darshan-core.c 48.3 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
};

Shane Snyder's avatar
Shane Snyder committed
58
59
60
#define DARSHAN_CORE_LOCK() pthread_mutex_lock(&darshan_core_mutex)
#define DARSHAN_CORE_UNLOCK() pthread_mutex_unlock(&darshan_core_mutex)

61
62
63
64
65
66
/* FS mount information */
#define DARSHAN_MAX_MNTS 64
#define DARSHAN_MAX_MNT_PATH 256
#define DARSHAN_MAX_MNT_TYPE 32
struct mnt_data
{
67
    int block_size;
68
69
70
71
72
73
    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;

74
75
76
77
/* 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(
78
79
80
81
82
83
    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);
84
85
static void darshan_block_size_from_path(
    const char *path, int *block_size);
86
static void darshan_get_shared_records(
87
    struct darshan_core_runtime *core, darshan_record_id *shared_recs);
88
static int darshan_log_open_all(
89
    char *logfile_name, MPI_File *log_fh);
90
static int darshan_deflate_buffer(
Shane Snyder's avatar
Shane Snyder committed
91
92
    void **pointers, int *lengths, int count, char *comp_buf,
    int *comp_buf_length);
93
static int darshan_log_write_record_hash(
94
    MPI_File log_fh, struct darshan_core_runtime *core,
95
96
97
    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
98
    int count, uint64_t *inout_off);
Shane Snyder's avatar
Shane Snyder committed
99
100
static void darshan_core_cleanup(
    struct darshan_core_runtime* core);
101

102
103
/* *********************************** */

Shane Snyder's avatar
Shane Snyder committed
104
void darshan_core_initialize(int argc, char **argv)
105
106
107
108
{
    int i;
    int internal_timing_flag = 0;
    double init_start, init_time, init_max;
109
    char *envstr;
110
111
112
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;
113
114
    int ret;
    int tmpval;
115
116

    DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
117
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &my_rank);
118
119
120
121

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

122
    if(internal_timing_flag)
123
124
125
        init_start = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* setup darshan runtime if darshan is enabled and hasn't been initialized already */
126
    if(!getenv("DARSHAN_DISABLE") && !darshan_core)
127
    {
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
        #if (__CP_MEM_ALIGNMENT < 1)
            #error Darshan must be configured with a positive value for --with-mem-align
        #endif
        envstr = getenv("DARSHAN_MEMALIGN");
        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
        {
            darshan_mem_alignment = __CP_MEM_ALIGNMENT;
        }

        /* avoid floating point errors on faulty input */
        if (darshan_mem_alignment < 1)
        {
            darshan_mem_alignment = 1;
        }
151
152
153
154

        /* allocate structure to track darshan_core_runtime information */
        darshan_core = malloc(sizeof(*darshan_core));
        if(darshan_core)
155
        {
156
            memset(darshan_core, 0, sizeof(*darshan_core));
157

158
159
160
161
            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)();
162
163

            /* record exe and arguments */
164
            for(i=0; i<argc; i++)
165
            {
166
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
167
168
                strncat(darshan_core->exe, argv[i], chars_left);
                if(i < (argc-1))
169
                {
170
                    chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
171
                    strncat(darshan_core->exe, " ", chars_left);
172
173
174
175
176
177
178
179
                }
            }

            /* if we don't see any arguments, then use glibc symbol to get
             * program name at least (this happens in fortran)
             */
            if(argc == 0)
            {
180
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
181
                strncat(darshan_core->exe, __progname_full, chars_left);
182
                chars_left = DARSHAN_EXE_LEN-strlen(darshan_core->exe);
183
                strncat(darshan_core->exe, " <unknown args>", chars_left);
184
185
186
187
188
            }

            if(chars_left == 0)
            {
                /* we ran out of room; mark that string was truncated */
189
                truncate_offset = DARSHAN_EXE_LEN - strlen(truncate_string);
190
                sprintf(&darshan_core->exe[truncate_offset], "%s",
191
192
                    truncate_string);
            }
193
194
195

