[n];
        ret = PMIx_Get(&tmp_proc, key, NULL, 0, &val_p);
        if (ret != PMIX_SUCCESS) goto fini;

        addr_strs[n] = val_p->data.string;
    }
Shane Snyder's avatar
Shane Snyder committed
396

Shane Snyder's avatar
Shane Snyder committed
397 398 399
    /* invoke the generic group create routine using our list of addrs */
    group_id = ssg_group_create(group_name, addr_strs, nprocs,
        update_cb, update_cb_dat);
Shane Snyder's avatar
Shane Snyder committed
400 401

fini:
Shane Snyder's avatar
Shane Snyder committed
402 403 404 405
    /* cleanup before returning */
    free(self_addr_str);
    free(addr_strs);
    PMIX_VALUE_FREE(addr_vals, nprocs);
Shane Snyder's avatar
Shane Snyder committed
406 407 408 409 410

    return group_id;
}
#endif 

Shane Snyder's avatar
Shane Snyder committed
411 412
int ssg_group_destroy(
    ssg_group_id_t group_id)
413
{
Shane Snyder's avatar
Shane Snyder committed
414 415
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
    ssg_group_t *g;
416

Shane Snyder's avatar
Shane Snyder committed
417
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) return SSG_FAILURE;
418

Shane Snyder's avatar
Shane Snyder committed
419
    if (group_descriptor->owner_status != SSG_OWNER_IS_MEMBER)
420
    {
Shane Snyder's avatar
Shane Snyder committed
421 422
        fprintf(stderr, "Error: SSG unable to destroy a group it is not a member of\n");
        return SSG_FAILURE;
423
    }
424

425 426
    ABT_rwlock_wrlock(ssg_inst->lock);

Shane Snyder's avatar
Shane Snyder committed
427 428 429 430
    /* find the group structure and destroy it */
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
    if (!g)
431
    {
432
        ABT_rwlock_unlock(ssg_inst->lock);
Shane Snyder's avatar
Shane Snyder committed
433 434
        fprintf(stderr, "Error: SSG unable to find expected group reference\n");
        return SSG_FAILURE;
435
    }
436 437
    HASH_DELETE(hh, ssg_inst->group_table, g);
    ABT_rwlock_unlock(ssg_inst->lock);
Shane Snyder's avatar
Shane Snyder committed
438
    ssg_group_destroy_internal(g);
439

Shane Snyder's avatar
Shane Snyder committed
440
    return SSG_SUCCESS;
441 442
}

Shane Snyder's avatar
Shane Snyder committed
443 444
ssg_group_id_t ssg_group_join(
    ssg_group_id_t in_group_id,
445 446
    ssg_membership_update_cb update_cb,
    void * update_cb_dat)
447
{
Shane Snyder's avatar
Shane Snyder committed
448 449
    ssg_group_descriptor_t *in_group_descriptor = (ssg_group_descriptor_t *)in_group_id;
    char *self_addr_str = NULL;
450
    hg_addr_t group_target_addr = HG_ADDR_NULL;
Shane Snyder's avatar
Shane Snyder committed
451 452 453 454
    char *group_name = NULL;
    int group_size;
    void *view_buf = NULL;
    const char **addr_strs = NULL;
455
    hg_return_t hret;
456
    int sret;
Shane Snyder's avatar
Shane Snyder committed
457 458
    ssg_group_t *g = NULL;
    ssg_group_id_t g_id = SSG_GROUP_ID_NULL;
459

Shane Snyder's avatar
Shane Snyder committed
460
    if (!ssg_inst || in_group_id == SSG_GROUP_ID_NULL) goto fini;
461

Shane Snyder's avatar
Shane Snyder committed
462
    if (in_group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
463
    {
Shane Snyder's avatar
Shane Snyder committed
464
        fprintf(stderr, "Error: SSG unable to join a group it is already a member of\n");
465 466
        goto fini;
    }
Shane Snyder's avatar
Shane Snyder committed
467
    else if (in_group_descriptor->owner_status == SSG_OWNER_IS_ATTACHER)
468
    {
Shane Snyder's avatar
Shane Snyder committed
469
        fprintf(stderr, "Error: SSG unable to join a group it is attached to\n");
470 471
        goto fini;
    }
472

