darshan-merge.c 15.6 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_log_glob>\n", exename);
Shane Snyder's avatar
Shane Snyder committed
23
    fprintf(stderr, "This utility merges multiple Darshan log files into a single output log file.\n");
24
    fprintf(stderr, "<input_log_glob> is a pattern that matches all input log files (e.g., /log-path/*.darshan).\n");
Shane Snyder's avatar
Shane Snyder committed
25
    fprintf(stderr, "Options:\n");
26
    fprintf(stderr, "\t--output\t(REQUIRED) Full path of the output darshan log file.\n");
Shane Snyder's avatar
Shane Snyder committed
27
    fprintf(stderr, "\t--shared-redux\tReduce globally shared records into a single record.\n");
28
    fprintf(stderr, "\t--job-end-time\tSet the output log's job end time (requires argument of seconds since Epoch).\n");
29 30 31 32

    exit(1);
}

Shane Snyder's avatar
Shane Snyder committed
33
void parse_args(int argc, char **argv, char ***infile_list, int *n_files,
34
    char **outlog_path, int *shared_redux, int64_t *job_end_time)
35 36
{
    int index;
37
    char *check;
38 39
    static struct option long_opts[] =
    {
40
        {"output", required_argument, NULL, 'o'},
41 42
        {"shared-redux", no_argument, NULL, 's'},
        {"job-end-time", required_argument, NULL, 'e'},
43 44 45 46
        {0, 0, 0, 0}
    };

    *shared_redux = 0;
47
    *outlog_path = NULL;
48
    *job_end_time = 0;
49 50 51 52 53 54 55 56 57 58 59 60

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

        if(c == -1) break;

        switch(c)
        {
            case 's':
                *shared_redux = 1;
                break;
61 62
            case 'o':
                *outlog_path = optarg;
Shane Snyder's avatar
Shane Snyder committed
63
                break;
64 65 66 67 68 69 70 71
            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;
72 73 74 75 76 77 78
            case '?':
            default:
                usage(argv[0]);
                break;
        }
    }

79
    if(*outlog_path == NULL)
80 81 82 83
    {
        usage(argv[0]);
    }

Shane Snyder's avatar
Shane Snyder committed
84 85
    *infile_list = &argv[optind];
    *n_files = argc - optind;
86

Shane Snyder's avatar
Shane Snyder committed
87
    return;
88 89
}

Shane Snyder's avatar
Shane Snyder committed
90 91 92
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)
93 94 95 96
{
    darshan_fd in_fd;
    struct darshan_base_record *base_rec;
    struct darshan_shared_record_ref *ref, *tmp;
97
    int init_rank = -1;
98 99 100 101
    int ret;
    int i;

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

113
        while((ret = mod_logutils[mod_id]->log_get_record(in_fd, (void **)&mod_buf)) == 1)
114 115
        {
            base_rec = (struct darshan_base_record *)mod_buf;
116 117
            if(init_rank == -1)
                init_rank = base_rec->rank;
118 119

            /* initialize the hash with the first rank's records */
120
            if(base_rec->rank == init_rank)
121 122 123 124 125 126 127 128 129 130
            {
                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);
                }
131
                memset(ref, 0, sizeof(*ref));
132 133

                /* initialize the aggregate record with this rank's record */
134
                mod_logutils[mod_id]->log_agg_records(mod_buf, ref->agg_rec, 1);
135 136 137 138 139 140 141 142 143
                agg_base = (struct darshan_base_record *)ref->agg_rec;
                agg_base->id = base_rec->id;
                agg_base->rank = -1;

                ref->id = base_rec->id;
                ref->ref_cnt = 1;
                HASH_ADD(hlink, *shared_rec_hash, id, sizeof(darshan_record_id), ref);
            }
            else
144
           {
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159
                /* 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
160
                darshan_module_names[mod_id], infile_list[i]);
161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
            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);
}

181 182
int main(int argc, char *argv[])
{
Shane Snyder's avatar
Shane Snyder committed
183 184
    char **infile_list;
    int n_infiles;
185
    int shared_redux;
186
    int64_t job_end_time = 0;
187
    char *outlog_path;
Shane Snyder's avatar
Shane Snyder committed
188 189
    darshan_fd in_fd, merge_fd;
    struct darshan_job in_job, merge_job;
190
    char merge_exe[DARSHAN_EXE_LEN+1] = {0};
191
    struct darshan_mnt_info *merge_mnt_array;
Shane Snyder's avatar
Shane Snyder committed
192
    int merge_mnt_count = 0;
193 194 195
    struct darshan_name_record_ref *in_hash = NULL;
    struct darshan_name_record_ref *merge_hash = NULL;
    struct darshan_name_record_ref *ref, *tmp, *found;
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;
199
    char *mod_buf;
200 201
    int i, j;
    int ret;
202 203

