darshan-logutils.c 17.9 KB
Newer Older
1
2
3
4
5
/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

6
7
#include <stdio.h>
#include <string.h>
8
#include <assert.h>
9
#include <stdlib.h>
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
#include "darshan-logutils.h"

/* isn't there a clever c way to avoid this? */
char *darshan_names[] = {
    "CP_INDEP_OPENS",
    "CP_COLL_OPENS",               /* count of MPI collective opens */
    "CP_INDEP_READS",              /* count of independent MPI reads */
    "CP_INDEP_WRITES",             /* count of independent MPI writes */
    "CP_COLL_READS",               /* count of collective MPI reads */
    "CP_COLL_WRITES",              /* count of collective MPI writes */
    "CP_SPLIT_READS",              /* count of split collective MPI reads */
    "CP_SPLIT_WRITES",             /* count of split collective MPI writes */
    "CP_NB_READS",                 /* count of nonblocking MPI reads */
    "CP_NB_WRITES",                /* count of nonblocking MPI writes */
    "CP_SYNCS",                    /* count of MPI_File_sync */
    "CP_POSIX_READS",              /* count of posix reads */
    "CP_POSIX_WRITES",             /* count of posix writes */
    "CP_POSIX_OPENS",              /* count of posix opens */
    "CP_POSIX_SEEKS",              /* count of posix seeks */
    "CP_POSIX_STATS",              /* count of posix stat/lstat/fstats */
    "CP_POSIX_MMAPS",              /* count of posix mmaps */
    "CP_POSIX_FREADS",
    "CP_POSIX_FWRITES",
    "CP_POSIX_FOPENS",
    "CP_POSIX_FSEEKS",
    "CP_POSIX_FSYNCS",
    "CP_POSIX_FDSYNCS",
37
38
39
    "CP_INDEP_NC_OPENS",
    "CP_COLL_NC_OPENS",
    "CP_HDF5_OPENS",
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
    "CP_COMBINER_NAMED",           /* count of each MPI datatype category */
    "CP_COMBINER_DUP",
    "CP_COMBINER_CONTIGUOUS",
    "CP_COMBINER_VECTOR",
    "CP_COMBINER_HVECTOR_INTEGER",
    "CP_COMBINER_HVECTOR",
    "CP_COMBINER_INDEXED",
    "CP_COMBINER_HINDEXED_INTEGER",
    "CP_COMBINER_HINDEXED",
    "CP_COMBINER_INDEXED_BLOCK",
    "CP_COMBINER_STRUCT_INTEGER",
    "CP_COMBINER_STRUCT",
    "CP_COMBINER_SUBARRAY",
    "CP_COMBINER_DARRAY",
    "CP_COMBINER_F90_REAL",
    "CP_COMBINER_F90_COMPLEX",
    "CP_COMBINER_F90_INTEGER",
    "CP_COMBINER_RESIZED",
    "CP_HINTS",                     /* count of MPI hints used */
    "CP_VIEWS",                     /* count of MPI set view calls */
    "CP_MODE",                      /* mode of file */
    "CP_BYTES_READ",                /* total bytes read */
    "CP_BYTES_WRITTEN",             /* total bytes written */
    "CP_MAX_BYTE_READ",             /* highest offset byte read */
    "CP_MAX_BYTE_WRITTEN",          /* highest offset byte written */
    "CP_CONSEC_READS",              /* count of consecutive reads */
    "CP_CONSEC_WRITES",             /* count of consecutive writes */
    "CP_SEQ_READS",                 /* count of sequential reads */
    "CP_SEQ_WRITES",                /* count of sequential writes */
    "CP_RW_SWITCHES",
