darshan-merge.c 14.4 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 32 33 34

    exit(1);
}

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

    *shared_redux = 0;
47
    *outlog_path = NULL;
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
            case '?':
            default:
                usage(argv[0]);
                break;
        }
    }

70
    if(*outlog_path == NULL)
71 72 73 74
    {
        usage(argv[0]);
    }

Shane Snyder's avatar
Shane Snyder committed
75 76
    *infile_list = &argv[optind];
    *n_files = argc - optind;
77

Shane Snyder's avatar
Shane Snyder committed
78
    return;
79 80
}

Shane Snyder's avatar
Shane Snyder committed
81 82 83
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)
84 85 86 87 88 89 90 91 92
{
    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
93
    for(i = 0; i < n_infiles; i++)
94
    {
Shane Snyder's avatar
Shane Snyder committed
95
        in_fd = darshan_log_open(infile_list[i]);
96 97 98 99
        if(in_fd == NULL)
        {
            fprintf(stderr,
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
100
                infile_list[i]);
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119
            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);
                }
120
                memset(ref, 0, sizeof(*ref));
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

                /* 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
150
                darshan_module_names[mod_id], infile_list[i]);
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
            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);
}

171 172
int main(int argc, char *argv[])
{
Shane Snyder's avatar
Shane Snyder committed
173 174
    char **infile_list;
    int n_infiles;
175
    int shared_redux;
176
    char *outlog_path;
Shane Snyder's avatar
Shane Snyder committed
177 178
    darshan_fd in_fd, merge_fd;
    struct darshan_job in_job, merge_job;
179
    char merge_exe[DARSHAN_EXE_LEN+1] = {0};
Shane Snyder's avatar
Shane Snyder committed
180 181 182
    char **merge_mnt_pts;
    char **merge_fs_types;
    int merge_mnt_count = 0;
183
    struct darshan_record_ref *in_hash = NULL;
Shane Snyder's avatar
Shane Snyder committed
184
    struct darshan_record_ref *merge_hash = NULL;
185
    struct darshan_record_ref *ref, *tmp, *found;
186 187 188 189
    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];
190 191
    int i, j;
    int ret;
192 193

    /* grab command line arguments */
194
    parse_args(argc, argv, &infile_list, &n_infiles, &outlog_path, &shared_redux);
195

Shane Snyder's avatar
Shane Snyder committed
196
    memset(&merge_job, 0, sizeof(struct darshan_job));
197

Shane Snyder's avatar
Shane Snyder committed
198
    /* first pass at merging together logs:
199 200
     *      - compose output job-level metadata structure (including exe & mount data)
     *      - compose output record_id->file_name mapping 
201
     */
Shane Snyder's avatar
Shane Snyder committed
202
    for(i = 0; i < n_infiles; i++)
203 204 205
    {
        memset(&in_job, 0, sizeof(struct darshan_job));

Shane Snyder's avatar
Shane Snyder committed
206
        in_fd = darshan_log_open(infile_list[i]);
207
        if(in_fd == NULL)
208 209
        {
            fprintf(stderr,
210
                "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
211
                infile_list[i]);
212 213 214 215
            return(-1);
        }

        /* read job-level metadata from the input file */
216 217
        ret = darshan_log_getjob(in_fd, &in_job);
        if(ret < 0)
218 219
        {
            fprintf(stderr,
220
                "Error: unable to read job data from input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
221
                infile_list[i]);
222
            darshan_log_close(in_fd);
223 224 225
            return(-1);
        }

226 227 228 229 230 231 232 233 234
        /* 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
235
                infile_list[i]);
236 237 238 239
            darshan_log_close(in_fd);
            return(-1);
        }

240 241
        if(i == 0)
        {
242
            /* get job data, exe, & mounts directly from the first input log */
Shane Snyder's avatar
Shane Snyder committed
243
            memcpy(&merge_job, &in_job, sizeof(struct darshan_job));
244

Shane Snyder's avatar
Shane Snyder committed
245
            ret = darshan_log_getexe(in_fd, merge_exe);
246
            if(ret < 0)
247 248
            {
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
249 250
                    "Error: unable to read exe string from input Darshan log file %s.\n",
                    infile_list[i]);
251
                darshan_log_close(in_fd);
252 253 254
                return(-1);
            }

Shane Snyder's avatar
Shane Snyder committed
255 256
            ret = darshan_log_getmounts(in_fd, &merge_mnt_pts,
                &merge_fs_types, &merge_mnt_count);
257
            if(ret < 0)
258
            {
259
                fprintf(stderr,
Shane Snyder's avatar
Shane Snyder committed
260 261
                    "Error: unable to read mount info from input Darshan log file %s.\n",
                    infile_list[i]);
262 263
                darshan_log_close(in_fd);
                return(-1);
264 265 266 267
            }
        }
        else
        {
268
            /* potentially update job timestamps using remaining logs */
Shane Snyder's avatar
Shane Snyder committed
269 270 271 272
            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;
273 274
        }

