darshan-core.c 49.6 KB
Newer Older
1
/*
Shane Snyder's avatar
Shane Snyder committed
2 3 4
 * Copyright (C) 2015 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
5 6
 */

7
#define _XOPEN_SOURCE 500
8
#define _GNU_SOURCE
9

10 11 12 13 14 15
#include "darshan-runtime-config.h"

#include <stdio.h>
#ifdef HAVE_MNTENT_H
#include <mntent.h>
#endif
16 17 18 19 20
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <limits.h>
#include <pthread.h>
21 22 23
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/vfs.h>
24
#include <zlib.h>
25
#include <mpi.h>
26
#include <assert.h>
27

28
#include "uthash.h"
Shane Snyder's avatar
Shane Snyder committed
29
#include "darshan.h"
30
#include "darshan-core.h"
Shane Snyder's avatar
Shane Snyder committed
31
#include "darshan-dynamic.h"
32

33
extern char* __progname;
34
extern char* __progname_full;
35

36
/* internal variable delcarations */
37
static struct darshan_core_runtime *darshan_core = NULL;
38
static pthread_mutex_t darshan_core_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
39
static int my_rank = -1;
40
static int nprocs = -1;
41
static int darshan_mem_alignment = 1;
42

43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
/* paths prefixed with the following directories are not traced by darshan */
char* darshan_path_exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

58 59 60 61 62 63 64 65 66 67 68 69 70 71 72
#ifdef DARSHAN_BGQ
extern void bgq_runtime_initialize();
#endif

/* array of init functions for modules which need to be statically
 * initialized by darshan at startup time
 */
void (*mod_static_init_fns[])(void) =
{
#ifdef DARSHAN_BGQ
    &bgq_runtime_initialize,
#endif
    NULL
};

Shane Snyder's avatar
Shane Snyder committed
73 74 75
#define DARSHAN_CORE_LOCK() pthread_mutex_lock(&darshan_core_mutex)
#define DARSHAN_CORE_UNLOCK() pthread_mutex_unlock(&darshan_core_mutex)

76 77 78 79 80 81
/* FS mount information */
#define DARSHAN_MAX_MNTS 64
#define DARSHAN_MAX_MNT_PATH 256
#define DARSHAN_MAX_MNT_TYPE 32
struct mnt_data
{
82
    int block_size;
83 84 85 86 87 88
    char path[DARSHAN_MAX_MNT_PATH];
    char type[DARSHAN_MAX_MNT_TYPE];
};
static struct mnt_data mnt_data_array[DARSHAN_MAX_MNTS];
static int mnt_data_count = 0;

89 90 91 92
/* prototypes for internal helper functions */
static void darshan_get_logfile_name(
    char* logfile_name, int jobid, struct tm* start_tm);
static void darshan_log_record_hints_and_ver(
93 94 95 96 97 98
    struct darshan_core_runtime* core);
static void darshan_get_exe_and_mounts_root(
    struct darshan_core_runtime *core, char* trailing_data,
    int space_left);
static char* darshan_get_exe_and_mounts(
    struct darshan_core_runtime *core);
99 100
static void darshan_block_size_from_path(
    const char *path, int *block_size);
101
static void darshan_get_shared_records(
102
    struct darshan_core_runtime *core, darshan_record_id *shared_recs);
103
static int darshan_log_open_all(
104
    char *logfile_name, MPI_File *log_fh);
105
static int darshan_deflate_buffer(
Shane Snyder's avatar
Shane Snyder committed
106 107
    void **pointers, int *lengths, int count, char *comp_buf,
    int *comp_buf_length);
108
static int darshan_log_write_record_hash(
109
    MPI_File log_fh, struct darshan_core_runtime *core,
110 111 112
    uint64_t *inout_off);
static int darshan_log_append_all(
    MPI_File log_fh, struct darshan_core_runtime *core, void *buf,
Shane Snyder's avatar
Shane Snyder committed
113
    int count, uint64_t *inout_off);
Shane Snyder's avatar
Shane Snyder committed
114 115
static void darshan_core_cleanup(
    struct darshan_core_runtime* core);
116

117 118
/* *********************************** */

Shane Snyder's avatar
Shane Snyder committed
119
void darshan_core_initialize(int argc, char **argv)
120 121 122 123
{
    int i;
    int internal_timing_flag = 0;
    double init_start, init_time, init_max;
124
    char *envstr;
125 126 127
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;
128 129
    int ret;
    int tmpval;
130 131

    DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
132
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank);
133 134 135 136

    if(getenv("DARSHAN_INTERNAL_TIMING"))
        internal_timing_flag = 1;

137
    if(internal_timing_flag)
138 139 140
        init_start = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* setup darshan runtime if darshan is enabled and hasn't been initialized already */
141
    if(!getenv("DARSHAN_DISABLE") && !darshan_core)
142
    {
143
        #if (__DARSHAN_MEM_ALIGNMENT < 1)
144 145
            #error Darshan must be configured with a positive value for --with-mem-align
        #endif
146
        envstr = getenv(DARSHAN_MEM_ALIGNMENT_OVERRIDE);
147 148 149 150 151 152 153 154 155 156 157
        if(envstr)
        {
            ret = sscanf(envstr, "%d", &tmpval);
            /* silently ignore if the env variable is set poorly */
            if(ret == 1 && tmpval > 0)
            {
                darshan_mem_alignment = tmpval;
            }
        }
        else
        {
158
            darshan_mem_alignment = __DARSHAN_MEM_ALIGNMENT;
159 160 161 162 163 164 165
        }

        /* avoid floating point errors on faulty input */
        if (darshan_mem_alignment < 1)
        {
            darshan_mem_alignment = 1;
        }
166 167 168 169

        /* allocate structure to track darshan_core_runtime information */
        darshan_core = malloc(sizeof(*darshan_core));
        if(darshan_core)
170
        {
171
            memset(darshan_core, 0, sizeof(*darshan_core));
172

173 174 175 176
            darshan_core->log_job.uid = getuid();
            darshan_core->log_job.start_time = time(NULL);
            darshan_core->log_job.nprocs = nprocs;
            darshan_core->wtime_offset = DARSHAN_MPI_CALL(PMPI_Wtime)();
177 178

            /* record exe and arguments */
179
            for(i=0; i<argc; i++)
180
            {
181
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
182 183
                strncat(darshan_core->exe, argv[i], chars_left);
                if(i < (argc-1))
184
                {
185
                    chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
186
                    strncat(darshan_core->exe, " ", chars_left);
187 188 189 190 191 192 193 194
                }
            }

            /* if we don't see any arguments, then use glibc symbol to get
             * program name at least (this happens in fortran)
             */
            if(argc == 0)
            {
195
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
196
                strncat(darshan_core->exe, __progname_full, chars_left);
197
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
198
                strncat(darshan_core->exe, " <unknown args>", chars_left);
199 200 201 202 203
            }

            if(chars_left == 0)
            {
                /* we ran out of room; mark that string was truncated */
204
                truncate_offset = DARSHAN_EXE_LEN - strlen(truncate_string);
205
                sprintf(&darshan_core->exe[truncate_offset], "%s",
206 207
                    truncate_string);
            }
208 209 210

            /* collect information about command line and mounted file systems */
            darshan_core->trailing_data = darshan_get_exe_and_mounts(darshan_core);
211 212
        }

213 214 215 216 217 218 219
        /* maybe bootstrap modules with static initializers */
        i = 0;
        while(mod_static_init_fns[i])
        {
            (*mod_static_init_fns[i])();
            i++;
        }
220 221
    }

222 223 224 225 226
    if(internal_timing_flag)
    {
        init_time = DARSHAN_MPI_CALL(PMPI_Wtime)() - init_start;
        DARSHAN_MPI_CALL(PMPI_Reduce)(&init_time, &init_max, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
227
        if(my_rank == 0)
228
        {
229 230
            fprintf(stderr, "#darshan:<op>\t<nprocs>\t<time>\n");
            fprintf(stderr, "darshan:init\t%d\t%f\n", nprocs, init_max);
231 232 233 234 235 236
        }
    }

