lustre/tests/writemany.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #include <stdlib.h>
  38 #include <stdio.h>
  39 #include <unistd.h>
  40 #include <sys/types.h>
  41 #include <sys/stat.h>
  42 #include <signal.h>
  43 #include <fcntl.h>
  44 #include <errno.h>
  45 #include <string.h>
  46 #include <sys/wait.h>
  47 #include <time.h>
  48 #include <sys/time.h>
  49
  50
  51 #define difftime(a, b)                                          \
  52         ((double)(a)->tv_sec - (b)->tv_sec +                    \
  53          ((double)((a)->tv_usec - (b)->tv_usec) / 1000000))
  54
  55
  56
  57 char cmdname[512];
  58 int o_abort = 0;
  59 int o_quiet = 0;
  60
  61 void usage(char *name)
  62 {
  63         fprintf(stderr, "usage: %s [opts] <dirname> <seconds> <threads>\n",
  64                 name);
  65         fprintf(stderr, "  -q quiet\n");
  66         fprintf(stderr, "  -a abort other children on first err\n");
  67         exit(1);
  68 }
  69
  70
  71 struct kid_list_t {
  72         pid_t kid;
  73         struct kid_list_t *next;
  74 };
  75
  76 struct kid_list_t *head = NULL;
  77
  78 int push_kid(pid_t kid)
  79 {
  80         struct kid_list_t *new;
  81         new = (struct kid_list_t *)malloc(sizeof(struct kid_list_t));
  82         if (new == NULL)
  83                 return 1;
  84
  85         new->kid = kid;
  86         new->next = head;
  87         head = new;
  88         return 0;
  89 }
  90
  91 void kill_kids(void)
  92 {
  93         while(head) {
  94                 kill(head->kid, SIGTERM);
  95                 head = head->next;
  96         }
  97 }
  98
  99 static int usr1_received;
 100 void usr1_handler(int unused)
 101 {
 102         usr1_received = 1;
 103         kill_kids();
 104 }
 105
 106 int wait_for_threads(int live_threads)
 107 {
 108         int rc = 0;
 109
 110         while (live_threads > 0) {
 111                 int status;
 112                 pid_t ret;
 113
 114                 ret = waitpid(0, &status, 0);
 115                 if (ret == 0) {
 116                         continue;
 117                 }
 118
 119                 if (ret < 0) {
 120                         fprintf(stderr, "%s: error: wait - %s\n",
 121                                 cmdname, strerror(errno));
 122                         if (!rc)
 123                                 rc = errno;
 124                 } else {
 125                         /*
 126                          * This is a hack.  We _should_ be able to use
 127                          * WIFEXITED(status) to see if there was an
 128                          * error, but it appears to be broken and it
 129                          * always returns 1 (OK).  See wait(2).
 130                          */
 131                         int err = WEXITSTATUS(status);
 132                         if (err)
 133                                 fprintf(stderr,
 134                                         "%s: error: PID %d had rc=%d\n",
 135                                         cmdname, ret, err);
 136                         /* Record first error */
 137                         if (!rc)
 138                                 rc = err;
 139
 140                         /* Give up on first error */
 141                         if (rc && o_abort) {
 142                                 kill_kids();
 143                                 break;
 144                         }
 145
 146                         live_threads--;
 147                 }
 148         }
 149         if (!o_quiet)
 150                 printf("%s done, rc = %d\n", cmdname, rc);
 151         return rc;
 152 }
 153
 154 void print_err(char *op, char *filename, struct timeval *time, int err)
 155 {
 156         fprintf(stderr, "%s: %d.%.06d error: %s(%s): %s\n",
 157                 cmdname, (int)(time->tv_sec), (int)(time->tv_usec), op,
 158                 filename, strerror(errno));
 159 }
 160
 161 int run_one_child(char *file, int thread, int seconds)
 162 {
 163         struct timeval start, cur;
 164         double diff;
 165         char filename[1024];
 166         char buf[1024];
 167         int fd, rc = 0, rand, maxrand, len;
 168         long nfiles = 0, nbytes = 0;
 169
 170         if (!o_quiet)
 171                 printf("%s: running thread #%d\n", cmdname, thread);
 172
 173         srandom(thread);
 174         /* Higher thread numbers will produce bigger random files.
