darshan-lustre_old.c 17.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
/*
 * Copyright (C) 2015 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
 */

#define _XOPEN_SOURCE 500
#define _GNU_SOURCE

#include "darshan-runtime-config.h"
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <time.h>
#include <stdlib.h>
#include <assert.h>
#include <pthread.h>
#include <sys/ioctl.h>

/* XXX stick this into autoconf .h */
#include <lustre/lustre_user.h>

#include "uthash.h"

#include "darshan.h"
#include "darshan-dynamic.h"
#include "darshan-lustre.h"

struct lustre_runtime *lustre_runtime = NULL;
static pthread_mutex_t lustre_runtime_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
static int instrumentation_disabled = 0;
static int my_rank = -1;

static void lustre_runtime_initialize(void);

static void lustre_begin_shutdown(void);
static void lustre_get_output_data(MPI_Comm mod_comm, darshan_record_id *shared_recs,
    int shared_rec_count, void **lustre_buf, int *lustre_buf_sz);
static void lustre_shutdown(void);
static int lustre_record_compare(const void* a_p, const void* b_p);
static void lustre_record_reduction_op(void* infile_v, void* inoutfile_v,
    int *len, MPI_Datatype *datatype);

#define LUSTRE_LOCK() pthread_mutex_lock(&lustre_runtime_mutex)
#define LUSTRE_UNLOCK() pthread_mutex_unlock(&lustre_runtime_mutex)
#define LUSTRE_RECORD_SIZE( osts ) ( sizeof(struct darshan_lustre_record) + sizeof(int64_t) * (osts - 1) )

void darshan_instrument_lustre_file(const char* filepath, int fd)
{
    struct lustre_record_runtime *rec_rt;
    struct darshan_lustre_record *rec;
    struct darshan_fs_info fs_info;
    darshan_record_id rec_id;
    int limit_flag;
    int i;
    struct lov_user_md *lum;
    size_t lumsize = sizeof(struct lov_user_md) +
        LOV_MAX_STRIPE_COUNT * sizeof(struct lov_user_ost_data);
    size_t rec_size;

    LUSTRE_LOCK();
    /* make sure the lustre module is already initialized */
    lustre_runtime_initialize();

    /* if we can't issue ioctl, we have no counter data at all */
    if ( (lum = calloc(1, lumsize)) == NULL )
        return;

    /* find out the OST count of this file so we can allocate memory */
    lum->lmm_magic = LOV_USER_MAGIC;
    lum->lmm_stripe_count = LOV_MAX_STRIPE_COUNT;

    /* -1 means ioctl failed, likely because file isn't on Lustre */
    if ( ioctl( fd, LL_IOC_LOV_GETSTRIPE, (void *)lum ) == -1 )
    {
        free(lum);
        return;
    }

    rec_size = LUSTRE_RECORD_SIZE( lum->lmm_stripe_count );

    {
        /* broken out for clarity */
        void *end_of_new_record = (char*)lustre_runtime->next_free_record + rec_size;
        void *end_of_rec_buffer = (char*)lustre_runtime->record_buffer + lustre_runtime->record_buffer_max;
        limit_flag = ( end_of_new_record > end_of_rec_buffer );
    }

    /* register a Lustre file record with Darshan */
    fs_info.fs_type = -1;
    darshan_core_register_record(
        (void *)filepath,
        strlen(filepath),
        DARSHAN_LUSTRE_MOD,
        1,
        limit_flag,
        &rec_id,
        &fs_info);

    /* if record id is 0, darshan has no more memory for instrumenting */
    if(rec_id == 0)
    {
        free(lum);
        LUSTRE_UNLOCK();
        return;
    }

    /* search the hash table for this file record, and initialize if not found */
    HASH_FIND(hlink, lustre_runtime->record_runtime_hash, &rec_id, sizeof(darshan_record_id), rec_rt );
    if ( !rec_rt ) {
        /* allocate a new lustre record and append it to the array */
        rec_rt = &(lustre_runtime->record_runtime_array[lustre_runtime->record_count]);
        rec_rt->record = lustre_runtime->next_free_record;
        rec_rt->record_size = rec_size;
        lustre_runtime->next_free_record = (char*)(lustre_runtime->next_free_record) + rec_size;
        lustre_runtime->record_buffer_used += rec_size;
        rec = rec_rt->record;
        rec->rec_id = rec_id;
        rec->rank = my_rank;