    return;
}

Shane Snyder's avatar
Shane Snyder committed
237
void darshan_core_shutdown()
238
{
239
    int i;
240
    char *logfile_name;
241
    struct darshan_core_runtime *final_core;
242
    int internal_timing_flag = 0;
243 244
    char *envjobid;
    char *jobid_str;
245
    int jobid;
246
    struct tm *start_tm;
247
    time_t start_time_tmp;
248 249
    int ret = 0;
    int all_ret = 0;
250 251
    int64_t first_start_time;
    int64_t last_end_time;
252 253
    int local_mod_use[DARSHAN_MAX_MODS] = {0};
    int global_mod_use_count[DARSHAN_MAX_MODS] = {0};
254
    darshan_record_id shared_recs[DARSHAN_CORE_MAX_RECORDS] = {0};
255
    double start_log_time;
256 257 258 259 260 261 262
    double open1, open2;
    double job1, job2;
    double rec1, rec2;
    double mod1[DARSHAN_MAX_MODS] = {0};
    double mod2[DARSHAN_MAX_MODS] = {0};
    double header1, header2;
    double tm_end;
263
    uint64_t gz_fp = 0;
264
    unsigned char tmp_partial_flag;
265 266
    MPI_File log_fh;
    MPI_Status status;
267 268 269 270

    if(getenv("DARSHAN_INTERNAL_TIMING"))
        internal_timing_flag = 1;

271 272
    start_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();

Shane Snyder's avatar
Shane Snyder committed
273
    /* disable darhan-core while we shutdown */
274
    DARSHAN_CORE_LOCK();
275
    if(!darshan_core)
276
    {
277
        DARSHAN_CORE_UNLOCK();
278 279
        return;
    }
280 281
    final_core = darshan_core;
    darshan_core = NULL;
Shane Snyder's avatar
Shane Snyder committed
282

283
    /* we also need to set which modules were registered on this process and
284 285
     * call into those modules and give them a chance to perform any necessary
     * pre-shutdown steps.
Shane Snyder's avatar
Shane Snyder committed
286 287 288 289 290 291
     */
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
    {
        if(final_core->mod_array[i])
        {
            local_mod_use[i] = 1;
292
            final_core->mod_array[i]->mod_funcs.begin_shutdown();
Shane Snyder's avatar
Shane Snyder committed
293 294
        }
    }
295
    DARSHAN_CORE_UNLOCK();
296 297 298 299

    logfile_name = malloc(PATH_MAX);
    if(!logfile_name)
    {
300
        darshan_core_cleanup(final_core);
301 302 303
        return;
    }

304
    /* set darshan job id/metadata and constuct log file name on rank 0 */
305
    if(my_rank == 0)
306
    {
307
        /* Use DARSHAN_JOBID_OVERRIDE for the env var for __DARSHAN_JOBID */
308
        envjobid = getenv(DARSHAN_JOBID_OVERRIDE);
309
        if(!envjobid)
310
        {
311
            envjobid = __DARSHAN_JOBID;
312 313
        }

314
        /* find a job id */
315 316 317 318 319 320 321 322 323 324 325 326
        jobid_str = getenv(envjobid);
        if(jobid_str)
        {
            /* in cobalt we can find it in env var */
            ret = sscanf(jobid_str, "%d", &jobid);
        }
        if(!jobid_str || ret != 1)
        {
            /* use pid as fall back */
            jobid = getpid();
        }

327
        final_core->log_job.jobid = (int64_t)jobid;
328

329
        /* if we are using any hints to write the log file, then record those
330
         * hints with the darshan job information
331
         */
332
        darshan_log_record_hints_and_ver(final_core);
333

334
        /* use human readable start time format in log filename */
335
        start_time_tmp = final_core->log_job.start_time;
336
        start_tm = localtime(&start_time_tmp);
337

338 339
        /* construct log file name */
        darshan_get_logfile_name(logfile_name, jobid, start_tm);
340 341 342 343 344 345 346 347 348
    }

    /* broadcast log file name */
    DARSHAN_MPI_CALL(PMPI_Bcast)(logfile_name, PATH_MAX, MPI_CHAR, 0,
        MPI_COMM_WORLD);

    if(strlen(logfile_name) == 0)
    {
        /* failed to generate log file name */
349
        free(logfile_name);
350
        darshan_core_cleanup(final_core);
351 352 353
        return;
    }

354
    final_core->log_job.end_time = time(NULL);
355

356 357 358
    /* reduce to report first start time and last end time across all ranks
     * at rank 0
     */
359 360
    DARSHAN_MPI_CALL(PMPI_Reduce)(&final_core->log_job.start_time, &first_start_time, 1, MPI_LONG_LONG, MPI_MIN, 0, MPI_COMM_WORLD);
    DARSHAN_MPI_CALL(PMPI_Reduce)(&final_core->log_job.end_time, &last_end_time, 1, MPI_LONG_LONG, MPI_MAX, 0, MPI_COMM_WORLD);
361 362
    if(my_rank == 0)
    {
363 364
        final_core->log_job.start_time = first_start_time;
        final_core->log_job.end_time = last_end_time;
365
    }
366

