src/apps_plugin.c

   1 #ifdef HAVE_CONFIG_H
   2 #include <config.h>
   3 #endif
   4 #include <stdio.h>
   5 #include <stdlib.h>
   6 #include <string.h>
   7 #include <time.h>
   8 #include <unistd.h>
   9 #include <sys/types.h>
  10 #include <sys/time.h>
  11 #include <sys/wait.h>
  12 #include <sys/stat.h>
  13
  14 #include <sys/resource.h>
  15 #include <sys/stat.h>
  16
  17 #include <errno.h>
  18 #include <stdarg.h>
  19 #include <locale.h>
  20 #include <ctype.h>
  21 #include <fcntl.h>
  22
  23 #include <malloc.h>
  24 #include <dirent.h>
  25 #include <arpa/inet.h>
  26
  27 #include <sys/types.h>
  28 #include <pwd.h>
  29 #include <grp.h>
  30
  31 #include "avl.h"
  32
  33 #include "common.h"
  34 #include "log.h"
  35 #include "procfile.h"
  36 #include "../config.h"
  37 #include "web_buffer.h"
  38
  39 #ifdef NETDATA_INTERNAL_CHECKS
  40 #include <sys/prctl.h>
  41 #endif
  42
  43 #define MAX_COMPARE_NAME 100
  44 #define MAX_NAME 100
  45 #define MAX_CMDLINE 1024
  46
  47 // the rates we are going to send to netdata
  48 // will have this detail
  49 // a value of:
  50 // 1 will send just integer parts to netdata
  51 // 100 will send 2 decimal points
  52 // 1000 will send 3 decimal points
  53 // etc.
  54 #define RATES_DETAIL 10000ULL
  55
  56 int processors = 1;
  57 pid_t pid_max = 32768;
  58 int debug = 0;
  59
  60 int update_every = 1;
  61 unsigned long long global_iterations_counter = 1;
  62 unsigned long long file_counter = 0;
  63 int proc_pid_cmdline_is_needed = 0;
  64 int include_exited_childs = 1;
  65 char *host_prefix = "";
  66 char *config_dir = CONFIG_DIR;
  67
  68 pid_t *all_pids_sortlist = NULL;
  69
  70 // will be automatically set to 1, if guest values are collected
  71 int show_guest_time = 0;
  72 int show_guest_time_old = 0;
  73
  74 // ----------------------------------------------------------------------------
  75
  76 void netdata_cleanup_and_exit(int ret) {
  77         exit(ret);
  78 }
  79
  80
  81 // ----------------------------------------------------------------------------
  82 // system functions
  83 // to retrieve settings of the system
  84
  85 long get_system_cpus(void) {
  86         procfile *ff = NULL;
  87
  88         int processors = 0;
  89
  90         char filename[FILENAME_MAX + 1];
  91         snprintfz(filename, FILENAME_MAX, "%s/proc/stat", host_prefix);
  92
  93         ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);
  94         if(!ff) return 1;
  95
  96         ff = procfile_readall(ff);
  97         if(!ff) {
  98                 procfile_close(ff);
  99                 return 1;
 100         }
 101
 102         unsigned int i;
 103         for(i = 0; i < procfile_lines(ff); i++) {
 104                 if(!procfile_linewords(ff, i)) continue;
 105
 106                 if(strncmp(procfile_lineword(ff, i, 0), "cpu", 3) == 0) processors++;
 107         }
 108         processors--;
 109         if(processors < 1) processors = 1;
 110
 111         procfile_close(ff);
 112         return processors;
 113 }
 114
 115 pid_t get_system_pid_max(void) {
 116         procfile *ff = NULL;
 117         pid_t mpid = 32768;
 118
 119         char filename[FILENAME_MAX + 1];
 120         snprintfz(filename, FILENAME_MAX, "%s/proc/sys/kernel/pid_max", host_prefix);
 121         ff = procfile_open(filename, NULL, PROCFILE_FLAG_DEFAULT);
 122         if(!ff) return mpid;
 123
 124         ff = procfile_readall(ff);
 125         if(!ff) {
 126                 procfile_close(ff);
 127                 return mpid;
 128         }
 129
 130         mpid = (pid_t)atoi(procfile_lineword(ff, 0, 0));
 131         if(!mpid) mpid = 32768;
 132
 133         procfile_close(ff);
 134         return mpid;
 135 }
 136
 137 // ----------------------------------------------------------------------------
 138 // target
 139 // target is the structure that process data are aggregated
 140
 141 struct target {
 142         char compare[MAX_COMPARE_NAME + 1];
 143         uint32_t comparehash;
 144         size_t comparelen;
 145
 146         char id[MAX_NAME + 1];
 147         uint32_t idhash;
 148
 149         char name[MAX_NAME + 1];
 150
 151         uid_t uid;
 152         gid_t gid;
 153
 154         unsigned long long minflt;
 155         unsigned long long cminflt;
 156         unsigned long long majflt;
 157         unsigned long long cmajflt;
 158         unsigned long long utime;
 159         unsigned long long stime;
 160         unsigned long long gtime;
 161         unsigned long long cutime;
 162         unsigned long long cstime;
 163         unsigned long long cgtime;
 164         unsigned long long num_threads;
 165         unsigned long long rss;
 166
 167         unsigned long long statm_size;
 168         unsigned long long statm_resident;
 169         unsigned long long statm_share;
 170         unsigned long long statm_text;
 171         unsigned long long statm_lib;
 172         unsigned long long statm_data;
 173         unsigned long long statm_dirty;
 174
 175         unsigned long long io_logical_bytes_read;
 176         unsigned long long io_logical_bytes_written;
 177         unsigned long long io_read_calls;
 178         unsigned long long io_write_calls;
 179         unsigned long long io_storage_bytes_read;
 180         unsigned long long io_storage_bytes_written;
 181         unsigned long long io_cancelled_write_bytes;
 182
 183         int *fds;
 184         unsigned long long openfiles;
 185         unsigned long long openpipes;
 186         unsigned long long opensockets;
 187         unsigned long long openinotifies;
 188         unsigned long long openeventfds;
 189         unsigned long long opentimerfds;
 190         unsigned long long opensignalfds;
 191         unsigned long long openeventpolls;
 192         unsigned long long openother;
 193
 194         unsigned long processes;        // how many processes have been merged to this
 195         int exposed;                            // if set, we have sent this to netdata
 196         int hidden;                                     // if set, we set the hidden flag on the dimension
 197         int debug;
 198         int ends_with;
 199         int starts_with;            // if set, the compare string matches only the
 200                                                                 // beginning of the command
 201
 202         struct target *target;          // the one that will be reported to netdata
 203         struct target *next;
 204 };
 205
 206
 207 // ----------------------------------------------------------------------------
 208 // apps_groups.conf
 209 // aggregate all processes in groups, to have a limited number of dimensions
 210
 211 struct target *apps_groups_root_target = NULL;
 212 struct target *apps_groups_default_target = NULL;
 213 long apps_groups_targets = 0;
 214
 215 struct target *users_root_target = NULL;
 216 struct target *groups_root_target = NULL;
 217
 218 struct target *get_users_target(uid_t uid)
 219 {
 220         struct target *w;
 221         for(w = users_root_target ; w ; w = w->next)
 222                 if(w->uid == uid) return w;
 223
 224         w = calloc(sizeof(struct target), 1);
 225         if(unlikely(!w)) {
 226                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 227                 return NULL;
 228         }
 229
 230         snprintfz(w->compare, MAX_COMPARE_NAME, "%u", uid);
 231         w->comparehash = simple_hash(w->compare);
 232         w->comparelen = strlen(w->compare);
 233
 234         snprintfz(w->id, MAX_NAME, "%u", uid);
 235         w->idhash = simple_hash(w->id);
 236
 237         struct passwd *pw = getpwuid(uid);
 238         if(!pw)
 239                 snprintfz(w->name, MAX_NAME, "%u", uid);
 240         else
 241                 snprintfz(w->name, MAX_NAME, "%s", pw->pw_name);
 242
 243         netdata_fix_chart_name(w->name);
 244
 245         w->uid = uid;
 246
 247         w->next = users_root_target;
 248         users_root_target = w;
 249
 250         if(unlikely(debug))
 251                 fprintf(stderr, "apps.plugin: added uid %u ('%s') target\n", w->uid, w->name);
 252
 253         return w;
 254 }
 255
 256 struct target *get_groups_target(gid_t gid)
 257 {
 258         struct target *w;
 259         for(w = groups_root_target ; w ; w = w->next)
 260                 if(w->gid == gid) return w;
 261
 262         w = calloc(sizeof(struct target), 1);
 263         if(unlikely(!w)) {
 264                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 265                 return NULL;
 266         }
 267
 268         snprintfz(w->compare, MAX_COMPARE_NAME, "%u", gid);
 269         w->comparehash = simple_hash(w->compare);
 270         w->comparelen = strlen(w->compare);
 271
 272         snprintfz(w->id, MAX_NAME, "%u", gid);
 273         w->idhash = simple_hash(w->id);
 274
 275         struct group *gr = getgrgid(gid);
 276         if(!gr)
 277                 snprintfz(w->name, MAX_NAME, "%u", gid);
 278         else
 279                 snprintfz(w->name, MAX_NAME, "%s", gr->gr_name);
 280
 281         netdata_fix_chart_name(w->name);
 282
 283         w->gid = gid;
 284
 285         w->next = groups_root_target;
 286         groups_root_target = w;
 287
 288         if(unlikely(debug))
 289                 fprintf(stderr, "apps.plugin: added gid %u ('%s') target\n", w->gid, w->name);
 290
 291         return w;
 292 }
 293
 294 // find or create a new target
 295 // there are targets that are just aggregated to other target (the second argument)
 296 struct target *get_apps_groups_target(const char *id, struct target *target)
 297 {
 298         int tdebug = 0, thidden = 0, ends_with = 0;
 299         const char *nid = id;
 300
 301         while(nid[0] == '-' || nid[0] == '+' || nid[0] == '*') {
 302                 if(nid[0] == '-') thidden = 1;
 303                 if(nid[0] == '+') tdebug = 1;
 304                 if(nid[0] == '*') ends_with = 1;
 305                 nid++;
 306         }
 307         uint32_t hash = simple_hash(id);
 308
 309         struct target *w, *last = apps_groups_root_target;
 310         for(w = apps_groups_root_target ; w ; w = w->next) {
 311                 if(w->idhash == hash && strncmp(nid, w->id, MAX_NAME) == 0)
 312                         return w;
 313
 314                 last = w;
 315         }
 316
 317         w = calloc(sizeof(struct target), 1);
 318         if(unlikely(!w)) {
 319                 error("Cannot allocate %lu bytes of memory", (unsigned long)sizeof(struct target));
 320                 return NULL;
 321         }
 322
 323         strncpyz(w->id, nid, MAX_NAME);
 324         w->idhash = simple_hash(w->id);
 325
 326         strncpyz(w->name, nid, MAX_NAME);
 327
 328         strncpyz(w->compare, nid, MAX_COMPARE_NAME);
 329         int len = strlen(w->compare);
 330         if(w->compare[len - 1] == '*') {
 331                 w->compare[len - 1] = '\0';
 332                 w->starts_with = 1;
 333         }
 334         w->ends_with = ends_with;
 335
 336         if(w->starts_with && w->ends_with)
 337                 proc_pid_cmdline_is_needed = 1;
 338
 339         w->comparehash = simple_hash(w->compare);
 340         w->comparelen = strlen(w->compare);
 341
 342         w->hidden = thidden;
 343         w->debug = tdebug;
 344         w->target = target;
 345
 346         // append it, to maintain the order in apps_groups.conf
 347         if(last) last->next = w;
 348         else apps_groups_root_target = w;
 349
 350         if(unlikely(debug))
 351                 fprintf(stderr, "apps.plugin: ADDING TARGET ID '%s', process name '%s' (%s), aggregated on target '%s', options: %s %s\n"
 352                         , w->id
 353                                 , w->compare, (w->starts_with && w->ends_with)?"substring":((w->starts_with)?"prefix":((w->ends_with)?"suffix":"exact"))
 354                                 , w->target?w->target->id:w->id
 355                                 , (w->hidden)?"hidden":"-"
 356                                 , (w->debug)?"debug":"-"
 357                 );
 358
 359         return w;
 360 }
 361
 362 // read the apps_groups.conf file
 363 int read_apps_groups_conf(const char *name)
 364 {
 365         char filename[FILENAME_MAX + 1];
 366
 367         snprintfz(filename, FILENAME_MAX, "%s/apps_%s.conf", config_dir, name);
 368
 369         if(unlikely(debug))
 370                 fprintf(stderr, "apps.plugin: process groups file: '%s'\n", filename);
 371
 372         // ----------------------------------------
 373
 374         procfile *ff = procfile_open(filename, " :\t", PROCFILE_FLAG_DEFAULT);
 375         if(!ff) return 1;
 376
 377         procfile_set_quotes(ff, "'\"");
 378
 379         ff = procfile_readall(ff);
 380         if(!ff) {
 381                 procfile_close(ff);
 382                 return 1;
 383         }
 384
 385         unsigned long line, lines = procfile_lines(ff);
 386
 387         for(line = 0; line < lines ;line++) {
 388                 unsigned long word, words = procfile_linewords(ff, line);
 389                 struct target *w = NULL;
 390
 391                 char *t = procfile_lineword(ff, line, 0);
 392                 if(!t || !*t) continue;
 393
 394                 for(word = 0; word < words ;word++) {
 395                         char *s = procfile_lineword(ff, line, word);
 396                         if(!s || !*s) continue;
 397                         if(*s == '#') break;
 398
 399                         if(t == s) continue;
 400
 401                         struct target *n = get_apps_groups_target(s, w);
 402                         if(!n) {
 403                                 error("Cannot create target '%s' (line %lu, word %lu)", s, line, word);
 404                                 continue;
 405                         }
 406
 407                         if(!w) w = n;
 408                 }
 409
 410                 if(w) {
 411                         int tdebug = 0, thidden = 0;
 412
 413                         while(t[0] == '-' || t[0] == '+') {
 414                                 if(t[0] == '-') thidden = 1;
 415                                 if(t[0] == '+') tdebug = 1;
 416                                 t++;
 417                         }
 418
 419                         strncpyz(w->name, t, MAX_NAME);
 420                         w->hidden = thidden;
 421                         w->debug = tdebug;
 422
 423                         if(unlikely(debug))
 424                                 fprintf(stderr, "apps.plugin: AGGREGATION TARGET NAME '%s' on ID '%s', process name '%s' (%s), aggregated on target '%s', options: %s %s\n"
 425                                                 , w->name
 426                                                 , w->id
 427                                                 , w->compare, (w->starts_with && w->ends_with)?"substring":((w->starts_with)?"prefix":((w->ends_with)?"suffix":"exact"))
 428                                                 , w->target?w->target->id:w->id
 429                                                 , (w->hidden)?"hidden":"-"
 430                                                 , (w->debug)?"debug":"-"
 431                                 );
 432                 }
 433         }
 434
 435         procfile_close(ff);
 436
 437         apps_groups_default_target = get_apps_groups_target("p+!o@w#e$i^r&7*5(-i)l-o_", NULL); // match nothing
 438         if(!apps_groups_default_target)
 439                 error("Cannot create default target");
 440         else
 441                 strncpyz(apps_groups_default_target->name, "other", MAX_NAME);
 442
 443         return 0;
 444 }
 445
 446
 447 // ----------------------------------------------------------------------------
 448 // data to store for each pid
 449 // see: man proc
 450
 451 struct pid_stat {
 452         int32_t pid;
 453         char comm[MAX_COMPARE_NAME + 1];
 454         char cmdline[MAX_CMDLINE + 1];
 455
 456         // char state;
 457         int32_t ppid;
 458         // int32_t pgrp;
 459         // int32_t session;
 460         // int32_t tty_nr;
 461         // int32_t tpgid;
 462         // uint64_t flags;
 463
 464         // these are raw values collected
 465         unsigned long long minflt_raw;
 466         unsigned long long cminflt_raw;
 467         unsigned long long majflt_raw;
 468         unsigned long long cmajflt_raw;
 469         unsigned long long utime_raw;
 470         unsigned long long stime_raw;
 471         unsigned long long gtime_raw; // guest_time
 472         unsigned long long cutime_raw;
 473         unsigned long long cstime_raw;
 474         unsigned long long cgtime_raw; // cguest_time
 475
 476         // these are rates
