jobflow.c

   1 /*
   2 Copyright (C) 2012,2014,2016 rofl0r
   3
   4 This program is free software: you can redistribute it and/or modify
   5 it under the terms of the GNU General Public License as published by
   6 the Free Software Foundation, either version 3 of the License, or
   7 (at your option) any later version.
   8
   9 This program is distributed in the hope that it will be useful,
  10 but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 GNU General Public License for more details.
  13
  14 You should have received a copy of the GNU General Public License
  15 along with this program.  If not, see <http://www.gnu.org/licenses/>.
  16 */
  17
  18
  19 #define VERSION "1.1.0"
  20
  21
  22 #undef _POSIX_C_SOURCE
  23 #define _POSIX_C_SOURCE 200809L
  24 #undef _XOPEN_SOURCE
  25 #define _XOPEN_SOURCE 700
  26 #undef _GNU_SOURCE
  27 #define _GNU_SOURCE
  28
  29 #include "../lib/include/optparser.h"
  30 #include "../lib/include/stringptr.h"
  31 #include "../lib/include/stringptrlist.h"
  32 #include "../lib/include/sblist.h"
  33 #include "../lib/include/strlib.h"
  34 #include "../lib/include/timelib.h"
  35 #include "../lib/include/filelib.h"
  36
  37 #include <stdio.h>
  38 #include <stdlib.h>
  39 #include <unistd.h>
  40 #include <stdint.h>
  41 #include <stddef.h>
  42 #include <errno.h>
  43 #include <time.h>
  44 #include <assert.h>
  45
  46 /* process handling */
  47
  48 #include <fcntl.h>
  49 #include <spawn.h>
  50 #include <sys/wait.h>
  51 #include <sys/stat.h>
  52
  53 #include <sys/resource.h>
  54
  55 #if defined(__GLIBC__) && (__GLIBC__ < 3) && (__GLIBC_MINOR__ < 13)
  56 /* http://repo.or.cz/w/glibc.git/commitdiff/c08fb0d7bba4015078406b28d3906ccc5fda9d5a ,
  57  * http://repo.or.cz/w/glibc.git/commitdiff/052fa7b33ef5deeb4987e5264cf397b3161d8a01 */
  58 #warning to use prlimit() you have to use musl libc 0.8.4+ or glibc 2.13+
  59 static int prlimit(int pid, ...) {
  60         (void) pid;
  61         dprintf(2, "prlimit() not implemented on this system\n");
  62         errno = EINVAL;
  63         return -1;
  64 }
  65 #endif
  66
  67
  68 #include <sys/time.h>
  69
  70 typedef struct {
  71         pid_t pid;
  72         posix_spawn_file_actions_t fa;
  73 } job_info;
  74
  75 typedef struct {
  76         int limit;
  77         struct rlimit rl;
  78 } limit_rec;
  79
  80 typedef struct {
  81         int numthreads;
  82         unsigned threads_running;
  83         char* statefile;
  84         unsigned skip;
  85         sblist* job_infos;
  86         sblist* subst_entries;
  87         sblist* limits;
  88         unsigned cmd_startarg;
  89         char* tempdir;
  90         int delayedspinup_interval; /* use a random delay until the queue gets filled for the first time.
  91                                 the top value in ms can be supplied via a command line switch.
  92                                 this option makes only sense if the interval is somewhat smaller than the
  93                                 expected runtime of the average job.
  94                                 this option is useful to not overload a network app due to hundreds of
  95                                 parallel connection tries on startup.
  96                                 */
  97         int buffered:1; /* write stdout and stderr of each task into a file,
  98                         and print it to stdout once the process ends.
  99                         this prevents mixing up of the output of multiple tasks. */
 100         int delayedflush:1; /* only write to statefile whenever all processes are busy, and at program end.