    "CP_MEM_NOT_ALIGNED",           /* count of accesses not mem aligned */
    "CP_MEM_ALIGNMENT",             /* mem alignment in bytes */
    "CP_FILE_NOT_ALIGNED",          /* count of accesses not file aligned */
    "CP_FILE_ALIGNMENT",            /* file alignment in bytes */
74
75
    "CP_MAX_READ_TIME_SIZE",
    "CP_MAX_WRITE_TIME_SIZE",
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
    "CP_SIZE_READ_0_100",           /* count of posix read size ranges */
    "CP_SIZE_READ_100_1K",
    "CP_SIZE_READ_1K_10K",
    "CP_SIZE_READ_10K_100K",
    "CP_SIZE_READ_100K_1M",
    "CP_SIZE_READ_1M_4M",
    "CP_SIZE_READ_4M_10M",
    "CP_SIZE_READ_10M_100M",
    "CP_SIZE_READ_100M_1G",
    "CP_SIZE_READ_1G_PLUS",
    "CP_SIZE_WRITE_0_100",          /* count of posix write size ranges */
    "CP_SIZE_WRITE_100_1K",
    "CP_SIZE_WRITE_1K_10K",
    "CP_SIZE_WRITE_10K_100K",
    "CP_SIZE_WRITE_100K_1M",
    "CP_SIZE_WRITE_1M_4M",
    "CP_SIZE_WRITE_4M_10M",
    "CP_SIZE_WRITE_10M_100M",
    "CP_SIZE_WRITE_100M_1G",
    "CP_SIZE_WRITE_1G_PLUS",
    "CP_SIZE_READ_AGG_0_100",       /* count of MPI read size ranges */
    "CP_SIZE_READ_AGG_100_1K",
    "CP_SIZE_READ_AGG_1K_10K",
    "CP_SIZE_READ_AGG_10K_100K",
    "CP_SIZE_READ_AGG_100K_1M",
    "CP_SIZE_READ_AGG_1M_4M",
    "CP_SIZE_READ_AGG_4M_10M",
    "CP_SIZE_READ_AGG_10M_100M",
    "CP_SIZE_READ_AGG_100M_1G",
    "CP_SIZE_READ_AGG_1G_PLUS",
    "CP_SIZE_WRITE_AGG_0_100",      /* count of MPI write size ranges */
    "CP_SIZE_WRITE_AGG_100_1K",
    "CP_SIZE_WRITE_AGG_1K_10K",
    "CP_SIZE_WRITE_AGG_10K_100K",
    "CP_SIZE_WRITE_AGG_100K_1M",
    "CP_SIZE_WRITE_AGG_1M_4M",
    "CP_SIZE_WRITE_AGG_4M_10M",
    "CP_SIZE_WRITE_AGG_10M_100M",
    "CP_SIZE_WRITE_AGG_100M_1G",
    "CP_SIZE_WRITE_AGG_1G_PLUS",
    "CP_EXTENT_READ_0_100",          /* count of MPI read extent ranges */
    "CP_EXTENT_READ_100_1K",
    "CP_EXTENT_READ_1K_10K",
    "CP_EXTENT_READ_10K_100K",
    "CP_EXTENT_READ_100K_1M",
    "CP_EXTENT_READ_1M_4M",
    "CP_EXTENT_READ_4M_10M",
    "CP_EXTENT_READ_10M_100M",
    "CP_EXTENT_READ_100M_1G",
    "CP_EXTENT_READ_1G_PLUS",
    "CP_EXTENT_WRITE_0_100",         /* count of MPI write extent ranges */
    "CP_EXTENT_WRITE_100_1K",
    "CP_EXTENT_WRITE_1K_10K",
    "CP_EXTENT_WRITE_10K_100K",
    "CP_EXTENT_WRITE_100K_1M",
    "CP_EXTENT_WRITE_1M_4M",
    "CP_EXTENT_WRITE_4M_10M",
    "CP_EXTENT_WRITE_10M_100M",
    "CP_EXTENT_WRITE_100M_1G",
    "CP_EXTENT_WRITE_1G_PLUS",
    "CP_STRIDE1_STRIDE",             /* the four most frequently appearing strides */
    "CP_STRIDE2_STRIDE",
    "CP_STRIDE3_STRIDE",
    "CP_STRIDE4_STRIDE",
    "CP_STRIDE1_COUNT",              /* count of each of the most frequent strides */
    "CP_STRIDE2_COUNT",
    "CP_STRIDE3_COUNT",
    "CP_STRIDE4_COUNT",
    "CP_ACCESS1_ACCESS",
    "CP_ACCESS2_ACCESS",
    "CP_ACCESS3_ACCESS",
    "CP_ACCESS4_ACCESS",
    "CP_ACCESS1_COUNT",
    "CP_ACCESS2_COUNT",
    "CP_ACCESS3_COUNT",
    "CP_ACCESS4_COUNT",
152
    "CP_DEVICE",
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170

