darshan-core.c 13.8 KB
Newer Older
1
2
3
4
5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6
7
#define _XOPEN_SOURCE 500

8
9
10
11
12
13
#include "darshan-runtime-config.h"

#include <stdio.h>
#ifdef HAVE_MNTENT_H
#include <mntent.h>
#endif
14
15
16
17
18
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <limits.h>
#include <pthread.h>
19
20
21
22
23
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/vfs.h>

#include <mpi.h>
24

25
#include "darshan-core.h"
26
#include "utlist.h"
27
28
29

extern char* __progname_full;

30
31
static void darshan_core_initialize(int *argc, char ***argv);
static void darshan_core_shutdown(void);
32
static void darshan_core_cleanup(struct darshan_core_job_runtime* job);
33

34
/* internal variables */
35
static struct darshan_core_job_runtime *darshan_core_job = NULL;
36
static pthread_mutex_t darshan_mutex = PTHREAD_MUTEX_INITIALIZER;
37

38
39
40
#define DARSHAN_LOCK() pthread_mutex_lock(&darshan_mutex)
#define DARSHAN_UNLOCK() pthread_mutex_unlock(&darshan_mutex)

41
42
43
44
45
46
47
48
#define DARSHAN_MOD_REGISTER(__mod, __job) \
    LL_PREPEND(__job->mod_list_head, __mod)
#define DARSHAN_MOD_SEARCH(__mod, __tmp, __job) \
    LL_SEARCH(__job->mod_list_head, __mod, __tmp, mod_cmp)
#define DARSHAN_MOD_ITER(__mod, __tmp, __job) \
    LL_FOREACH_SAFE(__job->mod_list_head, __mod, __tmp)
#define DARSHAN_MOD_DELETE(__mod, __job) \
   LL_DELETE(__job->mod_list_head, __mod)
49

50
51
52
53
54
55
56
57
58
59
60
/* intercept MPI initialize and finalize to initialize darshan */
int MPI_Init(int *argc, char ***argv)
{
    int ret;

    ret = DARSHAN_MPI_CALL(PMPI_Init)(argc, argv);
    if(ret != MPI_SUCCESS)
    {
        return(ret);
    }

61
    darshan_core_initialize(argc, argv);
62
63
64
65

    return(ret);
}

66
int MPI_Init_thread(int *argc, char ***argv, int required, int *provided)
67
68
69
70
{
    int ret;

    ret = DARSHAN_MPI_CALL(PMPI_Init_thread)(argc, argv, required, provided);
71
    if(ret != MPI_SUCCESS)
72
73
74
75
    {
        return(ret);
    }

76
    darshan_core_initialize(argc, argv);
77
78
79
80
81
82
83
84

    return(ret);
}

int MPI_Finalize(void)
{
    int ret;

85
    darshan_core_shutdown();
86
87
88
89
90

    ret = DARSHAN_MPI_CALL(PMPI_Finalize)();
    return(ret);
}

91
92
/* *********************************** */

93
static void darshan_core_initialize(int *argc, char ***argv)
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
{
    int i;
    int nprocs;
    int rank;
    int internal_timing_flag = 0;
    double init_start, init_time, init_max;
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;

    DARSHAN_MPI_CALL(PMPI_Comm_size)(MPI_COMM_WORLD, &nprocs);
    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &rank);

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

110
    if(internal_timing_flag)
111
112
113
        init_start = DARSHAN_MPI_CALL(PMPI_Wtime)();

    /* setup darshan runtime if darshan is enabled and hasn't been initialized already */
114
    if(!getenv("DARSHAN_DISABLE") && !darshan_core_job)
115
    {
116
117
118
        /* allocate structure to track darshan_core_job information */
        darshan_core_job = malloc(sizeof(*darshan_core_job));
        if(darshan_core_job)
119
        {
120
            memset(darshan_core_job, 0, sizeof(*darshan_core_job));
121

122
            if(getenv("DARSHAN_DISABLE_TIMING"))
123
            {
124
                darshan_core_job->flags |= CP_FLAG_NOTIMING;
125
126
            }

127
128
129
130
131
132
            strcpy(darshan_core_job->log_job.version_string, CP_VERSION);
            darshan_core_job->log_job.magic_nr = CP_MAGIC_NR;
            darshan_core_job->log_job.uid = getuid();
            darshan_core_job->log_job.start_time = time(NULL);
            darshan_core_job->log_job.nprocs = nprocs;
            darshan_core_job->wtime_offset = DARSHAN_MPI_CALL(PMPI_Wtime)();
133
134
135
136

