src/proc_diskstats.c

   1 #include "common.h"
   2
   3 #define RRD_TYPE_DISK "disk"
   4
   5 #define DISK_TYPE_PHYSICAL  1
   6 #define DISK_TYPE_PARTITION 2
   7 #define DISK_TYPE_CONTAINER 3
   8
   9 #ifndef NETDATA_RELOAD_MOUNTINFO_EVERY
  10 #define NETDATA_RELOAD_MOUNTINFO_EVERY 10
  11 #endif
  12
  13 static struct disk {
  14     char *disk;             // the name of the disk (sda, sdb, etc)
  15     unsigned long major;
  16     unsigned long minor;
  17     int sector_size;
  18     int type;
  19
  20     char *mount_point;
  21     uint32_t mount_point_hash;
  22
  23     // disk options caching
  24     int configured;
  25     int do_io;
  26     int do_ops;
  27     int do_mops;
  28     int do_iotime;
  29     int do_qops;
  30     int do_util;
  31     int do_backlog;
  32     int do_space;
  33     int do_inodes;
  34
  35     struct disk *next;
  36 } *disk_root = NULL;
  37
  38 static struct mountinfo *disk_mountinfo_root = NULL;
  39
  40 static inline void mountinfo_reload(int force) {
  41     static time_t last_loaded = 0;
  42     time_t now = now_realtime_sec();
  43
  44     if(force || now - last_loaded >= NETDATA_RELOAD_MOUNTINFO_EVERY) {
  45 //#ifdef NETDATA_INTERNAL_CHECKS
  46 //        info("Reloading mountinfo");
  47 //#endif
  48
  49         // mountinfo_free() can be called with NULL disk_mountinfo_root
  50         mountinfo_free(disk_mountinfo_root);
  51
  52         // re-read mountinfo in case something changed
  53         disk_mountinfo_root = mountinfo_read();
  54
  55         last_loaded = now;
  56     }
  57 }
  58
  59
  60 // linked list of mount points that are by default disabled
  61 static struct excluded_mount_point {
  62     const char *prefix;
  63     size_t len;
  64     struct excluded_mount_point *next;
  65 } *excluded_mount_points = NULL;
  66
  67 static inline int is_mount_point_excluded(const char *mount_point) {
  68     static int initialized = 0;
  69
  70     if(unlikely(!initialized)) {
  71         initialized = 1;
  72
  73         char *a = config_get("plugin:proc:/proc/diskstats", "exclude space metrics on paths", "/proc/ /sys/ /var/run/user/ /run/user/");
  74         if(a && *a) {
  75             char *s = a;
  76
  77             while(s && *s) {
  78                 // skip all spaces
  79                 while(isspace(*s)) s++;
  80
  81                 // empty string
  82                 if(unlikely(!*s)) break;
  83
  84                 // find the next space
  85                 char *c = s;
  86                 while(*c && !isspace(*c)) c++;
  87
  88                 char *n;
  89                 if(likely(*c)) n = c + 1;
  90                 else n = NULL;
  91
  92                 // terminate our string
  93                 *c = '\0';
  94
  95                 // allocate the structure
  96                 struct excluded_mount_point *m = mallocz(sizeof(struct excluded_mount_point));
  97                 m->prefix = strdup(s);
  98                 m->len = strlen(m->prefix);
  99                 m->next = excluded_mount_points;
 100                 excluded_mount_points = m;
 101
 102                 // prepare for next loop
 103                 s = n;
 104                 if(likely(n)) *c = ' ';
 105             }
 106         }
 107     }
 108
 109     size_t len = strlen(mount_point);
 110     struct excluded_mount_point *m;
 111     for(m = excluded_mount_points; m ; m = m->next) {
 112         if(m->len <= len) {
 113             // fprintf(stderr, "SPACE: comparing '%s' with '%s'\n", mount_point, m->prefix);
 114             if(unlikely(strncmp(m->prefix, mount_point, m->len) == 0)) {
 115                 // fprintf(stderr, "SPACE: excluded '%s'\n", mount_point);
 116                 return 1;
 117             }
 118         }
 119     }
 120
 121     // fprintf(stderr, "SPACE: included '%s'\n", mount_point);
 122     return 0;
 123 }
 124
 125 // Data to be stored in DICTIONARY mount_points used by do_disk_space_stats().
 126 // This DICTIONARY is used to lookup the settings of the mount point on each iteration.