 477         unsigned long long minflt;
 478         unsigned long long cminflt;
 479         unsigned long long majflt;
 480         unsigned long long cmajflt;
 481         unsigned long long utime;
 482         unsigned long long stime;
 483         unsigned long long gtime;
 484         unsigned long long cutime;
 485         unsigned long long cstime;
 486         unsigned long long cgtime;
 487
 488         // int64_t priority;
 489         // int64_t nice;
 490         int32_t num_threads;
 491         // int64_t itrealvalue;
 492         // unsigned long long starttime;
 493         // unsigned long long vsize;
 494         unsigned long long rss;
 495         // unsigned long long rsslim;
 496         // unsigned long long starcode;
 497         // unsigned long long endcode;
 498         // unsigned long long startstack;
 499         // unsigned long long kstkesp;
 500         // unsigned long long kstkeip;
 501         // uint64_t signal;
 502         // uint64_t blocked;
 503         // uint64_t sigignore;
 504         // uint64_t sigcatch;
 505         // uint64_t wchan;
 506         // uint64_t nswap;
 507         // uint64_t cnswap;
 508         // int32_t exit_signal;
 509         // int32_t processor;
 510         // uint32_t rt_priority;
 511         // uint32_t policy;
 512         // unsigned long long delayacct_blkio_ticks;
 513
 514         uid_t uid;
 515         gid_t gid;
 516
 517         unsigned long long statm_size;
 518         unsigned long long statm_resident;
 519         unsigned long long statm_share;
 520         unsigned long long statm_text;
 521         unsigned long long statm_lib;
 522         unsigned long long statm_data;
 523         unsigned long long statm_dirty;
 524
 525         unsigned long long io_logical_bytes_read_raw;
 526         unsigned long long io_logical_bytes_written_raw;
 527         unsigned long long io_read_calls_raw;
 528         unsigned long long io_write_calls_raw;
 529         unsigned long long io_storage_bytes_read_raw;
 530         unsigned long long io_storage_bytes_written_raw;
 531         unsigned long long io_cancelled_write_bytes_raw;
 532
 533         unsigned long long io_logical_bytes_read;
 534         unsigned long long io_logical_bytes_written;
 535         unsigned long long io_read_calls;
 536         unsigned long long io_write_calls;
 537         unsigned long long io_storage_bytes_read;
 538         unsigned long long io_storage_bytes_written;
 539         unsigned long long io_cancelled_write_bytes;
 540
 541         int *fds;                                               // array of fds it uses
 542         int fds_size;                                   // the size of the fds array
 543
 544         int children_count;                             // number of processes directly referencing this
 545         int keep;                                               // 1 when we need to keep this process in memory even after it exited
 546         int keeploops;                                  // increases by 1 every time keep is 1 and updated 0
 547         int updated;                                    // 1 when the process is currently running
 548         int merged;                                             // 1 when it has been merged to its parent
 549         int new_entry;                                  // 1 when this is a new process, just saw for the first time
 550         int read;                                               // 1 when we have already read this process for this iteration
 551         int sortlist;                                   // higher numbers = top on the process tree
 552                                                                         // each process gets a unique number
 553
 554         struct target *target;                  // app_groups.conf targets
 555         struct target *user_target;             // uid based targets
 556         struct target *group_target;    // gid based targets
 557
 558         unsigned long long stat_collected_usec;
 559         unsigned long long last_stat_collected_usec;
 560
 561         unsigned long long io_collected_usec;
 562         unsigned long long last_io_collected_usec;
 563
 564         char *stat_filename;
 565         char *statm_filename;
 566         char *io_filename;
 567         char *cmdline_filename;
 568
 569         struct pid_stat *parent;
 570         struct pid_stat *prev;
 571         struct pid_stat *next;
 572 } *root_of_pids = NULL, **all_pids;
 573
 574 long all_pids_count = 0;
 575
 576 struct pid_stat *get_pid_entry(pid_t pid) {
 577         if(all_pids[pid]) {
 578                 all_pids[pid]->new_entry = 0;
 579                 return all_pids[pid];
 580         }
 581
 582         all_pids[pid] = calloc(sizeof(struct pid_stat), 1);
 583         if(!all_pids[pid]) {
 584                 error("Cannot allocate %zu bytes of memory", (size_t)sizeof(struct pid_stat));
 585                 return NULL;
 586         }
 587
 588         all_pids[pid]->fds = calloc(sizeof(int), 100);
 589         if(!all_pids[pid]->fds)
 590                 error("Cannot allocate %zu bytes of memory", (size_t)(sizeof(int) * 100));
 591         else all_pids[pid]->fds_size = 100;
 592
 593         if(root_of_pids) root_of_pids->prev = all_pids[pid];
 594         all_pids[pid]->next = root_of_pids;
 595         root_of_pids = all_pids[pid];
 596
 597         all_pids[pid]->pid = pid;
 598         all_pids[pid]->new_entry = 1;
 599
 600         all_pids_count++;
 601
 602         return all_pids[pid];
 603 }
 604
 605 void del_pid_entry(pid_t pid) {
 606         if(!all_pids[pid]) {
 607                 error("attempted to free pid %d that is not allocated.", pid);
 608                 return;
 609         }
 610
 611         if(unlikely(debug))
 612                 fprintf(stderr, "apps.plugin: process %d %s exited, deleting it.\n", pid, all_pids[pid]->comm);
 613
 614         if(root_of_pids == all_pids[pid]) root_of_pids = all_pids[pid]->next;
 615         if(all_pids[pid]->next) all_pids[pid]->next->prev = all_pids[pid]->prev;
 616         if(all_pids[pid]->prev) all_pids[pid]->prev->next = all_pids[pid]->next;
 617
 618         if(all_pids[pid]->fds) free(all_pids[pid]->fds);
 619         if(all_pids[pid]->stat_filename) free(all_pids[pid]->stat_filename);
 620         if(all_pids[pid]->statm_filename) free(all_pids[pid]->statm_filename);
 621         if(all_pids[pid]->io_filename) free(all_pids[pid]->io_filename);
 622         if(all_pids[pid]->cmdline_filename) free(all_pids[pid]->cmdline_filename);
 623         free(all_pids[pid]);
 624
 625         all_pids[pid] = NULL;
 626         all_pids_count--;
 627 }
 628
 629
 630 // ----------------------------------------------------------------------------
 631 // update pids from proc
 632
 633 int read_proc_pid_cmdline(struct pid_stat *p) {
 634
 635         if(unlikely(!p->cmdline_filename)) {
 636                 char filename[FILENAME_MAX + 1];
 637                 snprintfz(filename, FILENAME_MAX, "%s/proc/%d/cmdline", host_prefix, p->pid);
 638                 if(!(p->cmdline_filename = strdup(filename)))
 639                         fatal("Cannot allocate memory for filename '%s'", filename);
 640         }
 641
 642         int fd = open(p->cmdline_filename, O_RDONLY, 0666);
 643         if(unlikely(fd == -1)) goto cleanup;
 644
 645         int i, bytes = read(fd, p->cmdline, MAX_CMDLINE);
 646         close(fd);
 647
 648         if(unlikely(bytes <= 0)) goto cleanup;
 649
 650         p->cmdline[bytes] = '\0';
 651         for(i = 0; i < bytes ; i++)
 652                 if(unlikely(!p->cmdline[i])) p->cmdline[i] = ' ';
 653
 654         if(unlikely(debug))
 655                 fprintf(stderr, "Read file '%s' contents: %s\n", p->cmdline_filename, p->cmdline);
 656
 657         return 0;
 658
 659 cleanup:
 660         // copy the command to the command line
 661         strncpyz(p->cmdline, p->comm, MAX_CMDLINE);
 662         return 0;
 663 }
 664
 665 int read_proc_pid_ownership(struct pid_stat *p) {
 666         if(unlikely(!p->stat_filename)) {
 667                 error("pid %d does not have a stat_filename", p->pid);
 668                 return 1;
 669         }
 670
 671         // ----------------------------------------
 672         // read uid and gid
 673
 674         struct stat st;
 675         if(stat(p->stat_filename, &st) != 0) {
 676                 error("Cannot stat file '%s'", p->stat_filename);
 677                 return 1;
 678         }
 679
 680         p->uid = st.st_uid;
 681         p->gid = st.st_gid;
 682
 683         return 0;
 684 }
 685
 686 int read_proc_pid_stat(struct pid_stat *p) {
 687         static procfile *ff = NULL;
 688
 689         if(unlikely(!p->stat_filename)) {
 690                 char filename[FILENAME_MAX + 1];
 691                 snprintfz(filename, FILENAME_MAX, "%s/proc/%d/stat", host_prefix, p->pid);
 692                 if(!(p->stat_filename = strdup(filename)))
 693                         fatal("Cannot allocate memory for filename '%s'", filename);
 694         }
 695
 696         int set_quotes = (!ff)?1:0;
 697
 698         ff = procfile_reopen(ff, p->stat_filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 699         if(unlikely(!ff)) goto cleanup;
 700
 701         // if(set_quotes) procfile_set_quotes(ff, "()");
 702         if(set_quotes) procfile_set_open_close(ff, "(", ")");
 703
 704         ff = procfile_readall(ff);
 705         if(unlikely(!ff)) goto cleanup;
 706
 707         p->last_stat_collected_usec = p->stat_collected_usec;
 708         p->stat_collected_usec = timems();
 709         file_counter++;
 710
 711         // p->pid                       = atol(procfile_lineword(ff, 0, 0+i));
 712
 713         strncpyz(p->comm, procfile_lineword(ff, 0, 1), MAX_COMPARE_NAME);
 714
 715         // p->state                     = *(procfile_lineword(ff, 0, 2));
 716         p->ppid                         = (int32_t) atol(procfile_lineword(ff, 0, 3));
 717         // p->pgrp                      = atol(procfile_lineword(ff, 0, 4));
 718         // p->session           = atol(procfile_lineword(ff, 0, 5));
 719         // p->tty_nr            = atol(procfile_lineword(ff, 0, 6));
 720         // p->tpgid                     = atol(procfile_lineword(ff, 0, 7));
 721         // p->flags                     = strtoull(procfile_lineword(ff, 0, 8), NULL, 10);
 722
 723         unsigned long long last;
 724
 725         last = p->minflt_raw;
 726         p->minflt_raw           = strtoull(procfile_lineword(ff, 0, 9), NULL, 10);
 727         p->minflt = (p->minflt_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 728
 729         last = p->cminflt_raw;
 730         p->cminflt_raw          = strtoull(procfile_lineword(ff, 0, 10), NULL, 10);
 731         p->cminflt = (p->cminflt_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 732
 733         last = p->majflt_raw;
 734         p->majflt_raw           = strtoull(procfile_lineword(ff, 0, 11), NULL, 10);
 735         p->majflt = (p->majflt_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 736
 737         last = p->cmajflt_raw;
 738         p->cmajflt_raw          = strtoull(procfile_lineword(ff, 0, 12), NULL, 10);
 739         p->cmajflt = (p->cmajflt_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 740
 741         last = p->utime_raw;
 742         p->utime_raw            = strtoull(procfile_lineword(ff, 0, 13), NULL, 10);
 743         p->utime = (p->utime_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 744
 745         last = p->stime_raw;
 746         p->stime_raw            = strtoull(procfile_lineword(ff, 0, 14), NULL, 10);
 747         p->stime = (p->stime_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 748
 749         last = p->cutime_raw;
 750         p->cutime_raw           = strtoull(procfile_lineword(ff, 0, 15), NULL, 10);
 751         p->cutime = (p->cutime_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 752
 753         last = p->cstime_raw;
 754         p->cstime_raw           = strtoull(procfile_lineword(ff, 0, 16), NULL, 10);
 755         p->cstime = (p->cstime_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 756
 757         // p->priority          = strtoull(procfile_lineword(ff, 0, 17), NULL, 10);
 758         // p->nice                      = strtoull(procfile_lineword(ff, 0, 18), NULL, 10);
 759         p->num_threads          = (int32_t) atol(procfile_lineword(ff, 0, 19));
 760         // p->itrealvalue       = strtoull(procfile_lineword(ff, 0, 20), NULL, 10);
 761         // p->starttime         = strtoull(procfile_lineword(ff, 0, 21), NULL, 10);
 762         // p->vsize                     = strtoull(procfile_lineword(ff, 0, 22), NULL, 10);
 763         p->rss                          = strtoull(procfile_lineword(ff, 0, 23), NULL, 10);
 764         // p->rsslim            = strtoull(procfile_lineword(ff, 0, 24), NULL, 10);
 765         // p->starcode          = strtoull(procfile_lineword(ff, 0, 25), NULL, 10);
 766         // p->endcode           = strtoull(procfile_lineword(ff, 0, 26), NULL, 10);
 767         // p->startstack        = strtoull(procfile_lineword(ff, 0, 27), NULL, 10);
 768         // p->kstkesp           = strtoull(procfile_lineword(ff, 0, 28), NULL, 10);
 769         // p->kstkeip           = strtoull(procfile_lineword(ff, 0, 29), NULL, 10);
 770         // p->signal            = strtoull(procfile_lineword(ff, 0, 30), NULL, 10);
 771         // p->blocked           = strtoull(procfile_lineword(ff, 0, 31), NULL, 10);
 772         // p->sigignore         = strtoull(procfile_lineword(ff, 0, 32), NULL, 10);
 773         // p->sigcatch          = strtoull(procfile_lineword(ff, 0, 33), NULL, 10);
 774         // p->wchan                     = strtoull(procfile_lineword(ff, 0, 34), NULL, 10);
 775         // p->nswap                     = strtoull(procfile_lineword(ff, 0, 35), NULL, 10);
 776         // p->cnswap            = strtoull(procfile_lineword(ff, 0, 36), NULL, 10);
 777         // p->exit_signal       = atol(procfile_lineword(ff, 0, 37));
 778         // p->processor         = atol(procfile_lineword(ff, 0, 38));
 779         // p->rt_priority       = strtoul(procfile_lineword(ff, 0, 39), NULL, 10);
 780         // p->policy            = strtoul(procfile_lineword(ff, 0, 40), NULL, 10);
 781         // p->delayacct_blkio_ticks = strtoull(procfile_lineword(ff, 0, 41), NULL, 10);
 782
 783         last = p->gtime_raw;
 784         p->gtime_raw            = strtoull(procfile_lineword(ff, 0, 42), NULL, 10);
 785         p->gtime = (p->gtime_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 786
 787         last = p->cgtime_raw;
 788         p->cgtime_raw           = strtoull(procfile_lineword(ff, 0, 43), NULL, 10);
 789         p->cgtime = (p->cgtime_raw - last) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
 790
 791         if(show_guest_time || p->gtime || p->cgtime) {
 792                 p->utime  -= (p->utime  >= p->gtime )?p->gtime :p->utime;
 793                 p->cutime -= (p->cutime >= p->cgtime)?p->cgtime:p->cutime;
 794                 show_guest_time = 1;
 795         }
 796
 797         if(unlikely(debug || (p->target && p->target->debug)))
 798                 fprintf(stderr, "apps.plugin: READ PROC/PID/STAT: %s/proc/%d/stat, process: '%s' on target '%s' (dt=%llu) VALUES: utime=%llu, stime=%llu, cutime=%llu, cstime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu, threads=%d\n", host_prefix, p->pid, p->comm, (p->target)?p->target->name:"UNSET", p->stat_collected_usec - p->last_stat_collected_usec, p->utime, p->stime, p->cutime, p->cstime, p->minflt, p->majflt, p->cminflt, p->cmajflt, p->num_threads);
 799
 800         if(unlikely(global_iterations_counter == 1)) {
 801                 p->minflt                       = 0;
 802                 p->cminflt                      = 0;
 803                 p->majflt                       = 0;
 804                 p->cmajflt                      = 0;
 805                 p->utime                        = 0;
 806                 p->stime                        = 0;
 807                 p->gtime                        = 0;
 808                 p->cutime                       = 0;
 809                 p->cstime                       = 0;
 810                 p->cgtime                       = 0;
 811         }
 812
 813         return 0;
 814
 815 cleanup:
 816         p->minflt                       = 0;
 817         p->cminflt                      = 0;
 818         p->majflt                       = 0;
 819         p->cmajflt                      = 0;
 820         p->utime                        = 0;
 821         p->stime                        = 0;
 822         p->gtime                        = 0;
 823         p->cutime                       = 0;
