ssg.c 19.3 KB
Newer Older
1 2 3 4 5 6
/*
 * Copyright (c) 2016 UChicago Argonne, LLC
 *
 * See COPYRIGHT in top-level directory.
 */

Jonathan Jenkins's avatar
Jonathan Jenkins committed
7 8 9 10
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
11
#include <errno.h>
Jonathan Jenkins's avatar
Jonathan Jenkins committed
12 13 14 15 16 17 18 19 20 21 22 23
#include <stdlib.h>
#include <string.h>

#include <mercury_proc.h>

#include <ssg.h>
#include <ssg-config.h>
#include "def.h"

#ifdef HAVE_MPI
#include <ssg-mpi.h>
#endif
Jonathan Jenkins's avatar
Jonathan Jenkins committed
24 25 26
#ifdef HAVE_MARGO
#include <ssg-margo.h>
#endif
Jonathan Jenkins's avatar
Jonathan Jenkins committed
27 28

// helpers for looking up a server
Jonathan Jenkins's avatar
Jonathan Jenkins committed
29 30 31
static hg_return_t lookup_serv_addr_cb(const struct hg_cb_info *info);
static hg_addr_t lookup_serv_addr(
        hg_context_t *hgctx,
Jonathan Jenkins's avatar
Jonathan Jenkins committed
32 33 34 35
        const char *info_str);

static char** setup_addr_str_list(int num_addrs, char * buf);

36
ssg_t ssg_init_config(const char * fname, int is_member)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114
{
    // file to read
    int fd = -1;
    struct stat st;

    // file content to parse
    char *rdbuf = NULL;
    ssize_t rdsz;

    // parse metadata (strtok)
    char *tok;

    // vars to build up the addr string list
    int addr_cap = 128;
    int addr_len = 0;
    int num_addrs = 0;
    void *buf = NULL;
    char **addr_strs = NULL;

    // return var
    ssg_t s = NULL;

    // misc return codes
    int ret;

    // open file for reading
    fd = open(fname, O_RDONLY);
    if (fd == -1) goto fini;

    // get file size
    ret = fstat(fd, &st);
    if (ret == -1) goto fini;

    // slurp file in all at once
    rdbuf = malloc(st.st_size+1);
    if (rdbuf == NULL) goto fini;

    // load it all in one fell swoop
    rdsz = read(fd, rdbuf, st.st_size);
    if (rdsz <= 0) goto fini;
    if (rdsz != st.st_size) { free(rdbuf); close(fd); return NULL; }
    rdbuf[rdsz]='\0';

    // strtok the result - each space-delimited address is assumed to be
    // a unique mercury address
    tok = strtok(rdbuf, "\r\n\t ");
    if (tok == NULL) goto fini;

    // build up the address buffer
    buf = malloc(addr_cap);
    if (buf == NULL) goto fini;
    do {
        int tok_sz = strlen(tok);
        if (tok_sz + addr_len + 1 > addr_cap) {
            void * tmp;
            addr_cap *= 2;
            tmp = realloc(buf, addr_cap);
            if (tmp == NULL) goto fini;
            buf = tmp;
        }
        memcpy((char*)buf + addr_len, tok, tok_sz+1);
        addr_len += tok_sz+1;
        num_addrs++;
        tok = strtok(NULL, "\r\n\t ");
    } while (tok != NULL);

    // set up the list of addresses
    addr_strs = malloc(num_addrs * sizeof(*addr_strs));
    if (addr_strs == NULL) goto fini;
    tok = (char*)buf;
    for (int i = 0; i < num_addrs; i++) {
        addr_strs[i] = tok;
        tok += strlen(tok) + 1;
    }