            /* collect information about command line and mounted file systems */
            darshan_core->trailing_data = darshan_get_exe_and_mounts(darshan_core);
196
197
198
199
200
201
202
203
        }
    }

    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);
204
        if(my_rank == 0)
205
206
207
208
209
210
        {
            printf("#darshan:<op>\t<nprocs>\t<time>\n");
            printf("darshan:init\t%d\t%f\n", nprocs, init_max);
        }
    }

Kevin Harms's avatar
Kevin Harms committed
211
212
213
214
215
#ifdef __bgq__
    extern void bgq_runtime_initialize();
    bgq_runtime_initialize();
#endif

216
217
218
    return;
}

Shane Snyder's avatar
Shane Snyder committed
219
void darshan_core_shutdown()
220
{
221
    int i;
222
    char *logfile_name;
223
    struct darshan_core_runtime *final_core;
224
    int internal_timing_flag = 0;
225
226
    char *envjobid;
    char *jobid_str;
227
    int jobid;
228
    struct tm *start_tm;
229
    time_t start_time_tmp;
230
231
    int ret = 0;
    int all_ret = 0;
232
233
    int64_t first_start_time;
    int64_t last_end_time;
234
235
    int local_mod_use[DARSHAN_MAX_MODS] = {0};
    int global_mod_use_count[DARSHAN_MAX_MODS] = {0};
236
    darshan_record_id shared_recs[DARSHAN_CORE_MAX_RECORDS] = {0};
237
    double start_log_time;
238
239
240
241
242
243
244
    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;
245
    uint64_t gz_fp = 0;
246
247
    MPI_File log_fh;
    MPI_Status status;
248
249
250
251

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

252
253
    start_log_time = DARSHAN_MPI_CALL(PMPI_Wtime)();

Shane Snyder's avatar
Shane Snyder committed
254
    /* disable darhan-core while we shutdown */
255
    DARSHAN_CORE_LOCK();
256
    if(!darshan_core)
257
    {
258
        DARSHAN_CORE_UNLOCK();
259
260
        return;
    }
261
262
    final_core = darshan_core;
    darshan_core = NULL;
Shane Snyder's avatar
Shane Snyder committed
263

264
    /* we also need to set which modules were registered on this process and
265
266
     * call into those modules and give them a chance to perform any necessary
     * pre-shutdown steps.
Shane Snyder's avatar
Shane Snyder committed
267
268
269
270
271
272
     */
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
    {
        if(final_core->mod_array[i])
        {
            local_mod_use[i] = 1;
273
            final_core->mod_array[i]->mod_funcs.begin_shutdown();
Shane Snyder's avatar
Shane Snyder committed
274
275
        }
    }
276
    DARSHAN_CORE_UNLOCK();
277
278
279
280

    logfile_name = malloc(PATH_MAX);
    if(!logfile_name)
    {
281
        darshan_core_cleanup(final_core);
282
283
284
        return;
    }

285
    /* set darshan job id/metadata and constuct log file name on rank 0 */
286
    if(my_rank == 0)
287
288
289
    {
        /* Use CP_JOBID_OVERRIDE for the env var or CP_JOBID */
        envjobid = getenv(CP_JOBID_OVERRIDE);
290
        if(!envjobid)
291
292
293
294
        {
            envjobid = CP_JOBID;
        }

295
        /* find a job id */
296
297
298
299
300
301
302
303
304
305
306
307
        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();
        }

308
        final_core->log_job.jobid = (int64_t)jobid;
309

310
        /* if we are using any hints to write the log file, then record those
311
         * hints with the darshan job information
312
         */
313
        darshan_log_record_hints_and_ver(final_core);
314

315
        /* use human readable start time format in log filename */
316
        start_time_tmp = final_core->log_job.start_time;
317
        start_tm = localtime(&start_time_tmp);
318