367 368 369
    /* reduce the number of times a module was opened globally and bcast to everyone */   
    DARSHAN_MPI_CALL(PMPI_Allreduce)(local_mod_use, global_mod_use_count, DARSHAN_MAX_MODS, MPI_INT, MPI_SUM, MPI_COMM_WORLD);

370
    /* get a list of records which are shared across all processes */
371
    darshan_get_shared_records(final_core, shared_recs);
372

373 374
    if(internal_timing_flag)
        open1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
375
    /* collectively open the darshan log file */
376
    ret = darshan_log_open_all(logfile_name, &log_fh);
377 378
    if(internal_timing_flag)
        open2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
379 380 381 382 383 384 385 386

    /* error out if unable to open log file */
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
        MPI_LOR, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        if(my_rank == 0)
        {
387 388
            fprintf(stderr, "darshan library warning: unable to open log file %s\n",
                logfile_name);
389 390 391
            unlink(logfile_name);
        }
        free(logfile_name);
392
        darshan_core_cleanup(final_core);
393 394 395
        return;
    }

396 397
    if(internal_timing_flag)
        job1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
398
    /* rank 0 is responsible for writing the compressed darshan job information */
Shane Snyder's avatar
Shane Snyder committed
399
    if(my_rank == 0)
400
    {
401
        void *pointers[2] = {&final_core->log_job, final_core->trailing_data};
402
        int lengths[2] = {sizeof(struct darshan_job), strlen(final_core->trailing_data)};
403
        int comp_buf_sz = 0;
404

405
        /* compress the job info and the trailing mount/exe data */
Shane Snyder's avatar
Shane Snyder committed
406
        all_ret = darshan_deflate_buffer(pointers, lengths, 2,
407 408
            final_core->comp_buf, &comp_buf_sz);
        if(all_ret)
409
        {
410
            fprintf(stderr, "darshan library warning: unable to compress job data\n");
411
            unlink(logfile_name);
412
        }
413 414 415
        else
        {
            /* write the job information, preallocing space for the log header */
Shane Snyder's avatar
Shane Snyder committed
416
            gz_fp += sizeof(struct darshan_header);
417 418
            all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, gz_fp,
                final_core->comp_buf, comp_buf_sz, MPI_BYTE, &status);
419 420 421 422 423
            if(all_ret != MPI_SUCCESS)
            {
                fprintf(stderr, "darshan library warning: unable to write job data to log file %s\n",
                        logfile_name);
                unlink(logfile_name);
Shane Snyder's avatar
Shane Snyder committed
424
                
425
            }
426
            gz_fp += comp_buf_sz;
427
        }
428 429
    }

430 431 432 433 434
    /* error out if unable to write job information */
    DARSHAN_MPI_CALL(PMPI_Bcast)(&all_ret, 1, MPI_INT, 0, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        free(logfile_name);
435
        darshan_core_cleanup(final_core);
436 437
        return;
    }
438 439
    if(internal_timing_flag)
        job2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
440

441 442
    if(internal_timing_flag)
        rec1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
443
    /* write the record name->id hash to the log file */
Shane Snyder's avatar
Shane Snyder committed
444
    final_core->log_header.rec_map.off = gz_fp;
445
    ret = darshan_log_write_record_hash(log_fh, final_core, &gz_fp);
Shane Snyder's avatar
Shane Snyder committed
446
    final_core->log_header.rec_map.len = gz_fp - final_core->log_header.rec_map.off;
447

448
    /* error out if unable to write record hash */
449 450 451 452 453 454
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
        MPI_LOR, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        if(my_rank == 0)
        {
455
            fprintf(stderr, "darshan library warning: unable to write record hash to log file %s\n",
456
                logfile_name);
457
            unlink(logfile_name);
458 459
        }
        free(logfile_name);
460
        darshan_core_cleanup(final_core);
461 462
        return;
    }
Shane Snyder's avatar
Shane Snyder committed
463 464
    if(internal_timing_flag)
        rec2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
465 466