 127 struct mount_point_metadata {
 128     int do_space;
 129     int do_inodes;
 130 };
 131
 132 static inline void do_disk_space_stats(struct mountinfo *mi, int update_every, unsigned long long dt) {
 133     (void)dt;
 134
 135     const char *family = mi->mount_point;
 136     const char *disk = mi->persistent_id;
 137
 138     static DICTIONARY *mount_points = NULL;
 139     int do_space, do_inodes;
 140
 141     if(unlikely(!mount_points)) {
 142         mount_points = dictionary_create(DICTIONARY_FLAG_SINGLE_THREADED);
 143     }
 144
 145     struct mount_point_metadata *m = dictionary_get(mount_points, mi->mount_point);
 146     if(unlikely(!m)) {
 147         char var_name[4096 + 1];
 148         snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", mi->mount_point);
 149
 150         int def_space = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "space usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 151         int def_inodes = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "inodes usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 152
 153         if(unlikely(is_mount_point_excluded(mi->mount_point))) {
 154             def_space = CONFIG_ONDEMAND_NO;
 155             def_inodes = CONFIG_ONDEMAND_NO;
 156         }
 157
 158         do_space = config_get_boolean_ondemand(var_name, "space usage", def_space);
 159         do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", def_inodes);
 160
 161         struct mount_point_metadata mp = {
 162             .do_space = do_space,
 163             .do_inodes = do_inodes
 164         };
 165
 166         dictionary_set(mount_points, mi->mount_point, &mp, sizeof(struct mount_point_metadata));
 167     }
 168     else {
 169         do_space = m->do_space;
 170         do_inodes = m->do_inodes;
 171     }
 172
 173     if(unlikely(do_space == CONFIG_ONDEMAND_NO && do_inodes == CONFIG_ONDEMAND_NO))
 174         return;
 175
 176     struct statvfs buff_statvfs;
 177     if (statvfs(mi->mount_point, &buff_statvfs) < 0) {
 178         error("Failed statvfs() for '%s' (disk '%s')", mi->mount_point, disk);
 179         return;
 180     }
 181
 182     // taken from get_fs_usage() found in coreutils
 183     unsigned long bsize = (buff_statvfs.f_frsize) ? buff_statvfs.f_frsize : buff_statvfs.f_bsize;
 184
 185     fsblkcnt_t bavail         = buff_statvfs.f_bavail;
 186     fsblkcnt_t btotal         = buff_statvfs.f_blocks;
 187     fsblkcnt_t bavail_root    = buff_statvfs.f_bfree;
 188     fsblkcnt_t breserved_root = bavail_root - bavail;
 189     fsblkcnt_t bused;
 190     if(likely(btotal >= bavail_root))
 191         bused = btotal - bavail_root;
 192     else
 193         bused = bavail_root - btotal;
 194
 195 #ifdef NETDATA_INTERNAL_CHECKS
 196     if(unlikely(btotal != bavail + breserved_root + bused))
 197         error("Disk block statistics for '%s' (disk '%s') do not sum up: total = %llu, available = %llu, reserved = %llu, used = %llu", mi->mount_point, disk, (unsigned long long)btotal, (unsigned long long)bavail, (unsigned long long)breserved_root, (unsigned long long)bused);
 198 #endif
 199
 200     // --------------------------------------------------------------------------
 201
 202     fsfilcnt_t favail         = buff_statvfs.f_favail;
 203     fsfilcnt_t ftotal         = buff_statvfs.f_files;
 204     fsfilcnt_t favail_root    = buff_statvfs.f_ffree;
 205     fsfilcnt_t freserved_root = favail_root - favail;
 206     fsfilcnt_t fused          = ftotal - favail_root;
 207
 208 #ifdef NETDATA_INTERNAL_CHECKS
 209     if(unlikely(btotal != bavail + breserved_root + bused))
 210         error("Disk inode statistics for '%s' (disk '%s') do not sum up: total = %llu, available = %llu, reserved = %llu, used = %llu", mi->mount_point, disk, (unsigned long long)ftotal, (unsigned long long)favail, (unsigned long long)freserved_root, (unsigned long long)fused);
 211 #endif
 212
 213     // --------------------------------------------------------------------------
 214
 215     RRDSET *st;
 216
 217     if(do_space == CONFIG_ONDEMAND_YES || (do_space == CONFIG_ONDEMAND_ONDEMAND && (bavail || breserved_root || bused))) {
 218         st = rrdset_find_bytype("disk_space", disk);
 219         if(unlikely(!st)) {
 220             char title[4096 + 1];
 221             snprintfz(title, 4096, "Disk Space Usage for %s [%s]", family, mi->mount_source);
 222             st = rrdset_create("disk_space", disk, NULL, family, "disk.space", title, "GB", 2023, update_every, RRDSET_TYPE_STACKED);
 223
 224             rrddim_add(st, "avail", NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 225             rrddim_add(st, "used" , NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 226             rrddim_add(st, "reserved_for_root", "reserved for root", bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 227         }
 228         else rrdset_next(st);
 229
 230         rrddim_set(st, "avail", (collected_number)bavail);
 231         rrddim_set(st, "used", (collected_number)bused);
 232         rrddim_set(st, "reserved_for_root", (collected_number)breserved_root);
 233         rrdset_done(st);
 234     }
 235
 236     // --------------------------------------------------------------------------
 237
 238     if(do_inodes == CONFIG_ONDEMAND_YES || (do_inodes == CONFIG_ONDEMAND_ONDEMAND && (favail || freserved_root || fused))) {
 239         st = rrdset_find_bytype("disk_inodes", disk);
 240         if(unlikely(!st)) {
 241             char title[4096 + 1];
 242             snprintfz(title, 4096, "Disk Files (inodes) Usage for %s [%s]", family, mi->mount_source);