319
320
        /* construct log file name */
        darshan_get_logfile_name(logfile_name, jobid, start_tm);
321
322
323
324
325
326
327
328
329
    }

    /* 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 */
330
        free(logfile_name);
331
        darshan_core_cleanup(final_core);
332
333
334
        return;
    }

335
    final_core->log_job.end_time = time(NULL);
336

337
338
339
    /* reduce to report first start time and last end time across all ranks
     * at rank 0
     */
340
341
    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);
342
343
    if(my_rank == 0)
    {
344
345
        final_core->log_job.start_time = first_start_time;
        final_core->log_job.end_time = last_end_time;
346
    }
347

348
349
350
    /* 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);

351
    /* get a list of records which are shared across all processes */
352
    darshan_get_shared_records(final_core, shared_recs);
353

354
355
    if(internal_timing_flag)
        open1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
356
    /* collectively open the darshan log file */
357
    ret = darshan_log_open_all(logfile_name, &log_fh);
358
359
    if(internal_timing_flag)
        open2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
360
361
362
363
364
365
366
367

    /* 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)
        {
368
369
            fprintf(stderr, "darshan library warning: unable to open log file %s\n",
                logfile_name);
370
371
372
            unlink(logfile_name);
        }
        free(logfile_name);
373
        darshan_core_cleanup(final_core);
374
375
376
        return;
    }

377
378
    if(internal_timing_flag)
        job1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
379
    /* rank 0 is responsible for writing the compressed darshan job information */
Shane Snyder's avatar
Shane Snyder committed
380
    if(my_rank == 0)
381
    {
382
        void *pointers[2] = {&final_core->log_job, final_core->trailing_data};
383
        int lengths[2] = {sizeof(struct darshan_job), strlen(final_core->trailing_data)};
384
        int comp_buf_sz = 0;
385

386
        /* compress the job info and the trailing mount/exe data */
Shane Snyder's avatar
Shane Snyder committed
387
        all_ret = darshan_deflate_buffer(pointers, lengths, 2,
388
389
            final_core->comp_buf, &comp_buf_sz);
        if(all_ret)
390
        {
391
            fprintf(stderr, "darshan library warning: unable to compress job data\n");
392
            unlink(logfile_name);
393
        }
394
395
396
        else
        {
            /* write the job information, preallocing space for the log header */
Shane Snyder's avatar
Shane Snyder committed
397
            gz_fp += sizeof(struct darshan_header);
398
399
            all_ret = DARSHAN_MPI_CALL(PMPI_File_write_at)(log_fh, gz_fp,
                final_core->comp_buf, comp_buf_sz, MPI_BYTE, &status);
400
401
402
403
404
            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
405
                
406
            }
407
            gz_fp += comp_buf_sz;
408
        }
409
410
    }

411
412
413
414
415
    /* 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);
416
        darshan_core_cleanup(final_core);
417
418
        return;
    }
419
420
    if(internal_timing_flag)
        job2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
421

422
423
    if(internal_timing_flag)
        rec1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
424
    /* write the record name->id hash to the log file */
Shane Snyder's avatar
Shane Snyder committed
425
    final_core->log_header.rec_map.off = gz_fp;
426
    ret = darshan_log_write_record_hash(log_fh, final_core, &gz_fp);
Shane Snyder's avatar
Shane Snyder committed
427
    final_core->log_header.rec_map.len = gz_fp - final_core->log_header.rec_map.off;
428

429
    /* error out if unable to write record hash */
430
431
432
433
434
435
    DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
        MPI_LOR, MPI_COMM_WORLD);
    if(all_ret != 0)
    {
        if(my_rank == 0)
        {
436
            fprintf(stderr, "darshan library warning: unable to write record hash to log file %s\n",
437
                logfile_name);
438
            unlink(logfile_name);
439
440
        }
        free(logfile_name);
441
        darshan_core_cleanup(final_core);
442
443
        return;
    }
Shane Snyder's avatar
Shane Snyder committed
444
445
    if(internal_timing_flag)
        rec2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
446
447