    /* loop over globally used darshan modules and:
467
     *      - perform shared file reductions, if possible
468
     *      - get final output buffer
469
     *      - compress (zlib) provided output buffer
Shane Snyder's avatar
Shane Snyder committed
470
     *      - append compressed buffer to log file
471 472
     *      - add module index info (file offset/length) to log header
     *      - shutdown the module
473
     */
474
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
475
    {
476
        struct darshan_core_module* this_mod = final_core->mod_array[i];
477
        struct darshan_core_record_ref *ref = NULL;
478 479
        darshan_record_id mod_shared_recs[DARSHAN_CORE_MAX_RECORDS];
        int mod_shared_rec_cnt = 0;
480
        void* mod_buf = NULL;
481
        int mod_buf_sz = 0;
482
        int j;
483

484
        if(global_mod_use_count[i] == 0)
485 486
        {
            if(my_rank == 0)
487 488 489 490
            {
                final_core->log_header.mod_map[i].off = 0;
                final_core->log_header.mod_map[i].len = 0;
            }
491
            continue;
492
        }
493 494
 
        if(internal_timing_flag)
495
            mod1[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();
496

497 498 499 500 501 502 503 504
        /* set the shared file list for this module */
        memset(mod_shared_recs, 0, DARSHAN_CORE_MAX_RECORDS * sizeof(darshan_record_id));
        for(j = 0; j < DARSHAN_CORE_MAX_RECORDS && shared_recs[j] != 0; j++)
        {
            HASH_FIND(hlink, final_core->rec_hash, &shared_recs[j],
                sizeof(darshan_record_id), ref);
            assert(ref);
            if(DARSHAN_CORE_MOD_ISSET(ref->global_mod_flags, i))
505
            {
506
                mod_shared_recs[mod_shared_rec_cnt++] = shared_recs[j];
507
            }
508
        }
509

510 511 512 513 514
        /* if module is registered locally, get the corresponding output buffer
         * 
         * NOTE: this function can be used to run collective operations across
         * modules, if there are file records shared globally.
         */
515
        if(this_mod)
516
        {
517 518
            this_mod->mod_funcs.get_output_data(MPI_COMM_WORLD, mod_shared_recs,
                mod_shared_rec_cnt, &mod_buf, &mod_buf_sz);
519 520
        }

521
        /* append this module's data to the darshan log */
Shane Snyder's avatar
Shane Snyder committed
522 523 524 525
        final_core->log_header.mod_map[i].off = gz_fp;
        ret = darshan_log_append_all(log_fh, final_core, mod_buf, mod_buf_sz, &gz_fp);
        final_core->log_header.mod_map[i].len =
            gz_fp - final_core->log_header.mod_map[i].off;
526

527
        /* error out if the log append failed */
528 529 530
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
            MPI_LOR, MPI_COMM_WORLD);
        if(all_ret != 0)
531
        {
532 533 534 535 536 537 538 539
            if(my_rank == 0)
            {
                fprintf(stderr,
                    "darshan library warning: unable to write %s module data to log file %s\n",
                    darshan_module_names[i], logfile_name);
                unlink(logfile_name);
            }
            free(logfile_name);
540
            darshan_core_cleanup(final_core);
541
            return;
542 543 544
        }

        /* shutdown module if registered locally */
545
        if(this_mod)
546 547 548
        {
            this_mod->mod_funcs.shutdown();
        }
549 550
        if(internal_timing_flag)
            mod2[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();
551 552
    }

553 554 555 556 557 558 559
    /* run a reduction to determine if any application processes had to set the
     * partial flag. this happens when a process has tracked too many records
     * at once and cannot track new records
     */
    DARSHAN_MPI_CALL(PMPI_Reduce)(&(final_core->log_header.partial_flag),
        &tmp_partial_flag, 1, MPI_UNSIGNED_CHAR, MPI_MAX, 0, MPI_COMM_WORLD);

560 561
    if(internal_timing_flag)
        header1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
562
    /* rank 0 is responsible for writing the log header */
563 564
    if(my_rank == 0)
    {
565 566 567
        /* initialize the remaining header fields */
        strcpy(final_core->log_header.version_string, DARSHAN_LOG_VERSION);
        final_core->log_header.magic_nr = DARSHAN_MAGIC_NR;
568
        final_core->log_header.comp_type = DARSHAN_ZLIB_COMP;
569
        final_core->log_header.partial_flag = tmp_partial_flag;
570