    "CP_NUM_INDICES"
};

/* isn't there a clever c way to avoid this? */
char *darshan_f_names[] = {
    "CP_F_OPEN_TIMESTAMP",        /* timestamp of first open */
    "CP_F_READ_START_TIMESTAMP",  /* timestamp of first read */
    "CP_F_WRITE_START_TIMESTAMP", /* timestamp of first write */
    "CP_F_CLOSE_TIMESTAMP",       /* timestamp of last close */
    "CP_F_READ_END_TIMESTAMP",    /* timestamp of last read */
    "CP_F_WRITE_END_TIMESTAMP",   /* timestamp of last write */
    "CP_F_POSIX_READ_TIME",       /* cumulative posix read time */
    "CP_F_POSIX_WRITE_TIME",      /* cumulative posix write time */
    "CP_F_POSIX_META_TIME",       /* cumulative posix meta time */
    "CP_F_MPI_META_TIME",         /* cumulative mpi-io metadata time */
    "CP_F_MPI_READ_TIME",         /* cumulative mpi-io read time */
    "CP_F_MPI_WRITE_TIME",        /* cumulative mpi-io write time */
171
172
    "CP_F_MAX_READ_TIME",
    "CP_F_MAX_WRITE_TIME",
173
174
175
    "CP_F_NUM_INDICES",
};

176
/*******************************
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
 * version 1.22 to 1.23 differences
 *
 * - added:
 *   - CP_DEVICE
 * - changed params:
 *   - CP_FILE_RECORD_SIZE: 1240 to 1244
 *   - CP_NUM_INDICES: 138 to 139
 */
#define CP_NUM_INDICES_1_22 138
struct darshan_file_1_22
{
    uint64_t hash;
    int rank;
    int64_t counters[CP_NUM_INDICES_1_22];
    double fcounters[CP_F_NUM_INDICES];
    char name_suffix[CP_NAME_SUFFIX_LEN+1];
};

static void shift_missing_1_22(struct darshan_file* file);


/*******************************
 * version 1.21 to 1.23 differences 
200
201
202
203
204
205
 * - added:
 *   - CP_INDEP_NC_OPENS
 *   - CP_COLL_NC_OPENS
 *   - CP_HDF5_OPENS
 *   - CP_MAX_READ_TIME_SIZE
 *   - CP_MAX_WRITE_TIME_SIZE
206
 *   - CP_DEVICE
207
208
209
 *   - CP_F_MAX_READ_TIME
 *   - CP_F_MAX_WRITE_TIME
 * - changed params:
210
211
 *   - CP_FILE_RECORD_SIZE: 1184 to 1244
 *   - CP_NUM_INDICES: 133 to 139
212
 *   - CP_F_NUM_INDICES: 12 to 14
213
 * - so 60 bytes worth of new indices are the only difference
214
215
216
217
218
219
220
221
222
223
224
225
226
227
 */
#define CP_NUM_INDICES_1_21 133
#define CP_F_NUM_INDICES_1_21 12
struct darshan_file_1_21
{
    uint64_t hash;
    int rank;
    int64_t counters[CP_NUM_INDICES_1_21];
    double fcounters[CP_F_NUM_INDICES_1_21];
    char name_suffix[CP_NAME_SUFFIX_LEN+1];
};

static void shift_missing_1_21(struct darshan_file* file);

