darshan-merge.c 15.2 KB
Newer Older
1
#include <stdio.h>
2
3
4
#include <stdlib.h>
#include <fcntl.h>
#include <string.h>
5
6
#include <getopt.h>
#include <glob.h>
7

8
9
#include "uthash-1.9.2/src/uthash.h"

10
11
#include "darshan-logutils.h"

12
13
#define DEF_MOD_BUF_SIZE 1024 /* 1 KiB is enough for all current mod records ... */

14
15
16
17
18
19
20
21
struct darshan_shared_record_ref
{
    darshan_record_id id;
    int ref_cnt;
    char agg_rec[DEF_MOD_BUF_SIZE];
    UT_hash_handle hlink;
};

22
23
void usage(char *exename)
{
24
    fprintf(stderr, "Usage: %s --output <output_path> [options] <input-logs>\n", exename);
Shane Snyder's avatar
Shane Snyder committed
25
26
    fprintf(stderr, "This utility merges multiple Darshan log files into a single output log file.\n");
    fprintf(stderr, "Options:\n");
27
    fprintf(stderr, "\t--output\t(REQUIRED) Full path of the output darshan log file.\n");
Shane Snyder's avatar
Shane Snyder committed
28
    fprintf(stderr, "\t--shared-redux\tReduce globally shared records into a single record.\n");
29
    fprintf(stderr, "\t--job-end-time\tSet the output log's job end time (requires argument of seconds since Epoch).\n");
30
31
32
33

    exit(1);
}

Shane Snyder's avatar
Shane Snyder committed
34
void parse_args(int argc, char **argv, char ***infile_list, int *n_files,
35
    char **outlog_path, int *shared_redux, int64_t *job_end_time)
36
37
{
    int index;
38
    char *check;
39
40
    static struct option long_opts[] =
    {
41
        {"output", required_argument, NULL, 'o'},
42
43
        {"shared-redux", no_argument, NULL, 's'},
        {"job-end-time", required_argument, NULL, 'e'},
44
45
46
47
        {0, 0, 0, 0}
    };

    *shared_redux = 0;
48
    *outlog_path = NULL;
49
    *job_end_time = 0;
50
51
52
53
54
55
56
57
58
59
60
61

    while(1)
    {
        int c = getopt_long(argc, argv, "", long_opts, &index);

        if(c == -1) break;

        switch(c)
        {
            case 's':
                *shared_redux = 1;
                break;
62
63
            case 'o':
                *outlog_path = optarg;
Shane Snyder's avatar
Shane Snyder committed
64
                break;
65
66
67
68
69
70
71
72
            case 'e':
                *job_end_time = strtol(optarg, &check, 10);
                if(optarg == check)
                {
                    fprintf(stderr, "Error: unable to parse job end time value.\n");
                    exit(1);
                }
                break;
73
74
75
76
77
78
79
            case '?':
            default:
                usage(argv[0]);
                break;
        }
    }

80
    if(*outlog_path == NULL)
81
82
83
84
    {
        usage(argv[0]);
    }

Shane Snyder's avatar
Shane Snyder committed
85
86
    *infile_list = &argv[optind];
    *n_files = argc - optind;
87

Shane Snyder's avatar
Shane Snyder committed
88
    return;
89
90
}

Shane Snyder's avatar
Shane Snyder committed
91
92
93
int build_mod_shared_rec_hash(char **infile_list, int n_infiles,
    darshan_module_id mod_id, int nprocs, char *mod_buf,
    struct darshan_shared_record_ref **shared_rec_hash)
94
95
96
97
{
    darshan_fd in_fd;
    struct darshan_base_record *base_rec;
    struct darshan_shared_record_ref *ref, *tmp;
98
    int init_rank = -1;
99
100
101
102
    int ret;
    int i;

    /* loop over each input log file */
Shane Snyder's avatar
Shane Snyder committed
103
    for(i = 0; i < n_infiles; i++)
104
    {
Shane Snyder's avatar
Shane Snyder committed
105
        in_fd = darshan_log_open(infile_list[i]);
106
107
108
109
        if(in_fd == NULL)
        {
            fprintf(stderr,
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
110
                infile_list[i]);
111
112
113
114
115
116
            return(-1);
        }

        while((ret = mod_logutils[mod_id]->log_get_record(in_fd, mod_buf)) == 1)
        {
            base_rec = (struct darshan_base_record *)mod_buf;
117
118
            if(init_rank == -1)
                init_rank = base_rec->rank;
119
120

