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

14 15 16 17 18 19 20 21 22 23
/* TODO: set job end timestamp? */

struct darshan_shared_record_ref
{
    darshan_record_id id;
    int ref_cnt;
    char agg_rec[DEF_MOD_BUF_SIZE];
    UT_hash_handle hlink;
};

24 25
void usage(char *exename)
{
26
    fprintf(stderr, "Usage: %s --output <output_path> [options] <input-logs>\n", exename);
Shane Snyder's avatar
Shane Snyder committed
27 28
    fprintf(stderr, "This utility merges multiple Darshan log files into a single output log file.\n");
    fprintf(stderr, "Options:\n");
29
    fprintf(stderr, "\t--output\t(REQUIRED) Full path of the output darshan log file.\n");
Shane Snyder's avatar
Shane Snyder committed
30
    fprintf(stderr, "\t--shared-redux\tReduce globally shared records into a single record.\n");
31
    fprintf(stderr, "\t--job-end-time\tSet the output log's job end time (requires argument of seconds since Epoch).\n");
32 33 34 35

    exit(1);
}

Shane Snyder's avatar
Shane Snyder committed
36
void parse_args(int argc, char **argv, char ***infile_list, int *n_files,
37
    char **outlog_path, int *shared_redux, int64_t *job_end_time)
38 39
{
    int index;
40
    char *check;
41 42
    static struct option long_opts[] =
    {
43
        {"output", required_argument, NULL, 'o'},
44 45
        {"shared-redux", no_argument, NULL, 's'},
        {"job-end-time", required_argument, NULL, 'e'},
46 47 48 49
        {0, 0, 0, 0}
    };

    *shared_redux = 0;
50
    *outlog_path = NULL;
51
    *job_end_time = 0;
52 53 54 55 56 57 58 59 60 61 62 63

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

        if(c == -1) break;

        switch(c)
        {
            case 's':
                *shared_redux = 1;
                break;
64 65
            case 'o':
                *outlog_path = optarg;
Shane Snyder's avatar
Shane Snyder committed
66
                break;
67 68 69 70 71 72 73 74
            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;
75 76 77 78 79 80 81
            case '?':
            default:
                usage(argv[0]);
                break;
        }
    }

82
    if(*outlog_path == NULL)
83 84 85 86
    {
        usage(argv[0]);
    }

Shane Snyder's avatar
Shane Snyder committed
87 88
    *infile_list = &argv[optind];
    *n_files = argc - optind;
89

Shane Snyder's avatar
Shane Snyder committed
90
    return;
91 92
}

Shane Snyder's avatar
Shane Snyder committed
93 94 95
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)
96 97 98 99 100 101 102 103 104
{
    darshan_fd in_fd;
    struct darshan_base_record *base_rec;
    struct darshan_shared_record_ref *ref, *tmp;
    int init = 0;
    int ret;
    int i;

    /* loop over each input log file */
Shane Snyder's avatar
Shane Snyder committed
105
    for(i = 0; i < n_infiles; i++)
106
    {
Shane Snyder's avatar
Shane Snyder committed
107
        in_fd = darshan_log_open(infile_list[i]);
108 109 110 111
        if(in_fd == NULL)
        {
            fprintf(stderr,
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
112
                infile_list[i]);
113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
            return(-1);
        }

        while((ret = mod_logutils[mod_id]->log_get_record(in_fd, mod_buf)) == 1)
        {
            base_rec = (struct darshan_base_record *)mod_buf;

            /* initialize the hash with the first rank's records */
            if(!init)
            {
                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);
                }
132
                memset(ref, 0, sizeof(*ref));
133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161

                /* 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);
                init = 1;
            }
            else
            {
                /* 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
162
                darshan_module_names[mod_id], infile_list[i]);
163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
            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);
}