 243             st = rrdset_create("disk_inodes", disk, NULL, family, "disk.inodes", title, "Inodes", 2024, update_every, RRDSET_TYPE_STACKED);
 244
 245             rrddim_add(st, "avail", NULL, 1, 1, RRDDIM_ABSOLUTE);
 246             rrddim_add(st, "used" , NULL, 1, 1, RRDDIM_ABSOLUTE);
 247             rrddim_add(st, "reserved_for_root", "reserved for root", 1, 1, RRDDIM_ABSOLUTE);
 248         }
 249         else rrdset_next(st);
 250
 251         rrddim_set(st, "avail", (collected_number)favail);
 252         rrddim_set(st, "used", (collected_number)fused);
 253         rrddim_set(st, "reserved_for_root", (collected_number)freserved_root);
 254         rrdset_done(st);
 255     }
 256 }
 257
 258 static struct disk *get_disk(unsigned long major, unsigned long minor, char *disk) {
 259     static char path_to_get_hw_sector_size[FILENAME_MAX + 1] = "";
 260     static char path_to_get_hw_sector_size_partitions[FILENAME_MAX + 1] = "";
 261     static char path_find_block_device[FILENAME_MAX + 1] = "";
 262     struct disk *d;
 263
 264     // search for it in our RAM list.
 265     // this is sequential, but since we just walk through
 266     // and the number of disks / partitions in a system
 267     // should not be that many, it should be acceptable
 268     for(d = disk_root; d ; d = d->next)
 269         if(unlikely(d->major == major && d->minor == minor))
 270             break;
 271
 272     // if we found it, return it
 273     if(likely(d))
 274         return d;
 275
 276     // not found
 277     // create a new disk structure
 278     d = (struct disk *)mallocz(sizeof(struct disk));
 279
 280     d->disk = strdupz(disk);
 281     d->major = major;
 282     d->minor = minor;
 283     d->type = DISK_TYPE_PHYSICAL; // Default type. Changed later if not correct.
 284     d->configured = 0;
 285     d->sector_size = 512; // the default, will be changed below
 286     d->next = NULL;
 287
 288     // append it to the list
 289     if(unlikely(!disk_root))
 290         disk_root = d;
 291     else {
 292         struct disk *last;
 293         for(last = disk_root; last->next ;last = last->next);
 294         last->next = d;
 295     }
 296
 297     // ------------------------------------------------------------------------
 298     // find the type of the device
 299
 300     char buffer[FILENAME_MAX + 1];
 301
 302     // get the default path for finding info about the block device
 303     if(unlikely(!path_find_block_device[0])) {
 304         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/%s");
 305         snprintfz(path_find_block_device, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get block device infos", buffer));
 306     }
 307
 308     // find if it is a partition
 309     // by checking if /sys/dev/block/MAJOR:MINOR/partition is readable.
 310     snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "partition");
 311     if(likely(access(buffer, R_OK) == 0)) {
 312         d->type = DISK_TYPE_PARTITION;
 313     }
 314     else {
 315         // find if it is a container
 316         // by checking if /sys/dev/block/MAJOR:MINOR/slaves has entries
 317         snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "slaves/");
 318         DIR *dirp = opendir(buffer);
 319         if(likely(dirp != NULL)) {
 320             struct dirent *dp;
 321             while( (dp = readdir(dirp)) ) {
 322                 // . and .. are also files in empty folders.
 323                 if(unlikely(strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0)) {
 324                     continue;
 325                 }
 326
 327                 d->type = DISK_TYPE_CONTAINER;
 328
 329                 // Stop the loop after we found one file.
 330                 break;
 331             }
 332             if(unlikely(closedir(dirp) == -1))
 333                 error("Unable to close dir %s", buffer);
 334         }
 335     }
 336
 337     // ------------------------------------------------------------------------
 338     // check if we can find its mount point
 339
 340     // mountinfo_find() can be called with NULL disk_mountinfo_root
 341     struct mountinfo *mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 342 /*    if(unlikely(!mi)) {
 343         mountinfo_reload(1);
 344
 345         // search again for this disk
 346         mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 347     }
 348 */
 349     if(unlikely(mi)) {
 350         d->mount_point = strdupz(mi->mount_point);
 351         d->mount_point_hash = mi->mount_point_hash;
 352     }
 353     else {
 354         d->mount_point = NULL;
 355         d->mount_point_hash = 0;
 356     }
 357
 358     // ------------------------------------------------------------------------
 359     // find the disk sector size
 360
 361     if(unlikely(!path_to_get_hw_sector_size[0])) {
 362         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/block/%s/queue/hw_sector_size");
 363         snprintfz(path_to_get_hw_sector_size, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get h/w sector size", buffer));
 364     }
 365     if(unlikely(!path_to_get_hw_sector_size_partitions[0])) {
 366         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/subsystem/%s/../queue/hw_sector_size");
 367         snprintfz(path_to_get_hw_sector_size_partitions, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get h/w sector size for partitions", buffer));
 368     }
 369
 370     {
 371         char tf[FILENAME_MAX + 1], *t;
 372         strncpyz(tf, d->disk, FILENAME_MAX);
 373
 374         // replace all / with !