    // done parsing - setup the return structure
    s = malloc(sizeof(*s));
    if (s == NULL) goto fini;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
115
    s->hgcl = NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
116 117
    s->addrs = malloc(num_addrs*sizeof(*s->addrs));
    if (s->addrs == NULL) goto fini;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
118
    for (int i = 0; i < num_addrs; i++) s->addrs[i] = HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
119 120 121 122
    s->addr_strs = addr_strs; addr_strs = NULL;
    s->backing_buf = buf; buf = NULL;
    s->num_addrs = num_addrs;
    s->buf_size = addr_len;
123
    s->rank = is_member ? SSG_RANK_UNKNOWN : SSG_EXTERNAL_RANK;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
124 125 126 127 128 129

fini:
    if (fd != -1) close(fd);
    free(rdbuf);
    free(addr_strs);
    free(buf);
130
    if (s != NULL && s->addrs == NULL) { free(s); s = NULL; }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
131 132 133 134
    return s;
}

#ifdef HAVE_MPI
Jonathan Jenkins's avatar
Jonathan Jenkins committed
135
ssg_t ssg_init_mpi(hg_class_t *hgcl, MPI_Comm comm)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
136 137
{
    // my addr
Jonathan Jenkins's avatar
Jonathan Jenkins committed
138
    hg_addr_t self_addr = HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
139
    char * self_addr_str = NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
140
    hg_size_t self_addr_size = 0;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
141 142 143 144 145 146 147 148 149
    int self_addr_size_int = 0; // for mpi-friendly conversion

    // collective helpers
    char * buf = NULL;
    int * sizes = NULL;
    int * sizes_psum = NULL;
    int comm_size = 0;
    int comm_rank = 0;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
150 151
    // hg addresses
    hg_addr_t *addrs = NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
152 153 154 155 156 157

    // return data
    char **addr_strs = NULL;
    ssg_t s = NULL;

    // misc return codes
Jonathan Jenkins's avatar
Jonathan Jenkins committed
158
    hg_return_t hret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
159 160

    // get my address
Jonathan Jenkins's avatar
Jonathan Jenkins committed
161 162 163
    hret = HG_Addr_self(hgcl, &self_addr);
    if (hret != HG_SUCCESS) goto fini;
    hret = HG_Addr_to_string(hgcl, NULL, &self_addr_size, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
164 165 166
    if (self_addr == NULL) goto fini;
    self_addr_str = malloc(self_addr_size);
    if (self_addr_str == NULL) goto fini;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
167 168
    hret = HG_Addr_to_string(hgcl, self_addr_str, &self_addr_size, self_addr);
    if (hret != HG_SUCCESS) goto fini;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
    self_addr_size_int = (int)self_addr_size; // null char included in call

    // 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;
    sizes[comm_rank] = self_addr_size_int;
    MPI_Allgather(MPI_IN_PLACE, 0, MPI_BYTE, sizes, 1, MPI_INT, comm);

    // compute a exclusive prefix sum of the data sizes,
    // including the total at the end
    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];

    // allgather the addresses
    buf = malloc(sizes_psum[comm_size]);
    if (buf == NULL) goto fini;
    MPI_Allgatherv(self_addr_str, self_addr_size_int, MPI_BYTE,
            buf, sizes, sizes_psum, MPI_BYTE, comm);

    // set the addresses
    addr_strs = setup_addr_str_list(comm_size, buf);
    if (addr_strs == NULL) goto fini;

    // init peer addresses
    addrs = malloc(comm_size*sizeof(*addrs));
    if (addrs == NULL) goto fini;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
200
    for (int i = 0; i < comm_size; i++) addrs[i] = HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
201 202 203 204 205
    addrs[comm_rank] = self_addr;

