src/rrdpush.c

   1 #include "common.h"
   2
   3 /*
   4  * rrdpush
   5  *
   6  * 3 threads are involved for all stream operations
   7  *
   8  * 1. a random data collection thread, calling rrdset_done_push()
   9  *    this is called for each chart.
  10  *
  11  *    the output of this work is kept in a BUFFER in RRDHOST
  12  *    the sender thread is signalled via a pipe (also in RRDHOST)
  13  *
  14  * 2. a sender thread running at the sending netdata
  15  *    this is spawned automatically on the first chart to be pushed
  16  *
  17  *    It tries to push the metrics to the remote netdata, as fast
  18  *    as possible (i.e. immediately after they are collected).
  19  *
  20  * 3. a receiver thread, running at the receiving netdata
  21  *    this is spawned automatically when the sender connects to
  22  *    the receiver.
  23  *
  24  */
  25
  26 #define START_STREAMING_PROMPT "Hit me baby, push them over..."
  27
  28 int default_rrdpush_enabled = 0;
  29 char *default_rrdpush_destination = NULL;
  30 char *default_rrdpush_api_key = NULL;
  31
  32 int rrdpush_init() {
  33     default_rrdpush_enabled     = appconfig_get_boolean(&stream_config, CONFIG_SECTION_STREAM, "enabled", default_rrdpush_enabled);
  34     default_rrdpush_destination = appconfig_get(&stream_config, CONFIG_SECTION_STREAM, "destination", "");
  35     default_rrdpush_api_key     = appconfig_get(&stream_config, CONFIG_SECTION_STREAM, "api key", "");
  36     rrdhost_free_orphan_time    = appconfig_get_number(&stream_config, CONFIG_SECTION_STREAM, "free orphan hosts after seconds", rrdhost_free_orphan_time);
  37
  38     if(default_rrdpush_enabled && (!default_rrdpush_destination || !*default_rrdpush_destination || !default_rrdpush_api_key || !*default_rrdpush_api_key)) {
  39         error("STREAM [send]: cannot enable sending thread - information is missing.");
  40         default_rrdpush_enabled = 0;
  41     }
  42
  43     return default_rrdpush_enabled;
  44 }
  45
  46 #define CONNECTED_TO_SIZE 100
  47
  48 // data collection happens from multiple threads
  49 // each of these threads calls rrdset_done()
  50 // which in turn calls rrdset_done_push()
  51 // which uses this pipe to notify the streaming thread
  52 // that there are more data ready to be sent
  53 #define PIPE_READ 0
  54 #define PIPE_WRITE 1
  55
  56 // to have the remote netdata re-sync the charts
  57 // to its current clock, we send for this many
  58 // iterations a BEGIN line without microseconds
  59 // this is for the first iterations of each chart
  60 static unsigned int remote_clock_resync_iterations = 60;
  61
  62 #define rrdpush_lock(host) pthread_mutex_lock(&((host)->rrdpush_mutex))
  63 #define rrdpush_unlock(host) pthread_mutex_unlock(&((host)->rrdpush_mutex))
  64
  65 // checks if the current chart definition has been sent
  66 static inline int need_to_send_chart_definition(RRDSET *st) {
  67     RRDDIM *rd;
  68     rrddim_foreach_read(rd, st)
  69         if(!rd->exposed)
  70             return 1;
  71
  72     return 0;
  73 }
  74
  75 // sends the current chart definition
  76 static inline void send_chart_definition(RRDSET *st) {
  77     buffer_sprintf(st->rrdhost->rrdpush_buffer, "CHART '%s' '%s' '%s' '%s' '%s' '%s' '%s' %ld %d\n"
  78                 , st->id
  79                 , st->name
  80                 , st->title
  81                 , st->units
  82                 , st->family
  83                 , st->context
  84                 , rrdset_type_name(st->chart_type)
  85                 , st->priority
  86                 , st->update_every
  87     );
  88
  89     RRDDIM *rd;
  90     rrddim_foreach_read(rd, st) {
  91         buffer_sprintf(st->rrdhost->rrdpush_buffer, "DIMENSION '%s' '%s' '%s' " COLLECTED_NUMBER_FORMAT " " COLLECTED_NUMBER_FORMAT " '%s %s'\n"
  92                        , rd->id
  93                        , rd->name
  94                        , rrd_algorithm_name(rd->algorithm)
  95                        , rd->multiplier
  96                        , rd->divisor
  97                        , rrddim_flag_check(rd, RRDDIM_FLAG_HIDDEN)?"hidden":""