 375         for(t = tf; *t ;t++)
 376             if(unlikely(*t == '/')) *t = '!';
 377
 378         if(likely(d->type == DISK_TYPE_PARTITION))
 379             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size_partitions, d->major, d->minor, tf);
 380         else
 381             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size, tf);
 382
 383         FILE *fpss = fopen(buffer, "r");
 384         if(likely(fpss)) {
 385             char buffer2[1024 + 1];
 386             char *tmp = fgets(buffer2, 1024, fpss);
 387
 388             if(likely(tmp)) {
 389                 d->sector_size = atoi(tmp);
 390                 if(unlikely(d->sector_size <= 0)) {
 391                     error("Invalid sector size %d for device %s in %s. Assuming 512.", d->sector_size, d->disk, buffer);
 392                     d->sector_size = 512;
 393                 }
 394             }
 395             else error("Cannot read data for sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 396
 397             fclose(fpss);
 398         }
 399         else error("Cannot read sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 400     }
 401
 402     return d;
 403 }
 404
 405 static inline int select_positive_option(int option1, int option2) {
 406     if(unlikely(option1 == CONFIG_ONDEMAND_YES || option2 == CONFIG_ONDEMAND_YES))
 407         return CONFIG_ONDEMAND_YES;
 408     else if(unlikely(option1 == CONFIG_ONDEMAND_ONDEMAND || option2 == CONFIG_ONDEMAND_ONDEMAND))
 409         return CONFIG_ONDEMAND_ONDEMAND;
 410
 411     return CONFIG_ONDEMAND_NO;
 412 }
 413
 414 int do_proc_diskstats(int update_every, unsigned long long dt) {
 415     (void)dt;
 416
 417     static procfile *ff = NULL;
 418     static int  global_enable_new_disks_detected_at_runtime = CONFIG_ONDEMAND_YES,
 419                 global_enable_performance_for_physical_disks = CONFIG_ONDEMAND_ONDEMAND,
 420                 global_enable_performance_for_virtual_disks = CONFIG_ONDEMAND_NO,
 421                 global_enable_performance_for_partitions = CONFIG_ONDEMAND_NO,
 422                 global_enable_performance_for_mountpoints = CONFIG_ONDEMAND_NO,
 423                 global_enable_performance_for_virtual_mountpoints = CONFIG_ONDEMAND_ONDEMAND,
 424                 global_do_io = CONFIG_ONDEMAND_ONDEMAND,
 425                 global_do_ops = CONFIG_ONDEMAND_ONDEMAND,
 426                 global_do_mops = CONFIG_ONDEMAND_ONDEMAND,
 427                 global_do_iotime = CONFIG_ONDEMAND_ONDEMAND,
 428                 global_do_qops = CONFIG_ONDEMAND_ONDEMAND,
 429                 global_do_util = CONFIG_ONDEMAND_ONDEMAND,
 430                 global_do_backlog = CONFIG_ONDEMAND_ONDEMAND,
 431                 globals_initialized = 0;
 432
 433     if(unlikely(!globals_initialized)) {
 434         global_enable_new_disks_detected_at_runtime = config_get_boolean("plugin:proc:/proc/diskstats", "enable new disks detected at runtime", global_enable_new_disks_detected_at_runtime);
 435
 436         global_enable_performance_for_physical_disks = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for physical disks", global_enable_performance_for_physical_disks);
 437         global_enable_performance_for_virtual_disks = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for virtual disks", global_enable_performance_for_virtual_disks);
 438         global_enable_performance_for_partitions = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for partitions", global_enable_performance_for_partitions);
 439         global_enable_performance_for_mountpoints = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for mounted filesystems", global_enable_performance_for_mountpoints);
 440         global_enable_performance_for_virtual_mountpoints = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for mounted virtual disks", global_enable_performance_for_virtual_mountpoints);
 441
 442         global_do_io      = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "bandwidth for all disks", global_do_io);
 443         global_do_ops     = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "operations for all disks", global_do_ops);
 444         global_do_mops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "merged operations for all disks", global_do_mops);
 445         global_do_iotime  = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "i/o time for all disks", global_do_iotime);
 446         global_do_qops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "queued operations for all disks", global_do_qops);