 101                            this means faster program execution, but could also be imprecise if the number of
 102                            jobs is small or smaller than the available threadcount. */
 103         int join_output:1; /* join stdout and stderr of launched jobs into stdout */
 104 } prog_state_s;
 105
 106 prog_state_s prog_state;
 107
 108
 109 extern char** environ;
 110
 111 int makeLogfilename(char* buf, size_t bufsize, size_t jobindex, int is_stderr) {
 112         int ret = snprintf(buf, bufsize,
 113                            is_stderr ? "%s/jd_proc_%.5u_stdout.log" : "%s/jd_proc_%.5u_stderr.log",
 114                            prog_state.tempdir, (unsigned) jobindex);
 115         return ret > 0 && (size_t) ret < bufsize;
 116 }
 117
 118 void launch_job(size_t jobindex, char** argv) {
 119         char stdout_filename_buf[256];
 120         char stderr_filename_buf[256];
 121         job_info* job = sblist_get(prog_state.job_infos, jobindex);
 122
 123         if(job->pid != -1) return;
 124
 125         if(prog_state.buffered) {
 126                 if((!makeLogfilename(stdout_filename_buf, sizeof(stdout_filename_buf), jobindex, 0)) ||
 127                    ((!prog_state.join_output) && !makeLogfilename(stderr_filename_buf, sizeof(stderr_filename_buf), jobindex, 1)) ) {
 128                         dprintf(2, "temp filename too long!\n");
 129                         return;
 130                 }
 131         }
 132
 133         errno = posix_spawn_file_actions_init(&job->fa);
 134         if(errno) goto spawn_error;
 135         errno = posix_spawn_file_actions_addclose(&job->fa, 0);
 136         if(errno) goto spawn_error;
 137
 138         if(prog_state.buffered) {
 139                 errno = posix_spawn_file_actions_addclose(&job->fa, 1);
 140                 if(errno) goto spawn_error;
 141                 errno = posix_spawn_file_actions_addclose(&job->fa, 2);
 142                 if(errno) goto spawn_error;
 143         }
 144
 145         errno = posix_spawn_file_actions_addopen(&job->fa, 0, "/dev/null", O_RDONLY, 0);
 146         if(errno) goto spawn_error;
 147
 148         if(prog_state.buffered) {
 149                 errno = posix_spawn_file_actions_addopen(&job->fa, 1, stdout_filename_buf, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
 150                 if(errno) goto spawn_error;
 151                 if(prog_state.join_output)
 152                         errno = posix_spawn_file_actions_adddup2(&job->fa, 1, 2);
 153                 else
 154                         errno = posix_spawn_file_actions_addopen(&job->fa, 2, stderr_filename_buf, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
 155                 if(errno) goto spawn_error;
 156         }
 157
 158         errno = posix_spawnp(&job->pid, argv[0], &job->fa, NULL, argv, environ);
 159         if(errno) {
 160                 spawn_error:
 161                 job->pid = -1;
 162                 perror("posix_spawn");
 163         } else {
 164                 prog_state.threads_running++;
 165                 if(prog_state.limits) {
 166                         limit_rec* limit;
 167                         sblist_iter(prog_state.limits, limit) {
 168                                 if(prlimit(job->pid, limit->limit, &limit->rl, NULL) == -1)
 169                                         perror("prlimit");
 170                         }
 171                 }
 172         }
 173 }
 174
 175 static void dump_output(size_t job_id, int is_stderr) {
 176         char out_filename_buf[256];
 177         char buf[4096];
 178         FILE* dst, *out_stream = is_stderr ? stderr : stdout;
 179         size_t nread;
 180
 181         makeLogfilename(out_filename_buf, sizeof(out_filename_buf), job_id, is_stderr);
 182
 183         dst = fopen(out_filename_buf, "r");
 184         if(dst) {
 185                 while((nread = fread(buf, 1, sizeof(buf), dst))) {
 186                         fwrite(buf, 1, nread, out_stream);
 187                         if(nread < sizeof(buf)) break;
 188                 }
 189                 fclose(dst);
 190                 fflush(out_stream);
 191         }
 192 }
 193
 194 /* wait till a child exits, reap it, and return its job index for slot reuse */
 195 static size_t reap_child(void) {
 196         size_t i;
 197         job_info* job;
 198         int ret, retval;
 199
 200         do ret = waitpid(-1, &retval, 0);
 201         while(ret == -1 || !WIFEXITED(retval));
 202