    /* grab command line arguments */
204
    parse_args(argc, argv, &infile_list, &n_infiles, &outlog_path, &shared_redux, &job_end_time);
205

Shane Snyder's avatar
Shane Snyder committed
206
    memset(&merge_job, 0, sizeof(struct darshan_job));
207

Shane Snyder's avatar
Shane Snyder committed
208
    /* first pass at merging together logs:
209 210
     *      - compose output job-level metadata structure (including exe & mount data)
     *      - compose output record_id->file_name mapping 
211
     */
Shane Snyder's avatar
Shane Snyder committed
212
    for(i = 0; i < n_infiles; i++)
213 214 215
    {
        memset(&in_job, 0, sizeof(struct darshan_job));

Shane Snyder's avatar
Shane Snyder committed
216
        in_fd = darshan_log_open(infile_list[i]);
217
        if(in_fd == NULL)
218 219
        {
            fprintf(stderr,
220
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
221
                infile_list[i]);
222 223 224 225
            return(-1);
        }

        /* read job-level metadata from the input file */
226
        ret = darshan_log_get_job(in_fd, &in_job);
227
        if(ret < 0)
228 229
        {
            fprintf(stderr,
230
                "Error: unable to read job data from input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
231
                infile_list[i]);
232
            darshan_log_close(in_fd);
233 234 235
            return(-1);
        }

236 237 238 239 240 241 242 243 244
        /* 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
245
                infile_list[i]);
246 247 248 249
            darshan_log_close(in_fd);
            return(-1);
        }

250 251
        if(i == 0)
        {
252
            /* get job data, exe, & mounts directly from the first input log */
Shane Snyder's avatar
Shane Snyder committed
253
            memcpy(&merge_job, &in_job, sizeof(struct darshan_job));
254

255
            ret = darshan_log_get_exe(in_fd, merge_exe);
256
            if(ret < 0)
257 258
            {
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
259 260
                    "Error: unable to read exe string from input Darshan log file %s.\n",
                    infile_list[i]);
261
                darshan_log_close(in_fd);
262 263 264
                return(-1);
            }

265
            ret = darshan_log_get_mounts(in_fd, &merge_mnt_array, &merge_mnt_count);
266
            if(ret < 0)
267
            {
268
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
269 270
                    "Error: unable to read mount info from input Darshan log file %s.\n",
                    infile_list[i]);
271 272
                darshan_log_close(in_fd);
                return(-1);
273 274 275 276
            }
        }
        else
        {
277
            /* potentially update job timestamps using remaining logs */
Shane Snyder's avatar
Shane Snyder committed
278 279 280 281
            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;
282 283
        }

284
        /* read the hash of ids->names for the input log */
285
        ret = darshan_log_get_namehash(in_fd, &in_hash);
286 287 288 289
        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
290
                infile_list[i]);
291 292 293 294
            darshan_log_close(in_fd);
            return(-1);
        }

295
        /* iterate the input hash, copying over record id->name mappings
296 297 298 299
         * that have not already been copied to the output hash
         */
        HASH_ITER(hlink, in_hash, ref, tmp)
        {
300 301
            HASH_FIND(hlink, merge_hash, &(ref->name_record->id),
                sizeof(darshan_record_id), found);
302 303
            if(!found)
            {
304 305
                HASH_ADD(hlink, merge_hash, name_record->id,
                    sizeof(darshan_record_id), ref);
306
            }
307
            else if(strcmp(ref->name_record->name, found->name_record->name))
308 309 310 311 312 313 314 315 316
            {
                fprintf(stderr,
                    "Error: invalid Darshan record table entry.\n");
                darshan_log_close(in_fd);
                return(-1);
            }
        }

        darshan_log_close(in_fd);
317 318
    }

319 320 321 322
    /* 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
323 324 325
    /* create the output "merged" log */
    merge_fd = darshan_log_create(outlog_path, DARSHAN_ZLIB_COMP, 1);
    if(merge_fd == NULL)
326 327 328 329 330 331
    {
        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 */
332
    ret = darshan_log_put_job(merge_fd, &merge_job);
333 334 335
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write job data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
336 337
        darshan_log_close(merge_fd);
        unlink(outlog_path);
338 339 340
        return(-1);
    }