  98                        , rrddim_flag_check(rd, RRDDIM_FLAG_DONT_DETECT_RESETS_OR_OVERFLOWS)?"noreset":""
  99         );
 100         rd->exposed = 1;
 101     }
 102 }
 103
 104 // sends the current chart dimensions
 105 static inline void send_chart_metrics(RRDSET *st) {
 106     buffer_sprintf(st->rrdhost->rrdpush_buffer, "BEGIN %s %llu\n", st->id, (st->counter_done > remote_clock_resync_iterations)?st->usec_since_last_update:0);
 107
 108     RRDDIM *rd;
 109     rrddim_foreach_read(rd, st) {
 110         if(rd->updated && rd->exposed)
 111             buffer_sprintf(st->rrdhost->rrdpush_buffer, "SET %s = " COLLECTED_NUMBER_FORMAT "\n"
 112                        , rd->id
 113                        , rd->collected_value
 114         );
 115     }
 116
 117     buffer_strcat(st->rrdhost->rrdpush_buffer, "END\n");
 118 }
 119
 120 void rrdpush_sender_thread_spawn(RRDHOST *host);
 121
 122 void rrdset_done_push(RRDSET *st) {
 123     RRDHOST *host = st->rrdhost;
 124
 125     if(unlikely(!rrdset_flag_check(st, RRDSET_FLAG_ENABLED)))
 126         return;
 127
 128     rrdpush_lock(host);
 129
 130     if(unlikely(host->rrdpush_enabled && !host->rrdpush_spawn))
 131         rrdpush_sender_thread_spawn(host);
 132
 133     if(unlikely(!host->rrdpush_buffer || !host->rrdpush_connected)) {
 134         if(unlikely(!host->rrdpush_error_shown))
 135             error("STREAM %s [send]: not ready - discarding collected metrics.", host->hostname);
 136
 137         host->rrdpush_error_shown = 1;
 138
 139         rrdpush_unlock(host);
 140         return;
 141     }
 142     else if(unlikely(host->rrdpush_error_shown)) {
 143         info("STREAM %s [send]: ready - sending metrics...", host->hostname);
 144         host->rrdpush_error_shown = 0;
 145     }
 146
 147     if(need_to_send_chart_definition(st))
 148         send_chart_definition(st);
 149
 150     send_chart_metrics(st);
 151
 152     // signal the sender there are more data
 153     if(write(host->rrdpush_pipe[PIPE_WRITE], " ", 1) == -1)
 154         error("STREAM %s [send]: cannot write to internal pipe", host->hostname);
 155
 156     rrdpush_unlock(host);
 157 }
 158
 159 // ----------------------------------------------------------------------------
 160 // rrdpush sender thread
 161
 162 // resets all the chart, so that their definitions
 163 // will be resent to the central netdata
 164 static void rrdpush_sender_thread_reset_all_charts(RRDHOST *host) {
 165     rrdhost_rdlock(host);
 166
 167     RRDSET *st;
 168     rrdset_foreach_read(st, host) {
 169
 170         // make it re-align the current time
 171         // on the remote host
 172         st->counter_done = 0;
 173
 174         rrdset_rdlock(st);
 175
 176         RRDDIM *rd;
 177         rrddim_foreach_read(rd, st)
 178             rd->exposed = 0;
 179
 180         rrdset_unlock(st);
 181     }
 182
 183     rrdhost_unlock(host);
 184 }
 185
 186 static inline void rrdpush_sender_thread_data_flush(RRDHOST *host) {
 187     rrdpush_lock(host);
 188
 189     if(buffer_strlen(host->rrdpush_buffer))
 190         error("STREAM %s [send]: discarding %zu bytes of metrics already in the buffer.", host->hostname, buffer_strlen(host->rrdpush_buffer));
 191
 192     buffer_flush(host->rrdpush_buffer);
 193
 194     rrdpush_sender_thread_reset_all_charts(host);
 195
 196     rrdpush_unlock(host);
 197 }
 198
 199 static void rrdpush_sender_thread_cleanup_locked_all(RRDHOST *host) {
 200     host->rrdpush_connected = 0;
 201
 202     if(host->rrdpush_socket != -1) {
 203         close(host->rrdpush_socket);
 204         host->rrdpush_socket = -1;
 205     }
 206
 207     // close the pipe
 208     if(host->rrdpush_pipe[PIPE_READ] != -1) {
 209         close(host->rrdpush_pipe[PIPE_READ]);
 210         host->rrdpush_pipe[PIPE_READ] = -1;
 211     }
 212
 213     if(host->rrdpush_pipe[PIPE_WRITE] != -1) {
 214         close(host->rrdpush_pipe[PIPE_WRITE]);
 215         host->rrdpush_pipe[PIPE_WRITE] = -1;
 216     }
 217
 218     buffer_free(host->rrdpush_buffer);
 219     host->rrdpush_buffer = NULL;
 220
 221     host->rrdpush_spawn = 0;
 222
 223     rrdhost_flag_set(host, RRDHOST_ORPHAN);
 224 }
 225