275
        /* read the hash of ids->names for the input log */
276 277 278 279 280
        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
281
                infile_list[i]);
282 283 284 285 286 287 288 289 290
            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
291
            HASH_FIND(hlink, merge_hash, &(ref->id), sizeof(darshan_record_id), found);
292 293
            if(!found)
            {
Shane Snyder's avatar
Shane Snyder committed
294
                HASH_ADD(hlink, merge_hash, id, sizeof(darshan_record_id), ref);
295
            }
296
            else if(strcmp(ref->name, found->name))
297 298 299 300 301 302 303 304 305
            {
                fprintf(stderr,
                    "Error: invalid Darshan record table entry.\n");
                darshan_log_close(in_fd);
                return(-1);
            }
        }

        darshan_log_close(in_fd);
306 307
    }

Shane Snyder's avatar
Shane Snyder committed
308 309 310
    /* create the output "merged" log */
    merge_fd = darshan_log_create(outlog_path, DARSHAN_ZLIB_COMP, 1);
    if(merge_fd == NULL)
311 312 313 314 315 316
    {
        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
317
    ret = darshan_log_putjob(merge_fd, &merge_job);
318 319 320
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write job data to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
321 322
        darshan_log_close(merge_fd);
        unlink(outlog_path);
323 324 325
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
326
    ret = darshan_log_putexe(merge_fd, merge_exe);
327 328 329
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write exe string to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
330 331
        darshan_log_close(merge_fd);
        unlink(outlog_path);
332 333 334
        return(-1);
    }

Shane Snyder's avatar
Shane Snyder committed
335
    ret = darshan_log_putmounts(merge_fd, merge_mnt_pts, merge_fs_types, merge_mnt_count);
336 337 338
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write mount 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);
    }

Shane Snyder's avatar
Shane Snyder committed
344 345
    /* write the merged table of records to output file */
    ret = darshan_log_puthash(merge_fd, merge_hash);
346 347 348
    if(ret < 0)
    {
        fprintf(stderr, "Error: unable to write record table to output darshan log.\n");
Shane Snyder's avatar
Shane Snyder committed
349 350
        darshan_log_close(merge_fd);
        unlink(outlog_path);
351 352
        return(-1);
    }
353

354
    /* iterate over active darshan modules and gather module data to write
Shane Snyder's avatar
Shane Snyder committed
355
     * to the merged output log
356
     */
Shane Snyder's avatar
Shane Snyder committed
357
    for(i = 0; i < DARSHAN_MAX_MODS; i++)
358 359 360
    {
        if(!mod_logutils[i]) continue;

361
        if(shared_redux)
362
        {
363
            /* build the hash of records shared globally by this module */
Shane Snyder's avatar
Shane Snyder committed
364 365
            ret = build_mod_shared_rec_hash(infile_list, n_infiles, i,
                merge_job.nprocs, mod_buf, &shared_rec_hash);
366
            if(ret < 0)
367
            {
368 369 370
                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
371 372
                darshan_log_close(merge_fd);
                unlink(outlog_path);
373
                return(-1);
374
            }
375 376 377

        }

Shane Snyder's avatar
Shane Snyder committed
378
        for(j = 0; j < n_infiles; j++)
379
        {
Shane Snyder's avatar
Shane Snyder committed
380
            in_fd = darshan_log_open(infile_list[j]);
381 382 383 384
            if(in_fd == NULL)
            {
                fprintf(stderr,
                    "Error: unable to open input Darshan log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
385 386 387
                    infile_list[j]);
                darshan_log_close(merge_fd);
                unlink(outlog_path);
388 389 390
                return(-1);
            }

Shane Snyder's avatar
Shane Snyder committed
391 392 393 394 395
            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
396
                    ret = mod_logutils[i]->log_put_record(merge_fd, sref->agg_rec, in_fd->mod_ver[i]);
Shane Snyder's avatar
Shane Snyder committed
397 398 399 400 401
                    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
402 403 404
                        darshan_log_close(in_fd);
                        darshan_log_close(merge_fd);
                        unlink(outlog_path);
Shane Snyder's avatar
Shane Snyder committed
405 406 407 408 409
                        return(-1);
                    }
                }
            }

410
            /* loop over module records and write them to output file */
411
            while((ret = mod_logutils[i]->log_get_record(in_fd, mod_buf)) == 1)
412
            {
413 414 415 416 417 418
                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
419
                ret = mod_logutils[i]->log_put_record(merge_fd, mod_buf, in_fd->mod_ver[i]);
420 421 422 423
                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
424
                        darshan_module_names[i], infile_list[j]);
425
                    darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
426 427
                    darshan_log_close(merge_fd);
                    unlink(outlog_path);
428 429 430
                    return(-1);
                }
            }
431 432 433
            if(ret < 0)
            {
                fprintf(stderr,
434
                    "Error: unable to read %s module record from input log file %s.\n",
Shane Snyder's avatar
Shane Snyder committed
435
                    darshan_module_names[i], infile_list[j]);
436
                darshan_log_close(in_fd);
Shane Snyder's avatar
Shane Snyder committed
437 438
                darshan_log_close(merge_fd);
                unlink(outlog_path);
439 440 441
                return(-1);
            }

442
            darshan_log_close(in_fd);
443
        }
444 445 446 447 448 449 450 451 452 453

        /* 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);
            }
        }
454 455
    }

Shane Snyder's avatar
Shane Snyder committed
456
    darshan_log_close(merge_fd);
457 458 459

    return(0);
}
Shane Snyder's avatar
Shane Snyder committed
460 461 462 463 464 465 466 467 468

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