183 184
int main(int argc, char *argv[])
{
Shane Snyder's avatar
Shane Snyder committed
185 186
    char **infile_list;
    int n_infiles;
187
    int shared_redux;
188
    int64_t job_end_time = 0;
189
    char *outlog_path;
Shane Snyder's avatar
Shane Snyder committed
190 191
    darshan_fd in_fd, merge_fd;
    struct darshan_job in_job, merge_job;
192
    char merge_exe[DARSHAN_EXE_LEN+1] = {0};
Shane Snyder's avatar
Shane Snyder committed
193 194 195
    char **merge_mnt_pts;
    char **merge_fs_types;
    int merge_mnt_count = 0;
196
    struct darshan_record_ref *in_hash = NULL;
Shane Snyder's avatar
Shane Snyder committed
197
    struct darshan_record_ref *merge_hash = NULL;
198
    struct darshan_record_ref *ref, *tmp, *found;
199 200 201 202
    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];
203 204
    int i, j;
    int ret;
205 206

    /* grab command line arguments */
207
    parse_args(argc, argv, &infile_list, &n_infiles, &outlog_path, &shared_redux, &job_end_time);
208

Shane Snyder's avatar
Shane Snyder committed
209
    memset(&merge_job, 0, sizeof(struct darshan_job));
210

Shane Snyder's avatar
Shane Snyder committed
211
    /* first pass at merging together logs:
212 213
     *      - compose output job-level metadata structure (including exe & mount data)
     *      - compose output record_id->file_name mapping 
214
     */
Shane Snyder's avatar
Shane Snyder committed
215
    for(i = 0; i < n_infiles; i++)
216 217 218
    {
        memset(&in_job, 0, sizeof(struct darshan_job));

Shane Snyder's avatar
Shane Snyder committed
219
        in_fd = darshan_log_open(infile_list[i]);
220
        if(in_fd == NULL)
221 222
        {
            fprintf(stderr,
223
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
224
                infile_list[i]);
225 226 227 228
            return(-1);
        }

        /* read job-level metadata from the input file */
229 230
        ret = darshan_log_getjob(in_fd, &in_job);
        if(ret < 0)
231 232
        {
            fprintf(stderr,
233
                "Error: unable to read job data from input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
234
                infile_list[i]);
235
            darshan_log_close(in_fd);
236 237 238
            return(-1);
        }

239 240 241 242 243 244 245 246 247
        /* 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
248
                infile_list[i]);
249 250 251 252
            darshan_log_close(in_fd);
            return(-1);
        }

253 254
        if(i == 0)
        {
255
            /* get job data, exe, & mounts directly from the first input log */
Shane Snyder's avatar
Shane Snyder committed
256
            memcpy(&merge_job, &in_job, sizeof(struct darshan_job));
257

Shane Snyder's avatar
Shane Snyder committed
258
            ret = darshan_log_getexe(in_fd, merge_exe);
259
            if(ret < 0)
260 261
            {
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
262 263
                    "Error: unable to read exe string from input Darshan log file %s.\n",
                    infile_list[i]);
264
                darshan_log_close(in_fd);
265 266 267
                return(-1);
            }

Shane Snyder's avatar
Shane Snyder committed
268 269
            ret = darshan_log_getmounts(in_fd, &merge_mnt_pts,
                &merge_fs_types, &merge_mnt_count);
270
            if(ret < 0)
271
            {
272
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
273 274
                    "Error: unable to read mount info from input Darshan log file %s.\n",
                    infile_list[i]);
275 276
                darshan_log_close(in_fd);
                return(-1);
277 278 279 280
            }
        }
        else
        {
281
            /* potentially update job timestamps using remaining logs */
Shane Snyder's avatar
Shane Snyder committed
282 283 284 285
            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;
286 287
        }

288
        /* read the hash of ids->names for the input log */
289 290 291 292 293
        ret = darshan_log_gethash(in_fd, &in_hash);
        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
294
                infile_list[i]);
295 296 297 298 299 300 301 302 303
            darshan_log_close(in_fd);
            return(-1);
        }