            /* initialize the hash with the first rank's records */
121
            if(base_rec->rank == init_rank)
122
123
124
125
126
127
128
129
130
131
            {
                struct darshan_base_record *agg_base;

                /* create a new ref and add to the hash */
                ref = malloc(sizeof(*ref));
                if(!ref)
                {
                    darshan_log_close(in_fd);
                    return(-1);
                }
Shane Snyder's avatar
Shane Snyder committed
132
                memset(ref, 0, sizeof(*ref));
133
134
135
136
137
138
139
140
141
142
143
144

                /* initialize the aggregate record with this rank's record */
                agg_base = (struct darshan_base_record *)ref->agg_rec;
                agg_base->id = base_rec->id;
                agg_base->rank = -1;
                mod_logutils[mod_id]->log_agg_records(mod_buf, ref->agg_rec, 1);

                ref->id = base_rec->id;
                ref->ref_cnt = 1;
                HASH_ADD(hlink, *shared_rec_hash, id, sizeof(darshan_record_id), ref);
            }
            else
145
           {
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
                /* search for this record in shared record hash */
                HASH_FIND(hlink, *shared_rec_hash, &(base_rec->id),
                    sizeof(darshan_record_id), ref);
                if(ref)
                {
                    /* if found, aggregate this rank's record into the shared record */
                    mod_logutils[mod_id]->log_agg_records(mod_buf, ref->agg_rec, 0);
                    ref->ref_cnt++;
                }
            }
        }
        if(ret < 0)
        {
            fprintf(stderr,
                "Error: unable to read %s module record from input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
161
                darshan_module_names[mod_id], infile_list[i]);
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
            darshan_log_close(in_fd);
            return(-1);
        }

        darshan_log_close(in_fd);
    }

    /* prune any non-shared records from the hash one last time */
    HASH_ITER(hlink, *shared_rec_hash, ref, tmp)
    {
        if(ref->ref_cnt != nprocs)
        {
            HASH_DELETE(hlink, *shared_rec_hash, ref);
            free(ref);
        }
    }

    return(0);
}

182
183
int main(int argc, char *argv[])
{
Shane Snyder's avatar
Shane Snyder committed
184
185
    char **infile_list;
    int n_infiles;
186
    int shared_redux;
187
    int64_t job_end_time = 0;
188
    char *outlog_path;
Shane Snyder's avatar
Shane Snyder committed
189
190
    darshan_fd in_fd, merge_fd;
    struct darshan_job in_job, merge_job;
191
    char merge_exe[DARSHAN_EXE_LEN+1] = {0};
Shane Snyder's avatar
Shane Snyder committed
192
193
194
    char **merge_mnt_pts;
    char **merge_fs_types;
    int merge_mnt_count = 0;
195
196
197
    struct darshan_name_record_ref *in_hash = NULL;
    struct darshan_name_record_ref *merge_hash = NULL;
    struct darshan_name_record_ref *ref, *tmp, *found;
198
199
200
201
    struct darshan_shared_record_ref *shared_rec_hash = NULL;
    struct darshan_shared_record_ref *sref, *stmp;
    struct darshan_base_record *base_rec;
    char mod_buf[DEF_MOD_BUF_SIZE];
202
203
    int i, j;
    int ret;
204
205

    /* grab command line arguments */
206
    parse_args(argc, argv, &infile_list, &n_infiles, &outlog_path, &shared_redux, &job_end_time);
207

Shane Snyder's avatar
Shane Snyder committed
208
    memset(&merge_job, 0, sizeof(struct darshan_job));
209

Shane Snyder's avatar
Shane Snyder committed
210
    /* first pass at merging together logs:
211
212
     *      - compose output job-level metadata structure (including exe & mount data)
     *      - compose output record_id->file_name mapping 
213
     */
Shane Snyder's avatar
Shane Snyder committed
214
    for(i = 0; i < n_infiles; i++)
215
216
217
    {
        memset(&in_job, 0, sizeof(struct darshan_job));

Shane Snyder's avatar
Shane Snyder committed
218
        in_fd = darshan_log_open(infile_list[i]);
219
        if(in_fd == NULL)
220
221
        {
            fprintf(stderr,
222
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
223
                infile_list[i]);
224
225
226
227
            return(-1);
        }

        /* read job-level metadata from the input file */
228
        ret = darshan_log_get_job(in_fd, &in_job);
229
        if(ret < 0)
230
231
        {
            fprintf(stderr,
232
                "Error: unable to read job data from input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
233
                infile_list[i]);
234
            darshan_log_close(in_fd);
235
236
237
            return(-1);
        }