Shane Snyder's avatar
Shane Snyder committed
571 572 573
        all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, 0, &(final_core->log_header),
            sizeof(struct darshan_header), MPI_BYTE, &status);
        if(all_ret != MPI_SUCCESS)
574
        {
Shane Snyder's avatar
Shane Snyder committed
575 576
            fprintf(stderr, "darshan library warning: unable to write header to log file %s\n",
                    logfile_name);
577
            unlink(logfile_name);
578
        }
579 580
    }

581 582 583 584 585
    /* error out if unable to write log header */
    DARSHAN_MPI_CALL(PMPI_Bcast)(&all_ret, 1, MPI_INT, 0, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        free(logfile_name);
586
        darshan_core_cleanup(final_core);
587 588
        return;
    }
589 590
    if(internal_timing_flag)
        header2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
591

592 593 594
    DARSHAN_MPI_CALL(PMPI_File_close)(&log_fh);

    /* if we got this far, there are no errors, so rename from *.darshan_partial
595
     * to *-<logwritetime>.darshan, which indicates that this log file is
596 597
     * complete and ready for analysis
     */
598 599
    if(my_rank == 0)
    {
Shane Snyder's avatar
Shane Snyder committed
600
        if(getenv("DARSHAN_LOGFILE"))
601
        {
602
#ifdef __DARSHAN_GROUP_READABLE_LOGS
Shane Snyder's avatar
Shane Snyder committed
603
            chmod(logfile_name, (S_IRUSR|S_IRGRP));
604
#else
Shane Snyder's avatar
Shane Snyder committed
605
            chmod(logfile_name, (S_IRUSR));
606
#endif
Shane Snyder's avatar
Shane Snyder committed
607 608 609 610 611 612 613 614 615 616 617 618 619 620
        }
        else
        {
            char* tmp_index;
            double end_log_time;
            char* new_logfile_name;

            new_logfile_name = malloc(PATH_MAX);
            if(new_logfile_name)
            {
                new_logfile_name[0] = '\0';
                end_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();
                strcat(new_logfile_name, logfile_name);
                tmp_index = strstr(new_logfile_name, ".darshan_partial");
621
                sprintf(tmp_index, "_%d.darshan", (int)(end_log_time-start_log_time+1));
Shane Snyder's avatar
Shane Snyder committed
622 623
                rename(logfile_name, new_logfile_name);
                /* set permissions on log file */
624
#ifdef __DARSHAN_GROUP_READABLE_LOGS
Shane Snyder's avatar
Shane Snyder committed
625 626 627 628 629 630
                chmod(new_logfile_name, (S_IRUSR|S_IRGRP));
#else
                chmod(new_logfile_name, (S_IRUSR));
#endif
                free(new_logfile_name);
            }
631
        }
632
    }
633

634
    free(logfile_name);
635
    darshan_core_cleanup(final_core);
636

637
    if(internal_timing_flag)
638
    {
639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672
        double open_tm, open_slowest;
        double header_tm, header_slowest;
        double job_tm, job_slowest;
        double rec_tm, rec_slowest;
        double mod_tm[DARSHAN_MAX_MODS], mod_slowest[DARSHAN_MAX_MODS];
        double all_tm, all_slowest;

        tm_end = DARSHAN_MPI_CALL(PMPI_Wtime)();

        open_tm = open2 - open1;
        header_tm = header2 - header1;
        job_tm = job2 - job1;
        rec_tm = rec2 - rec1;
        all_tm = tm_end - start_log_time;
        for(i = 0;i < DARSHAN_MAX_MODS; i++)
        {
            mod_tm[i] = mod2[i] - mod1[i];
        }

        DARSHAN_MPI_CALL(PMPI_Reduce)(&open_tm, &open_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&header_tm, &header_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&job_tm, &job_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&rec_tm, &rec_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(&all_tm, &all_slowest, 1,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
        DARSHAN_MPI_CALL(PMPI_Reduce)(mod_tm, mod_slowest, DARSHAN_MAX_MODS,
            MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);