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

        darshan_log_close(in_fd);
319 320
    }

321 322 323 324
    /* 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
325 326 327
    /* create the output "merged" log */
    merge_fd = darshan_log_create(outlog_path, DARSHAN_ZLIB_COMP, 1);
    if(merge_fd == NULL)
328 329 330 331 332 333
    {
        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 */
Shane Snyder's avatar
Shane Snyder committed
334
    ret = darshan_log_putjob(merge_fd, &merge_job);
335 336 337
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write job data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
338 339
        darshan_log_close(merge_fd);
        unlink(outlog_path);
340 341 342
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
343
    ret = darshan_log_putexe(merge_fd, merge_exe);
344 345 346
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write exe string to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
347 348
        darshan_log_close(merge_fd);
        unlink(outlog_path);
349 350 351
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
352
    ret = darshan_log_putmounts(merge_fd, merge_mnt_pts, merge_fs_types, merge_mnt_count);
353 354 355
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write mount data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
356 357
        darshan_log_close(merge_fd);
        unlink(outlog_path);
358 359 360
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
361 362
    /* write the merged table of records to output file */
    ret = darshan_log_puthash(merge_fd, merge_hash);
363 364 365
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write record table to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
366 367
        darshan_log_close(merge_fd);
        unlink(outlog_path);
368 369
        return(-1);
    }
370

371
    /* iterate over active darshan modules and gather module data to write
Shane Snyder's avatar
Shane Snyder committed
372
     * to the merged output log
373
     */
Shane Snyder's avatar
Shane Snyder committed
374
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
375 376 377
    {
        if(!mod_logutils[i]) continue;

378
        if(shared_redux)
379
        {
380
            /* build the hash of records shared globally by this module */
Shane Snyder's avatar
Shane Snyder committed
381 382
            ret = build_mod_shared_rec_hash(infile_list, n_infiles, i,
                merge_job.nprocs, mod_buf, &shared_rec_hash);
383
            if(ret < 0)
384
            {
385 386 387
                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
388 389
                darshan_log_close(merge_fd);
                unlink(outlog_path);
390
                return(-1);
391
            }
392 393 394

        }

Shane Snyder's avatar
Shane Snyder committed
395
        for(j = 0; j < n_infiles; j++)
396
        {
Shane Snyder's avatar
Shane Snyder committed
397
            in_fd = darshan_log_open(infile_list[j]);
398 399 400 401
            if(in_fd == NULL)
            {
                fprintf(stderr,
                    "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
402 403 404
                    infile_list[j]);
                darshan_log_close(merge_fd);
                unlink(outlog_path);
405 406 407
                return(-1);
            }

Shane Snyder's avatar
Shane Snyder committed
408 409 410 411 412
            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
413
                    ret = mod_logutils[i]->log_put_record(merge_fd, sref->agg_rec, in_fd->mod_ver[i]);
Shane Snyder's avatar
Shane Snyder committed
414 415 416 417 418
                    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
419 420 421
                        darshan_log_close(in_fd);
                        darshan_log_close(merge_fd);
                        unlink(outlog_path);
Shane Snyder's avatar
Shane Snyder committed
422 423 424 425 426
                        return(-1);
                    }
                }
            }

427
            /* loop over module records and write them to output file */
428
            while((ret = mod_logutils[i]->log_get_record(in_fd, mod_buf)) == 1)
429
            {
430 431 432 433 434 435
                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
436
                ret = mod_logutils[i]->log_put_record(merge_fd, mod_buf, in_fd->mod_ver[i]);
437 438 439 440
                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
441
                        darshan_module_names[i], infile_list[j]);
442
                    darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
443 444
                    darshan_log_close(merge_fd);
                    unlink(outlog_path);
445 446 447
                    return(-1);
                }
            }
448 449 450
            if(ret < 0)
            {
                fprintf(stderr,
451
                    "Error: unable to read %s module record from input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
452
                    darshan_module_names[i], infile_list[j]);
453
                darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
454 455
                darshan_log_close(merge_fd);
                unlink(outlog_path);
456 457 458
                return(-1);
            }

459
            darshan_log_close(in_fd);
460
        }
461 462 463 464 465 466 467 468 469 470

        /* 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);
            }
        }
471 472
    }

Shane Snyder's avatar
Shane Snyder committed
473
    darshan_log_close(merge_fd);
474 475 476

    return(0);
}
Shane Snyder's avatar
Shane Snyder committed
477 478 479 480 481 482 483 484 485

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