473 474 475 476 477 478 479 480 481
    /* lookup the address of the target group member in the GID */
    hret = margo_addr_lookup(ssg_inst->mid, in_group_descriptor->addr_str,
        &group_target_addr);
    if (hret != HG_SUCCESS) goto fini;

    sret = ssg_group_join_send(in_group_descriptor, group_target_addr,
        &group_name, &group_size, &view_buf);
    if (sret != SSG_SUCCESS || !group_name || !view_buf) goto fini;

Shane Snyder's avatar
Shane Snyder committed
482
    /* get my address string */
483
    SSG_GET_SELF_ADDR_STR(ssg_inst->mid, self_addr_str);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
484 485
    if (self_addr_str == NULL) goto fini;

Shane Snyder's avatar
Shane Snyder committed
486 487
    /* set up address string array for all group members */
    addr_strs = ssg_addr_str_buf_to_list(view_buf, group_size);
488 489
    if (!addr_strs) goto fini;

Shane Snyder's avatar
Shane Snyder committed
490 491 492 493 494 495 496 497 498
    /* append self address string to list of group member address strings */
    addr_strs = realloc(addr_strs, (group_size+1)*sizeof(char *));
    if(!addr_strs) goto fini;
    addr_strs[group_size++] = self_addr_str;

    g = ssg_group_create_internal(group_name, addr_strs, group_size,
            update_cb, update_cb_dat);
    if (g)
    {
Shane Snyder's avatar
Shane Snyder committed
499
        g_id = (ssg_group_id_t)g->descriptor;
500 501 502

        /* don't free on success */
        group_name = NULL;
Shane Snyder's avatar
Shane Snyder committed
503
    }
Shane Snyder's avatar
Shane Snyder committed
504 505

fini:
506 507
    if (group_target_addr != HG_ADDR_NULL)
        margo_addr_free(ssg_inst->mid, group_target_addr);
Shane Snyder's avatar
Shane Snyder committed
508 509 510
    free(addr_strs);
    free(view_buf);
    free(group_name);
511
    free(self_addr_str);
512

Shane Snyder's avatar
Shane Snyder committed
513
    return g_id;
Shane Snyder's avatar
Shane Snyder committed
514 515
}

Shane Snyder's avatar
Shane Snyder committed
516
int ssg_group_leave(
517
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
518
{
Shane Snyder's avatar
Shane Snyder committed
519
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
520 521 522 523
    ssg_group_t *g = NULL;
    hg_addr_t group_target_addr = HG_ADDR_NULL;
    hg_return_t hret;
    int sret = SSG_FAILURE;
524

525
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
526

527 528
    if (group_descriptor->owner_status != SSG_OWNER_IS_MEMBER)
    {
Shane Snyder's avatar
Shane Snyder committed
529
        fprintf(stderr, "Error: SSG unable to leave group it is not a member of\n");
530
        goto fini;
531 532
    }

533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567
    ABT_rwlock_rdlock(ssg_inst->lock);
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
    if (!g)
    {
        ABT_rwlock_unlock(ssg_inst->lock);
        goto fini;
    }

    /* send the leave req to the first member in the view */
    hret = margo_addr_dup(ssg_inst->mid, g->view.member_map->addr, &group_target_addr);
    if (hret != HG_SUCCESS)
    {
        ABT_rwlock_unlock(ssg_inst->lock);
        goto fini;
    }
    ABT_rwlock_unlock(ssg_inst->lock);

    sret = ssg_group_leave_send(group_descriptor, g->self_id, group_target_addr);
    if (sret != SSG_SUCCESS) goto fini;

    /* at least one group member knows of the leave request -- safe to
     * shutdown the group locally
     */

    /* re-lookup the group as we don't hold the lock while sending the leave req */
    ABT_rwlock_wrlock(ssg_inst->lock);
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
    if (g)
    {
        HASH_DELETE(hh, ssg_inst->group_table, g);
        ABT_rwlock_unlock(ssg_inst->lock);
        ssg_group_destroy_internal(g);
    }
Shane Snyder's avatar
Shane Snyder committed
568 569
    else
        ABT_rwlock_unlock(ssg_inst->lock);
570 571 572 573 574 575 576 577

    sret = SSG_SUCCESS;

fini:
    if (group_target_addr != HG_ADDR_NULL)
        margo_addr_free(ssg_inst->mid, group_target_addr);

    return sret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
578 579
}

580
#if 0
581 582 583
int ssg_group_attach(
    ssg_group_id_t group_id)
{
Shane Snyder's avatar
Shane Snyder committed
584
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
585 586 587 588 589 590
    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;
591