 824         p->cstime                       = 0;
 825         p->cgtime                       = 0;
 826         p->num_threads          = 0;
 827         p->rss                          = 0;
 828         return 1;
 829 }
 830
 831 int read_proc_pid_statm(struct pid_stat *p) {
 832         static procfile *ff = NULL;
 833
 834         if(unlikely(!p->statm_filename)) {
 835                 char filename[FILENAME_MAX + 1];
 836                 snprintfz(filename, FILENAME_MAX, "%s/proc/%d/statm", host_prefix, p->pid);
 837                 if(!(p->statm_filename = strdup(filename)))
 838                         fatal("Cannot allocate memory for filename '%s'", filename);
 839         }
 840
 841         ff = procfile_reopen(ff, p->statm_filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 842         if(unlikely(!ff)) goto cleanup;
 843
 844         ff = procfile_readall(ff);
 845         if(unlikely(!ff)) goto cleanup;
 846
 847         file_counter++;
 848
 849         p->statm_size                   = strtoull(procfile_lineword(ff, 0, 0), NULL, 10);
 850         p->statm_resident               = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 851         p->statm_share                  = strtoull(procfile_lineword(ff, 0, 2), NULL, 10);
 852         p->statm_text                   = strtoull(procfile_lineword(ff, 0, 3), NULL, 10);
 853         p->statm_lib                    = strtoull(procfile_lineword(ff, 0, 4), NULL, 10);
 854         p->statm_data                   = strtoull(procfile_lineword(ff, 0, 5), NULL, 10);
 855         p->statm_dirty                  = strtoull(procfile_lineword(ff, 0, 6), NULL, 10);
 856
 857         return 0;
 858
 859 cleanup:
 860         p->statm_size                   = 0;
 861         p->statm_resident               = 0;
 862         p->statm_share                  = 0;
 863         p->statm_text                   = 0;
 864         p->statm_lib                    = 0;
 865         p->statm_data                   = 0;
 866         p->statm_dirty                  = 0;
 867         return 1;
 868 }
 869
 870 int read_proc_pid_io(struct pid_stat *p) {
 871         static procfile *ff = NULL;
 872
 873         if(unlikely(!p->io_filename)) {
 874                 char filename[FILENAME_MAX + 1];
 875                 snprintfz(filename, FILENAME_MAX, "%s/proc/%d/io", host_prefix, p->pid);
 876                 if(!(p->io_filename = strdup(filename)))
 877                         fatal("Cannot allocate memory for filename '%s'", filename);
 878         }
 879
 880         // open the file
 881         ff = procfile_reopen(ff, p->io_filename, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);
 882         if(unlikely(!ff)) goto cleanup;
 883
 884         ff = procfile_readall(ff);
 885         if(unlikely(!ff)) goto cleanup;
 886
 887         file_counter++;
 888
 889         p->last_io_collected_usec = p->io_collected_usec;
 890         p->io_collected_usec = timems();
 891
 892         unsigned long long last;
 893
 894         last = p->io_logical_bytes_read_raw;
 895         p->io_logical_bytes_read_raw = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 896         p->io_logical_bytes_read = (p->io_logical_bytes_read_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 897
 898         last = p->io_logical_bytes_written_raw;
 899         p->io_logical_bytes_written_raw = strtoull(procfile_lineword(ff, 1, 1), NULL, 10);
 900         p->io_logical_bytes_written = (p->io_logical_bytes_written_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 901
 902         last = p->io_read_calls_raw;
 903         p->io_read_calls_raw = strtoull(procfile_lineword(ff, 2, 1), NULL, 10);
 904         p->io_read_calls = (p->io_read_calls_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 905
 906         last = p->io_write_calls_raw;
 907         p->io_write_calls_raw = strtoull(procfile_lineword(ff, 3, 1), NULL, 10);
 908         p->io_write_calls = (p->io_write_calls_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 909
 910         last = p->io_storage_bytes_read_raw;
 911         p->io_storage_bytes_read_raw = strtoull(procfile_lineword(ff, 4, 1), NULL, 10);
 912         p->io_storage_bytes_read = (p->io_storage_bytes_read_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 913
 914         last = p->io_storage_bytes_written_raw;
 915         p->io_storage_bytes_written_raw = strtoull(procfile_lineword(ff, 5, 1), NULL, 10);
 916         p->io_storage_bytes_written = (p->io_storage_bytes_written_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 917
 918         last = p->io_cancelled_write_bytes_raw;
 919         p->io_cancelled_write_bytes_raw = strtoull(procfile_lineword(ff, 6, 1), NULL, 10);
 920         p->io_cancelled_write_bytes = (p->io_cancelled_write_bytes_raw - last) * (1000000ULL * RATES_DETAIL) / (p->io_collected_usec - p->last_io_collected_usec);
 921
 922         if(unlikely(global_iterations_counter == 1)) {
 923                 p->io_logical_bytes_read                = 0;
 924                 p->io_logical_bytes_written     = 0;
 925                 p->io_read_calls                                = 0;
 926                 p->io_write_calls                               = 0;
 927                 p->io_storage_bytes_read                = 0;
 928                 p->io_storage_bytes_written     = 0;
 929                 p->io_cancelled_write_bytes             = 0;
 930         }
 931
 932         return 0;
 933
 934 cleanup:
 935         p->io_logical_bytes_read                = 0;
 936         p->io_logical_bytes_written     = 0;
 937         p->io_read_calls                                = 0;
 938         p->io_write_calls                               = 0;
 939         p->io_storage_bytes_read                = 0;
 940         p->io_storage_bytes_written     = 0;
 941         p->io_cancelled_write_bytes             = 0;
 942         return 1;
 943 }
 944
 945 unsigned long long global_utime = 0;
 946 unsigned long long global_stime = 0;
 947 unsigned long long global_gtime = 0;
 948
 949 int read_proc_stat() {
 950         static char filename[FILENAME_MAX + 1] = "";
 951         static procfile *ff = NULL;
 952         static unsigned long long utime_raw = 0, stime_raw = 0, gtime_raw = 0, ntime_raw = 0, collected_usec = 0, last_collected_usec = 0;
 953
 954         if(unlikely(!ff)) {
 955                 snprintfz(filename, FILENAME_MAX, "%s/proc/stat", host_prefix);
 956                 ff = procfile_open(filename, " \t:", PROCFILE_FLAG_DEFAULT);
 957                 if(unlikely(!ff)) goto cleanup;
 958         }
 959
 960         ff = procfile_readall(ff);
 961         if(unlikely(!ff)) goto cleanup;
 962
 963         last_collected_usec = collected_usec;
 964         collected_usec = timems();
 965
 966         file_counter++;
 967
 968         unsigned long long last;
 969
 970         last = utime_raw;
 971         utime_raw = strtoull(procfile_lineword(ff, 0, 1), NULL, 10);
 972         global_utime = (utime_raw - last) * (1000000ULL * RATES_DETAIL) / (collected_usec - last_collected_usec);
 973
 974         last = ntime_raw;
 975         ntime_raw = strtoull(procfile_lineword(ff, 0, 2), NULL, 10);
 976         global_utime += (ntime_raw - last) * (1000000ULL * RATES_DETAIL) / (collected_usec - last_collected_usec);
 977
 978         last = stime_raw;
 979         stime_raw = strtoull(procfile_lineword(ff, 0, 3), NULL, 10);
 980         global_stime = (stime_raw - last) * (1000000ULL * RATES_DETAIL) / (collected_usec - last_collected_usec);
 981
 982         last = gtime_raw;
 983         gtime_raw = strtoull(procfile_lineword(ff, 0, 3), NULL, 10);
 984         global_gtime = (gtime_raw - last) * (1000000ULL * RATES_DETAIL) / (collected_usec - last_collected_usec);
 985
 986         if(unlikely(global_iterations_counter == 1)) {
 987                 global_utime = 0;
 988                 global_stime = 0;
 989                 global_gtime = 0;
 990         }
 991
 992         return 0;
 993
 994 cleanup:
 995         global_utime = 0;
 996         global_stime = 0;
 997         global_gtime = 0;
 998         return 1;
 999 }
1000
1001
1002 // ----------------------------------------------------------------------------
1003 // file descriptor
1004 // this is used to keep a global list of all open files of the system
1005 // it is needed in order to calculate the unique files processes have open
1006
1007 #define FILE_DESCRIPTORS_INCREASE_STEP 100
1008
1009 struct file_descriptor {
1010         avl avl;
1011 #ifdef NETDATA_INTERNAL_CHECKS
1012         uint32_t magic;
1013 #endif /* NETDATA_INTERNAL_CHECKS */
1014         uint32_t hash;
1015         const char *name;
1016         int type;
1017         int count;
1018         int pos;
1019 } *all_files = NULL;
1020
1021 int all_files_len = 0;
1022 int all_files_size = 0;
1023
1024 int file_descriptor_compare(void* a, void* b) {
1025 #ifdef NETDATA_INTERNAL_CHECKS
1026         if(((struct file_descriptor *)a)->magic != 0x0BADCAFE || ((struct file_descriptor *)b)->magic != 0x0BADCAFE)
1027                 error("Corrupted index data detected. Please report this.");
1028 #endif /* NETDATA_INTERNAL_CHECKS */
1029
1030         if(((struct file_descriptor *)a)->hash < ((struct file_descriptor *)b)->hash)
1031                 return -1;
1032
1033         else if(((struct file_descriptor *)a)->hash > ((struct file_descriptor *)b)->hash)
1034                 return 1;
1035
1036         else
1037                 return strcmp(((struct file_descriptor *)a)->name, ((struct file_descriptor *)b)->name);
1038 }
1039
1040 int file_descriptor_iterator(avl *a) { if(a) {}; return 0; }
1041
1042 avl_tree all_files_index = {
1043                 NULL,
1044                 file_descriptor_compare
1045 };
1046
1047 static struct file_descriptor *file_descriptor_find(const char *name, uint32_t hash) {
1048         struct file_descriptor tmp;
1049         tmp.hash = (hash)?hash:simple_hash(name);
1050         tmp.name = name;
1051         tmp.count = 0;
1052         tmp.pos = 0;
1053 #ifdef NETDATA_INTERNAL_CHECKS
1054         tmp.magic = 0x0BADCAFE;
1055 #endif /* NETDATA_INTERNAL_CHECKS */
1056
1057         return (struct file_descriptor *)avl_search(&all_files_index, (avl *) &tmp);
1058 }
1059
1060 #define file_descriptor_add(fd) avl_insert(&all_files_index, (avl *)(fd))
1061 #define file_descriptor_remove(fd) avl_remove(&all_files_index, (avl *)(fd))
1062
1063 #define FILETYPE_OTHER 0
1064 #define FILETYPE_FILE 1
1065 #define FILETYPE_PIPE 2
1066 #define FILETYPE_SOCKET 3
1067 #define FILETYPE_INOTIFY 4
1068 #define FILETYPE_EVENTFD 5
1069 #define FILETYPE_EVENTPOLL 6
1070 #define FILETYPE_TIMERFD 7
1071 #define FILETYPE_SIGNALFD 8
1072
1073 void file_descriptor_not_used(int id)
1074 {
1075         if(id > 0 && id < all_files_size) {
1076
1077 #ifdef NETDATA_INTERNAL_CHECKS
1078                 if(all_files[id].magic != 0x0BADCAFE) {
1079                         error("Ignoring request to remove empty file id %d.", id);
1080                         return;
1081                 }
1082 #endif /* NETDATA_INTERNAL_CHECKS */
1083
1084                 if(unlikely(debug))
1085                         fprintf(stderr, "apps.plugin: decreasing slot %d (count = %d).\n", id, all_files[id].count);
1086
1087                 if(all_files[id].count > 0) {
1088                         all_files[id].count--;
1089
1090                         if(!all_files[id].count) {
1091                                 if(unlikely(debug))
1092                                         fprintf(stderr, "apps.plugin:   >> slot %d is empty.\n", id);
1093
1094                                 file_descriptor_remove(&all_files[id]);
1095 #ifdef NETDATA_INTERNAL_CHECKS
1096                                 all_files[id].magic = 0x00000000;
1097 #endif /* NETDATA_INTERNAL_CHECKS */
1098                                 all_files_len--;
1099                         }
1100                 }
1101                 else
1102                         error("Request to decrease counter of fd %d (%s), while the use counter is 0", id, all_files[id].name);
1103         }
1104         else    error("Request to decrease counter of fd %d, which is outside the array size (1 to %d)", id, all_files_size);
1105 }
1106
1107 int file_descriptor_find_or_add(const char *name)
1108 {
1109         static int last_pos = 0;
1110         uint32_t hash = simple_hash(name);
1111
1112         if(unlikely(debug))
1113                 fprintf(stderr, "apps.plugin: adding or finding name '%s' with hash %u\n", name, hash);
1114
1115         struct file_descriptor *fd = file_descriptor_find(name, hash);
1116         if(fd) {
1117                 // found
1118                 if(unlikely(debug))
1119                         fprintf(stderr, "apps.plugin:   >> found on slot %d\n", fd->pos);
1120
1121                 fd->count++;
1122                 return fd->pos;
1123         }
1124         // not found
1125
1126         // check we have enough memory to add it
1127         if(!all_files || all_files_len == all_files_size) {
1128                 void *old = all_files;
1129                 int i;
1130
1131                 // there is no empty slot
1132                 if(unlikely(debug))
1133                         fprintf(stderr, "apps.plugin: extending fd array to %d entries\n", all_files_size + FILE_DESCRIPTORS_INCREASE_STEP);
1134
1135                 all_files = realloc(all_files, (all_files_size + FILE_DESCRIPTORS_INCREASE_STEP) * sizeof(struct file_descriptor));
1136
1137                 // if the address changed, we have to rebuild the index
1138                 // since all pointers are now invalid
1139                 if(old && old != (void *)all_files) {
1140                         if(unlikely(debug))
1141                                 fprintf(stderr, "apps.plugin:   >> re-indexing.\n");
1142
1143                         all_files_index.root = NULL;
1144                         for(i = 0; i < all_files_size; i++) {
1145                                 if(!all_files[i].count) continue;
1146                                 file_descriptor_add(&all_files[i]);
1147                         }
1148
1149                         if(unlikely(debug))
1150                                 fprintf(stderr, "apps.plugin:   >> re-indexing done.\n");
1151                 }
1152
1153                 for(i = all_files_size; i < (all_files_size + FILE_DESCRIPTORS_INCREASE_STEP); i++) {
1154                         all_files[i].count = 0;
1155                         all_files[i].name = NULL;
1156 #ifdef NETDATA_INTERNAL_CHECKS
1157                         all_files[i].magic = 0x00000000;
1158 #endif /* NETDATA_INTERNAL_CHECKS */
1159                         all_files[i].pos = i;
1160                 }
1161
1162                 if(!all_files_size) all_files_len = 1;
1163                 all_files_size += FILE_DESCRIPTORS_INCREASE_STEP;
1164         }
1165
1166         if(unlikely(debug))
1167                 fprintf(stderr, "apps.plugin:   >> searching for empty slot.\n");
1168
1169         // search for an empty slot
1170         int i, c;
1171         for(i = 0, c = last_pos ; i < all_files_size ; i++, c++) {
1172                 if(c >= all_files_size) c = 0;
1173                 if(c == 0) continue;
1174
1175                 if(!all_files[c].count) {
1176                         if(unlikely(debug))
1177                                 fprintf(stderr, "apps.plugin:   >> Examining slot %d.\n", c);
1178
1179 #ifdef NETDATA_INTERNAL_CHECKS
1180                         if(all_files[c].magic == 0x0BADCAFE && all_files[c].name && file_descriptor_find(all_files[c].name, all_files[c].hash))
1181                                 error("fd on position %d is not cleared properly. It still has %s in it.\n", c, all_files[c].name);
1182 #endif /* NETDATA_INTERNAL_CHECKS */
1183
1184                         if(unlikely(debug))
1185                                 fprintf(stderr, "apps.plugin:   >> %s fd position %d for %s (last name: %s)\n", all_files[c].name?"re-using":"using", c, name, all_files[c].name);
1186
1187                         if(all_files[c].name) free((void *)all_files[c].name);
1188                         all_files[c].name = NULL;
1189                         last_pos = c;
1190                         break;
1191                 }
1192         }
1193         if(i == all_files_size) {
1194                 fatal("We should find an empty slot, but there isn't any");
1195                 exit(1);
1196         }
1197
1198         if(unlikely(debug))
1199                 fprintf(stderr, "apps.plugin:   >> updating slot %d.\n", c);
1200
1201         all_files_len++;
1202
1203         // else we have an empty slot in 'c'
1204
1205         int type;
1206         if(name[0] == '/') type = FILETYPE_FILE;