    // set up the output
    s = malloc(sizeof(*s));
    if (s == NULL) goto fini;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
206
    s->hgcl = NULL; // set in ssg_lookup
Jonathan Jenkins's avatar
Jonathan Jenkins committed
207 208 209 210 211 212
    s->addr_strs = addr_strs; addr_strs = NULL;
    s->addrs = addrs; addrs = NULL;
    s->backing_buf = buf; buf = NULL;
    s->num_addrs = comm_size;
    s->buf_size = sizes_psum[comm_size];
    s->rank = comm_rank;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
213
    self_addr = HG_ADDR_NULL; // don't free this on success
Jonathan Jenkins's avatar
Jonathan Jenkins committed
214 215

fini:
Jonathan Jenkins's avatar
Jonathan Jenkins committed
216
    if (self_addr != HG_ADDR_NULL) HG_Addr_free(hgcl, self_addr);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
217 218 219 220 221 222 223 224
    free(buf);
    free(sizes);
    free(addr_strs);
    free(addrs);
    return s;
}
#endif

225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278
hg_return_t ssg_resolve_rank(ssg_t s, hg_class_t *hgcl)
{
    if (s->rank == SSG_EXTERNAL_RANK ||
            s->rank != SSG_RANK_UNKNOWN)
        return HG_SUCCESS;

    // helpers
    hg_addr_t self_addr = HG_ADDR_NULL;
    char * self_addr_str = NULL;
    const char * self_addr_substr = NULL;
    hg_size_t self_addr_size = 0;
    const char * addr_substr = NULL;
    int rank = 0;
    hg_return_t hret;

    // get my address
    hret = HG_Addr_self(hgcl, &self_addr);
    if (hret != HG_SUCCESS) goto end;
    hret = HG_Addr_to_string(hgcl, NULL, &self_addr_size, self_addr);
    if (self_addr == NULL) { hret = HG_NOMEM_ERROR; goto end; }
    self_addr_str = malloc(self_addr_size);
    if (self_addr_str == NULL) { hret = HG_NOMEM_ERROR; goto end; }
    hret = HG_Addr_to_string(hgcl, self_addr_str, &self_addr_size, self_addr);
    if (hret != HG_SUCCESS) goto end;

    // strstr is used here b/c there may be inconsistencies in whether the class
    // is included in the address or not (it's not in HG_Addr_to_string, it
    // should be in ssg_init_config)
    self_addr_substr = strstr(self_addr_str, "://");
    if (self_addr_substr == NULL) { hret = HG_INVALID_PARAM; goto end; }
    self_addr_substr += 3;
    for (rank = 0; rank < s->num_addrs; rank++) {
        addr_substr = strstr(s->addr_strs[rank], "://");
        if (addr_substr == NULL) { hret = HG_INVALID_PARAM; goto end; }
        addr_substr+= 3;
        if (strcmp(self_addr_substr, addr_substr) == 0)
            break;
    }
    if (rank == s->num_addrs) {
        hret = HG_INVALID_PARAM;
        goto end;
    }

    // success - set
    s->rank = rank;
    s->addrs[rank] = self_addr; self_addr = HG_ADDR_NULL;

end:
    if (self_addr != HG_ADDR_NULL) HG_Addr_free(hgcl, self_addr);
    free(self_addr_str);

    return hret;
}

279
hg_return_t ssg_lookup(ssg_t s, hg_context_t *hgctx)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
280 281 282 283
{
    // "effective" rank for the lookup loop
    int eff_rank = 0;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
284 285 286
    // set the hg class up front - need for destructing addrs
    s->hgcl = HG_Context_get_class(hgctx);
    if (s->hgcl == NULL) return HG_INVALID_PARAM;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
287 288 289

    // perform search for my rank if not already set
    if (s->rank == SSG_RANK_UNKNOWN) {
290
        hg_return_t hret = ssg_resolve_rank(s, s->hgcl);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
291
        if (hret != HG_SUCCESS) return hret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
292 293 294 295 296
    }

    if (s->rank == SSG_EXTERNAL_RANK) {
        // do a completely arbitrary effective rank determination to try and
        // prevent everyone talking to the same member at once
Jonathan Jenkins's avatar
Jonathan Jenkins committed
297
        eff_rank = (((intptr_t)hgctx)/sizeof(hgctx)) % s->num_addrs;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
298 299 300 301 302 303
    }
    else
        eff_rank = s->rank;