592
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) goto fini;
593

594 595 596 597 598 599 600 601 602 603 604
    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;
    }
605

606 607 608 609 610 611 612
    /* 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;

Shane Snyder's avatar
Shane Snyder committed
613
    /* set up address string array for all group members */
614 615 616 617 618 619 620
    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));
Shane Snyder's avatar
Shane Snyder committed
621
    ag->name = strdup(group_name);
622 623 624 625 626
    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 */
Shane Snyder's avatar
Shane Snyder committed
627
    sret = ssg_group_view_create(addr_strs, group_size, NULL, ag->lock, &ag->view, NULL);
628 629 630 631 632 633 634 635 636
    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 */
Shane Snyder's avatar
Shane Snyder committed
637
    group_name = NULL;
638 639
    ag = NULL;
fini:
Shane Snyder's avatar
Shane Snyder committed
640
    if (ag) ssg_attached_group_destroy(ag);
641
    free(addr_strs);
Shane Snyder's avatar
Shane Snyder committed
642 643
    free(view_buf);
    free(group_name);
644 645

    return sret;
646 647 648 649 650
}

int ssg_group_detach(
    ssg_group_id_t group_id)
{
651 652 653
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
    ssg_attached_group_t *ag;

Shane Snyder's avatar
Shane Snyder committed
654 655
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) return SSG_FAILURE;

656 657 658 659 660 661 662 663 664 665 666 667
    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)
    {
Shane Snyder's avatar
Shane Snyder committed
668
        fprintf(stderr, "Error: SSG unable to find expected group attached\n");
669 670 671 672 673
        return SSG_FAILURE;
    }
    HASH_DELETE(hh, ssg_inst->attached_group_table, ag);
    ssg_attached_group_destroy(ag);

674 675
    return SSG_SUCCESS;
}
676
#endif
677

678 679 680
/*********************************
 *** SSG group access routines ***
 *********************************/
Shane Snyder's avatar
Shane Snyder committed
681

682
ssg_member_id_t ssg_get_group_self_id(
683
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
684
{
Shane Snyder's avatar
Shane Snyder committed
685
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
686
    ssg_group_t *g;
687
    ssg_member_id_t self_id;
688

Shane Snyder's avatar
Shane Snyder committed
689
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL) return SSG_MEMBER_ID_INVALID;
690

691 692 693 694
    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");
695
        return SSG_MEMBER_ID_INVALID;
696 697
    }

698
    ABT_rwlock_rdlock(ssg_inst->lock);
Shane Snyder's avatar
Shane Snyder committed
699 700
    HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
        sizeof(uint64_t), g);
701 702 703 704 705
    if (g)
        self_id = g->self_id;
    else
        self_id = SSG_MEMBER_ID_INVALID;
    ABT_rwlock_unlock(ssg_inst->lock);
706

707
    return self_id;
Shane Snyder's avatar
Shane Snyder committed
708 709
}

710 711
int ssg_get_group_size(
    ssg_group_id_t group_id)
Shane Snyder's avatar
Shane Snyder committed
712
{
Shane Snyder's avatar
Shane Snyder committed
713
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
Shane Snyder's avatar
Shane Snyder committed
714
    int group_size = 0;
715

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

718 719 720 721
    if (group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
    {
        ssg_group_t *g;

722
        ABT_rwlock_rdlock(ssg_inst->lock);
723 724 725
        HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), g);
        if (g)
Shane Snyder's avatar
Shane Snyder committed
726 727 728 729 730
        {
            ABT_rwlock_rdlock(g->lock);
            group_size = g->view.size + 1; /* add ourself to view size */
            ABT_rwlock_unlock(g->lock);
        }
731
        ABT_rwlock_unlock(ssg_inst->lock);
732
    }
733
#if 0
734 735 736 737 738 739 740
    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)
Shane Snyder's avatar
Shane Snyder committed
741 742 743 744 745
        {
            ABT_rwlock_rdlock(ag->lock);
            group_size = ag->view.size;
            ABT_rwlock_unlock(ag->lock);
        }
746
    }
747
#endif
748 749 750 751
    else
    {
        fprintf(stderr, "Error: SSG can only obtain size of groups that the caller" \
            " is a member of or an attacher of\n");
752
        return 0;
753
    }
754