 447         global_do_util    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "utilization percentage for all disks", global_do_util);
 448         global_do_backlog = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "backlog for all disks", global_do_backlog);
 449
 450         globals_initialized = 1;
 451     }
 452
 453     if(unlikely(!ff)) {
 454         char filename[FILENAME_MAX + 1];
 455         snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, "/proc/diskstats");
 456         ff = procfile_open(config_get("plugin:proc:/proc/diskstats", "filename to monitor", filename), " \t", PROCFILE_FLAG_DEFAULT);
 457     }
 458     if(unlikely(!ff)) return 1;
 459
 460     ff = procfile_readall(ff);
 461     if(unlikely(!ff)) return 0; // we return 0, so that we will retry to open it next time
 462
 463     uint32_t lines = procfile_lines(ff), l;
 464
 465     // this is smart enough not to reload it every time
 466     mountinfo_reload(0);
 467
 468     for(l = 0; l < lines ;l++) {
 469         // --------------------------------------------------------------------------
 470         // Read parameters
 471
 472         char *disk;
 473         unsigned long       major = 0, minor = 0;
 474
 475         collected_number    reads = 0,  mreads = 0,  readsectors = 0,  readms = 0,
 476                             writes = 0, mwrites = 0, writesectors = 0, writems = 0,
 477                             queued_ios = 0, busy_ms = 0, backlog_ms = 0;
 478
 479         collected_number    last_reads = 0,  last_readsectors = 0,  last_readms = 0,
 480                             last_writes = 0, last_writesectors = 0, last_writems = 0,
 481                             last_busy_ms = 0;
 482
 483         uint32_t words = procfile_linewords(ff, l);
 484         if(unlikely(words < 14)) continue;
 485
 486         major           = strtoul(procfile_lineword(ff, l, 0), NULL, 10);
 487         minor           = strtoul(procfile_lineword(ff, l, 1), NULL, 10);
 488         disk            = procfile_lineword(ff, l, 2);
 489
 490         // # of reads completed # of writes completed
 491         // This is the total number of reads or writes completed successfully.
 492         reads           = strtoull(procfile_lineword(ff, l, 3), NULL, 10);  // rd_ios
 493         writes          = strtoull(procfile_lineword(ff, l, 7), NULL, 10);  // wr_ios
 494
 495         // # of reads merged # of writes merged
 496         // Reads and writes which are adjacent to each other may be merged for
 497         // efficiency.  Thus two 4K reads may become one 8K read before it is
 498         // ultimately handed to the disk, and so it will be counted (and queued)
 499         mreads          = strtoull(procfile_lineword(ff, l, 4), NULL, 10);  // rd_merges_or_rd_sec
 500         mwrites         = strtoull(procfile_lineword(ff, l, 8), NULL, 10);  // wr_merges
 501
 502         // # of sectors read # of sectors written
 503         // This is the total number of sectors read or written successfully.
 504         readsectors     = strtoull(procfile_lineword(ff, l, 5), NULL, 10);  // rd_sec_or_wr_ios
 505         writesectors    = strtoull(procfile_lineword(ff, l, 9), NULL, 10);  // wr_sec
 506
 507         // # of milliseconds spent reading # of milliseconds spent writing
 508         // This is the total number of milliseconds spent by all reads or writes (as
 509         // measured from __make_request() to end_that_request_last()).
 510         readms          = strtoull(procfile_lineword(ff, l, 6), NULL, 10);  // rd_ticks_or_wr_sec
 511         writems         = strtoull(procfile_lineword(ff, l, 10), NULL, 10); // wr_ticks
 512
 513         // # of I/Os currently in progress
 514         // The only field that should go to zero. Incremented as requests are
 515         // given to appropriate struct request_queue and decremented as they finish.
 516         queued_ios      = strtoull(procfile_lineword(ff, l, 11), NULL, 10); // ios_pgr
 517
 518         // # of milliseconds spent doing I/Os
 519         // This field increases so long as field queued_ios is nonzero.
 520         busy_ms         = strtoull(procfile_lineword(ff, l, 12), NULL, 10); // tot_ticks
 521
 522         // weighted # of milliseconds spent doing I/Os
 523         // This field is incremented at each I/O start, I/O completion, I/O
 524         // merge, or read of these stats by the number of I/Os in progress
 525         // (field queued_ios) times the number of milliseconds spent doing I/O since the
 526         // last update of this field.  This can provide an easy measure of both
 527         // I/O completion time and the backlog that may be accumulating.