    // rank is set, perform lookup
    for (int i = eff_rank+1; i < s->num_addrs; i++) {
Jonathan Jenkins's avatar
Jonathan Jenkins committed
304 305
        s->addrs[i] = lookup_serv_addr(hgctx, s->addr_strs[i]);
        if (s->addrs[i] == HG_ADDR_NULL) return HG_PROTOCOL_ERROR;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
306 307
    }
    for (int i = 0; i < eff_rank; i++) {
Jonathan Jenkins's avatar
Jonathan Jenkins committed
308 309
        s->addrs[i] = lookup_serv_addr(hgctx, s->addr_strs[i]);
        if (s->addrs[i] == HG_ADDR_NULL) return HG_PROTOCOL_ERROR;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
310 311 312
    }
    if (s->rank == SSG_EXTERNAL_RANK) {
        s->addrs[eff_rank] =
Jonathan Jenkins's avatar
Jonathan Jenkins committed
313 314
            lookup_serv_addr(hgctx, s->addr_strs[eff_rank]);
        if (s->addrs[eff_rank] == HG_ADDR_NULL) return HG_PROTOCOL_ERROR;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
315 316
    }

Jonathan Jenkins's avatar
Jonathan Jenkins committed
317
    return HG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
318 319
}

Jonathan Jenkins's avatar
Jonathan Jenkins committed
320
#ifdef HAVE_MARGO
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337

struct lookup_ult_args
{
    ssg_t ssg;
    margo_instance_id mid;
    int rank;
    hg_return_t out;
};

static void lookup_ult(void *arg)
{
    struct lookup_ult_args *l = arg;

    l->out = margo_addr_lookup(l->mid, l->ssg->addr_strs[l->rank],
            &l->ssg->addrs[l->rank]);
}

Jonathan Jenkins's avatar
Jonathan Jenkins committed
338 339 340 341
// TODO: refactor - code is mostly a copy of ssg_lookup
hg_return_t ssg_lookup_margo(ssg_t s, margo_instance_id mid)
{
    hg_context_t *hgctx;
342 343 344
    ABT_thread *ults;
    struct lookup_ult_args *args;
    hg_return_t hret = HG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
345 346 347 348 349 350 351 352 353 354 355 356

    // "effective" rank for the lookup loop
    int eff_rank = 0;

    // set the hg class up front - need for destructing addrs
    hgctx = margo_get_context(mid);
    if (hgctx == NULL) return HG_INVALID_PARAM;
    s->hgcl = margo_get_class(mid);
    if (s->hgcl == NULL) return HG_INVALID_PARAM;

    // perform search for my rank if not already set
    if (s->rank == SSG_RANK_UNKNOWN) {
357
        hret = ssg_resolve_rank(s, s->hgcl);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
358 359 360 361 362 363 364 365 366 367 368
        if (hret != HG_SUCCESS) return hret;
    }

    if (s->rank == SSG_EXTERNAL_RANK) {
        // do a completely arbitrary effective rank determination to try and
        // prevent everyone talking to the same member at once
        eff_rank = (((intptr_t)hgctx)/sizeof(hgctx)) % s->num_addrs;
    }
    else
        eff_rank = s->rank;

369 370 371 372 373 374 375 376 377 378 379
    // initialize ULTs
    ults = malloc(s->num_addrs * sizeof(*ults));
    if (ults == NULL) return HG_NOMEM_ERROR;
    args = malloc(s->num_addrs * sizeof(*args));
    if (args == NULL) {
        free(ults);
        return HG_NOMEM_ERROR;
    }
    for (int i = 0; i < s->num_addrs; i++)
        ults[i] = ABT_THREAD_NULL;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
380
    for (int i = eff_rank+1; i < s->num_addrs; i++) {
381 382 383 384 385 386 387 388 389 390
        args[i].ssg = s;
        args[i].mid = mid;
        args[i].rank = i;