Shane Snyder's avatar
Shane Snyder committed
755
    return group_size;
Shane Snyder's avatar
Shane Snyder committed
756 757
}

758 759 760
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
761
{
Shane Snyder's avatar
Shane Snyder committed
762
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
763
    ssg_member_state_t *member_state;
Shane Snyder's avatar
Shane Snyder committed
764
    hg_addr_t member_addr = HG_ADDR_NULL;
765

766 767
    if (!ssg_inst || group_id == SSG_GROUP_ID_NULL ||
            member_id == SSG_MEMBER_ID_INVALID)
768 769
        return HG_ADDR_NULL;

770 771 772 773
    if (group_descriptor->owner_status == SSG_OWNER_IS_MEMBER)
    {
        ssg_group_t *g;

774
        ABT_rwlock_rdlock(ssg_inst->lock);
775 776 777
        HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), g);
        if (g)
Shane Snyder's avatar
Shane Snyder committed
778 779 780 781 782 783 784 785
        {
            ABT_rwlock_rdlock(g->lock);
            HASH_FIND(hh, g->view.member_map, &member_id, sizeof(ssg_member_id_t),
                member_state);
            if (member_state) 
                member_addr = member_state->addr;
            ABT_rwlock_unlock(g->lock);
        }
786
        ABT_rwlock_unlock(ssg_inst->lock);
787
    }
788
#if 0
789 790 791 792 793 794 795
    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)
Shane Snyder's avatar
Shane Snyder committed
796 797 798 799 800 801 802 803
        {
            ABT_rwlock_rdlock(ag->lock);
            HASH_FIND(hh, ag->view.member_map, &member_id, sizeof(ssg_member_id_t),
                member_state);
            if (member_state) 
                member_addr = member_state->addr;
            ABT_rwlock_unlock(ag->lock);
        }
804
    }
805
#endif
806 807 808 809 810 811 812
    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;
    }

Shane Snyder's avatar
Shane Snyder committed
813
    return member_addr;
814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830
}

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
831
    descriptor = SSG_GROUP_ID_NULL;
832 833 834
    return;
}

835 836 837 838 839 840 841 842
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);
}

843 844 845 846 847 848 849
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;
850
    char *gid_buf, *p; 
851 852 853 854 855 856 857 858 859 860 861 862 863

    *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 */
864 865 866 867 868 869
    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);
870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
    /* 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;
}

1025 1026 1027 1028 1029
void ssg_group_dump(
    ssg_group_id_t group_id)
{
    ssg_group_descriptor_t *group_descriptor = (ssg_group_descriptor_t *)group_id;
    ssg_group_view_t *group_view = NULL;
1030 1031
    ABT_rwlock group_view_lock;
    int group_size;
1032 1033 1034 1035 1036 1037 1038 1039
    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;

1040
        ABT_rwlock_rdlock(ssg_inst->lock);
1041 1042 1043 1044 1045
        HASH_FIND(hh, ssg_inst->group_table, &group_descriptor->name_hash,
            sizeof(uint64_t), g);
        if (g)
        {
            group_view = &g->view;
1046 1047
            group_view_lock = g->lock;
            group_size = g->view.size + 1;
1048 1049
            group_name = g->name;
            strcpy(group_role, "member");
1050
            sprintf(group_self_id, "%lu", g->self_id);
1051
        }
1052
        ABT_rwlock_unlock(ssg_inst->lock);
1053
    }
1054
#if 0
1055 1056 1057 1058 1059 1060 1061 1062 1063
    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;
1064
            group_size = ag->view.size;
1065 1066 1067 1068
            group_name = ag->name;
            strcpy(group_role, "attacher");
        }
    }
1069
#endif
1070 1071 1072 1073 1074 1075 1076 1077 1078
    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)
    {
1079
        ssg_member_state_t *member_state, *tmp_ms;
1080 1081 1082 1083 1084

        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);
1085
        printf("\tsize: %d\n", group_size);
1086
        printf("\tview:\n");
1087
        ABT_rwlock_rdlock(group_view_lock);
1088
        HASH_ITER(hh, group_view->member_map, member_state, tmp_ms)
1089
        {
Shane Snyder's avatar
Shane Snyder committed
1090
            printf("\t\tid: %20lu\taddr: %s\n", member_state->id,
1091
                member_state->addr_str);
1092
        }
1093
        ABT_rwlock_unlock(group_view_lock);
1094 1095 1096 1097
    }
    else
        fprintf(stderr, "Error: SSG unable to find group view associated" \
            " with the given group ID\n");
1098