 226 void rrdpush_sender_thread_stop(RRDHOST *host) {
 227     rrdpush_lock(host);
 228     rrdhost_wrlock(host);
 229
 230     if(host->rrdpush_spawn) {
 231         info("STREAM %s [send]: stopping sending thread...", host->hostname);
 232         pthread_cancel(host->rrdpush_thread);
 233         rrdpush_sender_thread_cleanup_locked_all(host);
 234     }
 235
 236     rrdhost_unlock(host);
 237     rrdpush_unlock(host);
 238 }
 239
 240 void *rrdpush_sender_thread(void *ptr) {
 241     RRDHOST *host = (RRDHOST *)ptr;
 242
 243     info("STREAM %s [send]: thread created (task id %d)", host->hostname, gettid());
 244
 245     if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
 246         error("STREAM %s [send]: cannot set pthread cancel type to DEFERRED.", host->hostname);
 247
 248     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 249         error("STREAM %s [send]: cannot set pthread cancel state to ENABLE.", host->hostname);
 250
 251     int timeout = (int)appconfig_get_number(&stream_config, CONFIG_SECTION_STREAM, "timeout seconds", 60);
 252     int default_port = (int)appconfig_get_number(&stream_config, CONFIG_SECTION_STREAM, "default port", 19999);
 253     size_t max_size = (size_t)appconfig_get_number(&stream_config, CONFIG_SECTION_STREAM, "buffer size bytes", 1024 * 1024);
 254     unsigned int reconnect_delay = (unsigned int)appconfig_get_number(&stream_config, CONFIG_SECTION_STREAM, "reconnect delay seconds", 5);
 255     remote_clock_resync_iterations = (unsigned int)appconfig_get_number(&stream_config, CONFIG_SECTION_STREAM, "initial clock resync iterations", remote_clock_resync_iterations);
 256     char connected_to[CONNECTED_TO_SIZE + 1] = "";
 257
 258     if(!host->rrdpush_enabled || !host->rrdpush_destination || !*host->rrdpush_destination || !host->rrdpush_api_key || !*host->rrdpush_api_key)
 259         goto cleanup;
 260
 261     // initialize rrdpush globals
 262     host->rrdpush_buffer = buffer_create(1);
 263     host->rrdpush_connected = 0;
 264     if(pipe(host->rrdpush_pipe) == -1) fatal("STREAM %s [send]: cannot create required pipe.", host->hostname);
 265
 266     // initialize local variables
 267     size_t begin = 0;
 268     size_t reconnects_counter = 0;
 269     size_t sent_bytes = 0;
 270     size_t sent_connection = 0;
 271
 272     struct timeval tv = {
 273             .tv_sec = timeout,
 274             .tv_usec = 0
 275     };
 276
 277     struct pollfd fds[2], *ifd, *ofd;
 278     nfds_t fdmax;
 279
 280     ifd = &fds[0];
 281     ofd = &fds[1];
 282
 283     for(; host->rrdpush_enabled && !netdata_exit ;) {
 284
 285         if(unlikely(host->rrdpush_socket == -1)) {
 286             // stop appending data into rrdpush_buffer
 287             // they will be lost, so there is no point to do it
 288             host->rrdpush_connected = 0;
 289
 290             info("STREAM %s [send to %s]: connecting...", host->hostname, host->rrdpush_destination);
 291             host->rrdpush_socket = connect_to_one_of(host->rrdpush_destination, default_port, &tv, &reconnects_counter, connected_to, CONNECTED_TO_SIZE);
 292
 293             if(unlikely(host->rrdpush_socket == -1)) {
 294                 error("STREAM %s [send to %s]: failed to connect", host->hostname, host->rrdpush_destination);
 295                 sleep(reconnect_delay);
 296                 continue;
 297             }
 298
 299             info("STREAM %s [send to %s]: initializing communication...", host->hostname, connected_to);
 300
 301             char http[1000 + 1];
 302             snprintfz(http, 1000,
 303                     "STREAM key=%s&hostname=%s&machine_guid=%s&os=%s&update_every=%d HTTP/1.1\r\n"
 304                     "User-Agent: netdata-push-service/%s\r\n"
 305                     "Accept: */*\r\n\r\n"
 306                       , host->rrdpush_api_key
 307                       , host->hostname
 308                       , host->machine_guid
 309                       , host->os
 310                       , default_rrd_update_every
 311                       , program_version
 312             );
 313
 314             if(send_timeout(host->rrdpush_socket, http, strlen(http), 0, timeout) == -1) {