        int aret = ABT_thread_create(*margo_get_handler_pool(mid), &lookup_ult,
                &args[i], ABT_THREAD_ATTR_NULL, &ults[i]);
        if (aret != ABT_SUCCESS) {
            hret = HG_OTHER_ERROR;
            goto fin;
        }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
391 392
    }
    for (int i = 0; i < eff_rank; i++) {
393 394 395 396 397 398 399 400 401 402
        args[i].ssg = s;
        args[i].mid = mid;
        args[i].rank = i;

        int aret = ABT_thread_create(*margo_get_handler_pool(mid), &lookup_ult,
                &args[i], ABT_THREAD_ATTR_NULL, &ults[i]);
        if (aret != ABT_SUCCESS) {
            hret = HG_OTHER_ERROR;
            goto fin;
        }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
403 404
    }
    if (s->rank == SSG_EXTERNAL_RANK) {
405 406 407 408 409 410 411 412 413 414
        args[eff_rank].ssg = s;
        args[eff_rank].mid = mid;
        args[eff_rank].rank = eff_rank;

        int aret = ABT_thread_create(*margo_get_handler_pool(mid), &lookup_ult,
                &args[eff_rank], ABT_THREAD_ATTR_NULL, &ults[eff_rank]);
        if (aret != ABT_SUCCESS) {
            hret = HG_OTHER_ERROR;
            goto fin;
        }
Jonathan Jenkins's avatar
Jonathan Jenkins committed
415 416
    }

417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447
    // wait on all
    for (int i = 0; i < s->num_addrs; i++) {
        if (ults[i] != ABT_THREAD_NULL) {
            int aret = ABT_thread_join(ults[i]);
            ABT_thread_free(&ults[i]);
            ults[i] = ABT_THREAD_NULL; // in case of cascading failure from join
            if (aret != ABT_SUCCESS) {
                hret = HG_OTHER_ERROR;
                break;
            }
            else if (args[i].out != HG_SUCCESS) {
                hret = args[i].out;
                break;
            }
        }
    }

fin:
    // cleanup
    if (ults != NULL) {
        for (int i = 0; i < s->num_addrs; i++) {
            if (ults[i] != ABT_THREAD_NULL) {
                ABT_thread_cancel(ults[i]);
                ABT_thread_free(ults[i]);
            }
        }
        free(ults);
    }
    if (args != NULL) free(args);

    return hret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
448 449 450
}
#endif

Jonathan Jenkins's avatar
Jonathan Jenkins committed
451 452
void ssg_finalize(ssg_t s)
{
453 454
    if (s == SSG_NULL) return;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
455
    for (int i = 0; i < s->num_addrs; i++) {
Jonathan Jenkins's avatar
Jonathan Jenkins committed
456
        if (s->addrs[i] != HG_ADDR_NULL) HG_Addr_free(s->hgcl, s->addrs[i]);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
    }
    free(s->backing_buf);
    free(s->addr_strs);
    free(s->addrs);
    free(s);
}

int ssg_get_rank(const ssg_t s)
{
    return s->rank;
}

int ssg_get_count(const ssg_t s)
{
    return s->num_addrs;
}

Jonathan Jenkins's avatar
Jonathan Jenkins committed
474
hg_addr_t ssg_get_addr(const ssg_t s, int rank)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
475
{
476
    if (rank >= 0 && rank < s->num_addrs)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
477 478
        return s->addrs[rank];
    else
Jonathan Jenkins's avatar
Jonathan Jenkins committed
479
        return HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
480 481 482 483
}

const char * ssg_get_addr_str(const ssg_t s, int rank)
{
484
    if (rank >= 0 && rank < s->num_addrs)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
485 486 487 488 489 490 491
        return s->addr_strs[rank];
    else
        return NULL;
}