1207         else if(strncmp(name, "pipe:", 5) == 0) type = FILETYPE_PIPE;
1208         else if(strncmp(name, "socket:", 7) == 0) type = FILETYPE_SOCKET;
1209         else if(strcmp(name, "anon_inode:inotify") == 0 || strcmp(name, "inotify") == 0) type = FILETYPE_INOTIFY;
1210         else if(strcmp(name, "anon_inode:[eventfd]") == 0) type = FILETYPE_EVENTFD;
1211         else if(strcmp(name, "anon_inode:[eventpoll]") == 0) type = FILETYPE_EVENTPOLL;
1212         else if(strcmp(name, "anon_inode:[timerfd]") == 0) type = FILETYPE_TIMERFD;
1213         else if(strcmp(name, "anon_inode:[signalfd]") == 0) type = FILETYPE_SIGNALFD;
1214         else if(strncmp(name, "anon_inode:", 11) == 0) {
1215                 if(unlikely(debug))
1216                         fprintf(stderr, "apps.plugin: FIXME: unknown anonymous inode: %s\n", name);
1217
1218                 type = FILETYPE_OTHER;
1219         }
1220         else {
1221                 if(unlikely(debug))
1222                         fprintf(stderr, "apps.plugin: FIXME: cannot understand linkname: %s\n", name);
1223
1224                 type = FILETYPE_OTHER;
1225         }
1226
1227         all_files[c].name = strdup(name);
1228         all_files[c].hash = hash;
1229         all_files[c].type = type;
1230         all_files[c].pos  = c;
1231         all_files[c].count = 1;
1232 #ifdef NETDATA_INTERNAL_CHECKS
1233         all_files[c].magic = 0x0BADCAFE;
1234 #endif /* NETDATA_INTERNAL_CHECKS */
1235         file_descriptor_add(&all_files[c]);
1236
1237         if(unlikely(debug))
1238                 fprintf(stderr, "apps.plugin: using fd position %d (name: %s)\n", c, all_files[c].name);
1239
1240         return c;
1241 }
1242
1243 int read_pid_file_descriptors(struct pid_stat *p) {
1244         char dirname[FILENAME_MAX+1];
1245
1246         snprintfz(dirname, FILENAME_MAX, "%s/proc/%d/fd", host_prefix, p->pid);
1247         DIR *fds = opendir(dirname);
1248         if(fds) {
1249                 int c;
1250                 struct dirent *de;
1251                 char fdname[FILENAME_MAX + 1];
1252                 char linkname[FILENAME_MAX + 1];
1253
1254                 // make the array negative
1255                 for(c = 0 ; c < p->fds_size ; c++)
1256                         p->fds[c] = -p->fds[c];
1257
1258                 while((de = readdir(fds))) {
1259                         if(strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
1260                                 continue;
1261
1262                         // check if the fds array is small
1263                         int fdid = atoi(de->d_name);
1264                         if(fdid < 0) continue;
1265                         if(fdid >= p->fds_size) {
1266                                 // it is small, extend it
1267                                 if(unlikely(debug))
1268                                         fprintf(stderr, "apps.plugin: extending fd memory slots for %s from %d to %d\n", p->comm, p->fds_size, fdid + 100);
1269
1270                                 p->fds = realloc(p->fds, (fdid + 100) * sizeof(int));
1271                                 if(!p->fds) {
1272                                         fatal("Cannot re-allocate fds for %s", p->comm);
1273                                         break;
1274                                 }
1275
1276                                 // and initialize it
1277                                 for(c = p->fds_size ; c < (fdid + 100) ; c++) p->fds[c] = 0;
1278                                 p->fds_size = fdid + 100;
1279                         }
1280
1281                         if(p->fds[fdid] == 0) {
1282                                 // we don't know this fd, get it
1283
1284                                 sprintf(fdname, "%s/proc/%d/fd/%s", host_prefix, p->pid, de->d_name);
1285                                 ssize_t l = readlink(fdname, linkname, FILENAME_MAX);
1286                                 if(l == -1) {
1287                                         if(debug || (p->target && p->target->debug)) {
1288                                                 if(debug || (p->target && p->target->debug))
1289                                                         error("Cannot read link %s", fdname);
1290                                         }
1291                                         continue;
1292                                 }
1293                                 linkname[l] = '\0';
1294                                 file_counter++;
1295
1296                                 // if another process already has this, we will get
1297                                 // the same id
1298                                 p->fds[fdid] = file_descriptor_find_or_add(linkname);
1299                         }
1300
1301                         // else make it positive again, we need it
1302                         // of course, the actual file may have changed, but we don't care so much
1303                         // FIXME: we could compare the inode as returned by readdir direct structure
1304                         else p->fds[fdid] = -p->fds[fdid];
1305                 }
1306                 closedir(fds);
1307
1308                 // remove all the negative file descriptors
1309                 for(c = 0 ; c < p->fds_size ; c++) if(p->fds[c] < 0) {
1310                         file_descriptor_not_used(-p->fds[c]);
1311                         p->fds[c] = 0;
1312                 }
1313         }
1314         else return 1;
1315
1316         return 0;
1317 }
1318
1319 // ----------------------------------------------------------------------------
1320
1321 int print_process_and_parents(struct pid_stat *p, unsigned long long time) {
1322         char *prefix = "\\_ ";
1323         int indent = 0;
1324
1325         if(p->parent)
1326                 indent = print_process_and_parents(p->parent, p->stat_collected_usec);
1327         else
1328                 prefix = " > ";
1329
1330         char buffer[indent + 1];
1331         int i;
1332
1333         for(i = 0; i < indent ;i++) buffer[i] = ' ';
1334         buffer[i] = '\0';
1335
1336         fprintf(stderr, "  %s %s%s (%d %s %lld"
1337                 , buffer
1338                 , prefix
1339                 , p->comm
1340                 , p->pid
1341                 , p->updated?"running":"exited"
1342                 , (long long)p->stat_collected_usec - (long long)time
1343                 );
1344
1345         if(p->utime)   fprintf(stderr, " utime=%llu",   p->utime);
1346         if(p->stime)   fprintf(stderr, " stime=%llu",   p->stime);
1347         if(p->gtime)   fprintf(stderr, " gtime=%llu",   p->gtime);
1348         if(p->cutime)  fprintf(stderr, " cutime=%llu",  p->cutime);
1349         if(p->cstime)  fprintf(stderr, " cstime=%llu",  p->cstime);
1350         if(p->cgtime)  fprintf(stderr, " cgtime=%llu",  p->cgtime);
1351         if(p->minflt)  fprintf(stderr, " minflt=%llu",  p->minflt);
1352         if(p->cminflt) fprintf(stderr, " cminflt=%llu", p->cminflt);
1353         if(p->majflt)  fprintf(stderr, " majflt=%llu",  p->majflt);
1354         if(p->cmajflt) fprintf(stderr, " cmajflt=%llu", p->cmajflt);
1355         fprintf(stderr, ")\n");
1356
1357         return indent + 1;
1358 }
1359
1360 void print_process_tree(struct pid_stat *p, char *msg) {
1361         log_date(stderr);
1362         fprintf(stderr, "%s: process %s (%d, %s) with parents:\n", msg, p->comm, p->pid, p->updated?"running":"exited");
1363         print_process_and_parents(p, p->stat_collected_usec);
1364 }
1365
1366 void find_lost_child_debug(struct pid_stat *pe, unsigned long long lost, int type) {
1367         int found = 0;
1368         struct pid_stat *p = NULL;
1369
1370         for(p = root_of_pids; p ; p = p->next) {
1371                 if(p == pe) continue;
1372
1373                 switch(type) {
1374                         case 1:
1375                                 if(p->cminflt > lost) {
1376                                         fprintf(stderr, " > process %d (%s) could use the lost exited child minflt %llu of process %d (%s)\n", p->pid, p->comm, lost, pe->pid, pe->comm);
1377                                         found++;
1378                                 }
1379                                 break;
1380
1381                         case 2:
1382                                 if(p->cmajflt > lost) {
1383                                         fprintf(stderr, " > process %d (%s) could use the lost exited child majflt %llu of process %d (%s)\n", p->pid, p->comm, lost, pe->pid, pe->comm);
1384                                         found++;
1385                                 }
1386                                 break;
1387
1388                         case 3:
1389                                 if(p->cutime > lost) {
1390                                         fprintf(stderr, " > process %d (%s) could use the lost exited child utime %llu of process %d (%s)\n", p->pid, p->comm, lost, pe->pid, pe->comm);
1391                                         found++;
1392                                 }
1393                                 break;
1394
1395                         case 4:
1396                                 if(p->cstime > lost) {
1397                                         fprintf(stderr, " > process %d (%s) could use the lost exited child stime %llu of process %d (%s)\n", p->pid, p->comm, lost, pe->pid, pe->comm);
1398                                         found++;
1399                                 }
1400                                 break;
1401
1402                         case 5:
1403                                 if(p->cgtime > lost) {
1404                                         fprintf(stderr, " > process %d (%s) could use the lost exited child gtime %llu of process %d (%s)\n", p->pid, p->comm, lost, pe->pid, pe->comm);
1405                                         found++;
1406                                 }
1407                                 break;
1408                 }
1409         }
1410
1411         if(!found) {
1412                 switch(type) {
1413                         case 1:
1414                                 fprintf(stderr, " > cannot find any process to use the lost exited child minflt %llu of process %d (%s)\n", lost, pe->pid, pe->comm);
1415                                 break;
1416
1417                         case 2:
1418                                 fprintf(stderr, " > cannot find any process to use the lost exited child majflt %llu of process %d (%s)\n", lost, pe->pid, pe->comm);
1419                                 break;
1420
1421                         case 3:
1422                                 fprintf(stderr, " > cannot find any process to use the lost exited child utime %llu of process %d (%s)\n", lost, pe->pid, pe->comm);
1423                                 break;
1424
1425                         case 4:
1426                                 fprintf(stderr, " > cannot find any process to use the lost exited child stime %llu of process %d (%s)\n", lost, pe->pid, pe->comm);
1427                                 break;
1428
1429                         case 5:
1430                                 fprintf(stderr, " > cannot find any process to use the lost exited child gtime %llu of process %d (%s)\n", lost, pe->pid, pe->comm);
1431                                 break;
1432                 }
1433         }
1434 }
1435
1436 unsigned long long remove_exited_child_from_parent(unsigned long long *field, unsigned long long *pfield) {
1437         unsigned long long absorbed = 0;
1438
1439         if(*field > *pfield) {
1440                 absorbed += *pfield;
1441                 *field -= *pfield;
1442                 *pfield = 0;
1443         }
1444         else {
1445                 absorbed += *field;
1446                 *pfield -= *field;
1447                 *field = 0;
1448         }
1449
1450         return absorbed;
1451 }
1452
1453 void process_exited_processes() {
1454         struct pid_stat *p;
1455
1456         for(p = root_of_pids; p ; p = p->next) {
1457                 if(p->updated || !p->stat_collected_usec)
1458                         continue;
1459
1460                 struct pid_stat *pp = p->parent;
1461
1462                 unsigned long long utime  = (p->utime_raw + p->cutime_raw)   * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
1463                 unsigned long long stime  = (p->stime_raw + p->cstime_raw)   * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
1464                 unsigned long long gtime  = (p->gtime_raw + p->cgtime_raw)   * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
1465                 unsigned long long minflt = (p->minflt_raw + p->cminflt_raw) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
1466                 unsigned long long majflt = (p->majflt_raw + p->cmajflt_raw) * (1000000ULL * RATES_DETAIL) / (p->stat_collected_usec - p->last_stat_collected_usec);
1467
1468                 if(utime + stime + gtime + minflt + majflt == 0)
1469                         continue;
1470
1471                 if(unlikely(debug)) {
1472                         log_date(stderr);
1473                         fprintf(stderr, "Absorb %s (%d %s total resources: utime=%llu stime=%llu gtime=%llu minflt=%llu majflt=%llu)\n"
1474                                 , p->comm
1475                                 , p->pid
1476                                 , p->updated?"running":"exited"
1477                                 , utime
1478                                 , stime
1479                                 , gtime
1480                                 , minflt
1481                                 , majflt
1482                                 );
1483                         print_process_tree(p, "Searching parents");
1484                 }
1485
1486                 for(pp = p->parent; pp ; pp = pp->parent) {
1487                         if(!pp->updated) continue;
1488
1489                         unsigned long long absorbed;
1490                         absorbed = remove_exited_child_from_parent(&utime,  &pp->cutime);
1491                         if(unlikely(debug && absorbed))
1492                                 fprintf(stderr, " > process %s (%d %s) absorbed %llu utime (remaining: %llu)\n", pp->comm, pp->pid, pp->updated?"running":"exited", absorbed, utime);
1493
1494                         absorbed = remove_exited_child_from_parent(&stime,  &pp->cstime);
1495                         if(unlikely(debug && absorbed))
1496                                 fprintf(stderr, " > process %s (%d %s) absorbed %llu stime (remaining: %llu)\n", pp->comm, pp->pid, pp->updated?"running":"exited", absorbed, stime);
1497
1498                         absorbed = remove_exited_child_from_parent(&gtime,  &pp->cgtime);
1499                         if(unlikely(debug && absorbed))
1500                                 fprintf(stderr, " > process %s (%d %s) absorbed %llu gtime (remaining: %llu)\n", pp->comm, pp->pid, pp->updated?"running":"exited", absorbed, gtime);
1501
1502                         absorbed = remove_exited_child_from_parent(&minflt, &pp->cminflt);
1503                         if(unlikely(debug && absorbed))
1504                                 fprintf(stderr, " > process %s (%d %s) absorbed %llu minflt (remaining: %llu)\n", pp->comm, pp->pid, pp->updated?"running":"exited", absorbed, minflt);
1505
1506                         absorbed = remove_exited_child_from_parent(&majflt, &pp->cmajflt);
1507                         if(unlikely(debug && absorbed))
1508                                 fprintf(stderr, " > process %s (%d %s) absorbed %llu majflt (remaining: %llu)\n", pp->comm, pp->pid, pp->updated?"running":"exited", absorbed, majflt);
1509                 }
1510
1511                 if(unlikely(utime + stime + gtime + minflt + majflt > 0)) {
1512                         if(unlikely(debug)) {
1513                                 if(utime)  find_lost_child_debug(p, utime,  3);
1514                                 if(stime)  find_lost_child_debug(p, stime,  4);
1515                                 if(gtime)  find_lost_child_debug(p, gtime,  5);
1516                                 if(minflt) find_lost_child_debug(p, minflt, 1);
1517                                 if(majflt) find_lost_child_debug(p, majflt, 2);
1518                         }
1519
1520                         p->keep = 1;
1521
1522                         if(unlikely(debug))
1523                                 fprintf(stderr, " > remaining resources - KEEP - for another loop: %s (%d %s total resources: utime=%llu stime=%llu gtime=%llu minflt=%llu majflt=%llu)\n"
1524                                         , p->comm
1525                                         , p->pid
1526                                         , p->updated?"running":"exited"