    /* loop over globally used darshan modules and:
448
     *      - perform shared file reductions, if possible
449
     *      - get final output buffer
450
     *      - compress (zlib) provided output buffer
Shane Snyder's avatar
Shane Snyder committed
451
     *      - append compressed buffer to log file
452
453
     *      - add module index info (file offset/length) to log header
     *      - shutdown the module
454
     */
455
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
456
    {
457
        struct darshan_core_module* this_mod = final_core->mod_array[i];
458
        struct darshan_core_record_ref *ref = NULL;
459
460
        darshan_record_id mod_shared_recs[DARSHAN_CORE_MAX_RECORDS];
        int mod_shared_rec_cnt = 0;
461
        void* mod_buf = NULL;
462
        int mod_buf_sz = 0;
463
        int j;
464

465
        if(global_mod_use_count[i] == 0)
466
467
        {
            if(my_rank == 0)
468
469
470
471
            {
                final_core->log_header.mod_map[i].off = 0;
                final_core->log_header.mod_map[i].len = 0;
            }
472
            continue;
473
        }
474
475
 
        if(internal_timing_flag)
476
            mod1[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();
477

478
479
480
481
482
483
484
485
        /* 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))
486
            {
487
                mod_shared_recs[mod_shared_rec_cnt++] = shared_recs[j];
488
            }
489
        }
490

491
492
493
494
495
        /* 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.
         */
496
        if(this_mod)
497
        {
498
499
            this_mod->mod_funcs.get_output_data(MPI_COMM_WORLD, mod_shared_recs,
                mod_shared_rec_cnt, &mod_buf, &mod_buf_sz);
500
501
        }

502
        /* append this module's data to the darshan log */
Shane Snyder's avatar
Shane Snyder committed
503
504
505
506
        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;
507

508
        /* error out if the log append failed */
509
510
511
        DARSHAN_MPI_CALL(PMPI_Allreduce)(&ret, &all_ret, 1, MPI_INT,
            MPI_LOR, MPI_COMM_WORLD);
        if(all_ret != 0)
512
        {
513
514
515
516
517
518
519
520
            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);
521
            darshan_core_cleanup(final_core);
522
            return;
523
524
525
        }

        /* shutdown module if registered locally */
526
        if(this_mod)
527
528
529
        {
            this_mod->mod_funcs.shutdown();
        }
530
531
        if(internal_timing_flag)
            mod2[i] = DARSHAN_MPI_CALL(PMPI_Wtime)();
532
533
    }

534
535
    if(internal_timing_flag)
        header1 = DARSHAN_MPI_CALL(PMPI_Wtime)();
536
    /* rank 0 is responsible for writing the log header */
537
538
    if(my_rank == 0)
    {
539
540
541
        /* initialize the remaining header fields */
        strcpy(final_core->log_header.version_string, DARSHAN_LOG_VERSION);
        final_core->log_header.magic_nr = DARSHAN_MAGIC_NR;
542
        final_core->log_header.comp_type = DARSHAN_ZLIB_COMP;
543

Shane Snyder's avatar
Shane Snyder committed
544
545
546
        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)
547
        {
Shane Snyder's avatar
Shane Snyder committed
548
549
            fprintf(stderr, "darshan library warning: unable to write header to log file %s\n",
                    logfile_name);
550
            unlink(logfile_name);
551
        }
552
553
    }

554
555
556
557
558
    /* 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);
559
        darshan_core_cleanup(final_core);
560
561
        return;
    }
562
563
    if(internal_timing_flag)
        header2 = DARSHAN_MPI_CALL(PMPI_Wtime)();
564

565
566
567
    DARSHAN_MPI_CALL(PMPI_File_close)(&log_fh);