 315                 close(host->rrdpush_socket);
 316                 host->rrdpush_socket = -1;
 317                 error("STREAM %s [send to %s]: failed to send http header to netdata", host->hostname, connected_to);
 318                 sleep(reconnect_delay);
 319                 continue;
 320             }
 321
 322             info("STREAM %s [send to %s]: waiting response from remote netdata...", host->hostname, connected_to);
 323
 324             if(recv_timeout(host->rrdpush_socket, http, 1000, 0, timeout) == -1) {
 325                 close(host->rrdpush_socket);
 326                 host->rrdpush_socket = -1;
 327                 error("STREAM %s [send to %s]: failed to initialize communication", host->hostname, connected_to);
 328                 sleep(reconnect_delay);
 329                 continue;
 330             }
 331
 332             if(strncmp(http, START_STREAMING_PROMPT, strlen(START_STREAMING_PROMPT))) {
 333                 close(host->rrdpush_socket);
 334                 host->rrdpush_socket = -1;
 335                 error("STREAM %s [send to %s]: server is not replying properly.", host->hostname, connected_to);
 336                 sleep(reconnect_delay);
 337                 continue;
 338             }
 339
 340             info("STREAM %s [send to %s]: established communication - sending metrics...", host->hostname, connected_to);
 341
 342             if(fcntl(host->rrdpush_socket, F_SETFL, O_NONBLOCK) < 0)
 343                 error("STREAM %s [send to %s]: cannot set non-blocking mode for socket.", host->hostname, connected_to);
 344
 345             rrdpush_sender_thread_data_flush(host);
 346             sent_connection = 0;
 347
 348             // allow appending data into rrdpush_buffer
 349             host->rrdpush_connected = 1;
 350         }
 351
 352         ifd->fd = host->rrdpush_pipe[PIPE_READ];
 353         ifd->events = POLLIN;
 354         ifd->revents = 0;
 355
 356         ofd->fd = host->rrdpush_socket;
 357         ofd->revents = 0;
 358         if(begin < buffer_strlen(host->rrdpush_buffer)) {
 359             ofd->events = POLLOUT;
 360             fdmax = 2;
 361         }
 362         else {
 363             ofd->events = 0;
 364             fdmax = 1;
 365         }
 366
 367         if(netdata_exit) break;
 368         int retval = poll(fds, fdmax, timeout * 1000);
 369         if(netdata_exit) break;
 370
 371         if(unlikely(retval == -1)) {
 372             if(errno == EAGAIN || errno == EINTR)
 373                 continue;
 374
 375             error("STREAM %s [send to %s]: failed to poll().", host->hostname, connected_to);
 376             close(host->rrdpush_socket);
 377             host->rrdpush_socket = -1;
 378             break;
 379         }
 380         else if(unlikely(!retval)) {
 381             // timeout
 382             continue;
 383         }
 384
 385         if(ifd->revents & POLLIN) {
 386             char buffer[1000 + 1];
 387             if(read(host->rrdpush_pipe[PIPE_READ], buffer, 1000) == -1)
 388                 error("STREAM %s [send to %s]: cannot read from internal pipe.", host->hostname, connected_to);
 389         }
 390
 391         if(ofd->revents & POLLOUT && begin < buffer_strlen(host->rrdpush_buffer)) {
 392
 393             // BEGIN RRDPUSH LOCKED SESSION
 394
 395             // during this session, data collectors
 396             // will not be able to append data to our buffer
 397             // but the socket is in non-blocking mode
 398             // so, we will not block at send()
 399
 400             if(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL) != 0)
 401                 error("STREAM %s [send]: cannot set pthread cancel state to DISABLE.", host->hostname);
 402
 403             rrdpush_lock(host);
 404
 405             ssize_t ret = send(host->rrdpush_socket, &host->rrdpush_buffer->buffer[begin], buffer_strlen(host->rrdpush_buffer) - begin, MSG_DONTWAIT);
 406             if(ret == -1) {
 407                 if(errno != EAGAIN && errno != EINTR) {
 408                     error("STREAM %s [send to %s]: failed to send metrics - closing connection - we have sent %zu bytes on this connection.", host->hostname, connected_to, sent_connection);