238
239
240
241
242
243
244
245
246
        /* if the input darshan log has metadata set indicating the darshan
         * shutdown procedure was called on the log, then we error out. if the
         * shutdown procedure was started, then it's possible the log has
         * incomplete or corrupt data, so we just throw out the data for now.
         */
        if(strstr(in_job.metadata, "darshan_shutdown=yes"))
        {
            fprintf(stderr,
                "Error: potentially corrupt data found in input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
247
                infile_list[i]);
248
249
250
251
            darshan_log_close(in_fd);
            return(-1);
        }

252
253
        if(i == 0)
        {
254
            /* get job data, exe, & mounts directly from the first input log */
Shane Snyder's avatar
Shane Snyder committed
255
            memcpy(&merge_job, &in_job, sizeof(struct darshan_job));
256

257
            ret = darshan_log_get_exe(in_fd, merge_exe);
258
            if(ret < 0)
259
260
            {
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
261
262
                    "Error: unable to read exe string from input Darshan log file %s.\n",
                    infile_list[i]);
263
                darshan_log_close(in_fd);
264
265
266
                return(-1);
            }

267
            ret = darshan_log_get_mounts(in_fd, &merge_mnt_pts,
Shane Snyder's avatar
Shane Snyder committed
268
                &merge_fs_types, &merge_mnt_count);
269
            if(ret < 0)
270
            {
271
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
272
273
                    "Error: unable to read mount info from input Darshan log file %s.\n",
                    infile_list[i]);
274
275
                darshan_log_close(in_fd);
                return(-1);
276
277
278
279
            }
        }
        else
        {
280
            /* potentially update job timestamps using remaining logs */
Shane Snyder's avatar
Shane Snyder committed
281
282
283
284
            if(in_job.start_time < merge_job.start_time)
                merge_job.start_time = in_job.start_time;
            if(in_job.end_time > merge_job.end_time)
                merge_job.end_time = in_job.end_time;
285
286
        }

287
        /* read the hash of ids->names for the input log */
288
        ret = darshan_log_get_namehash(in_fd, &in_hash);
289
290
291
292
        if(ret < 0)
        {
            fprintf(stderr,
                "Error: unable to read job data from input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
293
                infile_list[i]);
294
295
296
297
            darshan_log_close(in_fd);
            return(-1);
        }

298
        /* iterate the input hash, copying over record id->name mappings
299
300
301
302
         * that have not already been copied to the output hash
         */
        HASH_ITER(hlink, in_hash, ref, tmp)
        {
303
304
            HASH_FIND(hlink, merge_hash, &(ref->name_record->id),
                sizeof(darshan_record_id), found);
305
306
            if(!found)
            {
307
308
                HASH_ADD(hlink, merge_hash, name_record->id,
                    sizeof(darshan_record_id), ref);
309
            }
310
            else if(strcmp(ref->name_record->name, found->name_record->name))
311
312
313
314
315
316
317
318
319
            {
                fprintf(stderr,
                    "Error: invalid Darshan record table entry.\n");
                darshan_log_close(in_fd);
                return(-1);
            }
        }

        darshan_log_close(in_fd);
320
321
    }

322
323
324
325
    /* if a job end time was passed in, apply it to the output job */
    if(job_end_time > 0)
        merge_job.end_time = job_end_time;

Shane Snyder's avatar
Shane Snyder committed
326
327
328
    /* create the output "merged" log */
    merge_fd = darshan_log_create(outlog_path, DARSHAN_ZLIB_COMP, 1);
    if(merge_fd == NULL)
329
330
331
332
333
334
    {
        fprintf(stderr, "Error: unable to create output darshan log.\n");
        return(-1);
    }

    /* write the darshan job info, exe string, and mount data to output file */
335
    ret = darshan_log_put_job(merge_fd, &merge_job);
336
337
338
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write job data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
339
340
        darshan_log_close(merge_fd);
        unlink(outlog_path);
341
342
343
        return(-1);
    }

344
    ret = darshan_log_put_exe(merge_fd, merge_exe);
345
346
347
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write exe string to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
348
349
        darshan_log_close(merge_fd);
        unlink(outlog_path);
350
351
352
        return(-1);
    }

353
    ret = darshan_log_put_mounts(merge_fd, merge_mnt_pts, merge_fs_types, merge_mnt_count);
354
355
356
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write mount data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
357
358
        darshan_log_close(merge_fd);
        unlink(outlog_path);
359
360
361
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
362
    /* write the merged table of records to output file */
363
    ret = darshan_log_put_namehash(merge_fd, merge_hash);
364
365
366
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write record table to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
367
368
        darshan_log_close(merge_fd);
        unlink(outlog_path);
369
370
        return(-1);
    }
371