1527                                         , utime
1528                                         , stime
1529                                         , gtime
1530                                         , minflt
1531                                         , majflt
1532                                         );
1533
1534                         for(pp = p->parent; pp ; pp = pp->parent) {
1535                                 if(pp->updated) break;
1536                                 pp->keep = 1;
1537
1538                                 if(unlikely(debug))
1539                                         fprintf(stderr, " > - KEEP - parent for another loop: %s (%d %s)\n"
1540                                                 , pp->comm
1541                                                 , pp->pid
1542                                                 , pp->updated?"running":"exited"
1543                                                 );
1544                         }
1545
1546                         p->utime_raw   = utime  * (p->stat_collected_usec - p->last_stat_collected_usec) / (1000000ULL * RATES_DETAIL);
1547                         p->stime_raw   = stime  * (p->stat_collected_usec - p->last_stat_collected_usec) / (1000000ULL * RATES_DETAIL);
1548                         p->gtime_raw   = gtime  * (p->stat_collected_usec - p->last_stat_collected_usec) / (1000000ULL * RATES_DETAIL);
1549                         p->minflt_raw  = minflt * (p->stat_collected_usec - p->last_stat_collected_usec) / (1000000ULL * RATES_DETAIL);
1550                         p->majflt_raw  = majflt * (p->stat_collected_usec - p->last_stat_collected_usec) / (1000000ULL * RATES_DETAIL);
1551                         p->cutime_raw = p->cstime_raw = p->cgtime_raw = p->cminflt_raw = p->cmajflt_raw = 0;
1552
1553                         if(unlikely(debug))
1554                                 fprintf(stderr, "\n");
1555                 }
1556                 else if(unlikely(debug)) {
1557                         fprintf(stderr, " > totally absorbed - DONE - %s (%d %s)\n"
1558                                 , p->comm
1559                                 , p->pid
1560                                 , p->updated?"running":"exited"
1561                                 );
1562                 }
1563         }
1564 }
1565
1566 void link_all_processes_to_their_parents(void) {
1567         struct pid_stat *p, *pp;
1568
1569         // link all children to their parents
1570         // and update children count on parents
1571         for(p = root_of_pids; p ; p = p->next) {
1572                 // for each process found
1573
1574                 p->sortlist = 0;
1575                 p->parent = NULL;
1576
1577                 if(unlikely(!p->ppid)) {
1578                         p->parent = NULL;
1579                         continue;
1580                 }
1581
1582                 pp = all_pids[p->ppid];
1583                 if(likely(pp)) {
1584                         p->parent = pp;
1585                         pp->children_count++;
1586
1587                         if(unlikely(debug || (p->target && p->target->debug)))
1588                                 fprintf(stderr, "apps.plugin: \tchild %d (%s, %s) on target '%s' has parent %d (%s, %s). Parent: utime=%llu, stime=%llu, gtime=%llu, minflt=%llu, majflt=%llu, cutime=%llu, cstime=%llu, cgtime=%llu, cminflt=%llu, cmajflt=%llu\n", p->pid, p->comm, p->updated?"running":"exited", (p->target)?p->target->name:"UNSET", pp->pid, pp->comm, pp->updated?"running":"exited", pp->utime, pp->stime, pp->gtime, pp->minflt, pp->majflt, pp->cutime, pp->cstime, pp->cgtime, pp->cminflt, pp->cmajflt);
1589                 }
1590                 else {
1591                         p->parent = NULL;
1592                         error("pid %d %s states parent %d, but the later does not exist.", p->pid, p->comm, p->ppid);
1593                 }
1594         }
1595 }
1596
1597 // ----------------------------------------------------------------------------
1598
1599 // 1. read all files in /proc
1600 // 2. for each numeric directory:
1601 //    i.   read /proc/pid/stat
1602 //    ii.  read /proc/pid/statm
1603 //    iii. read /proc/pid/io (requires root access)
1604 //    iii. read the entries in directory /proc/pid/fd (requires root access)
1605 //         for each entry:
1606 //         a. find or create a struct file_descriptor
1607 //         b. cleanup any old/unused file_descriptors
1608
1609 // after all these, some pids may be linked to targets, while others may not
1610
1611 // in case of errors, only 1 every 1000 errors is printed
1612 // to avoid filling up all disk space
1613 // if debug is enabled, all errors are printed
1614
1615 static int compar_pid(const void *pid1, const void *pid2) {
1616
1617         struct pid_stat *p1 = all_pids[*((pid_t *)pid1)];
1618         struct pid_stat *p2 = all_pids[*((pid_t *)pid2)];
1619
1620         if(p1->sortlist > p2->sortlist)
1621                 return -1;
1622         else
1623                 return 1;
1624 }
1625
1626 void collect_data_for_pid(pid_t pid) {
1627         if(unlikely(pid <= 0 || pid > pid_max)) {
1628                 error("Invalid pid %d read (expected 1 to %d). Ignoring process.", pid, pid_max);
1629                 return;
1630         }
1631
1632         struct pid_stat *p = get_pid_entry(pid);
1633         if(unlikely(!p || p->read)) return;
1634         p->read             = 1;
1635
1636         // fprintf(stderr, "Reading process %d (%s), sortlist %d\n", p->pid, p->comm, p->sortlist);
1637
1638         // --------------------------------------------------------------------
1639         // /proc/<pid>/stat
1640
1641         if(unlikely(read_proc_pid_stat(p))) {
1642                 error("Cannot process %s/proc/%d/stat", host_prefix, pid);
1643                 // there is no reason to proceed if we cannot get its status
1644                 return;
1645         }
1646
1647         read_proc_pid_ownership(p);
1648
1649         // check its parent pid
1650         if(unlikely(p->ppid < 0 || p->ppid > pid_max)) {
1651                 error("Pid %d states invalid parent pid %d. Using 0.", pid, p->ppid);
1652                 p->ppid = 0;
1653         }
1654
1655         // --------------------------------------------------------------------
1656         // /proc/<pid>/io
1657
1658         if(unlikely(read_proc_pid_io(p)))
1659                 error("Cannot process %s/proc/%d/io", host_prefix, pid);
1660
1661         // --------------------------------------------------------------------
1662         // /proc/<pid>/statm
1663
1664         if(unlikely(read_proc_pid_statm(p))) {
1665                 error("Cannot process %s/proc/%d/statm", host_prefix, pid);
1666                 // there is no reason to proceed if we cannot get its memory status
1667                 return;
1668         }
1669
1670         // --------------------------------------------------------------------
1671         // link it
1672
1673         // check if it is target
1674         // we do this only once, the first time this pid is loaded
1675         if(unlikely(p->new_entry)) {
1676                 // /proc/<pid>/cmdline
1677                 if(likely(proc_pid_cmdline_is_needed)) {
1678                         if(unlikely(read_proc_pid_cmdline(p)))
1679                                 error("Cannot process %s/proc/%d/cmdline", host_prefix, pid);
1680                 }
1681
1682                 if(unlikely(debug))
1683                         fprintf(stderr, "apps.plugin: \tJust added %d (%s)\n", pid, p->comm);
1684
1685                 uint32_t hash = simple_hash(p->comm);
1686                 size_t pclen  = strlen(p->comm);
1687
1688                 struct target *w;
1689                 for(w = apps_groups_root_target; w ; w = w->next) {
1690                         // if(debug || (p->target && p->target->debug)) fprintf(stderr, "apps.plugin: \t\tcomparing '%s' with '%s'\n", w->compare, p->comm);
1691
1692                         // find it - 4 cases:
1693                         // 1. the target is not a pattern
1694                         // 2. the target has the prefix
1695                         // 3. the target has the suffix
1696                         // 4. the target is something inside cmdline
1697                         if(     (!w->starts_with && !w->ends_with && w->comparehash == hash && !strcmp(w->compare, p->comm))
1698                                || (w->starts_with && !w->ends_with && !strncmp(w->compare, p->comm, w->comparelen))
1699                                || (!w->starts_with && w->ends_with && pclen >= w->comparelen && !strcmp(w->compare, &p->comm[pclen - w->comparelen]))
1700                                || (proc_pid_cmdline_is_needed && w->starts_with && w->ends_with && strstr(p->cmdline, w->compare))
1701                                         ) {
1702                                 if(w->target) p->target = w->target;
1703                                 else p->target = w;
1704
1705                                 if(debug || (p->target && p->target->debug))
1706                                         fprintf(stderr, "apps.plugin: \t\t%s linked to target %s\n", p->comm, p->target->name);
1707
1708                                 break;
1709                         }
1710                 }
1711         }
1712
1713         // --------------------------------------------------------------------
1714         // /proc/<pid>/fd
1715
1716         if(unlikely(read_pid_file_descriptors(p))) {
1717                 error("Cannot process entries in %s/proc/%d/fd", host_prefix, pid);
1718         }
1719
1720         // --------------------------------------------------------------------
1721         // done!
1722
1723         if(unlikely(debug && include_exited_childs && all_pids_count && p->ppid && all_pids[p->ppid] && !all_pids[p->ppid]->read))
1724                 fprintf(stderr, "Read process %d (%s) sortlisted %d, but its parent %d (%s) sortlisted %d, is not read\n", p->pid, p->comm, p->sortlist, all_pids[p->ppid]->pid, all_pids[p->ppid]->comm, all_pids[p->ppid]->sortlist);
1725
1726         // mark it as updated
1727         p->updated = 1;
1728         p->keep = 0;
1729         p->keeploops = 0;
1730 }
1731
1732 int collect_data_for_all_processes_from_proc(void) {
1733         struct pid_stat *p = NULL;
1734
1735         if(all_pids_count) {
1736                 // read parents before childs
1737                 // this is needed to prevent a situation where
1738                 // a child is found running, but until we read
1739                 // its parent, it has exited and its parent
1740                 // has accumulated its resources
1741
1742                 long slc = 0;
1743                 for(p = root_of_pids; p ; p = p->next) {
1744                         p->read             = 0;
1745                         p->updated          = 0;
1746                         p->new_entry        = 0;
1747                         p->merged           = 0;
1748                         p->children_count   = 0;
1749                         p->parent           = NULL;
1750
1751                         all_pids_sortlist[slc++] = p->pid;
1752                 }
1753
1754                 if(unlikely(slc != all_pids_count)) {
1755                         error("Internal error: I was thinking I had %ld processes in my arrays, but it seems there are more.", all_pids_count);
1756                         all_pids_count = slc;
1757                 }
1758
1759                 if(include_exited_childs) {
1760                         qsort((void *)all_pids_sortlist, all_pids_count, sizeof(pid_t), compar_pid);
1761                         for(slc = 0; slc < all_pids_count; slc++)
1762                                 collect_data_for_pid(all_pids_sortlist[slc]);
1763                 }
1764         }
1765
1766         char dirname[FILENAME_MAX + 1];
1767
1768         snprintfz(dirname, FILENAME_MAX, "%s/proc", host_prefix);
1769         DIR *dir = opendir(dirname);
1770         if(!dir) return 0;
1771
1772         struct dirent *file = NULL;
1773
1774         while((file = readdir(dir))) {
1775                 char *endptr = file->d_name;
1776                 pid_t pid = (pid_t) strtoul(file->d_name, &endptr, 10);
1777
1778                 // make sure we read a valid number
1779                 if(unlikely(endptr == file->d_name || *endptr != '\0'))
1780                         continue;
1781
1782                 collect_data_for_pid(pid);
1783         }
1784         closedir(dir);
1785
1786         // normally this is done
1787         // however we may have processes exited while we collected values
1788         // so let's find the exited ones
1789         // we do this by collecting the ownership of process
1790         // if we manage to get the ownership, the process still runs
1791
1792         read_proc_stat();
1793         link_all_processes_to_their_parents();
1794         process_exited_processes();
1795
1796         return 1;
1797 }
1798
1799 // ----------------------------------------------------------------------------
1800 // update statistics on the targets
1801
1802 // 1. link all childs to their parents
1803 // 2. go from bottom to top, marking as merged all childs to their parents
1804 //    this step links all parents without a target to the child target, if any
1805 // 3. link all top level processes (the ones not merged) to the default target
1806 // 4. go from top to bottom, linking all childs without a target, to their parent target
1807 //    after this step, all processes have a target
1808 // [5. for each killed pid (updated = 0), remove its usage from its target]
1809 // 6. zero all apps_groups_targets
1810 // 7. concentrate all values on the apps_groups_targets
1811 // 8. remove all killed processes
1812 // 9. find the unique file count for each target
1813 // check: update_apps_groups_statistics()
1814
1815 void cleanup_exited_pids(void) {
1816         int c;
1817         struct pid_stat *p = NULL;
1818
1819         for(p = root_of_pids; p ;) {
1820                 if(!p->updated && (!p->keep || p->keeploops > 0)) {
1821 //                      fprintf(stderr, "\tEXITED %d %s [parent %d %s, target %s] utime=%llu, stime=%llu, gtime=%llu, cutime=%llu, cstime=%llu, cgtime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu\n", p->pid, p->comm, p->parent->pid, p->parent->comm, p->target->name,  p->utime, p->stime, p->gtime, p->cutime, p->cstime, p->cgtime, p->minflt, p->majflt, p->cminflt, p->cmajflt);
1822
1823                         if(unlikely(debug && (p->keep || p->keeploops)))
1824                                 fprintf(stderr, " > CLEANUP cannot keep exited process %d (%s) anymore - removing it.\n", p->pid, p->comm);
1825
1826                         for(c = 0 ; c < p->fds_size ; c++) if(p->fds[c] > 0) {
1827                                 file_descriptor_not_used(p->fds[c]);
1828                                 p->fds[c] = 0;
1829                         }
1830
1831                         pid_t r = p->pid;
1832                         p = p->next;
1833                         del_pid_entry(r);
1834                 }
1835                 else {
1836                         if(unlikely(p->keep)) p->keeploops++;
1837                         p->keep = 0;
1838                         p = p->next;
1839                 }
1840         }
1841 }
1842
1843 void apply_apps_groups_targets_inheritance(void) {
1844         struct pid_stat *p = NULL;
1845
1846         // children that do not have a target
1847         // inherit their target from their parent
1848         int found = 1, loops = 0;
1849         while(found) {
1850                 if(unlikely(debug)) loops++;
1851                 found = 0;
1852                 for(p = root_of_pids; p ; p = p->next) {
1853                         // if this process does not have a target
1854                         // and it has a parent
1855                         // and its parent has a target
1856                         // then, set the parent's target to this process
1857                         if(unlikely(!p->target && p->parent && p->parent->target)) {
1858                                 p->target = p->parent->target;
1859                                 found++;
1860
1861                                 if(debug || (p->target && p->target->debug))
1862                                         fprintf(stderr, "apps.plugin: \t\tTARGET INHERITANCE: %s is inherited by %d (%s) from its parent %d (%s).\n", p->target->name, p->pid, p->comm, p->parent->pid, p->parent->comm);
1863                         }
1864                 }
1865         }
1866
1867         // find all the procs with 0 childs and merge them to their parents
1868         // repeat, until nothing more can be done.