228
/* a rather crude API for accessing raw binary darshan files */
229
darshan_fd darshan_log_open(const char *name)
230
231
232
233
{
    return gzopen(name, "r");
}

234
/* darshan_log_getjob()
235
236
237
 *
 * returns 0 on success, -1 on failure
 */
238
int darshan_log_getjob(darshan_fd file, struct darshan_job *job)
239
240
{
    int ret;
241
242
243

    gzseek(file, 0, SEEK_SET);

244
245
246
247
248
249
250
251
252
253
    ret = gzread(file, job, sizeof(*job));
    if (ret < sizeof(*job))
    {
        if(gzeof(file))
        {
            fprintf(stderr, "Error: invalid log file (too short).\n");
        }
        perror("darshan_job_init");
        return(-1);
    }
254
255
256
257
    if(strcmp(job->version_string, "1.21"))
        return(0);
    if(strcmp(job->version_string, "1.22"))
        return(0);
258

259
260
261
262
    fprintf(stderr, "Error: incompatible darshan file.\n");
    fprintf(stderr, "Error: expected version %s, but got %s\n", 
            CP_VERSION, job->version_string);
    return(-1);
263
264
}

265
/* darshan_log_getfile()
266
267
268
 *
 * return 1 if file record found, 0 on eof, and -1 on error
 */
269
int darshan_log_getfile(darshan_fd fd, struct darshan_job *job, struct darshan_file *file)
270
{
271
272
    int ret;
    const char* err_string;
273
    struct darshan_file_1_21 file_1_21;
274
    struct darshan_file_1_22 file_1_22;
275
276
277
278
279
    
    if(gztell(file) < CP_JOB_RECORD_SIZE)
        gzseek(file, CP_JOB_RECORD_SIZE, SEEK_SET);

    gzseek(file, sizeof(struct darshan_job), SEEK_SET);
280
281
282
283
284
285

    /* reset file record, so that diff compares against a zero'd out record
     * if file is missing
     */
    memset(file, 0, sizeof(&file));

286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
    if(strcmp(job->version_string, "1.21") == 0)
    {
        ret = gzread(fd, &file_1_21, sizeof(file_1_21));
        if(ret == sizeof(file_1_21))
        {
            /* convert to new file record structure */
            file->hash = file_1_21.hash;
            file->rank = file_1_21.rank;
            strcpy(file->name_suffix, file_1_21.name_suffix);
            memcpy(file->counters, file_1_21.counters,
                (CP_NUM_INDICES_1_21*sizeof(int64_t)));
            memcpy(file->fcounters, file_1_21.fcounters,
                (CP_F_NUM_INDICES_1_21*sizeof(double)));
            shift_missing_1_21(file);

            /* got exactly one, correct size record */
            return(1);
        }
    }
    else if(strcmp(job->version_string, "1.22") == 0)
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
    {
        ret = gzread(fd, &file_1_22, sizeof(file_1_22));
        if(ret == sizeof(file_1_22))
        {
            /* convert to new file record structure */
            file->hash = file_1_22.hash;
            file->rank = file_1_22.rank;
            strcpy(file->name_suffix, file_1_22.name_suffix);
            memcpy(file->counters, file_1_22.counters,
                (CP_NUM_INDICES_1_22*sizeof(int64_t)));
            memcpy(file->fcounters, file_1_22.fcounters,
                (CP_F_NUM_INDICES*sizeof(double)));
            shift_missing_1_22(file);