 203         for(i = 0; i < sblist_getsize(prog_state.job_infos); i++) {
 204                 job = sblist_get(prog_state.job_infos, i);
 205                 if(job->pid == ret) {
 206                         job->pid = -1;
 207                         posix_spawn_file_actions_destroy(&job->fa);
 208                         prog_state.threads_running--;
 209                         if(prog_state.buffered) {
 210                                 dump_output(i, 0);
 211                                 if(!prog_state.join_output)
 212                                         dump_output(i, 1);
 213                         }
 214                         return i;
 215                 }
 216         }
 217         assert(0);
 218         return -1;
 219 }
 220
 221 static size_t free_slots(void) {
 222         return prog_state.numthreads - prog_state.threads_running;
 223 }
 224
 225 static void add_job(char **argv) {
 226         if(free_slots())
 227                 launch_job(prog_state.threads_running, argv);
 228         else
 229                 launch_job(reap_child(), argv);
 230 }
 231
 232 __attribute__((noreturn))
 233 static void die(const char* msg) {
 234         dprintf(2, msg);
 235         exit(1);
 236 }
 237
 238 static unsigned long parse_human_number(stringptr* num) {
 239         unsigned long ret = 0;
 240         static const unsigned long mul[] = {1024, 1024 * 1024, 1024 * 1024 * 1024};
 241         const char* kmg = "KMG";
 242         char* kmgind;
 243         if(num && num->size) {
 244                 ret = atol(num->ptr);
 245                 if((kmgind = strchr(kmg, num->ptr[num->size -1])))
 246                         ret *= mul[kmgind - kmg];
 247         }
 248         return ret;
 249 }
 250
 251 static int syntax(void) {
 252         dprintf(2,
 253                 "jobflow " VERSION " (C) rofl0r\n"
 254                 "------------------\n"
 255                 "this program is intended to be used as a recipient of another programs output\n"
 256                 "it launches processes to which the current line can be passed as an argument\n"
 257                 "using {} for substitution (as in find -exec).\n"
 258                 "\n"
 259                 "available options:\n\n"
 260                 "-skip=XXX -threads=XXX -resume -statefile=/tmp/state -delayedflush\n"
 261                 "-delayedspinup=XXX -buffered -joinoutput -limits=mem=16M,cpu=10\n"
 262                 "-exec ./mycommand {}\n"
 263                 "\n"
 264                 "-skip=XXX\n"
 265                 "    XXX=number of entries to skip\n"
 266                 "-threads=XXX\n"
 267                 "    XXX=number of parallel processes to spawn]\n"
 268                 "-resume\n"
 269                 "    resume from last jobnumber stored in statefile\n"
 270                 "-statefile=XXX\n"
 271                 "    XXX=filename\n"
 272                 "    saves last launched jobnumber into a file\n"
 273                 "-delayedflush\n"
 274                 "    only write to statefile whenever all processes are busy,\n"
 275                 "    and at program end\n"
 276                 "-delayedspinup=XXX\n"
 277                 "    XXX=maximum amount of milliseconds\n"
 278                 "    ...to wait when spinning up a fresh set of processes\n"
 279                 "    a random value between 0 and the chosen amount is used to delay initial\n"
 280                 "    spinup.\n"
 281                 "    this can be handy to circumvent an I/O lockdown because of a burst of \n"
 282                 "    activity on program startup\n"
 283                 "-buffered\n"
 284                 "    store the stdout and stderr of launched processes into a temporary file\n"
 285                 "    which will be printed after a process has finished.\n"
 286                 "    this prevents mixing up of output of different processes.\n"
 287                 "-joinoutput\n"
 288                 "    if -buffered, write both stdout and stderr into the same file.\n"
 289                 "    this saves the chronological order of the output, and the combined output\n"
 290                 "    will only be printed to stdout.\n"
 291                 "-limits=[mem=XXX,cpu=XXX,stack=XXX,fsize=XXX,nofiles=XXX]\n"
 292                 "    sets the rlimit of the new created processes.\n"
 293                 "    see \"man setrlimit\" for an explanation. the suffixes G/M/K are detected.\n"
 294                 "-exec command with args\n"