372
    /* iterate over active darshan modules and gather module data to write
Shane Snyder's avatar
Shane Snyder committed
373
     * to the merged output log
374
     */
Shane Snyder's avatar
Shane Snyder committed
375
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
376
377
378
    {
        if(!mod_logutils[i]) continue;

379
        if(shared_redux)
380
        {
381
            /* build the hash of records shared globally by this module */
Shane Snyder's avatar
Shane Snyder committed
382
383
            ret = build_mod_shared_rec_hash(infile_list, n_infiles, i,
                merge_job.nprocs, mod_buf, &shared_rec_hash);
384
            if(ret < 0)
385
            {
386
387
388
                fprintf(stderr,
                    "Error: unable to build list of %s module's shared records.\n",
                    darshan_module_names[i]);
Shane Snyder's avatar
Shane Snyder committed
389
390
                darshan_log_close(merge_fd);
                unlink(outlog_path);
391
                return(-1);
392
            }
393
394
395

        }

Shane Snyder's avatar
Shane Snyder committed
396
        for(j = 0; j < n_infiles; j++)
397
        {
Shane Snyder's avatar
Shane Snyder committed
398
            in_fd = darshan_log_open(infile_list[j]);
399
400
401
402
            if(in_fd == NULL)
            {
                fprintf(stderr,
                    "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
403
404
405
                    infile_list[j]);
                darshan_log_close(merge_fd);
                unlink(outlog_path);
406
407
408
                return(-1);
            }

Shane Snyder's avatar
Shane Snyder committed
409
410
411
412
413
            if(j == 0 && shared_rec_hash)
            {
                /* write out the shared records first */
                HASH_ITER(hlink, shared_rec_hash, sref, stmp)
                {
Shane Snyder's avatar
Shane Snyder committed
414
                    ret = mod_logutils[i]->log_put_record(merge_fd, sref->agg_rec, in_fd->mod_ver[i]);
Shane Snyder's avatar
Shane Snyder committed
415
416
417
418
419
                    if(ret < 0)
                    {
                        fprintf(stderr,
                            "Error: unable to write %s module record to output darshan log.\n",
                            darshan_module_names[i]);
Shane Snyder's avatar
Shane Snyder committed
420
421
422
                        darshan_log_close(in_fd);
                        darshan_log_close(merge_fd);
                        unlink(outlog_path);
Shane Snyder's avatar
Shane Snyder committed
423
424
425
426
427
                        return(-1);
                    }
                }
            }

428
            /* loop over module records and write them to output file */
429
            while((ret = mod_logutils[i]->log_get_record(in_fd, mod_buf)) == 1)
430
            {
431
432
433
434
435
436
                base_rec = (struct darshan_base_record *)mod_buf;

                HASH_FIND(hlink, shared_rec_hash, &(base_rec->id), sizeof(darshan_record_id), sref);
                if(sref)
                    continue; /* skip shared records */

Shane Snyder's avatar
Shane Snyder committed
437
                ret = mod_logutils[i]->log_put_record(merge_fd, mod_buf, in_fd->mod_ver[i]);
438
439
440
441
                if(ret < 0)
                {
                    fprintf(stderr,
                        "Error: unable to write %s module record to output log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
442
                        darshan_module_names[i], infile_list[j]);
443
                    darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
444
445
                    darshan_log_close(merge_fd);
                    unlink(outlog_path);
446
447
448
                    return(-1);
                }
            }
449
450
451
            if(ret < 0)
            {
                fprintf(stderr,
452
                    "Error: unable to read %s module record from input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
453
                    darshan_module_names[i], infile_list[j]);
454
                darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
455
456
                darshan_log_close(merge_fd);
                unlink(outlog_path);
457
458
459
                return(-1);
            }

460
            darshan_log_close(in_fd);
461
        }
462
463
464
465
466
467
468
469
470
471

        /* clear the shared record hash for the next module */
        if(shared_redux)
        {
            HASH_ITER(hlink, shared_rec_hash, sref, stmp)
            {
                HASH_DELETE(hlink, shared_rec_hash, sref);
                free(sref);
            }
        }
472
473
    }

Shane Snyder's avatar
Shane Snyder committed
474
    darshan_log_close(merge_fd);
475
476
477

    return(0);
}
Shane Snyder's avatar
Shane Snyder committed
478
479
480
481
482
483
484
485
486

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