        /* implicit assumption here that none of these counters will change
         * after the first time a file is opened.  This may not always be
         * true in the future */
        if ( fs_info.fs_type != -1 ) 
        {
            rec->counters[LUSTRE_OSTS] = fs_info.ost_count;
            rec->counters[LUSTRE_MDTS] = fs_info.mdt_count;
        }
        else
        {
            rec->counters[LUSTRE_OSTS] = -1;
            rec->counters[LUSTRE_MDTS] = -1;
        }

        rec->counters[LUSTRE_STRIPE_SIZE] = lum->lmm_stripe_size;
        rec->counters[LUSTRE_STRIPE_WIDTH] = lum->lmm_stripe_count;
        rec->counters[LUSTRE_STRIPE_OFFSET] = lum->lmm_stripe_offset;
        for ( i = 0; i < lum->lmm_stripe_count; i++ )
            rec->ost_ids[i] = lum->lmm_objects[i].l_ost_idx;
        free(lum);

        HASH_ADD(hlink, lustre_runtime->record_runtime_hash, record->rec_id, sizeof(darshan_record_id), rec_rt);

        lustre_runtime->record_count++;
    }

    LUSTRE_UNLOCK();
    return;
}

static void lustre_runtime_initialize()
{
    int mem_limit;
    int max_records;
    struct darshan_module_funcs lustre_mod_fns =
    {
        .begin_shutdown = &lustre_begin_shutdown,
        .get_output_data = &lustre_get_output_data,
        .shutdown = &lustre_shutdown
    };

    /* don't do anything if already initialized or instrumenation is disabled */
    if(lustre_runtime || instrumentation_disabled)
        return;

    /* register the lustre module with darshan-core */
    darshan_core_register_module(
        DARSHAN_LUSTRE_MOD,
        &lustre_mod_fns,
        &my_rank,
        &mem_limit,
        NULL);

    /* return if no memory assigned by darshan core */
    if(mem_limit == 0)
        return;

    lustre_runtime = malloc(sizeof(*lustre_runtime));
    if(!lustre_runtime)
        return;
    memset(lustre_runtime, 0, sizeof(*lustre_runtime));

    /* allocate the full size of the memory limit we are given */
    lustre_runtime->record_buffer= malloc(mem_limit);
    if(!lustre_runtime->record_buffer)
    {
        lustre_runtime->record_buffer_max = 0;
        return;
    }
    lustre_runtime->record_buffer_max = mem_limit;
    lustre_runtime->next_free_record = lustre_runtime->record_buffer;
    memset(lustre_runtime->record_buffer, 0, lustre_runtime->record_buffer_max);

    /* Allocate array of Lustre runtime data.  We calculate the maximum possible
     * number of records that will fit into mem_limit by assuming that each
     * record has the minimum possible OST count, then allocate that many 
     * runtime records.  record_buffer will always run out of memory before
     * we overflow record_runtime_array.
     */
    max_records = mem_limit / sizeof(struct darshan_lustre_record);
    lustre_runtime->record_runtime_array =
        malloc( max_records * sizeof(struct lustre_record_runtime));
    if(!lustre_runtime->record_runtime_array)
    {
        lustre_runtime->record_buffer_max = 0;
        free( lustre_runtime->record_buffer );
        return;
    }
    memset(lustre_runtime->record_runtime_array, 0,
        max_records * sizeof(struct lustre_record_runtime));

    return;
}

/**************************************************************************
 * Functions exported by Lustre module for coordinating with darshan-core *
 **************************************************************************/

static void lustre_begin_shutdown(void)
{
    assert(lustre_runtime);

    LUSTRE_LOCK();
    /* disable further instrumentation while Darshan shuts down */
    instrumentation_disabled = 1;
    LUSTRE_UNLOCK();

    return;
}

static void lustre_get_output_data(
    MPI_Comm mod_comm,
    darshan_record_id *shared_recs,
    int shared_rec_count,
    void **lustre_buf,
    int *lustre_buf_sz)
{
    struct lustre_record_runtime *file;
    int i, ishared;
    int *rec_lengths;
    size_t shared_rec_size;
    struct darshan_lustre_record *red_send_buf = NULL;
    struct darshan_lustre_record *red_recv_buf = NULL;
    MPI_Datatype red_type;
    MPI_Aint *rec_offsets;
    MPI_Op red_op;

    assert(lustre_runtime);

    /* if there are globally shared files, do a shared file reduction */
    /* NOTE: the shared file reduction is also skipped if the 
     * DARSHAN_DISABLE_SHARED_REDUCTION environment variable is set.
     */
    if (shared_rec_count && !getenv("DARSHAN_DISABLE_SHARED_REDUCTION"))
    {
        /* necessary initialization of shared records */
        for(i = 0; i < shared_rec_count; i++)
        {
            HASH_FIND(hlink, lustre_runtime->record_runtime_hash, &shared_recs[i],
                sizeof(darshan_record_id), file);
            assert(file);

            file->record->rank = -1;
        }

        /* sort the array of files descending by rank so that we get all of the 
         * shared files (marked by rank -1) in a contiguous portion at end 
         * of the array
         */
        sort_lustre_records();

