darshan-merge.c 15.1 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
14
15
16
17
18
19
struct darshan_shared_record_ref
{
    darshan_record_id id;
    int ref_cnt;
    char agg_rec[DEF_MOD_BUF_SIZE];
    UT_hash_handle hlink;
};

20
21
void usage(char *exename)
{
22
    fprintf(stderr, "Usage: %s --output <output_path> [options] <input-logs>\n", exename);
Shane Snyder's avatar
Shane Snyder committed
23
24
    fprintf(stderr, "This utility merges multiple Darshan log files into a single output log file.\n");
    fprintf(stderr, "Options:\n");
25
    fprintf(stderr, "\t--output\t(REQUIRED) Full path of the output darshan log file.\n");
Shane Snyder's avatar
Shane Snyder committed
26
    fprintf(stderr, "\t--shared-redux\tReduce globally shared records into a single record.\n");
27
    fprintf(stderr, "\t--job-end-time\tSet the output log's job end time (requires argument of seconds since Epoch).\n");
28
29
30
31

    exit(1);
}

Shane Snyder's avatar
Shane Snyder committed
32
void parse_args(int argc, char **argv, char ***infile_list, int *n_files,
33
    char **outlog_path, int *shared_redux, int64_t *job_end_time)
34
35
{
    int index;
36
    char *check;
37
38
    static struct option long_opts[] =
    {
39
        {"output", required_argument, NULL, 'o'},
40
41
        {"shared-redux", no_argument, NULL, 's'},
        {"job-end-time", required_argument, NULL, 'e'},
42
43
44
45
        {0, 0, 0, 0}
    };

    *shared_redux = 0;
46
    *outlog_path = NULL;
47
    *job_end_time = 0;
48
49
50
51
52
53
54
55
56
57
58
59

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

        if(c == -1) break;

        switch(c)
        {
            case 's':
                *shared_redux = 1;
                break;
60
61
            case 'o':
                *outlog_path = optarg;
Shane Snyder's avatar
Shane Snyder committed
62
                break;
63
64
65
66
67
68
69
70
            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;
71
72
73
74
75
76
77
            case '?':
            default:
                usage(argv[0]);
                break;
        }
    }

78
    if(*outlog_path == NULL)
79
80
81
82
    {
        usage(argv[0]);
    }

Shane Snyder's avatar
Shane Snyder committed
83
84
    *infile_list = &argv[optind];
    *n_files = argc - optind;
85

Shane Snyder's avatar
Shane Snyder committed
86
    return;
87
88
}

Shane Snyder's avatar
Shane Snyder committed
89
90
91
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)
92
93
94
95
{
    darshan_fd in_fd;
    struct darshan_base_record *base_rec;
    struct darshan_shared_record_ref *ref, *tmp;
96
    int init_rank = -1;
97
98
99
100
    int ret;
    int i;

    /* loop over each input log file */
Shane Snyder's avatar
Shane Snyder committed
101
    for(i = 0; i < n_infiles; i++)
102
    {
Shane Snyder's avatar
Shane Snyder committed
103
        in_fd = darshan_log_open(infile_list[i]);
104
105
106
107
        if(in_fd == NULL)
        {
            fprintf(stderr,
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
108
                infile_list[i]);
109
110
111
112
113
114
            return(-1);
        }

        while((ret = mod_logutils[mod_id]->log_get_record(in_fd, mod_buf)) == 1)
        {
            base_rec = (struct darshan_base_record *)mod_buf;
115
116
            if(init_rank == -1)
                init_rank = base_rec->rank;
117
118

            /* initialize the hash with the first rank's records */
119
            if(base_rec->rank == init_rank)
120
121
122
123
124
125
126
127
128
129
            {
                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
130
                memset(ref, 0, sizeof(*ref));
131
132
133
134
135
136
137
138
139
140
141
142

                /* 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
143
           {
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
                /* 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
159
                darshan_module_names[mod_id], infile_list[i]);
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
            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);
}