1099
    return;
Shane Snyder's avatar
Shane Snyder committed
1100 1101
}

1102 1103 1104
/************************************
 *** SSG internal helper routines ***
 ************************************/
1105

Shane Snyder's avatar
Shane Snyder committed
1106 1107 1108
static ssg_group_t * ssg_group_create_internal(
    const char * group_name, const char * const group_addr_strs[],
    int group_size, ssg_membership_update_cb update_cb, void *update_cb_dat)
1109
{
Shane Snyder's avatar
Shane Snyder committed
1110 1111 1112 1113
    uint64_t name_hash;
    char *self_addr_str = NULL;
    int sret;
    int success = 0;
1114
    ssg_group_t *g = NULL, *check_g;
1115

Shane Snyder's avatar
Shane Snyder committed
1116
    if (!ssg_inst) return NULL;
Shane Snyder's avatar
Shane Snyder committed
1117

Shane Snyder's avatar
Shane Snyder committed
1118 1119 1120
    name_hash = ssg_hash64_str(group_name);

    /* get my address string */
1121
    SSG_GET_SELF_ADDR_STR(ssg_inst->mid, self_addr_str);
Shane Snyder's avatar
Shane Snyder committed
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143
    if (self_addr_str == NULL) goto fini;

    /* 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->update_cb = update_cb;
    g->update_cb_dat = update_cb_dat;
    ABT_rwlock_create(&g->lock);

    /* generate unique descriptor for this group  */
    g->descriptor = ssg_group_descriptor_create(name_hash, self_addr_str,
        SSG_OWNER_IS_MEMBER);
    if (g->descriptor == NULL) goto fini;

    /* initialize the group view */
    sret = ssg_group_view_create(group_addr_strs, group_size, self_addr_str,
        g->lock, &g->view, &g->self_id);
    if (sret != SSG_SUCCESS) goto fini;
    if (g->self_id == SSG_MEMBER_ID_INVALID)
Shane Snyder's avatar
Shane Snyder committed
1144
    {
Shane Snyder's avatar
Shane Snyder committed
1145 1146 1147 1148
        /* if unable to resolve my rank within the group, error out */
        fprintf(stderr, "Error: SSG unable to resolve rank in group %s\n",
            group_name);
        goto fini;
Shane Snyder's avatar
Shane Snyder committed
1149
    }
1150

1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
#ifdef DEBUG
    /* set debug output pointer */
    char *dbg_log_dir = getenv("SSG_DEBUG_LOGDIR");
    if (dbg_log_dir)
    {
        char dbg_log_path[PATH_MAX];
        snprintf(dbg_log_path, PATH_MAX, "%s/ssg-%s-%lu.log",
            dbg_log_dir, g->name, g->self_id);
        g->dbg_log = fopen(dbg_log_path, "a");
        if (!g->dbg_log) goto fini;
    }
    else
    {
        g->dbg_log = stdout;
    }
#endif

1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178
    /* make sure we aren't re-creating an existing group */
    ABT_rwlock_wrlock(ssg_inst->lock);
    HASH_FIND(hh, ssg_inst->group_table, &name_hash, sizeof(uint64_t), check_g);
    if (check_g) goto fini;

    /* add this group reference to the group table */
    HASH_ADD(hh, ssg_inst->group_table, descriptor->name_hash,
        sizeof(uint64_t), g);
    ABT_rwlock_unlock(ssg_inst->lock);

    /* initialize swim failure detector */
Shane Snyder's avatar
Shane Snyder committed
1179 1180
    sret = swim_init(g, ssg_inst->mid, 1);
    if (sret != SSG_SUCCESS)
1181 1182 1183 1184 1185 1186 1187 1188
    {
        ABT_rwlock_wrlock(ssg_inst->lock);
        HASH_DELETE(hh, ssg_inst->group_table, g);
        ABT_rwlock_unlock(ssg_inst->lock);
        goto fini;
    }

    SSG_DEBUG(g, "group create successful (size=%d, self=%s)\n", group_size, self_addr_str);
Shane Snyder's avatar
Shane Snyder committed
1189
    success = 1;
1190