 528         backlog_ms      = strtoull(procfile_lineword(ff, l, 13), NULL, 10); // rq_ticks
 529
 530
 531         // --------------------------------------------------------------------------
 532         // remove slashes from disk names
 533         char *s;
 534         for(s = disk; *s ;s++)
 535             if(*s == '/') *s = '_';
 536
 537         // --------------------------------------------------------------------------
 538         // get a disk structure for the disk
 539
 540         struct disk *d = get_disk(major, minor, disk);
 541
 542
 543         // --------------------------------------------------------------------------
 544         // Set its family based on mount point
 545
 546         char *family = d->mount_point;
 547         if(!family) family = disk;
 548
 549
 550         // --------------------------------------------------------------------------
 551         // Check the configuration for the device
 552
 553         if(unlikely(!d->configured)) {
 554             char var_name[4096 + 1];
 555             snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", disk);
 556
 557             int def_enable = config_get_boolean_ondemand(var_name, "enable", global_enable_new_disks_detected_at_runtime);
 558             if(unlikely(def_enable == CONFIG_ONDEMAND_NO)) {
 559                 // the user does not want any metrics for this disk
 560                 d->do_io = CONFIG_ONDEMAND_NO;
 561                 d->do_ops = CONFIG_ONDEMAND_NO;
 562                 d->do_mops = CONFIG_ONDEMAND_NO;
 563                 d->do_iotime = CONFIG_ONDEMAND_NO;
 564                 d->do_qops = CONFIG_ONDEMAND_NO;
 565                 d->do_util = CONFIG_ONDEMAND_NO;
 566                 d->do_backlog = CONFIG_ONDEMAND_NO;
 567                 d->do_space = CONFIG_ONDEMAND_NO;
 568                 d->do_inodes = CONFIG_ONDEMAND_NO;
 569             }
 570             else {
 571                 // this disk is enabled
 572                 // check its direct settings
 573
 574                 int def_performance = CONFIG_ONDEMAND_ONDEMAND;
 575                 int def_space = (d->mount_point)?CONFIG_ONDEMAND_ONDEMAND:CONFIG_ONDEMAND_NO;
 576
 577                 // since this is 'on demand' we can figure the performance settings
 578                 // based on the type of disk
 579
 580                 switch(d->type) {
 581                     case DISK_TYPE_PHYSICAL:
 582                         def_performance = global_enable_performance_for_physical_disks;
 583                         break;
 584
 585                     case DISK_TYPE_PARTITION:
 586                         def_performance = global_enable_performance_for_partitions;
 587                         break;
 588
 589                     case DISK_TYPE_CONTAINER:
 590                         def_performance = global_enable_performance_for_virtual_disks;
 591
 592                         if(d->mount_point)
 593                             def_performance = select_positive_option(def_performance, global_enable_performance_for_virtual_mountpoints);
 594                         break;
 595                 }
 596
 597                 if(d->mount_point)
 598                     def_performance = select_positive_option(def_performance, global_enable_performance_for_mountpoints);
 599
 600                 // ------------------------------------------------------------
 601                 // now we have def_performance and def_space
 602                 // to work further
 603
 604                 // def_performance
 605                 // check the user configuration (this will also show our 'on demand' decision)
 606                 def_performance = config_get_boolean_ondemand(var_name, "enable performance metrics", def_performance);
 607
 608                 int ddo_io = CONFIG_ONDEMAND_NO,
 609                     ddo_ops = CONFIG_ONDEMAND_NO,
 610                     ddo_mops = CONFIG_ONDEMAND_NO,
 611                     ddo_iotime = CONFIG_ONDEMAND_NO,
 612                     ddo_qops = CONFIG_ONDEMAND_NO,
 613                     ddo_util = CONFIG_ONDEMAND_NO,
 614                     ddo_backlog = CONFIG_ONDEMAND_NO;
 615
 616                 // we enable individual performance charts only when def_performance is not disabled
 617                 if(unlikely(def_performance != CONFIG_ONDEMAND_NO)) {
 618                     ddo_io = global_do_io,
 619                     ddo_ops = global_do_ops,
 620                     ddo_mops = global_do_mops,
 621                     ddo_iotime = global_do_iotime,
 622                     ddo_qops = global_do_qops,
 623                     ddo_util = global_do_util,
 624                     ddo_backlog = global_do_backlog;
 625                 }
 626
 627                 d->do_io      = config_get_boolean_ondemand(var_name, "bandwidth", ddo_io);
 628                 d->do_ops     = config_get_boolean_ondemand(var_name, "operations", ddo_ops);
 629                 d->do_mops    = config_get_boolean_ondemand(var_name, "merged operations", ddo_mops);
 630                 d->do_iotime  = config_get_boolean_ondemand(var_name, "i/o time", ddo_iotime);
 631                 d->do_qops    = config_get_boolean_ondemand(var_name, "queued operations", ddo_qops);
 632                 d->do_util    = config_get_boolean_ondemand(var_name, "utilization percentage", ddo_util);
 633                 d->do_backlog = config_get_boolean_ondemand(var_name, "backlog", ddo_backlog);
 634
 635                 // def_space
 636                 if(unlikely(d->mount_point)) {