180
181
int main(int argc, char *argv[])
{
Shane Snyder's avatar
Shane Snyder committed
182
183
    char **infile_list;
    int n_infiles;
184
    int shared_redux;
185
    int64_t job_end_time = 0;
186
    char *outlog_path;
Shane Snyder's avatar
Shane Snyder committed
187
188
    darshan_fd in_fd, merge_fd;
    struct darshan_job in_job, merge_job;
189
    char merge_exe[DARSHAN_EXE_LEN+1] = {0};
190
    struct darshan_mnt_info *merge_mnt_array;
Shane Snyder's avatar
Shane Snyder committed
191
    int merge_mnt_count = 0;
192
193
194
    struct darshan_name_record_ref *in_hash = NULL;
    struct darshan_name_record_ref *merge_hash = NULL;
    struct darshan_name_record_ref *ref, *tmp, *found;
195
196
197
198
    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];
199
200
    int i, j;
    int ret;
201
202

    /* grab command line arguments */
203
    parse_args(argc, argv, &infile_list, &n_infiles, &outlog_path, &shared_redux, &job_end_time);
204

Shane Snyder's avatar
Shane Snyder committed
205
    memset(&merge_job, 0, sizeof(struct darshan_job));
206

Shane Snyder's avatar
Shane Snyder committed
207
    /* first pass at merging together logs:
208
209
     *      - compose output job-level metadata structure (including exe & mount data)
     *      - compose output record_id->file_name mapping 
210
     */
Shane Snyder's avatar
Shane Snyder committed
211
    for(i = 0; i < n_infiles; i++)
212
213
214
    {
        memset(&in_job, 0, sizeof(struct darshan_job));

Shane Snyder's avatar
Shane Snyder committed
215
        in_fd = darshan_log_open(infile_list[i]);
216
        if(in_fd == NULL)
217
218
        {
            fprintf(stderr,
219
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
220
                infile_list[i]);
221
222
223
224
            return(-1);
        }

        /* read job-level metadata from the input file */
225
        ret = darshan_log_get_job(in_fd, &in_job);
226
        if(ret < 0)
227
228
        {
            fprintf(stderr,
229
                "Error: unable to read job data from input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
230
                infile_list[i]);
231
            darshan_log_close(in_fd);
232
233
234
            return(-1);
        }

235
236
237
238
239
240
241
242
243
        /* 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
244
                infile_list[i]);
245
246
247
248
            darshan_log_close(in_fd);
            return(-1);
        }

249
250
        if(i == 0)
        {
251
            /* get job data, exe, & mounts directly from the first input log */
Shane Snyder's avatar
Shane Snyder committed
252
            memcpy(&merge_job, &in_job, sizeof(struct darshan_job));
253

254
            ret = darshan_log_get_exe(in_fd, merge_exe);
255
            if(ret < 0)
256
257
            {
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
258
259
                    "Error: unable to read exe string from input Darshan log file %s.\n",
                    infile_list[i]);
260
                darshan_log_close(in_fd);
261
262
263
                return(-1);
            }

264
            ret = darshan_log_get_mounts(in_fd, &merge_mnt_array, &merge_mnt_count);
265
            if(ret < 0)
266
            {
267
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
268
269
                    "Error: unable to read mount info from input Darshan log file %s.\n",
                    infile_list[i]);
270
271
                darshan_log_close(in_fd);
                return(-1);
272
273
274
275
            }
        }
        else
        {
276
            /* potentially update job timestamps using remaining logs */
Shane Snyder's avatar
Shane Snyder committed
277
278
279
280
            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;
281
282
        }

283
        /* read the hash of ids->names for the input log */
284
        ret = darshan_log_get_namehash(in_fd, &in_hash);
285
286
287
288
        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
289
                infile_list[i]);
290
291
292
293
            darshan_log_close(in_fd);
            return(-1);
        }

294
        /* iterate the input hash, copying over record id->name mappings
295
296
297
298
         * that have not already been copied to the output hash
         */
        HASH_ITER(hlink, in_hash, ref, tmp)
        {
299
300
            HASH_FIND(hlink, merge_hash, &(ref->name_record->id),
                sizeof(darshan_record_id), found);
301
302
            if(!found)
            {
303
304
                HASH_ADD(hlink, merge_hash, name_record->id,
                    sizeof(darshan_record_id), ref);
305
            }
306
            else if(strcmp(ref->name_record->name, found->name_record->name))
307
308
309
310
311
312
313
314
315
            {
                fprintf(stderr,
                    "Error: invalid Darshan record table entry.\n");
                darshan_log_close(in_fd);
                return(-1);
            }
        }

        darshan_log_close(in_fd);
316
317
    }

318
319
320
321
    /* 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
322
323
324
    /* create the output "merged" log */
    merge_fd = darshan_log_create(outlog_path, DARSHAN_ZLIB_COMP, 1);
    if(merge_fd == NULL)
325
326
327
328
329
330
    {
        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 */
331
    ret = darshan_log_put_job(merge_fd, &merge_job);
332
333
334
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write job data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
335
336
        darshan_log_close(merge_fd);
        unlink(outlog_path);
337
338
339
        return(-1);
    }