            /* record exe and arguments */
            for(i=0; i<(*argc); i++)
            {
137
138
                chars_left = CP_EXE_LEN-strlen(darshan_core_job->exe);
                strncat(darshan_core_job->exe, *(argv[i]), chars_left);
139
140
                if(i < ((*argc)-1))
                {
141
142
                    chars_left = CP_EXE_LEN-strlen(darshan_core_job->exe);
                    strncat(darshan_core_job->exe, " ", chars_left);
143
144
145
146
147
148
149
150
                }
            }

            /* if we don't see any arguments, then use glibc symbol to get
             * program name at least (this happens in fortran)
             */
            if(argc == 0)
            {
151
152
153
154
                chars_left = CP_EXE_LEN-strlen(darshan_core_job->exe);
                strncat(darshan_core_job->exe, __progname_full, chars_left);
                chars_left = CP_EXE_LEN-strlen(darshan_core_job->exe);
                strncat(darshan_core_job->exe, " <unknown args>", chars_left);
155
156
157
158
159
160
            }

            if(chars_left == 0)
            {
                /* we ran out of room; mark that string was truncated */
                truncate_offset = CP_EXE_LEN - strlen(truncate_string);
161
                sprintf(&darshan_core_job->exe[truncate_offset], "%s",
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
                    truncate_string);
            }
        }
    }

    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);
        if(rank == 0)
        {
            printf("#darshan:<op>\t<nprocs>\t<time>\n");
            printf("darshan:init\t%d\t%f\n", nprocs, init_max);
        }
    }

    return;
}

182
static void darshan_core_shutdown()
183
{
184
185
186
    int rank;
    char *logfile_name;
    struct darshan_core_job_runtime* final_job;
187
    struct darshan_core_module *mod, *tmp;
188
    int internal_timing_flag = 0;
189
190
191
192
193
194
195
196
197
198
199
200
201
    int jobid;
    char* jobid_str;
    char* envjobid;
    char* logpath;
    int ret;
    uint64_t hlevel;
    char hname[HOST_NAME_MAX];
    uint64_t logmod;
    char* logpath_override = NULL;
#ifdef __CP_LOG_ENV
    char env_check[256];
    char* env_tok;
#endif
202
203
204
205

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

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
    DARSHAN_LOCK();
    if(!darshan_core_job)
    {
        DARSHAN_UNLOCK();
        return;
    }
    /* disable further tracing while hanging onto the data so that we can
     * write it out
     */
    final_job = darshan_core_job;
    darshan_core_job = NULL;
    DARSHAN_UNLOCK();

    logfile_name = malloc(PATH_MAX);
    if(!logfile_name)
    {
        darshan_core_cleanup(final_job);
        return;
    }

    DARSHAN_MPI_CALL(PMPI_Comm_rank)(MPI_COMM_WORLD, &rank);

    /* construct log file name */
    if(rank == 0)
    {
        char cuser[L_cuserid] = {0};
        struct tm* my_tm;
        time_t start_time_tmp;

        /* Use CP_JOBID_OVERRIDE for the env var or CP_JOBID */
        envjobid = getenv(CP_JOBID_OVERRIDE);
        if (!envjobid)
        {
            envjobid = CP_JOBID;
        }

        /* Use CP_LOG_PATH_OVERRIDE for the value or __CP_LOG_PATH */
        logpath = getenv(CP_LOG_PATH_OVERRIDE);
        if (!logpath)
        {
#ifdef __CP_LOG_PATH
            logpath = __CP_LOG_PATH;
#endif
        }

        /* find a job id */
        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();
        }

        /* break out time into something human readable */
        start_time_tmp = final_job->log_job.start_time;
        my_tm = localtime(&start_time_tmp);

        /* 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.
         */

#ifndef DARSHAN_DISABLE_CUSERID
        cuserid(cuser);
#endif

        /* if cuserid() didn't work, then check the environment */
        if (strcmp(cuser, "") == 0)
        {
            char* logname_string;
            logname_string = getenv("LOGNAME");
            if(logname_string)
            {
                strncpy(cuser, logname_string, (L_cuserid-1));
            }

        }

        /* 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);
        }

        /* 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);