        /* make red_send_buf point to the first shared-file record */
        ishared = lustre_runtime->record_count - shared_rec_count;
        red_send_buf =
            (lustre_runtime->record_runtime_array[ishared]).record;

        /* allocate memory for the reduction output on rank 0 */
        if (my_rank == 0)
        {
            shared_rec_size = lustre_runtime->record_buffer_used - ((char*)red_send_buf - (char*)lustre_runtime->record_buffer);
            red_recv_buf = malloc(shared_rec_size);
            if (!red_recv_buf)
                return;
        }

        /* need to build rec_lengths (array of ints) and rec_offsets (array of ints) */
        rec_lengths = malloc(sizeof(*rec_lengths) * shared_rec_count);
        rec_offsets = malloc(sizeof(*rec_offsets) * shared_rec_count);
        for ( i = ishared; i < shared_rec_count; i ++ )
        {
            rec_lengths[i] = (lustre_runtime->record_runtime_array[i]).record_size;
            rec_offsets[i] = (char*)((lustre_runtime->record_runtime_array[i]).record) -
                (char*)((lustre_runtime->record_runtime_array[ishared]).record);
        }

        /* ... */
        DARSHAN_MPI_CALL(PMPI_Type_hindexed)(
            shared_rec_count,
            rec_lengths,
            rec_offsets,
            MPI_BYTE,
            &red_type
        );
        DARSHAN_MPI_CALL(PMPI_Type_commit)(&red_type);
        DARSHAN_MPI_CALL(PMPI_Op_create)(lustre_record_reduction_op, 1, &red_op);
        DARSHAN_MPI_CALL(PMPI_Reduce)(red_send_buf, red_recv_buf,
            shared_rec_count, red_type, red_op, 0, mod_comm);

        /* clean up reduction state */
        if (my_rank == 0)
        {
            memcpy(&(lustre_runtime->record_buffer[ishared]), red_recv_buf,
                shared_rec_size);
            free(red_recv_buf);
        }
        else
        {
            lustre_runtime->record_count -= shared_rec_count;
        }
        free(rec_lengths);
        free(rec_offsets);
        DARSHAN_MPI_CALL(PMPI_Type_free)(&red_type);
        DARSHAN_MPI_CALL(PMPI_Op_free)(&red_op);
    }

    *lustre_buf = (void *)(lustre_runtime->record_buffer);
    *lustre_buf_sz = lustre_runtime->record_buffer_used;

    return;
}

static void lustre_shutdown(void)
{
    assert(lustre_runtime);

    HASH_CLEAR(hlink, lustre_runtime->record_runtime_hash);
    free(lustre_runtime->record_runtime_array);
    free(lustre_runtime->record_buffer);
    free(lustre_runtime);
    lustre_runtime = NULL;

    return;
}

/* compare function for sorting file records by descending rank */
static int lustre_record_compare(const void* a_p, const void* b_p)
{
    const struct lustre_record_runtime* a = a_p;
    const struct lustre_record_runtime* b = b_p;

    if (a->record->rank < b->record->rank)
        return 1;
    if (a->record->rank > b->record->rank)
        return -1;

    /* if ( a->record->rank == b->record->rank ) we MUST do a secondary
     * sort so that the order of qsort is fully deterministic and consistent
     * across all MPI ranks.  Without a secondary sort, the sort order can
     * be affected by rank-specific variations (e.g., the order in which
     * files are first opened).
     */

    return 0;
}

/*
 * Sort the record_runtimes and records by MPI rank to facilitate shared redux.
 * This requires craftiness and additional heap utilization because the records
 * (but not record_runtimes) have variable size.  Currently has to temporarily
 * duplicate the entire record_buffer; there is room for more memory-efficient
 * optimization if this becomes a scalability issue.
 */
int sort_lustre_records()
{
    int i;
    struct darshan_lustre_record *rec;
    struct lustre_record_runtime *rec_rt, *tmp_rec_rt;
    char  *new_buf, *p;

    /* Create a new buffer to store an entire replica of record_buffer.  Since
     * we know the exact size of record_buffer's useful data at this point, we
     * can allocate the exact amount we need instead of record_buffer_max */
    new_buf = malloc(lustre_runtime->record_buffer_used);
    p = new_buf;
    if ( !new_buf )
        return 1;

    /* qsort breaks the hash table, so delete it now to free its memory buffers
     * and prevent later confusion */
    HASH_ITER( hlink, lustre_runtime->record_runtime_hash, rec_rt, tmp_rec_rt )
        HASH_DELETE( hlink, lustre_runtime->record_runtime_hash, rec_rt );