// serialization format looks like:
// < num members, buffer size, buffer... >
Jonathan Jenkins's avatar
Jonathan Jenkins committed
492
// doesn't attempt to grab hg_addr's, string buffers, etc. - client will be
Jonathan Jenkins's avatar
Jonathan Jenkins committed
493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563
// responsible for doing a separate address lookup routine
hg_return_t hg_proc_ssg_t(hg_proc_t proc, ssg_t *s)
{
    // error and return handling
    hg_return_t hret = HG_SUCCESS;
    char * err_str = NULL;

    // input/output vars + helpers for ssg decode setup
    ssg_t ss = NULL;

    switch(hg_proc_get_op(proc)) {
        case HG_ENCODE:
            ss = *s;
            // encode address count
            hret = hg_proc_int32_t(proc, &ss->num_addrs);
            if (hret != HG_SUCCESS) { err_str = "ssg num addrs"; goto end; }
            // encode addr
            hret = hg_proc_int32_t(proc, &ss->buf_size);
            if (hret != HG_SUCCESS) { err_str = "ssg buf size"; goto end; }
            // encode addr string, simple as blitting the backing buffer
            hret = hg_proc_memcpy(proc, ss->backing_buf, ss->buf_size);
            if (hret != HG_SUCCESS) { err_str = "ssg addr buf"; goto end; }
            break;

        case HG_DECODE:
            // create the output
            *s = NULL;
            ss = malloc(sizeof(*ss));
            if (ss == NULL) {
                err_str = "ssg alloc";
                hret = HG_NOMEM_ERROR;
                goto end;
            }
            ss->addr_strs = NULL;
            ss->addrs = NULL;
            ss->backing_buf = NULL;
            // get address count
            hret = hg_proc_int32_t(proc, &ss->num_addrs);
            if (hret != HG_SUCCESS) { err_str = "ssg num addrs"; goto end; }
            // get number of bytes for the address
            hret = hg_proc_int32_t(proc, &ss->buf_size);
            if (hret != HG_SUCCESS) { err_str = "ssg buf size"; goto end; }
            // allocate output buffer
            ss->backing_buf = malloc(ss->buf_size);
            if (hret != HG_SUCCESS) {
                err_str = "ssg buf alloc";
                hret = HG_NOMEM_ERROR;
                goto end;
            }
            hret = hg_proc_memcpy(proc, ss->backing_buf, ss->buf_size);
            if (hret != HG_SUCCESS) { err_str = "ssg addr buf"; goto end; }

            // set the remaining ssg vars

            ss->addr_strs = NULL; ss->addrs = NULL;
            ss->rank = -1; // receivers aren't part of the group

            ss->addr_strs = setup_addr_str_list(ss->num_addrs, ss->backing_buf);
            if (ss->addr_strs == NULL) {
                err_str = "ssg addr strs alloc";
                hret = HG_NOMEM_ERROR;
                goto end;
            }

            ss->addrs = malloc(ss->num_addrs * sizeof(*ss->addrs));
            if (ss->addrs == NULL) {
                err_str = "ssg addrs alloc";
                hret = HG_NOMEM_ERROR;
                goto end;
            }
            for (int i = 0; i < ss->num_addrs; i++) {
Jonathan Jenkins's avatar
Jonathan Jenkins committed
564
                ss->addrs[i] = HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
            }

            // success: set the output
            *s = ss;
            break;

        case HG_FREE:
            if (s != NULL && *s != NULL) {
                err_str = "ssg shouldn't be freed via HG_Free_*";
                hret = HG_INVALID_PARAM;
            }
            goto end;

        default:
            err_str = "bad proc mode";
            hret = HG_INVALID_PARAM;
    }
end:
    if (err_str) {
        HG_LOG_ERROR("Proc error: %s", err_str);
        if (hg_proc_get_op(proc) == HG_DECODE) {
            free(ss->addr_strs);
            free(ss->addrs);
            free(ss->backing_buf);
            free(ss);
        }
    }
    return hret;
}