    /* if we got this far, there are no errors, so rename from *.darshan_partial
568
     * to *-<logwritetime>.darshan, which indicates that this log file is
569
570
     * complete and ready for analysis
     */
571
572
    if(my_rank == 0)
    {
Shane Snyder's avatar
Shane Snyder committed
573
        if(getenv("DARSHAN_LOGFILE"))
574
        {
575
#ifdef __CP_GROUP_READABLE_LOGS
Shane Snyder's avatar
Shane Snyder committed
576
            chmod(logfile_name, (S_IRUSR|S_IRGRP));
577
#else
Shane Snyder's avatar
Shane Snyder committed
578
            chmod(logfile_name, (S_IRUSR));
579
#endif
Shane Snyder's avatar
Shane Snyder committed
580
581
582
583
584
585
586
587
588
589
590
591
592
593
        }
        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");
594
                sprintf(tmp_index, "_%d.darshan", (int)(end_log_time-start_log_time+1));
Shane Snyder's avatar
Shane Snyder committed
595
596
597
598
599
600
601
602
603
                rename(logfile_name, new_logfile_name);
                /* set permissions on log file */
#ifdef __CP_GROUP_READABLE_LOGS
                chmod(new_logfile_name, (S_IRUSR|S_IRGRP));
#else
                chmod(new_logfile_name, (S_IRUSR));
#endif
                free(new_logfile_name);
            }
604
        }
605
    }
606

607
    free(logfile_name);
608
    darshan_core_cleanup(final_core);
609

610
    if(internal_timing_flag)
611
    {
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
        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)
        {
            printf("#darshan:<op>\t<nprocs>\t<time>\n");
            printf("darshan:log_open\t%d\t%f\n", nprocs, open_slowest);
            printf("darshan:job_write\t%d\t%f\n", nprocs, job_slowest);
            printf("darshan:hash_write\t%d\t%f\n", nprocs, rec_slowest);
Shane Snyder's avatar
Shane Snyder committed
650
            printf("darshan:header_write\t%d\t%f\n", nprocs, header_slowest);
651
652
653
654
            for(i = 0; i < DARSHAN_MAX_MODS; i++)
            {
                if(global_mod_use_count[i])
                    printf("darshan:%s_shutdown\t%d\t%f\n", darshan_module_names[i],
Shane Snyder's avatar
Shane Snyder committed
655
                        nprocs, mod_slowest[i]);
656
657
658
            }
            printf("darshan:core_shutdown\t%d\t%f\n", nprocs, all_slowest);
        }
659
660
661
662
    }
    
    return;
}
663

Shane Snyder's avatar
Shane Snyder committed
664
/* *********************************** */
665

666
/* construct the darshan log file name */
667
static void darshan_get_logfile_name(char* logfile_name, int jobid, struct tm* start_tm)
668
{
Shane Snyder's avatar
Shane Snyder committed
669
    char* user_logfile_name;
670
671
672
673
674
675
676
677
678
679
680
681
682
    char* logpath;
    char* logname_string;
    char* logpath_override = NULL;
#ifdef __CP_LOG_ENV
    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
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
    /* 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
698
    {
Shane Snyder's avatar
Shane Snyder committed
699
700
701
702
703
704
        /* otherwise, generate the log path automatically */

        /* Use CP_LOG_PATH_OVERRIDE for the value or __CP_LOG_PATH */
        logpath = getenv(CP_LOG_PATH_OVERRIDE);
        if(!logpath)
        {
705
#ifdef __CP_LOG_PATH
Shane Snyder's avatar
Shane Snyder committed
706
            logpath = __CP_LOG_PATH;
707
#endif
Shane Snyder's avatar
Shane Snyder committed
708
        }
709

Shane Snyder's avatar
Shane Snyder committed
710
711
712
713
714
715
716
717
718
719
        /* 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.
         */
720
721

#ifndef DARSHAN_DISABLE_CUSERID
Shane Snyder's avatar
Shane Snyder committed
722
        cuserid(cuser);
723
724
#endif