341
    ret = darshan_log_put_exe(merge_fd, merge_exe);
342 343 344
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write exe string to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
345 346
        darshan_log_close(merge_fd);
        unlink(outlog_path);
347 348 349
        return(-1);
    }

350
    ret = darshan_log_put_mounts(merge_fd, merge_mnt_array, merge_mnt_count);
351 352 353
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write mount data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
354 355
        darshan_log_close(merge_fd);
        unlink(outlog_path);
356 357 358
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
359
    /* write the merged table of records to output file */
360
    ret = darshan_log_put_namehash(merge_fd, merge_hash);
361 362 363
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write record table to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
364 365
        darshan_log_close(merge_fd);
        unlink(outlog_path);
366 367
        return(-1);
    }
368

369 370 371 372 373 374 375 376 377
    mod_buf = malloc(DEF_MOD_BUF_SIZE);
    if(!mod_buf)
    {
        fprintf(stderr, "Error: unable to write record table to output darshan log.\n");
        darshan_log_close(merge_fd);
        unlink(outlog_path);
        return(-1);
    }

378
    /* iterate over active darshan modules and gather module data to write
Shane Snyder's avatar
Shane Snyder committed
379
     * to the merged output log
380
     */
381
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
382 383 384
    {
        if(!mod_logutils[i]) continue;

385
        if(shared_redux)
386
        {
387
            /* build the hash of records shared globally by this module */
Shane Snyder's avatar
Shane Snyder committed
388 389
            ret = build_mod_shared_rec_hash(infile_list, n_infiles, i,
                merge_job.nprocs, mod_buf, &shared_rec_hash);
390
            if(ret < 0)
391
            {
392 393 394
                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
395 396
                darshan_log_close(merge_fd);
                unlink(outlog_path);
397
                free(mod_buf);
398
                return(-1);
399
            }
400 401 402

        }

Shane Snyder's avatar
Shane Snyder committed
403
        for(j = 0; j < n_infiles; j++)
404
        {
Shane Snyder's avatar
Shane Snyder committed
405
            in_fd = darshan_log_open(infile_list[j]);
406 407 408 409
            if(in_fd == NULL)
            {
                fprintf(stderr,
                    "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
410 411 412
                    infile_list[j]);
                darshan_log_close(merge_fd);
                unlink(outlog_path);
413
                free(mod_buf);
414 415 416
                return(-1);
            }

417 418 419 420 421
            if(j == 0 && shared_rec_hash)
            {
                /* write out the shared records first */
                HASH_ITER(hlink, shared_rec_hash, sref, stmp)
                {
422
                    ret = mod_logutils[i]->log_put_record(merge_fd, sref->agg_rec);
423 424 425 426 427
                    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
428 429 430
                        darshan_log_close(in_fd);
                        darshan_log_close(merge_fd);
                        unlink(outlog_path);
431
                        free(mod_buf);
432 433 434 435 436
                        return(-1);
                    }
                }
            }

437
            /* loop over module records and write them to output file */
438
            while((ret = mod_logutils[i]->log_get_record(in_fd, (void **)&mod_buf)) == 1)
439
            {
440 441 442 443 444 445
                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 */

446
                ret = mod_logutils[i]->log_put_record(merge_fd, mod_buf);
447 448 449 450
                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
451
                        darshan_module_names[i], infile_list[j]);
452
                    darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
453 454
                    darshan_log_close(merge_fd);
                    unlink(outlog_path);
455
                    free(mod_buf);
456 457 458
                    return(-1);
                }
            }
459 460 461
            if(ret < 0)
            {
                fprintf(stderr,
462
                    "Error: unable to read %s module record from input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
463
                    darshan_module_names[i], infile_list[j]);
464
                darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
465 466
                darshan_log_close(merge_fd);
                unlink(outlog_path);
467
                free(mod_buf);
468 469 470
                return(-1);
            }

471
            darshan_log_close(in_fd);
472
        }
473 474 475 476 477 478 479 480 481 482

        /* 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);
            }
        }
483 484
    }

485 486
    free(mod_buf);

Shane Snyder's avatar
Shane Snyder committed
487
    darshan_log_close(merge_fd);
488 489 490

    return(0);
}
Shane Snyder's avatar
Shane Snyder committed
491 492 493 494 495 496 497 498 499

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