        if(my_rank == 0)
        {
673 674 675 676 677
            fprintf(stderr, "#darshan:<op>\t<nprocs>\t<time>\n");
            fprintf(stderr, "darshan:log_open\t%d\t%f\n", nprocs, open_slowest);
            fprintf(stderr, "darshan:job_write\t%d\t%f\n", nprocs, job_slowest);
            fprintf(stderr, "darshan:hash_write\t%d\t%f\n", nprocs, rec_slowest);
            fprintf(stderr, "darshan:header_write\t%d\t%f\n", nprocs, header_slowest);
678 679 680
            for(i = 0; i < DARSHAN_MAX_MODS; i++)
            {
                if(global_mod_use_count[i])
681
                    fprintf(stderr, "darshan:%s_shutdown\t%d\t%f\n", darshan_module_names[i],
Shane Snyder's avatar
Shane Snyder committed
682
                        nprocs, mod_slowest[i]);
683
            }
684
            fprintf(stderr, "darshan:core_shutdown\t%d\t%f\n", nprocs, all_slowest);
685
        }
686 687 688 689
    }
    
    return;
}
690

Shane Snyder's avatar
Shane Snyder committed
691
/* *********************************** */
692

693
/* construct the darshan log file name */
694
static void darshan_get_logfile_name(char* logfile_name, int jobid, struct tm* start_tm)
695
{
Shane Snyder's avatar
Shane Snyder committed
696
    char* user_logfile_name;
697 698 699
    char* logpath;
    char* logname_string;
    char* logpath_override = NULL;
700
#ifdef __DARSHAN_LOG_ENV
701 702 703 704 705 706 707 708 709
    char env_check[256];
    char* env_tok;
#endif
    uint64_t hlevel;
    char hname[HOST_NAME_MAX];
    uint64_t logmod;
    char cuser[L_cuserid] = {0};
    int ret;

Shane Snyder's avatar
Shane Snyder committed
710 711 712 713 714 715 716 717 718 719 720 721 722 723 724
    /* first, check if user specifies a complete logpath to use */
    user_logfile_name = getenv("DARSHAN_LOGFILE");
    if(user_logfile_name)
    {
        if(strlen(user_logfile_name) >= (PATH_MAX-1))
        {
            fprintf(stderr, "darshan library warning: user log file name too long.\n");
            logfile_name[0] = '\0';
        }
        else
        {
            strcpy(logfile_name, user_logfile_name);
        }
    }
    else
725
    {
Shane Snyder's avatar
Shane Snyder committed
726 727
        /* otherwise, generate the log path automatically */

728 729
        /* Use DARSHAN_LOG_PATH_OVERRIDE for the value or __DARSHAN_LOG_PATH */
        logpath = getenv(DARSHAN_LOG_PATH_OVERRIDE);
Shane Snyder's avatar
Shane Snyder committed
730 731
        if(!logpath)
        {
732 733
#ifdef __DARSHAN_LOG_PATH
            logpath = __DARSHAN_LOG_PATH;
734
#endif
Shane Snyder's avatar
Shane Snyder committed
735
        }
736

Shane Snyder's avatar
Shane Snyder committed
737 738 739 740 741 742 743 744 745 746
        /* get the username for this job.  In order we will try each of the
         * following until one of them succeeds:
         *
         * - cuserid()
         * - getenv("LOGNAME")
         * - snprintf(..., geteuid());
         *
         * Note that we do not use getpwuid() because it generally will not
         * work in statically compiled binaries.
         */
747 748

#ifndef DARSHAN_DISABLE_CUSERID
Shane Snyder's avatar
Shane Snyder committed
749
        cuserid(cuser);
750 751
#endif

Shane Snyder's avatar
Shane Snyder committed
752 753
        /* if cuserid() didn't work, then check the environment */
        if(strcmp(cuser, "") == 0)
754
        {
Shane Snyder's avatar
Shane Snyder committed
755 756 757 758 759
            logname_string = getenv("LOGNAME");
            if(logname_string)
            {
                strncpy(cuser, logname_string, (L_cuserid-1));
            }
760 761
        }

Shane Snyder's avatar
Shane Snyder committed
762 763 764 765 766 767
        /* if cuserid() and environment both fail, then fall back to uid */
        if(strcmp(cuser, "") == 0)
        {
            uid_t uid = geteuid();
            snprintf(cuser, sizeof(cuser), "%u", uid);
        }
768

Shane Snyder's avatar
Shane Snyder committed
769 770 771 772
        /* generate a random number to help differentiate the log */
        hlevel=DARSHAN_MPI_CALL(PMPI_Wtime)() * 1000000;
        (void)gethostname(hname, sizeof(hname));
        logmod = darshan_hash((void*)hname,strlen(hname),hlevel);
773