        /* 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.
         */
#ifdef __CP_LOG_ENV
        /* just silently skip if the environment variable list is too big */
        if(strlen(__CP_LOG_ENV) < 256)
        {
            /* 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)
            {
                do
                {
                    /* 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, ",")));
            }
        }
#endif

        if(logpath_override)
        {
            ret = snprintf(logfile_name, PATH_MAX,
                "%s/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath_override,
                cuser, __progname, jobid,
                (my_tm->tm_mon+1),
                my_tm->tm_mday,
                (my_tm->tm_hour*60*60 + my_tm->tm_min*60 + my_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);
            }
        }
        else if(logpath)
        {
            ret = snprintf(logfile_name, PATH_MAX,
                "%s/%d/%d/%d/%s_%s_id%d_%d-%d-%d-%" PRIu64 ".darshan_partial",
                logpath, (my_tm->tm_year+1900),
                (my_tm->tm_mon+1), my_tm->tm_mday,
                cuser, __progname, jobid,
                (my_tm->tm_mon+1),
                my_tm->tm_mday,
                (my_tm->tm_hour*60*60 + my_tm->tm_min*60 + my_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';
        }

        /* add jobid */
        final_job->log_job.jobid = (int64_t)jobid;
    }

    /* 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 */
        darshan_core_cleanup(final_job);
        return;
    }

    final_job->log_job.end_time = time(NULL);

394
    /* TODO: coordinate shutdown accross all registered modules */
395
396
397
398
399
400
401



    free(logfile_name);
    darshan_core_cleanup(final_job);

    if (internal_timing_flag)
402
    {
403
404
405
406
407
        /* TODO: what do we want to time in new darshan version? */
    }
    
    return;
}
408

409
410
411
412
413
414
415
static void darshan_core_cleanup(struct darshan_core_job_runtime* job)
{
    struct darshan_core_module *mod, *tmp;

    DARSHAN_MOD_ITER(mod, tmp, job)
    {
        DARSHAN_MOD_DELETE(mod, job);
416
417
418
        free(mod);
    };

419
    free(job);
420
421
422
423

    return;
}

424
425
426
427
428
static int mod_cmp(struct darshan_core_module* a, struct darshan_core_module* b)
{
    return strcmp(a->name, b->name);
}

429
/* ********************************************************* */
430
431
432
433
434
435

void darshan_core_register_module(
    char *name,
    struct darshan_module_funcs *funcs,
    int *runtime_mem_limit)
{
436
437
    struct darshan_core_module tmp;
    struct darshan_core_module* mod;
438

439
440
    DARSHAN_LOCK();

441
    *runtime_mem_limit = 0;
442
    if(!darshan_core_job)
443
444
    {
        DARSHAN_UNLOCK();
445
        return;
446
    }
447
448
449

    /* see if this module is already registered */
    strncpy(tmp.name, name, DARSHAN_MOD_NAME_LEN);
450
    DARSHAN_MOD_SEARCH(mod, &tmp, darshan_core_job);
451
    if(mod)
452
    {
453
454
455
456
457
458
        /* if module is already registered, update module_funcs and return */
        /* NOTE: we do not recalculate memory limit here, just set to 0 */
        mod->mod_funcs = *funcs;

        DARSHAN_UNLOCK();
        return;
459
460
    }

461
462
463
    /* this module has not been registered yet, allocate and register it */
    mod = malloc(sizeof(*mod));
    if(!mod)
464
465
466
467
    {
        DARSHAN_UNLOCK();
        return;
    }
468
    memset(mod, 0, sizeof(*mod));
469

470
471
    strncpy(mod->name, name, DARSHAN_MOD_NAME_LEN);
    mod->mod_funcs = *funcs;
472
    DARSHAN_MOD_REGISTER(mod, darshan_core_job);
473

474
475
    /* TODO: something smarter than just 2 MiB per module */
    *runtime_mem_limit = 2 * 1024 * 1024;
476

477
478
    DARSHAN_UNLOCK();

479
480
481
482
483
484
485
486
487
    return;
}

void darshan_core_lookup_id(
    void *name,
    int len,
    int printable_flag,
    darshan_file_id *id)
{
488
    darshan_file_id tmp_id;
489

490
    if(!darshan_core_job)
491
492
493
494
495
496
497
498
499
500
        return;

    /* TODO: what do you do with printable flag? */

    /* hash the input name to get a unique id for this record */
    tmp_id = darshan_hash(name, len, 0);
    
    /* TODO: how to store the filename to hash mapping? */

    *id = tmp_id;
501
502
503
    return;
}

504
505
double darshan_core_wtime()
{
506
    if(!darshan_core_job || darshan_core_job->flags & CP_FLAG_NOTIMING)
507
508
509
510
511
512
    {
        return(0);
    }

    return DARSHAN_MPI_CALL(PMPI_Wtime)();
}
513
514
515
516
517
518
519
520
521

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