 637                     // check the user configuration (this will also show our 'on demand' decision)
 638                     def_space = config_get_boolean_ondemand(var_name, "enable space metrics", def_space);
 639
 640                     int ddo_space = def_space,
 641                         ddo_inodes = def_space;
 642
 643                     d->do_space = config_get_boolean_ondemand(var_name, "space usage", ddo_space);
 644                     d->do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", ddo_inodes);
 645                 }
 646                 else {
 647                     // don't show settings for this disk
 648                     d->do_space = CONFIG_ONDEMAND_NO;
 649                     d->do_inodes = CONFIG_ONDEMAND_NO;
 650                 }
 651             }
 652
 653             d->configured = 1;
 654         }
 655
 656         RRDSET *st;
 657
 658         // --------------------------------------------------------------------------
 659         // Do performance metrics
 660
 661         if(d->do_io == CONFIG_ONDEMAND_YES || (d->do_io == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) {
 662             d->do_io = CONFIG_ONDEMAND_YES;
 663
 664             st = rrdset_find_bytype(RRD_TYPE_DISK, disk);
 665             if(unlikely(!st)) {
 666                 st = rrdset_create(RRD_TYPE_DISK, disk, NULL, family, "disk.io", "Disk I/O Bandwidth", "kilobytes/s", 2000, update_every, RRDSET_TYPE_AREA);
 667
 668                 rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_INCREMENTAL);
 669                 rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_INCREMENTAL);
 670             }
 671             else rrdset_next(st);
 672
 673             last_readsectors  = rrddim_set(st, "reads", readsectors);
 674             last_writesectors = rrddim_set(st, "writes", writesectors);
 675             rrdset_done(st);
 676         }
 677
 678         // --------------------------------------------------------------------
 679
 680         if(d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes))) {
 681             d->do_ops = CONFIG_ONDEMAND_YES;
 682
 683             st = rrdset_find_bytype("disk_ops", disk);
 684             if(unlikely(!st)) {
 685                 st = rrdset_create("disk_ops", disk, NULL, family, "disk.ops", "Disk Completed I/O Operations", "operations/s", 2001, update_every, RRDSET_TYPE_LINE);
 686                 st->isdetail = 1;
 687
 688                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 689                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 690             }
 691             else rrdset_next(st);
 692
 693             last_reads  = rrddim_set(st, "reads", reads);
 694             last_writes = rrddim_set(st, "writes", writes);
 695             rrdset_done(st);
 696         }
 697
 698         // --------------------------------------------------------------------
 699
 700         if(d->do_qops == CONFIG_ONDEMAND_YES || (d->do_qops == CONFIG_ONDEMAND_ONDEMAND && queued_ios)) {
 701             d->do_qops = CONFIG_ONDEMAND_YES;
 702
 703             st = rrdset_find_bytype("disk_qops", disk);
 704             if(unlikely(!st)) {
 705                 st = rrdset_create("disk_qops", disk, NULL, family, "disk.qops", "Disk Current I/O Operations", "operations", 2002, update_every, RRDSET_TYPE_LINE);
 706                 st->isdetail = 1;
 707
 708                 rrddim_add(st, "operations", NULL, 1, 1, RRDDIM_ABSOLUTE);
 709             }
 710             else rrdset_next(st);
 711
 712             rrddim_set(st, "operations", queued_ios);
 713             rrdset_done(st);
 714         }
 715
 716         // --------------------------------------------------------------------
 717
 718         if(d->do_backlog == CONFIG_ONDEMAND_YES || (d->do_backlog == CONFIG_ONDEMAND_ONDEMAND && backlog_ms)) {
 719             d->do_backlog = CONFIG_ONDEMAND_YES;
 720
 721             st = rrdset_find_bytype("disk_backlog", disk);
 722             if(unlikely(!st)) {
 723                 st = rrdset_create("disk_backlog", disk, NULL, family, "disk.backlog", "Disk Backlog", "backlog (ms)", 2003, update_every, RRDSET_TYPE_AREA);
 724                 st->isdetail = 1;
 725
 726                 rrddim_add(st, "backlog", NULL, 1, 10, RRDDIM_INCREMENTAL);
 727             }
 728             else rrdset_next(st);
 729
 730             rrddim_set(st, "backlog", backlog_ms);
 731             rrdset_done(st);
 732         }
 733
 734         // --------------------------------------------------------------------
 735
 736         if(d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) {
 737             d->do_util = CONFIG_ONDEMAND_YES;
 738
 739             st = rrdset_find_bytype("disk_util", disk);
 740             if(unlikely(!st)) {
 741                 st = rrdset_create("disk_util", disk, NULL, family, "disk.util", "Disk Utilization Time", "% of time working", 2004, update_every, RRDSET_TYPE_AREA);
 742                 st->isdetail = 1;
 743
 744                 rrddim_add(st, "utilization", NULL, 1, 10, RRDDIM_INCREMENTAL);
 745             }
 746             else rrdset_next(st);
 747
 748             last_busy_ms = rrddim_set(st, "utilization", busy_ms);
 749             rrdset_done(st);
 750         }
 751
 752         // --------------------------------------------------------------------
 753
 754         if(d->do_mops == CONFIG_ONDEMAND_YES || (d->do_mops == CONFIG_ONDEMAND_ONDEMAND && (mreads || mwrites))) {
 755             d->do_mops = CONFIG_ONDEMAND_YES;