    /* sort the runtime records, which is has fixed-length elements */
    qsort(
        lustre_runtime->record_runtime_array,
        lustre_runtime->record_count,
        sizeof(struct lustre_record_runtime),
        lustre_record_compare
    );

    /* rebuild the hash and array with the qsorted runtime records */
    for ( i = 0; i < lustre_runtime->record_count; i++ )
    {
        rec_rt = &(lustre_runtime->record_runtime_array[i]);
        HASH_ADD(hlink, lustre_runtime->record_runtime_hash, record->rec_id, sizeof(darshan_record_id), rec_rt );
    }

    /* create reordered record buffer, then copy it back in place */
    for ( i = 0; i < lustre_runtime->record_count; i++ )
    {
        rec_rt = &(lustre_runtime->record_runtime_array[i]);
        memcpy( p, rec_rt->record, rec_rt->record_size );
        /* fix record pointers within each runtime record too - pre-emptively
         * point them at where they will live in record_buffer after we memcpy
         * below */
        rec_rt->record = (struct darshan_lustre_record *)((char*)(lustre_runtime->record_buffer) + (p - new_buf));

        p += rec_rt->record_size;
    }
    memcpy( 
        lustre_runtime->record_buffer, 
        new_buf, 
        lustre_runtime->record_buffer_used );

    free(new_buf);
    return 0;
}

/* this is just boilerplate reduction code that isn't currently used */
static void lustre_record_reduction_op(void* infile_v, void* inoutfile_v,
    int *len, MPI_Datatype *datatype)
{
    struct darshan_lustre_record tmp_record;
    struct darshan_lustre_record *infile = infile_v;
    struct darshan_lustre_record *inoutfile = inoutfile_v;
    int i, j;

    assert(lustre_runtime);

    for( i=0; i<*len; i++ )
    {
        memset(&tmp_record, 0, sizeof(struct darshan_lustre_record));
        tmp_record.rec_id = infile->rec_id;
        tmp_record.rank = -1;

        /* preserve only rank 0's value */
        for( j = LUSTRE_OSTS; j < LUSTRE_NUM_INDICES; j++)
        {
            if ( my_rank == 0 ) 
            {
                tmp_record.counters[j] = infile->counters[j];
            }
            else
            {
                tmp_record.counters[j] = inoutfile->counters[j];
            }
        }

        /* update pointers */
        *inoutfile = tmp_record;
        inoutfile++;
        infile++;
    }

    return;
}

/*
 *  Dump the memory structure of our records and runtime records
 */
void print_lustre_runtime( void )
{
    int i, j;
    struct darshan_lustre_record *rec;

    /* print what we just loaded */
    for ( i = 0; i < lustre_runtime->record_count; i++ )
    {
        rec = (lustre_runtime->record_runtime_array[i]).record;
        printf( "File %2d\n", i );
        for ( j = 0; j < LUSTRE_NUM_INDICES; j++ )
        {
            printf( "  Counter %-2d: %10ld, addr %ld\n", 
                j, 
                rec->counters[j],
                (char*)(&(rec->counters[j])) - (char*)(lustre_runtime->record_buffer) );
        }
        for ( j = 0; j < rec->counters[LUSTRE_STRIPE_WIDTH]; j++ )
        {
            if ( j > 0 && j % 2 == 0 ) printf("\n");
            printf( "  Stripe  %-2d: %10ld, addr %-9d", 
                j, 
                rec->ost_ids[j],
                (char*)(&(rec->ost_ids[j])) - (char*)(lustre_runtime->record_buffer) );
        }
        printf( "\n" );
    }
    return;
}

/*
 *  Dump the order in which records appear in memory
 */
void print_array( void )
{
    int i;
    struct lustre_record_runtime *rec_rt;
    printf("*** DUMPING RECORD LIST BY ARRAY SEQUENCE\n");
    for ( i = 0; i < lustre_runtime->record_count; i++ )
    {
        rec_rt = &(lustre_runtime->record_runtime_array[i]);
        printf( "*** record %d rank %d osts %d\n", 
            rec_rt->record->rec_id, 
            rec_rt->record->rank,
            rec_rt->record->counters[LUSTRE_STRIPE_WIDTH]);
    }
}
void print_hash( void )
{
    struct lustre_record_runtime *rec_rt, *tmp_rec_rt;
    printf("*** DUMPING RECORD LIST BY HASH SEQUENCE\n");
    HASH_ITER( hlink, lustre_runtime->record_runtime_hash, rec_rt, tmp_rec_rt )
    {
        printf( "*** record %d rank %d osts %d\n", 
            rec_rt->record->rec_id, 
            rec_rt->record->rank,
            rec_rt->record->counters[LUSTRE_STRIPE_WIDTH]);
    }
    return;
}




/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ts=8 sts=4 sw=4 expandtab
 */