1869         int sortlist = 1;
1870         found = 1;
1871         while(found) {
1872                 if(unlikely(debug)) loops++;
1873                 found = 0;
1874
1875                 for(p = root_of_pids; p ; p = p->next) {
1876                         if(unlikely(!p->sortlist && !p->children_count))
1877                                 p->sortlist = sortlist++;
1878
1879                         // if this process does not have any children
1880                         // and is not already merged
1881                         // and has a parent
1882                         // and its parent has children
1883                         // and the target of this process and its parent is the same, or the parent does not have a target
1884                         // and its parent is not init
1885                         // then, mark them as merged.
1886                         if(unlikely(
1887                                         !p->children_count
1888                                         && !p->merged
1889                                         && p->parent
1890                                         && p->parent->children_count
1891                                         && (p->target == p->parent->target || !p->parent->target)
1892                                         && p->ppid != 1
1893                                 )) {
1894                                 p->parent->children_count--;
1895                                 p->merged = 1;
1896
1897                                 // the parent inherits the child's target, if it does not have a target itself
1898                                 if(unlikely(p->target && !p->parent->target)) {
1899                                         p->parent->target = p->target;
1900
1901                                         if(debug || (p->target && p->target->debug))
1902                                                 fprintf(stderr, "apps.plugin: \t\tTARGET INHERITANCE: %s is inherited by %d (%s) from its child %d (%s).\n", p->target->name, p->parent->pid, p->parent->comm, p->pid, p->comm);
1903                                 }
1904
1905                                 found++;
1906                         }
1907                 }
1908
1909                 if(unlikely(debug))
1910                         fprintf(stderr, "apps.plugin: TARGET INHERITANCE: merged %d processes\n", found);
1911         }
1912
1913         // init goes always to default target
1914         if(all_pids[1])
1915                 all_pids[1]->target = apps_groups_default_target;
1916
1917         // give a default target on all top level processes
1918         if(unlikely(debug)) loops++;
1919         for(p = root_of_pids; p ; p = p->next) {
1920                 // if the process is not merged itself
1921                 // then is is a top level process
1922                 if(unlikely(!p->merged && !p->target))
1923                         p->target = apps_groups_default_target;
1924
1925                 // make sure all processes have a sortlist
1926                 if(unlikely(!p->sortlist))
1927                         p->sortlist = sortlist++;
1928         }
1929
1930         if(all_pids[1])
1931                 all_pids[1]->sortlist = sortlist++;
1932
1933         // give a target to all merged child processes
1934         found = 1;
1935         while(found) {
1936                 if(unlikely(debug)) loops++;
1937                 found = 0;
1938                 for(p = root_of_pids; p ; p = p->next) {
1939                         if(unlikely(!p->target && p->merged && p->parent && p->parent->target)) {
1940                                 p->target = p->parent->target;
1941                                 found++;
1942
1943                                 if(debug || (p->target && p->target->debug))
1944                                         fprintf(stderr, "apps.plugin: \t\tTARGET INHERITANCE: %s is inherited by %d (%s) from its parent %d (%s) at phase 2.\n", p->target->name, p->pid, p->comm, p->parent->pid, p->parent->comm);
1945                         }
1946                 }
1947         }
1948
1949         if(unlikely(debug))
1950                 fprintf(stderr, "apps.plugin: apply_apps_groups_targets_inheritance() made %d loops on the process tree\n", loops);
1951 }
1952
1953 long zero_all_targets(struct target *root) {
1954         struct target *w;
1955         long count = 0;
1956
1957         for (w = root; w ; w = w->next) {
1958                 count++;
1959
1960                 if(w->fds) free(w->fds);
1961                 w->fds = NULL;
1962
1963                 w->minflt = 0;
1964                 w->majflt = 0;
1965                 w->utime = 0;
1966                 w->stime = 0;
1967                 w->gtime = 0;
1968                 w->cminflt = 0;
1969                 w->cmajflt = 0;
1970                 w->cutime = 0;
1971                 w->cstime = 0;
1972                 w->cgtime = 0;
1973                 w->num_threads = 0;
1974                 w->rss = 0;
1975                 w->processes = 0;
1976
1977                 w->statm_size = 0;
1978                 w->statm_resident = 0;
1979                 w->statm_share = 0;
1980                 w->statm_text = 0;
1981                 w->statm_lib = 0;
1982                 w->statm_data = 0;
1983                 w->statm_dirty = 0;
1984
1985                 w->io_logical_bytes_read = 0;
1986                 w->io_logical_bytes_written = 0;
1987                 w->io_read_calls = 0;
1988                 w->io_write_calls = 0;
1989                 w->io_storage_bytes_read = 0;
1990                 w->io_storage_bytes_written = 0;
1991                 w->io_cancelled_write_bytes = 0;
1992         }
1993
1994         return count;
1995 }
1996
1997 void aggregate_pid_on_target(struct target *w, struct pid_stat *p, struct target *o) {
1998         (void)o;
1999
2000         if(unlikely(!w->fds)) {
2001                 w->fds = calloc(sizeof(int), (size_t) all_files_size);
2002                 if(unlikely(!w->fds))
2003                         error("Cannot allocate memory for fds in %s", w->name);
2004         }
2005
2006         if(likely(p->updated)) {
2007                 w->cutime  += p->cutime;
2008                 w->cstime  += p->cstime;
2009                 w->cgtime  += p->cgtime;
2010                 w->cminflt += p->cminflt;
2011                 w->cmajflt += p->cmajflt;
2012
2013                 w->utime  += p->utime;
2014                 w->stime  += p->stime;
2015                 w->gtime  += p->gtime;
2016                 w->minflt += p->minflt;
2017                 w->majflt += p->majflt;
2018
2019                 w->rss += p->rss;
2020
2021                 w->statm_size += p->statm_size;
2022                 w->statm_resident += p->statm_resident;
2023                 w->statm_share += p->statm_share;
2024                 w->statm_text += p->statm_text;
2025                 w->statm_lib += p->statm_lib;
2026                 w->statm_data += p->statm_data;
2027                 w->statm_dirty += p->statm_dirty;
2028
2029                 w->io_logical_bytes_read    += p->io_logical_bytes_read;
2030                 w->io_logical_bytes_written += p->io_logical_bytes_written;
2031                 w->io_read_calls            += p->io_read_calls;
2032                 w->io_write_calls           += p->io_write_calls;
2033                 w->io_storage_bytes_read    += p->io_storage_bytes_read;
2034                 w->io_storage_bytes_written += p->io_storage_bytes_written;
2035                 w->io_cancelled_write_bytes += p->io_cancelled_write_bytes;
2036
2037                 w->processes++;
2038                 w->num_threads += p->num_threads;
2039
2040                 if(likely(w->fds)) {
2041                         int c;
2042                         for(c = 0; c < p->fds_size ;c++) {
2043                                 if(p->fds[c] == 0) continue;
2044
2045                                 if(likely(p->fds[c] < all_files_size)) {
2046                                         if(w->fds) w->fds[p->fds[c]]++;
2047                                 }
2048                                 else
2049                                         error("Invalid fd number %d", p->fds[c]);
2050                         }
2051                 }
2052
2053                 if(unlikely(debug || w->debug))
2054                         fprintf(stderr, "apps.plugin: \taggregating '%s' pid %d on target '%s' utime=%llu, stime=%llu, gtime=%llu, cutime=%llu, cstime=%llu, cgtime=%llu, minflt=%llu, majflt=%llu, cminflt=%llu, cmajflt=%llu\n", p->comm, p->pid, w->name, p->utime, p->stime, p->gtime, p->cutime, p->cstime, p->cgtime, p->minflt, p->majflt, p->cminflt, p->cmajflt);
2055         }
2056 }
2057
2058 void count_targets_fds(struct target *root) {
2059         int c;
2060         struct target *w;
2061
2062         for (w = root; w ; w = w->next) {
2063                 if(!w->fds) continue;
2064
2065                 w->openfiles = 0;
2066                 w->openpipes = 0;
2067                 w->opensockets = 0;
2068                 w->openinotifies = 0;
2069                 w->openeventfds = 0;
2070                 w->opentimerfds = 0;
2071                 w->opensignalfds = 0;
2072                 w->openeventpolls = 0;
2073                 w->openother = 0;
2074
2075                 for(c = 1; c < all_files_size ;c++) {
2076                         if(w->fds[c] > 0)
2077                                 switch(all_files[c].type) {
2078                                 case FILETYPE_FILE:
2079                                         w->openfiles++;
2080                                         break;
2081
2082                                 case FILETYPE_PIPE:
2083                                         w->openpipes++;
2084                                         break;
2085
2086                                 case FILETYPE_SOCKET:
2087                                         w->opensockets++;
2088                                         break;
2089
2090                                 case FILETYPE_INOTIFY:
2091                                         w->openinotifies++;
2092                                         break;
2093
2094                                 case FILETYPE_EVENTFD:
2095                                         w->openeventfds++;
2096                                         break;
2097
2098                                 case FILETYPE_TIMERFD:
2099                                         w->opentimerfds++;
2100                                         break;
2101
2102                                 case FILETYPE_SIGNALFD:
2103                                         w->opensignalfds++;
2104                                         break;
2105
2106                                 case FILETYPE_EVENTPOLL:
2107                                         w->openeventpolls++;
2108                                         break;
2109
2110                                 default:
2111                                         w->openother++;
2112                         }
2113                 }
2114
2115                 free(w->fds);
2116                 w->fds = NULL;
2117         }
2118 }
2119
2120 void calculate_netdata_statistics(void) {
2121         apply_apps_groups_targets_inheritance();
2122
2123         zero_all_targets(users_root_target);
2124         zero_all_targets(groups_root_target);
2125         apps_groups_targets = zero_all_targets(apps_groups_root_target);
2126
2127         // this has to be done, before the cleanup
2128         struct pid_stat *p = NULL;
2129         struct target *w = NULL, *o = NULL;
2130
2131         // concentrate everything on the apps_groups_targets
2132         for(p = root_of_pids; p ; p = p->next) {
2133
2134                 // --------------------------------------------------------------------
2135                 // apps_groups targets
2136                 if(likely(p->target))
2137                         aggregate_pid_on_target(p->target, p, NULL);
2138                 else
2139                         error("pid %d %s was left without a target!", p->pid, p->comm);
2140
2141
2142                 // --------------------------------------------------------------------
2143                 // user targets
2144                 o = p->user_target;
2145                 if(likely(p->user_target && p->user_target->uid == p->uid))
2146                         w = p->user_target;
2147                 else {
2148                         if(unlikely(debug && p->user_target))
2149                                         fprintf(stderr, "apps.plugin: \t\tpid %d (%s) switched user from %u (%s) to %u.\n", p->pid, p->comm, p->user_target->uid, p->user_target->name, p->uid);
2150
2151                         w = p->user_target = get_users_target(p->uid);
2152                 }
2153
2154                 if(likely(w))
2155                         aggregate_pid_on_target(w, p, o);
2156                 else
2157                         error("pid %d %s was left without a user target!", p->pid, p->comm);
2158
2159
2160                 // --------------------------------------------------------------------
2161                 // group targets
2162                 o = p->group_target;
2163                 if(likely(p->group_target && p->group_target->gid == p->gid))
2164                         w = p->group_target;
2165                 else {
2166                         if(unlikely(debug && p->group_target))
2167                                         fprintf(stderr, "apps.plugin: \t\tpid %d (%s) switched group from %u (%s) to %u.\n", p->pid, p->comm, p->group_target->gid, p->group_target->name, p->gid);
2168
2169                         w = p->group_target = get_groups_target(p->gid);
2170                 }
2171
2172                 if(likely(w))
2173                         aggregate_pid_on_target(w, p, o);
2174                 else
2175                         error("pid %d %s was left without a group target!", p->pid, p->comm);
2176
2177         }
2178
2179         count_targets_fds(apps_groups_root_target);
2180         count_targets_fds(users_root_target);
2181         count_targets_fds(groups_root_target);
2182
2183         cleanup_exited_pids();
2184 }
2185
2186 // ----------------------------------------------------------------------------
2187 // update chart dimensions
2188
2189 BUFFER *output = NULL;
2190 int print_calculated_number(char *str, calculated_number value) { (void)str; (void)value; return 0; }
2191
2192 static inline void send_BEGIN(const char *type, const char *id, unsigned long long usec) {
2193         // fprintf(stdout, "BEGIN %s.%s %llu\n", type, id, usec);
2194         buffer_strcat(output, "BEGIN ");
2195         buffer_strcat(output, type);
2196         buffer_strcat(output, ".");
2197         buffer_strcat(output, id);
2198         buffer_strcat(output, " ");
2199         buffer_print_llu(output, usec);
2200         buffer_strcat(output, "\n");
2201 }
2202
2203 static inline void send_SET(const char *name, unsigned long long value) {
2204         // fprintf(stdout, "SET %s = %llu\n", name, value);
2205         buffer_strcat(output, "SET ");
2206         buffer_strcat(output, name);
2207         buffer_strcat(output, " = ");
2208         buffer_print_llu(output, value);
2209         buffer_strcat(output, "\n");
2210 }
2211
2212 static inline void send_END(void) {
2213         // fprintf(stdout, "END\n");
2214         buffer_strcat(output, "END\n");
2215 }
2216
2217 double utime_fix_ratio = 1.0, stime_fix_ratio = 1.0, gtime_fix_ratio = 1.0, cutime_fix_ratio = 1.0, cstime_fix_ratio = 1.0, cgtime_fix_ratio = 1.0;
2218 double minflt_fix_ratio = 1.0, majflt_fix_ratio = 1.0, cminflt_fix_ratio = 1.0, cmajflt_fix_ratio = 1.0;
2219
2220 unsigned long long send_resource_usage_to_netdata() {
2221         static struct timeval last = { 0, 0 };
2222         static struct rusage me_last;
2223
2224         struct timeval now;
2225         struct rusage me;
2226
2227         unsigned long long usec;
2228         unsigned long long cpuuser;
2229         unsigned long long cpusyst;
2230
2231         if(!last.tv_sec) {
2232                 gettimeofday(&last, NULL);
2233                 getrusage(RUSAGE_SELF, &me_last);
2234
2235                 // the first time, give a zero to allow
2236                 // netdata calibrate to the current time
2237                 // usec = update_every * 1000000ULL;
2238                 usec = 0ULL;
2239                 cpuuser = 0;
2240                 cpusyst = 0;
2241         }
2242         else {
2243                 gettimeofday(&now, NULL);
2244                 getrusage(RUSAGE_SELF, &me);
2245
2246                 usec = usecdiff(&now, &last);
2247                 cpuuser = me.ru_utime.tv_sec * 1000000ULL + me.ru_utime.tv_usec;
2248                 cpusyst = me.ru_stime.tv_sec * 1000000ULL + me.ru_stime.tv_usec;
2249
2250                 bcopy(&now, &last, sizeof(struct timeval));
2251                 bcopy(&me, &me_last, sizeof(struct rusage));
2252         }
2253
2254         buffer_sprintf(output,
2255                 "BEGIN netdata.apps_cpu %llu\n"
2256                 "SET user = %llu\n"
2257                 "SET system = %llu\n"
2258                 "END\n"
2259                 "BEGIN netdata.apps_files %llu\n"
2260                 "SET files = %llu\n"
2261                 "SET pids = %ld\n"
2262                 "SET fds = %d\n"
2263                 "SET targets = %ld\n"
2264                 "END\n"
2265                 "BEGIN netdata.apps_fix %llu\n"
2266                 "SET utime = %llu\n"
2267                 "SET stime = %llu\n"
2268                 "SET gtime = %llu\n"
2269                 "SET minflt = %llu\n"
2270                 "SET majflt = %llu\n"
2271                 "END\n"
2272                 , usec
2273                 , cpuuser
2274                 , cpusyst
2275                 , usec
2276                 , file_counter
2277                 , all_pids_count
2278                 , all_files_len
2279                 , apps_groups_targets
2280                 , usec
2281                 , (unsigned long long)(utime_fix_ratio   * 100 * RATES_DETAIL)
2282                 , (unsigned long long)(stime_fix_ratio   * 100 * RATES_DETAIL)
2283                 , (unsigned long long)(gtime_fix_ratio   * 100 * RATES_DETAIL)
2284                 , (unsigned long long)(minflt_fix_ratio  * 100 * RATES_DETAIL)
2285                 , (unsigned long long)(majflt_fix_ratio  * 100 * RATES_DETAIL)
2286                 );
2287
2288         if(include_exited_childs)
2289                 buffer_sprintf(output,
2290                         "BEGIN netdata.apps_children_fix %llu\n"
2291                         "SET cutime = %llu\n"
2292                         "SET cstime = %llu\n"
2293                         "SET cgtime = %llu\n"
2294                         "SET cminflt = %llu\n"
2295                         "SET cmajflt = %llu\n"
2296                         "END\n"
2297                         , usec
2298                         , (unsigned long long)(cutime_fix_ratio  * 100 * RATES_DETAIL)
2299                         , (unsigned long long)(cstime_fix_ratio  * 100 * RATES_DETAIL)
2300                         , (unsigned long long)(cgtime_fix_ratio  * 100 * RATES_DETAIL)
2301                         , (unsigned long long)(cminflt_fix_ratio * 100 * RATES_DETAIL)
2302                         , (unsigned long long)(cmajflt_fix_ratio * 100 * RATES_DETAIL)
2303                         );
2304
2305         return usec;
2306 }
2307
2308 void normalize_data(struct target *root) {
2309         struct target *w;
2310
2311         // childs processing introduces spikes
2312         // here we try to eliminate them by disabling childs processing either for specific dimensions
2313         // or entirely. Of course, either way, we disable it just a single iteration.