 295                 "    everything past -exec is treated as the command to execute on each line of\n"
 296                 "    stdin received. the line can be passed as an argument using {}.\n"
 297                 "    {.} passes everything before the last dot in a line as an argument.\n"
 298                 "    it is possible to use multiple substitutions inside a single argument,\n"
 299                 "    but currently only of one type.\n"
 300                 "\n"
 301         );
 302         return 1;
 303 }
 304
 305 static int parse_args(int argc, char** argv) {
 306         op_state op_b, *op = &op_b;
 307         op_init(op, argc, argv);
 308         char *op_temp;
 309         if(argc == 1 || op_hasflag(op, SPL("-help")))
 310                 return syntax();
 311         op_temp = op_get(op, SPL("threads"));
 312         prog_state.numthreads = op_temp ? atoi(op_temp) : 1;
 313         if(prog_state.numthreads <= 0) die("threadcount must be >= 1\n");
 314         op_temp = op_get(op, SPL("statefile"));
 315         prog_state.statefile = op_temp;
 316
 317         op_temp = op_get(op, SPL("skip"));
 318         prog_state.skip = op_temp ? atoi(op_temp) : 0;
 319         if(op_hasflag(op, SPL("resume"))) {
 320                 if(!prog_state.statefile) die("-resume needs -statefile\n");
 321                 if(access(prog_state.statefile, W_OK | R_OK) != -1) {
 322                         stringptr* fc = stringptr_fromfile(prog_state.statefile);
 323                         prog_state.skip = atoi(fc->ptr);
 324                 }
 325         }
 326
 327         prog_state.delayedflush = 0;
 328         if(op_hasflag(op, SPL("delayedflush"))) {
 329                 if(!prog_state.statefile) die("-delayedflush needs -statefile\n");
 330                 prog_state.delayedflush = 1;
 331         }
 332
 333         op_temp = op_get(op, SPL("delayedspinup"));
 334         prog_state.delayedspinup_interval = op_temp ? atoi(op_temp) : 0;
 335
 336         prog_state.cmd_startarg = 0;
 337         prog_state.subst_entries = NULL;
 338
 339         if(op_hasflag(op, SPL("exec"))) {
 340                 uint32_t subst_ent;
 341                 unsigned i, r = 0;
 342                 for(i = 1; i < (unsigned) argc; i++) {
 343                         if(str_equal(argv[i], "-exec")) {
 344                                 r = i + 1;
 345                                 break;
 346                         }
 347                 }
 348                 if(r && r < (unsigned) argc) {
 349                         prog_state.cmd_startarg = r;
 350                 }
 351
 352                 // save entries which must be substituted, to save some cycles.
 353                 prog_state.subst_entries = sblist_new(sizeof(uint32_t), 16);
 354                 for(i = r; i < (unsigned) argc; i++) {
 355                         subst_ent = i - r;
 356                         if(strstr(argv[i], "{}") || strstr(argv[i], "{.}")) sblist_add(prog_state.subst_entries, &subst_ent);
 357                 }
 358         }
 359
 360         prog_state.buffered = 0;
 361         if(op_hasflag(op, SPL("buffered"))) {
 362                 prog_state.buffered = 1;
 363         }
 364
 365         prog_state.join_output = 0;
 366         if(op_hasflag(op, SPL("joinoutput"))) {
 367                 if(!prog_state.buffered) die("-joinoutput needs -buffered\n");
 368                 prog_state.join_output = 1;
 369         }
 370
 371         prog_state.limits = NULL;
 372         op_temp = op_get(op, SPL("limits"));
 373         if(op_temp) {
 374                 unsigned i;
 375                 SPDECLAREC(limits, op_temp);
 376                 stringptrlist* limit_list = stringptr_splitc(limits, ',');
 377                 stringptrlist* kv;
 378                 stringptr* key, *value;
 379                 limit_rec lim;
 380                 if(stringptrlist_getsize(limit_list)) {
 381                         prog_state.limits = sblist_new(sizeof(limit_rec), stringptrlist_getsize(limit_list));
 382                         for(i = 0; i < stringptrlist_getsize(limit_list); i++) {
 383                                 kv = stringptr_splitc(stringptrlist_get(limit_list, i), '=');
 384                                 if(stringptrlist_getsize(kv) != 2) continue;
 385                                 key = stringptrlist_get(kv, 0);
 386                                 value = stringptrlist_get(kv, 1);
 387                                 if(EQ(key, SPL("mem")))
 388                                         lim.limit = RLIMIT_AS;