            /* got exactly one, correct size record */
            return(1);
        }
    }
    else if(strcmp(job->version_string, "1.23") == 0)
325
326
    {
        /* make sure this is the current version */
327
        assert(strcmp("1.23", CP_VERSION) == 0);
328
329
330
331
332
333
334
335
336

        ret = gzread(fd, file, sizeof(*file));
        if(ret == sizeof(*file))
        {
            /* got exactly one, correct size record */
            return(1);
        }
    }
    else
337
    {
338
339
        fprintf(stderr, "Error: invalid log file version.\n");
        return(-1);
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
    }

    if(ret > 0)
    {
        /* got a short read */
        fprintf(stderr, "Error: invalid file record (too small)\n");
        return(-1);
    }

    if(ret == 0 && gzeof(fd))
    {
        /* hit end of file */
        return(0);
    }

    /* all other errors */
    err_string = gzerror(fd, &ret);
    fprintf(stderr, "Error: %s\n", err_string);
    return(-1);
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
/* darshan_log_getmounts()
 * 
 * retrieves mount table information from the log.  Note that devs, mnt_pts,
 * and fs_types are arrays that will be allocated by the function and must
 * be freed by the caller.  count will indicate the size of the arrays
 */
int darshan_log_getmounts(darshan_fd fd, int** devs, char*** mnt_pts, char***
    fs_types, int* count, int *flag)
{
    int ret;
    char* pos;
    int array_index = 0;
    char buf[CP_EXE_LEN+1];

    gzseek(fd, sizeof(struct darshan_job), SEEK_SET);

    ret = gzread(fd, buf, (CP_EXE_LEN + 1));
    if (ret < (CP_EXE_LEN + 1))
    {
        perror("gzread");
        return(-1);
    }
    if (gzeof(fd))
        *flag = 1;
    else
        *flag = 0;

    /* count entries */
    *count = 0;
    pos = buf;
    while((pos = strchr(pos, '\n')) != NULL)
    {
        pos++;
        (*count)++;
    }

    if(*count == 0)
    {
        /* no mount entries present */
        return(0);
    }

    /* allocate output arrays */
    *devs = malloc((*count)*sizeof(int));
    assert(*devs);
    *mnt_pts = malloc((*count)*sizeof(char*));
    assert(*mnt_pts);
    *fs_types = malloc((*count)*sizeof(char*));
    assert(*fs_types);
    
    /* work backwards through the table and parse each line (except for
     * first, which holds command line information)
     */
    while((pos = strrchr(buf, '\n')) != NULL)
    {
        /* overestimate string lengths */
        (*mnt_pts)[array_index] = malloc(CP_EXE_LEN);
        assert((*mnt_pts)[array_index]);
        (*fs_types)[array_index] = malloc(CP_EXE_LEN);
        assert((*fs_types)[array_index]);
        
        ret = sscanf(++pos, "%d\t%s\t%s", &(*devs)[array_index],
            (*fs_types)[array_index], (*mnt_pts)[array_index]);
        if(ret != 3)
        {
            fprintf(stderr, "Error: poorly formatted mount table in log file.\n");
            return(-1);
        }
        pos--;
        *pos = '\0';
        array_index++;
    }
   
    return (0);
}


438
int darshan_log_getexe(darshan_fd fd, char *buf, int *flag)
439
440
{
    int ret;
441
    char* newline;
442

443
444
    gzseek(fd, sizeof(struct darshan_job), SEEK_SET);

445
446
447
448
449
450
    ret = gzread(fd, buf, (CP_EXE_LEN + 1));
    if (ret < (CP_EXE_LEN + 1))
    {
        perror("gzread");
        return(-1);
    }
451
452
453
454
455
    if (gzeof(fd))
        *flag = 1;
    else
        *flag = 0;

456
457
458
459
460
461
462
463
    /* this call is only supposed to return the exe string, but starting in
     * log format 1.23 there could be a table of mount entry information
     * after the exe.  Look for newline character and truncate there.
     */
    newline = strchr(buf, '\n');
    if(newline)
        *newline = '\0';

464
465
466
    return (0);
}

467
void darshan_log_close(darshan_fd file)
468
469
470
471
{
    gzclose(file);
}