2314
2315         unsigned long long max = processors * hz * RATES_DETAIL;
2316         unsigned long long utime = 0, cutime = 0, stime = 0, cstime = 0, gtime = 0, cgtime = 0, minflt = 0, cminflt = 0, majflt = 0, cmajflt = 0;
2317
2318         if(global_utime > max) global_utime = max;
2319         if(global_stime > max) global_stime = max;
2320         if(global_gtime > max) global_gtime = max;
2321
2322         for(w = root; w ; w = w->next) {
2323                 if(w->target || (!w->processes && !w->exposed)) continue;
2324
2325                 utime   += w->utime;
2326                 stime   += w->stime;
2327                 gtime   += w->gtime;
2328                 cutime  += w->cutime;
2329                 cstime  += w->cstime;
2330                 cgtime  += w->cgtime;
2331
2332                 minflt  += w->minflt;
2333                 majflt  += w->majflt;
2334                 cminflt += w->cminflt;
2335                 cmajflt += w->cmajflt;
2336         }
2337
2338         if((global_utime || global_stime || global_gtime) && (utime || stime || gtime)) {
2339                 if(global_utime + global_stime + global_gtime > utime + cutime + stime + cstime + gtime + cgtime) {
2340                         // everything we collected fits
2341                         utime_fix_ratio  =
2342                         stime_fix_ratio  =
2343                         gtime_fix_ratio  =
2344                         cutime_fix_ratio =
2345                         cstime_fix_ratio =
2346                         cgtime_fix_ratio = 1.0; //(double)(global_utime + global_stime) / (double)(utime + cutime + stime + cstime);
2347                 }
2348                 else if(global_utime + global_stime > utime + stime) {
2349                         // childrens resources are too high
2350                         // lower only the children resources
2351                         utime_fix_ratio  =
2352                         stime_fix_ratio  =
2353                         gtime_fix_ratio  = 1.0;
2354                         cutime_fix_ratio =
2355                         cstime_fix_ratio =
2356                         cgtime_fix_ratio = (double)((global_utime + global_stime) - (utime + stime)) / (double)(cutime + cstime);
2357                 }
2358                 else {
2359                         // even running processes are unrealistic
2360                         // zero the children resources
2361                         // lower the running processes resources
2362                         utime_fix_ratio  =
2363                         stime_fix_ratio  =
2364                         gtime_fix_ratio  = (double)(global_utime + global_stime) / (double)(utime + stime);
2365                         cutime_fix_ratio =
2366                         cstime_fix_ratio =
2367                         cgtime_fix_ratio = 0.0;
2368                 }
2369         }
2370         else {
2371                 utime_fix_ratio  =
2372                 stime_fix_ratio  =
2373                 gtime_fix_ratio  =
2374                 cutime_fix_ratio =
2375                 cstime_fix_ratio =
2376                 cgtime_fix_ratio = 0.0;
2377         }
2378
2379         if(utime_fix_ratio  > 1.0) utime_fix_ratio  = 1.0;
2380         if(cutime_fix_ratio > 1.0) cutime_fix_ratio = 1.0;
2381         if(stime_fix_ratio  > 1.0) stime_fix_ratio  = 1.0;
2382         if(cstime_fix_ratio > 1.0) cstime_fix_ratio = 1.0;
2383         if(gtime_fix_ratio  > 1.0) gtime_fix_ratio  = 1.0;
2384         if(cgtime_fix_ratio > 1.0) cgtime_fix_ratio = 1.0;
2385
2386         // if(utime_fix_ratio  < 0.0) utime_fix_ratio  = 0.0;
2387         // if(cutime_fix_ratio < 0.0) cutime_fix_ratio = 0.0;
2388         // if(stime_fix_ratio  < 0.0) stime_fix_ratio  = 0.0;
2389         // if(cstime_fix_ratio < 0.0) cstime_fix_ratio = 0.0;
2390         // if(gtime_fix_ratio  < 0.0) gtime_fix_ratio  = 0.0;
2391         // if(cgtime_fix_ratio < 0.0) cgtime_fix_ratio = 0.0;
2392
2393         // FIXME
2394         // we use cpu time to normalize page faults
2395         // the problem is that to find the proper max values
2396         // for page faults we have to parse /proc/vmstat
2397         // which is quite big to do it again (netdata does it already)
2398         //
2399         // a better solution could be to somehow have netdata
2400         // do this normalization for us
2401
2402         if(utime || stime || gtime)
2403                 majflt_fix_ratio =
2404                 minflt_fix_ratio = (double)(utime * utime_fix_ratio + stime * stime_fix_ratio + gtime * gtime_fix_ratio) / (double)(utime + stime + gtime);
2405         else
2406                 minflt_fix_ratio =
2407                 majflt_fix_ratio = 1.0;
2408
2409         if(cutime || cstime || cgtime)
2410                 cmajflt_fix_ratio =
2411                 cminflt_fix_ratio = (double)(cutime * cutime_fix_ratio + cstime * cstime_fix_ratio + cgtime * cgtime_fix_ratio) / (double)(cutime + cstime + cgtime);
2412         else
2413                 cminflt_fix_ratio =
2414                 cmajflt_fix_ratio = 1.0;
2415
2416         // the report
2417
2418         if(unlikely(debug)) {
2419                 fprintf(stderr,
2420                         "SYSTEM: u=%llu s=%llu g=%llu "
2421                         "COLLECTED: u=%llu s=%llu g=%llu cu=%llu cs=%llu cg=%llu "
2422                         "DELTA: u=%lld s=%lld g=%lld "
2423                         "FIX: u=%0.2f s=%0.2f g=%0.2f cu=%0.2f cs=%0.2f cg=%0.2f "
2424                         "FINALLY: u=%llu s=%llu g=%llu cu=%llu cs=%llu cg=%llu "
2425                         "\n"
2426                         , global_utime
2427                         , global_stime
2428                         , global_gtime
2429                         , utime
2430                         , stime
2431                         , gtime
2432                         , cutime
2433                         , cstime
2434                         , cgtime
2435                         , (long long)utime + (long long)cutime - (long long)global_utime
2436                         , (long long)stime + (long long)cstime - (long long)global_stime
2437                         , (long long)gtime + (long long)cgtime - (long long)global_gtime
2438                         , utime_fix_ratio
2439                         , stime_fix_ratio
2440                         , gtime_fix_ratio
2441                         , cutime_fix_ratio
2442                         , cstime_fix_ratio
2443                         , cgtime_fix_ratio
2444                         , (unsigned long long)(utime * utime_fix_ratio)
2445                         , (unsigned long long)(stime * stime_fix_ratio)
2446                         , (unsigned long long)(gtime * gtime_fix_ratio)
2447                         , (unsigned long long)(cutime * cutime_fix_ratio)
2448                         , (unsigned long long)(cstime * cstime_fix_ratio)
2449                         , (unsigned long long)(cgtime * cgtime_fix_ratio)
2450                         );
2451         }
2452 }
2453
2454 void send_collected_data_to_netdata(struct target *root, const char *type, unsigned long long usec) {
2455         struct target *w;
2456
2457         send_BEGIN(type, "cpu", usec);
2458         for (w = root; w ; w = w->next) {
2459                 if(unlikely(w->exposed))
2460                         send_SET(w->name, (unsigned long long)(w->utime * utime_fix_ratio) + (unsigned long long)(w->stime * stime_fix_ratio) + (unsigned long long)(w->gtime * gtime_fix_ratio) + (include_exited_childs?((unsigned long long)(w->cutime * cutime_fix_ratio) + (unsigned long long)(w->cstime * cstime_fix_ratio) + (unsigned long long)(w->cgtime * cgtime_fix_ratio)):0ULL));
2461         }
2462         send_END();
2463
2464         send_BEGIN(type, "cpu_user", usec);
2465         for (w = root; w ; w = w->next) {
2466                 if(unlikely(w->exposed))
2467                         send_SET(w->name, (unsigned long long)(w->utime * utime_fix_ratio) + (include_exited_childs?((unsigned long long)(w->cutime * cutime_fix_ratio)):0ULL));
2468         }
2469         send_END();
2470
2471         send_BEGIN(type, "cpu_system", usec);
2472         for (w = root; w ; w = w->next) {
2473                 if(unlikely(w->exposed))
2474                         send_SET(w->name, (unsigned long long)(w->stime * stime_fix_ratio) + (include_exited_childs?((unsigned long long)(w->cstime * cstime_fix_ratio)):0ULL));
2475         }
2476         send_END();
2477
2478         if(show_guest_time) {
2479                 send_BEGIN(type, "cpu_guest", usec);
2480                 for (w = root; w ; w = w->next) {
2481                         if(unlikely(w->exposed))
2482                                 send_SET(w->name, (unsigned long long)(w->gtime * gtime_fix_ratio) + (include_exited_childs?((unsigned long long)(w->cgtime * cgtime_fix_ratio)):0ULL));
2483                 }
2484                 send_END();
2485         }
2486
2487         send_BEGIN(type, "threads", usec);
2488         for (w = root; w ; w = w->next) {
2489                 if(unlikely(w->exposed))
2490                         send_SET(w->name, w->num_threads);
2491         }
2492         send_END();
2493
2494         send_BEGIN(type, "processes", usec);
2495         for (w = root; w ; w = w->next) {
2496                 if(unlikely(w->exposed))
2497                         send_SET(w->name, w->processes);
2498         }
2499         send_END();
2500
2501         send_BEGIN(type, "mem", usec);
2502         for (w = root; w ; w = w->next) {
2503                 if(unlikely(w->exposed))
2504                         send_SET(w->name, (w->statm_resident > w->statm_share)?(w->statm_resident - w->statm_share):0ULL);
2505         }
2506         send_END();
2507
2508         send_BEGIN(type, "minor_faults", usec);
2509         for (w = root; w ; w = w->next) {
2510                 if(unlikely(w->exposed))
2511                         send_SET(w->name, (unsigned long long)(w->minflt * minflt_fix_ratio) + (include_exited_childs?((unsigned long long)(w->cminflt * cminflt_fix_ratio)):0ULL));
2512         }
2513         send_END();
2514
2515         send_BEGIN(type, "major_faults", usec);
2516         for (w = root; w ; w = w->next) {
2517                 if(unlikely(w->exposed))
2518                         send_SET(w->name, (unsigned long long)(w->majflt * majflt_fix_ratio) + (include_exited_childs?((unsigned long long)(w->cmajflt * cmajflt_fix_ratio)):0ULL));
2519         }
2520         send_END();
2521
2522         send_BEGIN(type, "lreads", usec);
2523         for (w = root; w ; w = w->next) {
2524                 if(unlikely(w->exposed))
2525                         send_SET(w->name, w->io_logical_bytes_read);
2526         }
2527         send_END();
2528
2529         send_BEGIN(type, "lwrites", usec);
2530         for (w = root; w ; w = w->next) {
2531                 if(unlikely(w->exposed))
2532                         send_SET(w->name, w->io_logical_bytes_written);
2533         }
2534         send_END();
2535
2536         send_BEGIN(type, "preads", usec);
2537         for (w = root; w ; w = w->next) {
2538                 if(unlikely(w->exposed))
2539                         send_SET(w->name, w->io_storage_bytes_read);
2540         }
2541         send_END();
2542
2543         send_BEGIN(type, "pwrites", usec);
2544         for (w = root; w ; w = w->next) {
2545                 if(unlikely(w->exposed))
2546                         send_SET(w->name, w->io_storage_bytes_written);
2547         }
2548         send_END();
2549
2550         send_BEGIN(type, "files", usec);
2551         for (w = root; w ; w = w->next) {
2552                 if(unlikely(w->exposed))
2553                         send_SET(w->name, w->openfiles);
2554         }
2555         send_END();
2556
2557         send_BEGIN(type, "sockets", usec);
2558         for (w = root; w ; w = w->next) {
2559                 if(unlikely(w->exposed))
2560                         send_SET(w->name, w->opensockets);
2561         }
2562         send_END();
2563
2564         send_BEGIN(type, "pipes", usec);
2565         for (w = root; w ; w = w->next) {
2566                 if(unlikely(w->exposed))
2567                         send_SET(w->name, w->openpipes);
2568         }
2569         send_END();
2570 }
2571
2572
2573 // ----------------------------------------------------------------------------
2574 // generate the charts
2575
2576 void send_charts_updates_to_netdata(struct target *root, const char *type, const char *title)
2577 {
2578         struct target *w;
2579         int newly_added = 0;
2580
2581         for(w = root ; w ; w = w->next) {
2582                 if (w->target) continue;
2583
2584                 if (!w->exposed && w->processes) {
2585                         newly_added++;
2586                         w->exposed = 1;
2587                         if (debug || w->debug) fprintf(stderr, "apps.plugin: %s just added - regenerating charts.\n", w->name);
2588                 }
2589         }
2590
2591         // nothing more to show
2592         if(!newly_added && show_guest_time == show_guest_time_old) return;
2593
2594         // we have something new to show
2595         // update the charts
2596         buffer_sprintf(output, "CHART %s.cpu '' '%s CPU Time (%d%% = %d core%s)' 'cpu time %%' cpu %s.cpu stacked 20001 %d\n", type, title, (processors * 100), processors, (processors>1)?"s":"", type, update_every);
2597         for (w = root; w ; w = w->next) {
2598                 if(unlikely(w->exposed))
2599                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu %s\n", w->name, hz * RATES_DETAIL / 100, w->hidden ? "hidden" : "");
2600         }
2601
2602         buffer_sprintf(output, "CHART %s.mem '' '%s Dedicated Memory (w/o shared)' 'MB' mem %s.mem stacked 20003 %d\n", type, title, type, update_every);
2603         for (w = root; w ; w = w->next) {
2604                 if(unlikely(w->exposed))
2605                         buffer_sprintf(output, "DIMENSION %s '' absolute %ld %ld\n", w->name, sysconf(_SC_PAGESIZE), 1024L*1024L);
2606         }
2607
2608         buffer_sprintf(output, "CHART %s.threads '' '%s Threads' 'threads' processes %s.threads stacked 20005 %d\n", type, title, type, update_every);