595 596 597 598 599 600 601 602 603 604 605 606 607 608
int ssg_dump(const ssg_t s, const char *fname)
{
    // file to write to
    int fd = -1;
    ssize_t written;

    // string to xform and dump
    char * addrs_dup = NULL;
    char * tok = NULL;
    char * addrs_dup_end = NULL;

    // return code
    int ret = 0;

609 610
    // copy the backing buffer, replacing all null chars with
    // newlines
611 612 613 614 615
    addrs_dup = malloc(s->buf_size);
    if (addrs_dup == NULL) { errno = ENOMEM; ret = -1; goto end; }
    memcpy(addrs_dup, s->backing_buf, s->buf_size);
    tok = addrs_dup;
    addrs_dup_end = addrs_dup + s->buf_size;
616
    for (int i = 0; i < s->num_addrs; i++) {
617 618 619 620 621 622
        tok = memchr(tok, '\0', addrs_dup_end - tok);
        if (tok == NULL) { errno = EINVAL; ret = -1; goto end; }
        *tok = '\n';
    }

    // open the file and dump in a single call
623
    fd = open(fname, O_WRONLY | O_CREAT | O_EXCL, 0644);
624 625
    if (fd == -1) { ret = -1; goto end; }
    // don't include the null char at the end
626 627
    written = write(fd, addrs_dup, s->buf_size);
    if (written != s->buf_size) ret = -1;
628 629 630 631 632 633 634 635

end:
    free(addrs_dup);
    if (fd != -1) close(fd);

    return ret;
}

Jonathan Jenkins's avatar
Jonathan Jenkins committed
636 637
typedef struct serv_addr_out
{
Jonathan Jenkins's avatar
Jonathan Jenkins committed
638
    hg_addr_t addr;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
639 640 641
    int set;
} serv_addr_out_t;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
642
static hg_return_t lookup_serv_addr_cb(const struct hg_cb_info *info)
Jonathan Jenkins's avatar
Jonathan Jenkins committed
643 644 645 646
{
    serv_addr_out_t *out = info->arg;
    out->addr = info->info.lookup.addr;
    out->set = 1;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
647
    return HG_SUCCESS;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
648 649
}

Jonathan Jenkins's avatar
Jonathan Jenkins committed
650 651
static hg_addr_t lookup_serv_addr(
        hg_context_t *hgctx,
Jonathan Jenkins's avatar
Jonathan Jenkins committed
652 653 654
        const char *info_str)
{
    serv_addr_out_t out;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
655
    hg_return_t hret;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
656

Jonathan Jenkins's avatar
Jonathan Jenkins committed
657
    out.addr = HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
658 659
    out.set = 0;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
660 661 662
    hret = HG_Addr_lookup(hgctx, &lookup_serv_addr_cb, &out,
            info_str, HG_OP_ID_IGNORE);
    if (hret != HG_SUCCESS) return HG_ADDR_NULL;
Jonathan Jenkins's avatar
Jonathan Jenkins committed
663 664 665 666 667

    // run the progress loop until we've got the output
    do {
        unsigned int count = 0;
        do {
Jonathan Jenkins's avatar
Jonathan Jenkins committed
668 669
            hret = HG_Trigger(hgctx, 0, 1, &count);
        } while (hret == HG_SUCCESS && count > 0);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
670 671 672

        if (out.set != 0) break;

Jonathan Jenkins's avatar
Jonathan Jenkins committed
673 674
        hret = HG_Progress(hgctx, 5000);
    } while(hret == HG_SUCCESS || hret == HG_TIMEOUT);
Jonathan Jenkins's avatar
Jonathan Jenkins committed
675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690

    return out.addr;
}

static char** setup_addr_str_list(int num_addrs, char * buf)
{
    char ** ret = malloc(num_addrs * sizeof(*ret));
    if (ret == NULL) return NULL;

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