340
    ret = darshan_log_put_exe(merge_fd, merge_exe);
341
342
343
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write exe string to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
344
345
        darshan_log_close(merge_fd);
        unlink(outlog_path);
346
347
348
        return(-1);
    }

349
    ret = darshan_log_put_mounts(merge_fd, merge_mnt_array, merge_mnt_count);
350
351
352
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write mount data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
353
354
        darshan_log_close(merge_fd);
        unlink(outlog_path);
355
356
357
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
358
    /* write the merged table of records to output file */
359
    ret = darshan_log_put_namehash(merge_fd, merge_hash);
360
361
362
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write record table to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
363
364
        darshan_log_close(merge_fd);
        unlink(outlog_path);
365
366
        return(-1);
    }
367

368
    /* iterate over active darshan modules and gather module data to write
Shane Snyder's avatar
Shane Snyder committed
369
     * to the merged output log
370
     */
Shane Snyder's avatar
Shane Snyder committed
371
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
372
373
374
    {
        if(!mod_logutils[i]) continue;

375
        if(shared_redux)
376
        {
377
            /* build the hash of records shared globally by this module */
Shane Snyder's avatar
Shane Snyder committed
378
379
            ret = build_mod_shared_rec_hash(infile_list, n_infiles, i,
                merge_job.nprocs, mod_buf, &shared_rec_hash);
380
            if(ret < 0)
381
            {
382
383
384
                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
385
386
                darshan_log_close(merge_fd);
                unlink(outlog_path);
387
                return(-1);
388
            }
389
390
391

        }

Shane Snyder's avatar
Shane Snyder committed
392
        for(j = 0; j < n_infiles; j++)
393
        {
Shane Snyder's avatar
Shane Snyder committed
394
            in_fd = darshan_log_open(infile_list[j]);
395
396
397
398
            if(in_fd == NULL)
            {
                fprintf(stderr,
                    "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
399
400
401
                    infile_list[j]);
                darshan_log_close(merge_fd);
                unlink(outlog_path);
402
403
404
                return(-1);
            }

Shane Snyder's avatar
Shane Snyder committed
405
406
407
408
409
            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
410
                    ret = mod_logutils[i]->log_put_record(merge_fd, sref->agg_rec, in_fd->mod_ver[i]);
Shane Snyder's avatar
Shane Snyder committed
411
412
413
414
415
                    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
416
417
418
                        darshan_log_close(in_fd);
                        darshan_log_close(merge_fd);
                        unlink(outlog_path);
Shane Snyder's avatar
Shane Snyder committed
419
420
421
422
423
                        return(-1);
                    }
                }
            }

424
            /* loop over module records and write them to output file */
425
            while((ret = mod_logutils[i]->log_get_record(in_fd, mod_buf)) == 1)
426
            {
427
428
429
430
431
432
                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
433
                ret = mod_logutils[i]->log_put_record(merge_fd, mod_buf, in_fd->mod_ver[i]);
434
435
436
437
                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
438
                        darshan_module_names[i], infile_list[j]);
439
                    darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
440
441
                    darshan_log_close(merge_fd);
                    unlink(outlog_path);
442
443
444
                    return(-1);
                }
            }
445
446
447
            if(ret < 0)
            {
                fprintf(stderr,
448
                    "Error: unable to read %s module record from input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
449
                    darshan_module_names[i], infile_list[j]);
450
                darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
451
452
                darshan_log_close(merge_fd);
                unlink(outlog_path);
453
454
455
                return(-1);
            }

456
            darshan_log_close(in_fd);
457
        }
458
459
460
461
462
463
464
465
466
467

        /* 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);
            }
        }
468
469
    }

Shane Snyder's avatar
Shane Snyder committed
470
    darshan_log_close(merge_fd);
471
472
473

    return(0);
}
Shane Snyder's avatar
Shane Snyder committed
474
475
476
477
478
479
480
481
482

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