 389                                 else if(EQ(key, SPL("cpu")))
 390                                         lim.limit = RLIMIT_CPU;
 391                                 else if(EQ(key, SPL("stack")))
 392                                         lim.limit = RLIMIT_STACK;
 393                                 else if(EQ(key, SPL("fsize")))
 394                                         lim.limit = RLIMIT_FSIZE;
 395                                 else if(EQ(key, SPL("nofiles")))
 396                                         lim.limit = RLIMIT_NOFILE;
 397                                 else
 398                                         die("unknown option passed to -limits");
 399
 400                                 if(getrlimit(lim.limit, &lim.rl) == -1) {
 401                                         perror("getrlimit");
 402                                         die("could not query rlimits");
 403                                 }
 404                                 lim.rl.rlim_cur = parse_human_number(value);
 405                                 sblist_add(prog_state.limits, &lim);
 406                                 stringptrlist_free(kv);
 407                         }
 408                         stringptrlist_free(limit_list);
 409                 }
 410         }
 411         return 0;
 412 }
 413
 414 static void init_queue(void) {
 415         int i;
 416         job_info ji = {.pid = -1};
 417
 418         for(i = 0; i < prog_state.numthreads; i++)
 419                 sblist_add(prog_state.job_infos, &ji);
 420 }
 421
 422 static void write_statefile(uint64_t n, const char* tempfile) {
 423         char numbuf[64];
 424         stringptr num_b, *num = &num_b;
 425
 426         num_b.ptr = uint64ToString(n + 1, numbuf);
 427         num_b.size = strlen(numbuf);
 428         stringptr_tofile((char*) tempfile, num);
 429         if(rename(tempfile, prog_state.statefile) == -1)
 430                 perror("rename");
 431 }
 432
 433 // returns numbers of substitutions done, -1 on out of buffer.
 434 // dest is always overwritten. if not substitutions were done, it contains a copy of source.
 435 int substitute_all(char* dest, ssize_t dest_size, stringptr* source, stringptr* what, stringptr* whit) {
 436         size_t i;
 437         int ret = 0;
 438         for(i = 0; dest_size > 0 && i < source->size; ) {
 439                 if(stringptr_here(source, i, what)) {
 440                         if(dest_size < (ssize_t) whit->size) return -1;
 441                         memcpy(dest, whit->ptr, whit->size);
 442                         dest += whit->size;
 443                         dest_size -= whit->size;
 444                         ret++;
 445                         i += what->size;
 446                 } else {
 447                         *dest = source->ptr[i];
 448                         dest++;
 449                         dest_size--;
 450                         i++;
 451                 }
 452         }
 453         if(!dest_size) return -1;
 454         *dest = 0;
 455         return ret;
 456 }
 457
 458 int main(int argc, char** argv) {
 459         char inbuf[4096]; char* fgets_result;
 460         stringptr line_b, *line = &line_b;
 461         char* cmd_argv[4096];
 462         char subst_buf[16][4096];
 463         unsigned max_subst;
 464
 465         uint64_t lineno = 0;
 466         unsigned i;
 467         unsigned spinup_counter = 0;
 468
 469         char tempdir_buf[256];
 470         char temp_state[256];
 471
 472         srand(time(NULL));
 473
 474         if(argc > 4096) argc = 4096;
 475
 476         prog_state.threads_running = 0;
 477
 478         if(parse_args(argc, argv)) return 1;
 479
 480         if(prog_state.statefile)
 481                 snprintf(temp_state, sizeof(temp_state), "%s.%u", prog_state.statefile, (unsigned) getpid());
 482
 483         prog_state.tempdir = NULL;
 484
 485         if(prog_state.buffered) {
 486                 prog_state.tempdir = tempdir_buf;
 487                 if(mktempdir("jobflow", tempdir_buf, sizeof(tempdir_buf)) == 0) {
 488                         perror("mkdtemp");
 489                         die("could not create tempdir\n");
 490                 }
 491         } else {
 492                 /* if the stdout/stderr fds are not in O_APPEND mode,
 493                    the dup()'s of the fds in posix_spawn can cause different
 494                    file positions, causing the different processes to overwrite each others output.