472
473
474
475
476
477
/* darshan_log_print_version_warnings()
 *
 * Print summary of any problems with the detected log format
 */
void darshan_log_print_version_warnings(struct darshan_job *job)
{
478
    if(strcmp(job->version_string, "1.23") == 0)
479
480
481
482
    {
        /* nothing to do, this is the current version */
        return;
    }
483
484
485
486
487
488
489
490
    
    if(strcmp(job->version_string, "1.22") == 0)
    {
        printf("# WARNING: version 1.22 log format does not support the following parameters:\n");
        printf("#   CP_DEVICE\n");
        printf("# It also does not record mounted file systems, mount points, or fs types.\n");
        return;
    }
491
492
493
494
495
496
497
498
499

    if(strcmp(job->version_string, "1.21") == 0)
    {
        printf("# WARNING: version 1.21 log format does not support the following parameters:\n");
        printf("#   CP_INDEP_NC_OPENS\n");
        printf("#   CP_COLL_NC_OPENS\n");
        printf("#   CP_HDF5_OPENS\n");
        printf("#   CP_MAX_READ_TIME_SIZE\n");
        printf("#   CP_MAX_WRITE_TIME_SIZE\n");
500
        printf("#   CP_DEVICE\n");
501
502
        printf("#   CP_F_MAX_READ_TIME\n");
        printf("#   CP_F_MAX_WRITE_TIME\n");
503
        printf("# It also does not record mounted file systems, mount points, or fs types.\n");
504
        printf("#\n");
505
506
507
508
509
510
511
512
        return;
    }

    fprintf(stderr, "Error: version %s not supported by parser.\n",
        job->version_string);
    return;
}

513
514
515
516
/* shift_missing_1_21()
 *
 * translates indices to account for counters that weren't present in log
 * format 1.21
517
 */
518
519
520
521
522
523
524
525
526
static void shift_missing_1_21(struct darshan_file* file)
{
    int c_index = 0;
    int missing_counters[] = {
        CP_INDEP_NC_OPENS,
        CP_COLL_NC_OPENS,
        CP_HDF5_OPENS,
        CP_MAX_READ_TIME_SIZE,
        CP_MAX_WRITE_TIME_SIZE,
527
        CP_DEVICE,
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
        -1};
    int missing_f_counters[] = {
        CP_F_MAX_READ_TIME,
        CP_F_MAX_WRITE_TIME,
        -1};

    c_index = 0;
    while(missing_counters[c_index] != -1)
    {
        int missing_counter = missing_counters[c_index];
        c_index++;
        if(missing_counter < (CP_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->counters[missing_counter+1],
                &file->counters[missing_counter],
                (CP_NUM_INDICES-missing_counter-1)*sizeof(int64_t));
        }
        /* zero out missing counter */
        file->counters[missing_counter] = 0;
    }

    c_index = 0;
    while(missing_f_counters[c_index] != -1)
    {
        int missing_counter = missing_f_counters[c_index];
        c_index++;
        if(missing_counter < (CP_F_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->fcounters[missing_counter+1],
                &file->fcounters[missing_counter],
                (CP_F_NUM_INDICES-missing_counter-1)*sizeof(double));
        }
        /* zero out missing counter */
        file->fcounters[missing_counter] = 0;
    }

    return;
}
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598

/* shift_missing_1_22()
 *
 * translates indices to account for counters that weren't present in log
 * format 1.22
 */
static void shift_missing_1_22(struct darshan_file* file)
{
    int c_index = 0;
    int missing_counters[] = {
        CP_DEVICE,
        -1};

    c_index = 0;
    while(missing_counters[c_index] != -1)
    {
        int missing_counter = missing_counters[c_index];
        c_index++;
        if(missing_counter < (CP_NUM_INDICES - 1))
        {
            /* shift down */
            memmove(&file->counters[missing_counter+1],
                &file->counters[missing_counter],
                (CP_NUM_INDICES-missing_counter-1)*sizeof(int64_t));
        }
        /* zero out missing counter */
        file->counters[missing_counter] = 0;
    }

    return;
}