 756
 757             st = rrdset_find_bytype("disk_mops", disk);
 758             if(unlikely(!st)) {
 759                 st = rrdset_create("disk_mops", disk, NULL, family, "disk.mops", "Disk Merged Operations", "merged operations/s", 2021, update_every, RRDSET_TYPE_LINE);
 760                 st->isdetail = 1;
 761
 762                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 763                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 764             }
 765             else rrdset_next(st);
 766
 767             rrddim_set(st, "reads", mreads);
 768             rrddim_set(st, "writes", mwrites);
 769             rrdset_done(st);
 770         }
 771
 772         // --------------------------------------------------------------------
 773
 774         if(d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) {
 775             d->do_iotime = CONFIG_ONDEMAND_YES;
 776
 777             st = rrdset_find_bytype("disk_iotime", disk);
 778             if(unlikely(!st)) {
 779                 st = rrdset_create("disk_iotime", disk, NULL, family, "disk.iotime", "Disk Total I/O Time", "milliseconds/s", 2022, update_every, RRDSET_TYPE_LINE);
 780                 st->isdetail = 1;
 781
 782                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 783                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 784             }
 785             else rrdset_next(st);
 786
 787             last_readms  = rrddim_set(st, "reads", readms);
 788             last_writems = rrddim_set(st, "writes", writems);
 789             rrdset_done(st);
 790         }
 791
 792         // --------------------------------------------------------------------
 793         // calculate differential charts
 794         // only if this is not the first time we run
 795
 796         if(likely(dt)) {
 797             if( (d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) &&
 798                 (d->do_ops    == CONFIG_ONDEMAND_YES || (d->do_ops    == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 799                 st = rrdset_find_bytype("disk_await", disk);
 800                 if(unlikely(!st)) {
 801                     st = rrdset_create("disk_await", disk, NULL, family, "disk.await", "Average Completed I/O Operation Time", "ms per operation", 2005, update_every, RRDSET_TYPE_LINE);
 802                     st->isdetail = 1;
 803
 804                     rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_ABSOLUTE);
 805                     rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_ABSOLUTE);
 806                 }
 807                 else rrdset_next(st);
 808
 809                 rrddim_set(st, "reads", (reads - last_reads) ? (readms - last_readms) / (reads - last_reads) : 0);
 810                 rrddim_set(st, "writes", (writes - last_writes) ? (writems - last_writems) / (writes - last_writes) : 0);
 811                 rrdset_done(st);
 812             }
 813
 814             if( (d->do_io  == CONFIG_ONDEMAND_YES || (d->do_io  == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) &&
 815                 (d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 816                 st = rrdset_find_bytype("disk_avgsz", disk);
 817                 if(unlikely(!st)) {
 818                     st = rrdset_create("disk_avgsz", disk, NULL, family, "disk.avgsz", "Average Completed I/O Operation Bandwidth", "kilobytes per operation", 2006, update_every, RRDSET_TYPE_AREA);
 819                     st->isdetail = 1;
 820
 821                     rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_ABSOLUTE);
 822                     rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_ABSOLUTE);
 823                 }
 824                 else rrdset_next(st);
 825
 826                 rrddim_set(st, "reads", (reads - last_reads) ? (readsectors - last_readsectors) / (reads - last_reads) : 0);
 827                 rrddim_set(st, "writes", (writes - last_writes) ? (writesectors - last_writesectors) / (writes - last_writes) : 0);
 828                 rrdset_done(st);
 829             }
 830
 831             if( (d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) &&
 832                 (d->do_ops  == CONFIG_ONDEMAND_YES || (d->do_ops  == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 833                 st = rrdset_find_bytype("disk_svctm", disk);
 834                 if(unlikely(!st)) {
 835                     st = rrdset_create("disk_svctm", disk, NULL, family, "disk.svctm", "Average Service Time", "ms per operation", 2007, update_every, RRDSET_TYPE_LINE);
 836                     st->isdetail = 1;
 837
 838                     rrddim_add(st, "svctm", NULL, 1, 1, RRDDIM_ABSOLUTE);
 839                 }
 840                 else rrdset_next(st);
 841
 842                 rrddim_set(st, "svctm", ((reads - last_reads) + (writes - last_writes)) ? (busy_ms - last_busy_ms) / ((reads - last_reads) + (writes - last_writes)) : 0);
 843                 rrdset_done(st);
 844             }
 845         }
 846     }
 847
 848     // --------------------------------------------------------------------------
 849     // space metrics for non-block devices
 850
 851     struct mountinfo *mi;
 852     for(mi = disk_mountinfo_root; mi ;mi = mi->next) {
 853         if(unlikely(mi->flags & (MOUNTINFO_IS_DUMMY|MOUNTINFO_IS_BIND|MOUNTINFO_IS_SAME_DEV|MOUNTINFO_NO_STAT|MOUNTINFO_NO_SIZE|MOUNTINFO_READONLY)))
 854             continue;
 855
 856         do_disk_space_stats(mi, update_every, dt);
 857     }
 858
 859     return 0;
 860 }