Shane Snyder's avatar
Shane Snyder committed
774 775 776 777
        /* see if darshan was configured using the --with-logpath-by-env
         * argument, which allows the user to specify an absolute path to
         * place logs via an env variable.
         */
778
#ifdef __DARSHAN_LOG_ENV
Shane Snyder's avatar
Shane Snyder committed
779
        /* just silently skip if the environment variable list is too big */
780
        if(strlen(__DARSHAN_LOG_ENV) < 256)
781
        {
Shane Snyder's avatar
Shane Snyder committed
782
            /* copy env variable list to a temporary buffer */
783
            strcpy(env_check, __DARSHAN_LOG_ENV);
Shane Snyder's avatar
Shane Snyder committed
784 785 786
            /* tokenize the comma-separated list */
            env_tok = strtok(env_check, ",");
            if(env_tok)
787
            {
Shane Snyder's avatar
Shane Snyder committed
788
                do
789
                {
Shane Snyder's avatar
Shane Snyder committed
790 791 792 793 794 795 796 797 798
                    /* check each env variable in order */
                    logpath_override = getenv(env_tok);
                    if(logpath_override)
                    {
                        /* stop as soon as we find a match */
                        break;
                    }
                }while((env_tok = strtok(NULL, ",")));
            }
799 800 801
        }
#endif

Shane Snyder's avatar
Shane Snyder committed
802
        if(logpath_override)
803
        {
Shane Snyder's avatar
Shane Snyder committed
804 805 806 807 808 809 810 811 812 813 814 815 816 817 818
            ret = snprintf(logfile_name, PATH_MAX,
                "%s/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath_override,
                cuser, __progname, jobid,
                (start_tm->tm_mon+1),
                start_tm->tm_mday,
                (start_tm->tm_hour*60*60 + start_tm->tm_min*60 + start_tm->tm_sec),
                logmod);
            if(ret == (PATH_MAX-1))
            {
                /* file name was too big; squish it down */
                snprintf(logfile_name, PATH_MAX,
                    "%s/id%d.darshan_partial",
                    logpath_override, jobid);
            }
819
        }
Shane Snyder's avatar
Shane Snyder committed
820
        else if(logpath)
821
        {
Shane Snyder's avatar
Shane Snyder committed
822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841
            ret = snprintf(logfile_name, PATH_MAX,
                "%s/%d/%d/%d/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath, (start_tm->tm_year+1900),
                (start_tm->tm_mon+1), start_tm->tm_mday,
                cuser, __progname, jobid,
                (start_tm->tm_mon+1),
                start_tm->tm_mday,
                (start_tm->tm_hour*60*60 + start_tm->tm_min*60 + start_tm->tm_sec),
                logmod);
            if(ret == (PATH_MAX-1))
            {
                /* file name was too big; squish it down */
                snprintf(logfile_name, PATH_MAX,
                    "%s/id%d.darshan_partial",
                    logpath, jobid);
            }
        }
        else
        {
            logfile_name[0] = '\0';
842 843 844 845
        }
    }

    return;
846 847
}

848
/* record any hints used to write the darshan log in the log header */
849
static void darshan_log_record_hints_and_ver(struct darshan_core_runtime* core)
850 851 852 853 854 855 856 857 858
{
    char* hints;
    char* header_hints;
    int meta_remain = 0;
    char* m;

    /* check environment variable to see if the default MPI file hints have
     * been overridden
     */
859
    hints = getenv(DARSHAN_LOG_HINTS_OVERRIDE);
860 861
    if(!hints)
    {
862
        hints = __DARSHAN_LOG_HINTS;
863 864 865 866 867 868 869 870 871 872
    }

    if(!hints || strlen(hints) < 1)
        return;

    header_hints = strdup(hints);
    if(!header_hints)
        return;

    meta_remain = DARSHAN_JOB_METADATA_LEN -
873
        strlen(core->log_job.metadata) - 1;
874 875
    if(meta_remain >= (strlen(PACKAGE_VERSION) + 9))
    {
876
        sprintf(core->log_job.metadata, "lib_ver=%s\n", PACKAGE_VERSION);
877 878 879 880
        meta_remain -= (strlen(PACKAGE_VERSION) + 9);
    }
    if(meta_remain >= (3 + strlen(header_hints)))
    {
881
        m = core->log_job.metadata + strlen(core->log_job.metadata);
882 883 884 885 886 887 888 889 890 891