2609         for (w = root; w ; w = w->next) {
2610                 if(unlikely(w->exposed))
2611                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 1\n", w->name);
2612         }
2613
2614         buffer_sprintf(output, "CHART %s.processes '' '%s Processes' 'processes' processes %s.processes stacked 20004 %d\n", type, title, type, update_every);
2615         for (w = root; w ; w = w->next) {
2616                 if(unlikely(w->exposed))
2617                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 1\n", w->name);
2618         }
2619
2620         buffer_sprintf(output, "CHART %s.cpu_user '' '%s CPU User Time (%d%% = %d core%s)' 'cpu time %%' cpu %s.cpu_user stacked 20020 %d\n", type, title, (processors * 100), processors, (processors>1)?"s":"", type, update_every);
2621         for (w = root; w ; w = w->next) {
2622                 if(unlikely(w->exposed))
2623                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, hz * RATES_DETAIL / 100LLU);
2624         }
2625
2626         buffer_sprintf(output, "CHART %s.cpu_system '' '%s CPU System Time (%d%% = %d core%s)' 'cpu time %%' cpu %s.cpu_system stacked 20021 %d\n", type, title, (processors * 100), processors, (processors>1)?"s":"", type, update_every);
2627         for (w = root; w ; w = w->next) {
2628                 if(unlikely(w->exposed))
2629                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, hz * RATES_DETAIL / 100LLU);
2630         }
2631
2632         if(show_guest_time) {
2633                 buffer_sprintf(output, "CHART %s.cpu_guest '' '%s CPU Guest Time (%d%% = %d core%s)' 'cpu time %%' cpu %s.cpu_system stacked 20022 %d\n", type, title, (processors * 100), processors, (processors > 1) ? "s" : "", type, update_every);
2634                 for (w = root; w; w = w->next) {
2635                         if(unlikely(w->exposed))
2636                                 buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, hz * RATES_DETAIL / 100LLU);
2637                 }
2638         }
2639
2640         buffer_sprintf(output, "CHART %s.major_faults '' '%s Major Page Faults (swap read)' 'page faults/s' swap %s.major_faults stacked 20010 %d\n", type, title, type, update_every);
2641         for (w = root; w ; w = w->next) {
2642                 if(unlikely(w->exposed))
2643                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, RATES_DETAIL);
2644         }
2645
2646         buffer_sprintf(output, "CHART %s.minor_faults '' '%s Minor Page Faults' 'page faults/s' mem %s.minor_faults stacked 20011 %d\n", type, title, type, update_every);
2647         for (w = root; w ; w = w->next) {
2648                 if(unlikely(w->exposed))
2649                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, RATES_DETAIL);
2650         }
2651
2652         buffer_sprintf(output, "CHART %s.lreads '' '%s Disk Logical Reads' 'kilobytes/s' disk %s.lreads stacked 20042 %d\n", type, title, type, update_every);
2653         for (w = root; w ; w = w->next) {
2654                 if(unlikely(w->exposed))
2655                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, 1024LLU * RATES_DETAIL);
2656         }
2657
2658         buffer_sprintf(output, "CHART %s.lwrites '' '%s I/O Logical Writes' 'kilobytes/s' disk %s.lwrites stacked 20042 %d\n", type, title, type, update_every);
2659         for (w = root; w ; w = w->next) {
2660                 if(unlikely(w->exposed))
2661                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, 1024LLU * RATES_DETAIL);
2662         }
2663
2664         buffer_sprintf(output, "CHART %s.preads '' '%s Disk Reads' 'kilobytes/s' disk %s.preads stacked 20002 %d\n", type, title, type, update_every);
2665         for (w = root; w ; w = w->next) {
2666                 if(unlikely(w->exposed))
2667                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, 1024LLU * RATES_DETAIL);
2668         }
2669
2670         buffer_sprintf(output, "CHART %s.pwrites '' '%s Disk Writes' 'kilobytes/s' disk %s.pwrites stacked 20002 %d\n", type, title, type, update_every);
2671         for (w = root; w ; w = w->next) {
2672                 if(unlikely(w->exposed))
2673                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 %llu\n", w->name, 1024LLU * RATES_DETAIL);
2674         }
2675
2676         buffer_sprintf(output, "CHART %s.files '' '%s Open Files' 'open files' disk %s.files stacked 20050 %d\n", type, title, type, update_every);
2677         for (w = root; w ; w = w->next) {
2678                 if(unlikely(w->exposed))
2679                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 1\n", w->name);
2680         }
2681
2682         buffer_sprintf(output, "CHART %s.sockets '' '%s Open Sockets' 'open sockets' net %s.sockets stacked 20051 %d\n", type, title, type, update_every);
2683         for (w = root; w ; w = w->next) {
2684                 if(unlikely(w->exposed))
2685                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 1\n", w->name);
2686         }
2687
2688         buffer_sprintf(output, "CHART %s.pipes '' '%s Pipes' 'open pipes' processes %s.pipes stacked 20053 %d\n", type, title, type, update_every);
2689         for (w = root; w ; w = w->next) {
2690                 if(unlikely(w->exposed))
2691                         buffer_sprintf(output, "DIMENSION %s '' absolute 1 1\n", w->name);
2692         }
2693 }
2694
2695
2696 // ----------------------------------------------------------------------------
2697 // parse command line arguments
2698
2699 void parse_args(int argc, char **argv)
2700 {
2701         int i, freq = 0;
2702         char *name = NULL;
2703
2704         for(i = 1; i < argc; i++) {
2705                 if(!freq) {
2706                         int n = atoi(argv[i]);
2707                         if(n > 0) {
2708                                 freq = n;
2709                                 continue;
2710                         }
2711                 }
2712
2713                 if(strcmp("debug", argv[i]) == 0) {
2714                         debug = 1;
2715                         // debug_flags = 0xffffffff;
2716                         continue;
2717                 }
2718
2719                 if(strcmp("no-childs", argv[i]) == 0) {
2720                         include_exited_childs = 0;
2721                         continue;
2722                 }
2723
2724                 if(strcmp("with-childs", argv[i]) == 0) {
2725                         include_exited_childs = 1;
2726                         continue;
2727                 }
2728
2729                 if(!name) {
2730                         name = argv[i];
2731                         continue;
2732                 }
2733
2734                 error("Cannot understand option %s", argv[i]);
2735                 exit(1);
2736         }
2737
2738         if(freq > 0) update_every = freq;
2739         if(!name) name = "groups";
2740
2741         if(read_apps_groups_conf(name)) {
2742                 error("Cannot read process groups %s", name);
2743                 exit(1);
2744         }
2745 }
2746
2747 int main(int argc, char **argv)
2748 {
2749         // debug_flags = D_PROCFILE;
2750
2751         // set the name for logging
2752         program_name = "apps.plugin";
2753
2754         // disable syslog for apps.plugin
2755         error_log_syslog = 0;
2756
2757         // set errors flood protection to 100 logs per hour
2758         error_log_errors_per_period = 100;
2759         error_log_throttle_period = 3600;
2760
2761         host_prefix = getenv("NETDATA_HOST_PREFIX");
2762         if(host_prefix == NULL) {
2763                 info("NETDATA_HOST_PREFIX is not passed from netdata");
2764                 host_prefix = "";
2765         }
2766         else info("Found NETDATA_HOST_PREFIX='%s'", host_prefix);
2767
2768         config_dir = getenv("NETDATA_CONFIG_DIR");
2769         if(config_dir == NULL) {
2770                 info("NETDATA_CONFIG_DIR is not passed from netdata");
2771                 config_dir = CONFIG_DIR;
2772         }
2773         else info("Found NETDATA_CONFIG_DIR='%s'", config_dir);
2774
2775 #ifdef NETDATA_INTERNAL_CHECKS
2776         if(debug_flags != 0) {
2777                 struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
2778                 if(setrlimit(RLIMIT_CORE, &rl) != 0)
2779                         info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
2780                 prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
2781         }
2782 #endif /* NETDATA_INTERNAL_CHECKS */
2783
2784         procfile_adaptive_initial_allocation = 1;
2785
2786         time_t started_t = time(NULL);
2787         time_t current_t;
2788         get_HZ();
2789         pid_max = get_system_pid_max();
2790         processors = get_system_cpus();
2791
2792         parse_args(argc, argv);
2793
2794         all_pids_sortlist = calloc(sizeof(pid_t), (size_t)pid_max);
2795         if(!all_pids_sortlist) {
2796                 error("Cannot allocate %zu bytes of memory.", sizeof(pid_t) * pid_max);
2797                 printf("DISABLE\n");
2798                 exit(1);
2799         }
2800
2801         all_pids = calloc(sizeof(struct pid_stat *), (size_t) pid_max);
2802         if(!all_pids) {
2803                 error("Cannot allocate %zu bytes of memory.", sizeof(struct pid_stat *) * pid_max);
2804                 printf("DISABLE\n");
2805                 exit(1);
2806         }
2807
2808         output = buffer_create(1024);
2809         if(!output)
2810                 fatal("Cannot create BUFFER.");
2811
2812         buffer_sprintf(output,
2813                 "CHART netdata.apps_cpu '' 'Apps Plugin CPU' 'milliseconds/s' apps.plugin netdata.apps_cpu stacked 140000 %1$d\n"
2814                 "DIMENSION user '' incremental 1 1000\n"
2815                 "DIMENSION system '' incremental 1 1000\n"
2816                 "CHART netdata.apps_files '' 'Apps Plugin Files' 'files/s' apps.plugin netdata.apps_files line 140001 %1$d\n"
2817                 "DIMENSION files '' incremental 1 1\n"
2818                 "DIMENSION pids '' absolute 1 1\n"
2819                 "DIMENSION fds '' absolute 1 1\n"
2820                 "DIMENSION targets '' absolute 1 1\n"
2821                 "CHART netdata.apps_fix '' 'Apps Plugin Normalization Ratios' 'percentage' apps.plugin netdata.apps_fix line 140002 %1$d\n"
2822                 "DIMENSION utime '' absolute 1 %2$llu\n"
2823                 "DIMENSION stime '' absolute 1 %2$llu\n"
2824                 "DIMENSION gtime '' absolute 1 %2$llu\n"
2825                 "DIMENSION minflt '' absolute 1 %2$llu\n"
2826                 "DIMENSION majflt '' absolute 1 %2$llu\n"
2827                 , update_every
2828                 , RATES_DETAIL
2829                 );
2830
2831         if(include_exited_childs)
2832                 buffer_sprintf(output,
2833                         "CHART netdata.apps_children_fix '' 'Apps Plugin Exited Children Normalization Ratios' 'percentage' apps.plugin netdata.apps_children_fix line 140003 %1$d\n"
2834                         "DIMENSION cutime '' absolute 1 %2$llu\n"
2835                         "DIMENSION cstime '' absolute 1 %2$llu\n"
2836                         "DIMENSION cgtime '' absolute 1 %2$llu\n"
2837                         "DIMENSION cminflt '' absolute 1 %2$llu\n"
2838                         "DIMENSION cmajflt '' absolute 1 %2$llu\n"
2839                         , update_every
2840                         , RATES_DETAIL
2841                         );
2842
2843 #ifndef PROFILING_MODE
2844         unsigned long long sunext = (time(NULL) - (time(NULL) % update_every) + update_every) * 1000000ULL;
2845         unsigned long long sunow;
2846 #endif /* PROFILING_MODE */
2847
2848         global_iterations_counter = 1;
2849         for(;1; global_iterations_counter++) {
2850 #ifndef PROFILING_MODE
2851                 // delay until it is our time to run
2852                 while((sunow = timems()) < sunext)
2853                         usecsleep(sunext - sunow);
2854
2855                 // find the next time we need to run
2856                 while(timems() > sunext)
2857                         sunext += update_every * 1000000ULL;
2858 #endif /* PROFILING_MODE */
2859
2860                 if(!collect_data_for_all_processes_from_proc()) {
2861                         error("Cannot collect /proc data for running processes. Disabling apps.plugin...");
2862                         printf("DISABLE\n");
2863                         exit(1);
2864                 }
2865
2866                 calculate_netdata_statistics();
2867                 normalize_data(apps_groups_root_target);
2868
2869                 unsigned long long dt = send_resource_usage_to_netdata();
2870
2871                 // this is smart enough to show only newly added apps, when needed
2872                 send_charts_updates_to_netdata(apps_groups_root_target, "apps", "Apps");
2873                 send_charts_updates_to_netdata(users_root_target, "users", "Users");
2874                 send_charts_updates_to_netdata(groups_root_target, "groups", "User Groups");
2875
2876                 send_collected_data_to_netdata(apps_groups_root_target, "apps", dt);
2877                 send_collected_data_to_netdata(users_root_target, "users", dt);
2878                 send_collected_data_to_netdata(groups_root_target, "groups", dt);
2879
2880                 show_guest_time_old = show_guest_time;
2881
2882                 //if(puts(buffer_tostring(output)) == EOF)
2883                 if(write(STDOUT_FILENO, buffer_tostring(output), buffer_strlen(output)) == -1)
2884                         fatal("Cannot send chart values to netdata.");
2885
2886                 // fflush(stdout);
2887                 buffer_flush(output);
2888
2889                 if(unlikely(debug))
2890                         fprintf(stderr, "apps.plugin: done Loop No %llu\n", global_iterations_counter);
2891
2892                 current_t = time(NULL);
2893
2894 #ifndef PROFILING_MODE
2895                 // restart check (14400 seconds)
2896                 if(current_t - started_t > 14400) exit(0);
2897 #else
2898                 if(current_t - started_t > 10) exit(0);
2899 #endif /* PROFILING_MODE */
2900         }
2901 }