 175            Thread 1 will produce only 0-len files. */
 176         maxrand = 1; rand = thread;
 177         while (--rand)
 178                 maxrand *= 10;
 179
 180         gettimeofday(&start, NULL);
 181
 182         while(!rc) {
 183                 if (usr1_received)
 184                         break;
 185
 186                 gettimeofday(&cur, NULL);
 187                 if (seconds) {
 188                         if (cur.tv_sec > (start.tv_sec + seconds))
 189                                 break;
 190                 }
 191
 192                 sprintf(filename, "%s-%d-%ld", file, thread, nfiles);
 193
 194                 fd = open(filename, O_RDWR | O_CREAT, 0666);
 195                 if (fd < 0) {
 196                         print_err("open", filename, &cur, errno);
 197                         rc = errno;
 198                         break;
 199                 }
 200
 201                 sprintf(buf, "%s %010ld %.19s.%012d\n", cmdname,
 202                         nfiles++, ctime(&cur.tv_sec), (int)cur.tv_usec);
 203                 len = strlen(buf);
 204
 205                 rand = random() % maxrand;
 206                 while (rand-- > 0) {
 207                         if (write(fd, buf, len) != len) {
 208                                 print_err("write", filename, &cur, errno);
 209                                 rc = errno;
 210                                 break;
 211                         }
 212                         nbytes += len;
 213                 }
 214
 215                 if (close(fd) < 0) {
 216                         print_err("close", filename, &cur, errno);
 217                         rc = errno;
 218                         break;
 219                 }
 220                 if (unlink(filename) < 0) {
 221                         print_err("unlink", filename, &cur, errno);
 222                         if (errno == ENOENT) {
 223                                 printf("Ignoring known bug 6082\n");
 224                         } else {
 225                                 rc = errno;
 226                                 break;
 227                         }
 228                 }
 229         }
 230
 231         diff = difftime(&cur, &start);
 232         if (!o_quiet)
 233                 printf("%s: %7ld files, %4ld MB in %.2fs (%7.2f files/s, "
 234                        "%5.2f MB/s): rc = %d\n",
 235                        cmdname, nfiles, nbytes >> 20, diff,
 236                        (double)nfiles / diff, (double)nbytes/1024/1024 / diff,
 237                        rc);
 238
 239         return rc;
 240 }
 241
 242 int main(int argc, char *argv[])
 243 {
 244         unsigned long duration;
 245         int threads = 0;
 246         char *end;
 247         char *directory;
 248         int i = 1, rc = 0;
 249
 250         sprintf(cmdname, "%s", argv[0]);
 251
 252         while((i < argc) && (argv[i][0] == '-')) {
 253                 switch (argv[i][1]) {
 254                 case 'q':
 255                         o_quiet++;
 256                         break;
 257                 case 'a':
 258                         o_abort++;
 259                         break;
 260                 }
 261                 i++;
 262         }
 263
 264         if ((argc - i) < 3)
 265                 usage(argv[0]);
 266
 267         directory = argv[i];
 268         duration = strtoul(argv[++i], &end, 0);
 269         if (*end) {
 270                 fprintf(stderr, "%s: error: bad number of seconds '%s'\n",
 271                         cmdname, argv[i]);
 272                 exit(2);
 273         }
 274
 275         threads = strtoul(argv[++i], &end, 0);
 276         if (*end) {
 277                 fprintf(stderr, "%s: error: bad thread count '%s'\n",
 278                         cmdname, argv[i]);
 279                 exit(2);
 280         }
 281
 282         signal(SIGUSR1, usr1_handler);
 283
 284         for (i = 1; i <= threads; i++) {
 285                 rc = fork();
 286                 if (rc < 0) {
 287                         if (!o_quiet)
 288                                 fprintf(stderr, "%s: error: #%d - %s\n",
 289                                         cmdname, i, strerror(rc = errno));
 290                         return (rc);
 291                 }
 292                 if (rc == 0) {
 293                         /* children */
 294                         sprintf(cmdname, "%s-%d", argv[0], i);
 295                         return (run_one_child(directory, i, duration));
 296                 } else {
 297                         /* parent */
 298                         rc = push_kid(rc);
 299                         if (rc != 0) {
 300                                 kill_kids();
 301                                 exit(3);
 302                         }
 303                 }
 304         }
 305         /* parent process */
 306         if (!o_quiet)
 307                 printf("%s will run for %ld minutes\n", cmdname, duration/60);
 308         return (wait_for_threads(threads));
 309 }