Shane Snyder's avatar
Shane Snyder committed
1191 1192
fini:
    if (!success && g)
1193
    {
1194 1195 1196 1197 1198
#ifdef DEBUG
        /* if using logfile debug output, close the stream */
        if (getenv("SSG_DEBUG_LOGDIR"))
            fclose(g->dbg_log);
#endif
Shane Snyder's avatar
Shane Snyder committed
1199 1200
        if (g->descriptor) ssg_group_descriptor_free(g->descriptor);
        ssg_group_view_destroy(&g->view);
1201
        ABT_rwlock_free(&g->lock);
Shane Snyder's avatar
Shane Snyder committed
1202 1203 1204
        free(g->name);
        free(g);
        g = NULL;
1205
    }
Shane Snyder's avatar
Shane Snyder committed
1206 1207 1208
    free(self_addr_str);

    return g;
1209
}
1210

1211
static int ssg_group_view_create(
1212
    const char * const group_addr_strs[], int group_size,
Shane Snyder's avatar
Shane Snyder committed
1213 1214
    const char * self_addr_str, ABT_rwlock view_lock,
    ssg_group_view_t * view, ssg_member_id_t * self_id)
1215 1216
{
    int i, j, r;
1217 1218
    ABT_thread *lookup_ults = NULL;
    struct ssg_group_lookup_ult_args *lookup_ult_args = NULL;
1219 1220 1221
    const char *self_addr_substr = NULL;
    const char *addr_substr = NULL;
    int aret;
1222
    int sret = SSG_FAILURE;
1223

1224 1225 1226
    if (self_id)
        *self_id = SSG_MEMBER_ID_INVALID;
        
Shane Snyder's avatar
Shane Snyder committed
1227
    if ((self_id != NULL && self_addr_str == NULL) || !view) goto fini;
Shane Snyder's avatar
Shane Snyder committed
1228

1229 1230
    /* allocate lookup ULTs */
    lookup_ults = malloc(group_size * sizeof(*lookup_ults));
1231 1232
    if (lookup_ults == NULL) goto fini;
    for (i = 0; i < group_size; i++) lookup_ults[i] = ABT_THREAD_NULL;
1233
    lookup_ult_args = malloc(group_size * sizeof(*lookup_ult_args));
1234
    if (lookup_ult_args == NULL) goto fini;
Shane Snyder's avatar
Shane Snyder committed
1235

Shane Snyder's avatar
Shane Snyder committed
1236
    if(self_addr_str)
1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248
    {
        /* 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;
    }

1249
    /* construct view using ULTs to lookup the address of each group member */
Shane Snyder's avatar
Shane Snyder committed
1250
    r = rand() % group_size;
1251 1252 1253 1254 1255
    for (i = 0; i < group_size; i++)
    {
        /* randomize our starting index so all group members aren't looking
         * up other group members in the same order
         */
Shane Snyder's avatar
Shane Snyder committed
1256
        j = (r + i) % group_size;
1257

1258 1259
        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;

1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270
        /* 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)
            {
1271
                if (self_id)               
1272
                    *self_id = ssg_gen_member_id(group_addr_strs[j]);
1273 1274 1275

                /* don't look up our own address, we already know it */
                continue;
1276 1277 1278 1279
            }
        }

        /* XXX limit outstanding lookups to some max */
1280
        lookup_ult_args[j].addr_str = group_addr_strs[j];
1281
        lookup_ult_args[j].view = view;
Shane Snyder's avatar
Shane Snyder committed
1282
        lookup_ult_args[j].lock = view_lock;
1283 1284
        ABT_pool pool;
        margo_get_handler_pool(ssg_inst->mid, &pool);
1285 1286
        aret = ABT_thread_create(pool, &ssg_group_lookup_ult,
            &lookup_ult_args[j], ABT_THREAD_ATTR_NULL,
1287
            &lookup_ults[j]);
1288
        if (aret != ABT_SUCCESS) goto fini;
1289 1290 1291 1292 1293 1294 1295 1296 1297 1298
    }

    /* 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;
1299
        if (aret != ABT_SUCCESS) goto fini;
1300
        else if (lookup_ult_args[i].out != SSG_SUCCESS)
1301
        {
1302 1303
            fprintf(stderr, "Error: SSG unable to lookup HG address %s\n",
                lookup_ult_args[i].addr_str);
1304
            goto fini;
1305 1306 1307
        }
    }

1308 1309 1310
    /* clean exit */
    sret = SSG_SUCCESS;

1311 1312