Shane Snyder's avatar
Shane Snyder committed
725
726
        /* if cuserid() didn't work, then check the environment */
        if(strcmp(cuser, "") == 0)
727
        {
Shane Snyder's avatar
Shane Snyder committed
728
729
730
731
732
            logname_string = getenv("LOGNAME");
            if(logname_string)
            {
                strncpy(cuser, logname_string, (L_cuserid-1));
            }
733
734
        }

Shane Snyder's avatar
Shane Snyder committed
735
736
737
738
739
740
        /* 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);
        }
741

Shane Snyder's avatar
Shane Snyder committed
742
743
744
745
        /* 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);
746

Shane Snyder's avatar
Shane Snyder committed
747
748
749
750
        /* 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.
         */
751
#ifdef __CP_LOG_ENV
Shane Snyder's avatar
Shane Snyder committed
752
753
        /* just silently skip if the environment variable list is too big */
        if(strlen(__CP_LOG_ENV) < 256)
754
        {
Shane Snyder's avatar
Shane Snyder committed
755
756
757
758
759
            /* copy env variable list to a temporary buffer */
            strcpy(env_check, __CP_LOG_ENV);
            /* tokenize the comma-separated list */
            env_tok = strtok(env_check, ",");
            if(env_tok)
760
            {
Shane Snyder's avatar
Shane Snyder committed
761
                do
762
                {
Shane Snyder's avatar
Shane Snyder committed
763
764
765
766
767
768
769
770
771
                    /* 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, ",")));
            }
772
773
774
        }
#endif

Shane Snyder's avatar
Shane Snyder committed
775
        if(logpath_override)
776
        {
Shane Snyder's avatar
Shane Snyder committed
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
            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);
            }
792
        }
Shane Snyder's avatar
Shane Snyder committed
793
        else if(logpath)
794
        {
Shane Snyder's avatar
Shane Snyder committed
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
            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';
815
816
817
818
        }
    }

    return;
819
820
}

821
/* record any hints used to write the darshan log in the log header */
822
static void darshan_log_record_hints_and_ver(struct darshan_core_runtime* core)
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
{
    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
     */
    hints = getenv(CP_LOG_HINTS_OVERRIDE);
    if(!hints)
    {
        hints = __CP_LOG_HINTS;
    }

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

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

    meta_remain = DARSHAN_JOB_METADATA_LEN -
846
        strlen(core->log_job.metadata) - 1;
847
848
    if(meta_remain >= (strlen(PACKAGE_VERSION) + 9))
    {
849
        sprintf(core->log_job.metadata, "lib_ver=%s\n", PACKAGE_VERSION);
850
851
852
853
        meta_remain -= (strlen(PACKAGE_VERSION) + 9);
    }
    if(meta_remain >= (3 + strlen(header_hints)))
    {
854
        m = core->log_job.metadata + strlen(core->log_job.metadata);
855
856
857
858
859
860
861
862
863
864
865
866
        /* We have room to store the hints in the metadata portion of
         * the job header.  We just prepend an h= to the hints list.  The
         * metadata parser will ignore = characters that appear in the value
         * portion of the metadata key/value pair.
         */
        sprintf(m, "h=%s\n", header_hints);
    }
    free(header_hints);

    return;
}

867
868
869
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
static int mnt_data_cmp(const void* a, const void* b)
{
    const struct mnt_data *d_a = (const struct mnt_data*)a;
    const struct mnt_data *d_b = (const struct mnt_data*)b;

    if(strlen(d_a->path) > strlen(d_b->path))
        return(-1);
    else if(strlen(d_a->path) < strlen(d_b->path))
        return(1);
    else
        return(0);
}