 495                    testcase:
 496                    seq 100 | ./jobflow.out -threads=100 -exec echo {} > test.tmp ; wc -l test.tmp
 497                 */
 498                 if(fcntl(1, F_SETFL, O_APPEND) == -1) perror("fcntl");
 499                 if(fcntl(2, F_SETFL, O_APPEND) == -1) perror("fcntl");
 500         }
 501
 502         if(prog_state.cmd_startarg) {
 503                 for(i = prog_state.cmd_startarg; i < (unsigned) argc; i++) {
 504                         cmd_argv[i - prog_state.cmd_startarg] = argv[i];
 505                 }
 506                 cmd_argv[argc - prog_state.cmd_startarg] = NULL;
 507         }
 508
 509         prog_state.job_infos = sblist_new(sizeof(job_info), prog_state.numthreads);
 510         init_queue();
 511
 512         for(;(fgets_result = fgets(inbuf, sizeof(inbuf), stdin));lineno++) {
 513                 if(prog_state.skip) {
 514                         prog_state.skip--;
 515                         continue;
 516                 }
 517                 if(!prog_state.cmd_startarg) {
 518                         dprintf(1, fgets_result);
 519                         continue;
 520                 }
 521                 stringptr_fromchar(fgets_result, line);
 522                 stringptr_chomp(line);
 523
 524                 max_subst = 0;
 525                 if(prog_state.subst_entries) {
 526                         uint32_t* index;
 527                         sblist_iter(prog_state.subst_entries, index) {
 528                                 SPDECLAREC(source, argv[*index + prog_state.cmd_startarg]);
 529                                 int ret;
 530                                 ret = substitute_all(subst_buf[max_subst], 4096, source, SPL("{}"), line);
 531                                 if(ret == -1) {
 532                                         too_long:
 533                                         dprintf(2, "fatal: line too long for substitution: %s\n", line->ptr);
 534                                         goto out;
 535                                 } else if(!ret) {
 536                                         char* lastdot = stringptr_rchr(line, '.');
 537                                         stringptr tilLastDot = *line;
 538                                         if(lastdot) tilLastDot.size = lastdot - line->ptr;
 539                                         ret = substitute_all(subst_buf[max_subst], 4096, source, SPL("{.}"), &tilLastDot);
 540                                         if(ret == -1) goto too_long;
 541                                 }
 542                                 if(ret) {
 543                                         cmd_argv[*index] = subst_buf[max_subst];
 544                                         max_subst++;
 545                                 }
 546                         }
 547                 }
 548
 549
 550                 if(prog_state.delayedspinup_interval && spinup_counter < (prog_state.numthreads * 2)) {
 551                         msleep(rand() % (prog_state.delayedspinup_interval + 1));
 552                         spinup_counter++;
 553                 }
 554
 555                 add_job(cmd_argv);
 556
 557                 if(prog_state.statefile && (prog_state.delayedflush == 0 || free_slots() == 0)) {
 558                         write_statefile(lineno, temp_state);
 559                 }
 560         }
 561
 562         out:
 563
 564         if(prog_state.delayedflush)
 565                 write_statefile(lineno - 1, temp_state);
 566
 567         while(prog_state.threads_running) reap_child();
 568
 569         if(prog_state.subst_entries) sblist_free(prog_state.subst_entries);
 570         if(prog_state.job_infos) sblist_free(prog_state.job_infos);
 571         if(prog_state.limits) sblist_free(prog_state.limits);
 572
 573         if(prog_state.tempdir)
 574                 rmdir(prog_state.tempdir);
 575
 576
 577         fflush(stdout);
 578         fflush(stderr);
 579
 580
 581         return 0;
 582 }