/* adds an entry to table of mounted file systems */
static void add_entry(char* trailing_data, int* space_left, struct mntent *entry)
{
    int ret;
    char tmp_mnt[256];
    struct statfs statfsbuf;

    strncpy(mnt_data_array[mnt_data_count].path, entry->mnt_dir,
        DARSHAN_MAX_MNT_PATH-1);
    strncpy(mnt_data_array[mnt_data_count].type, entry->mnt_type,
        DARSHAN_MAX_MNT_TYPE-1);
    /* NOTE: we now try to detect the preferred block size for each file 
     * system using fstatfs().  On Lustre we assume a size of 1 MiB 
     * because fstatfs() reports 4 KiB. 
     */
#ifndef LL_SUPER_MAGIC
#define LL_SUPER_MAGIC 0x0BD00BD0
#endif
    ret = statfs(entry->mnt_dir, &statfsbuf);
    if(ret == 0 && statfsbuf.f_type != LL_SUPER_MAGIC)
        mnt_data_array[mnt_data_count].block_size = statfsbuf.f_bsize;
    else if(ret == 0 && statfsbuf.f_type == LL_SUPER_MAGIC)
        mnt_data_array[mnt_data_count].block_size = 1024*1024;
    else
        mnt_data_array[mnt_data_count].block_size = 4096;

    /* store mount information for use in header of darshan log */
907
    ret = snprintf(tmp_mnt, 256, "\n%s\t%s",
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
        entry->mnt_type, entry->mnt_dir);
    if(ret < 256 && strlen(tmp_mnt) <= (*space_left))
    {
        strcat(trailing_data, tmp_mnt);
        (*space_left) -= strlen(tmp_mnt);
    }

    mnt_data_count++;
    return;
}

/* darshan_get_exe_and_mounts_root()
 *
 * collects command line and list of mounted file systems into a string that
 * will be stored with the job header
 */
static void darshan_get_exe_and_mounts_root(struct darshan_core_runtime *core,
    char* trailing_data, int space_left)
{
    FILE* tab;
    struct mntent *entry;
    char* exclude;
    int tmp_index = 0;
    int skip = 0;

    /* skip these fs types */
    static char* fs_exclusions[] = {
        "tmpfs",
        "proc",
        "sysfs",
        "devpts",
        "binfmt_misc",
        "fusectl",
        "debugfs",
        "securityfs",
        "nfsd",
        "none",
        "rpc_pipefs",
        "hugetlbfs",
        "cgroup",
        NULL
    };

    /* length of exe has already been safety checked in darshan-posix.c */
    strcat(trailing_data, core->exe);
953
    space_left = DARSHAN_EXE_LEN - strlen(trailing_data);
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

    /* we make two passes through mounted file systems; in the first pass we
     * grab any non-nfs mount points, then on the second pass we grab nfs
     * mount points
     */

    tab = setmntent("/etc/mtab", "r");
    if(!tab)
        return;
    /* loop through list of mounted file systems */
    while(mnt_data_count<DARSHAN_MAX_MNTS && (entry = getmntent(tab)) != NULL)
    {
        /* filter out excluded fs types */
        tmp_index = 0;
        skip = 0;
        while((exclude = fs_exclusions[tmp_index]))
        {
            if(!(strcmp(exclude, entry->mnt_type)))
            {
                skip =1;
                break;
            }
            tmp_index++;
        }

        if(skip || (strcmp(entry->mnt_type, "nfs") == 0))
            continue;

        add_entry(trailing_data, &space_left, entry);
    }
    endmntent(tab);

    tab = setmntent("/etc/mtab", "r");
    if(!tab)
        return;
    /* loop through list of mounted file systems */
    while(mnt_data_count<DARSHAN_MAX_MNTS && (entry = getmntent(tab)) != NULL)
    {
        if(strcmp(entry->mnt_type, "nfs") != 0)
            continue;

        add_entry(trailing_data, &space_left, entry);
    }
    endmntent(tab);

    /* Sort mount points in order of longest path to shortest path.  This is
     * necessary so that if we try to match file paths to mount points later
For faster browsing, not all history is shown. View entire blame