 409                     close(host->rrdpush_socket);
 410                     host->rrdpush_socket = -1;
 411                 }
 412             }
 413             else {
 414                 sent_connection += ret;
 415                 sent_bytes += ret;
 416                 begin += ret;
 417                 if(begin == buffer_strlen(host->rrdpush_buffer)) {
 418                     // we send it all
 419
 420                     buffer_flush(host->rrdpush_buffer);
 421                     begin = 0;
 422                 }
 423             }
 424
 425             rrdpush_unlock(host);
 426
 427             if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 428                 error("STREAM %s [send]: cannot set pthread cancel state to ENABLE.", host->hostname);
 429
 430             // END RRDPUSH LOCKED SESSION
 431         }
 432
 433         // protection from overflow
 434         if(host->rrdpush_buffer->len > max_size) {
 435             errno = 0;
 436             error("STREAM %s [send to %s]: too many data pending - buffer is %zu bytes long, %zu unsent - we have sent %zu bytes in total, %zu on this connection. Closing connection to flush the data.", host->hostname, connected_to, host->rrdpush_buffer->len, host->rrdpush_buffer->len - begin, sent_bytes, sent_connection);
 437             if(host->rrdpush_socket != -1) {
 438                 close(host->rrdpush_socket);
 439                 host->rrdpush_socket = -1;
 440             }
 441         }
 442     }
 443
 444 cleanup:
 445     debug(D_WEB_CLIENT, "STREAM %s [send]: sending thread exits.", host->hostname);
 446
 447     if(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL) != 0)
 448         error("STREAM %s [send]: cannot set pthread cancel state to DISABLE.", host->hostname);
 449
 450     rrdpush_lock(host);
 451     rrdhost_wrlock(host);
 452     rrdpush_sender_thread_cleanup_locked_all(host);
 453     rrdhost_unlock(host);
 454     rrdpush_unlock(host);
 455
 456     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 457         error("STREAM %s [send]: cannot set pthread cancel state to ENABLE.", host->hostname);
 458
 459     pthread_exit(NULL);
 460     return NULL;
 461 }
 462
 463
 464 // ----------------------------------------------------------------------------
 465 // rrdpush receiver thread
 466
 467 int rrdpush_receive(int fd, const char *key, const char *hostname, const char *machine_guid, const char *os, int update_every, char *client_ip, char *client_port) {
 468     RRDHOST *host;
 469     int history = default_rrd_history_entries;
 470     RRD_MEMORY_MODE mode = default_rrd_memory_mode;
 471     int health_enabled = default_health_enabled;
 472     int rrdpush_enabled = default_rrdpush_enabled;
 473     char *rrdpush_destination = default_rrdpush_destination;
 474     char *rrdpush_api_key = default_rrdpush_api_key;
 475     time_t alarms_delay = 60;
 476
 477     update_every = (int)appconfig_get_number(&stream_config, machine_guid, "update every", update_every);
 478     if(update_every < 0) update_every = 1;
 479
 480     history = (int)appconfig_get_number(&stream_config, key, "default history", history);
 481     history = (int)appconfig_get_number(&stream_config, machine_guid, "history", history);
 482     if(history < 5) history = 5;
 483
 484     mode = rrd_memory_mode_id(appconfig_get(&stream_config, key, "default memory mode", rrd_memory_mode_name(mode)));
 485     mode = rrd_memory_mode_id(appconfig_get(&stream_config, machine_guid, "memory mode", rrd_memory_mode_name(mode)));
 486
 487     health_enabled = appconfig_get_boolean_ondemand(&stream_config, key, "health enabled by default", health_enabled);
 488     health_enabled = appconfig_get_boolean_ondemand(&stream_config, machine_guid, "health enabled", health_enabled);
 489
 490     alarms_delay = appconfig_get_number(&stream_config, key, "default postpone alarms on connect seconds", alarms_delay);
 491     alarms_delay = appconfig_get_number(&stream_config, machine_guid, "postpone alarms on connect seconds", alarms_delay);
 492
 493     rrdpush_enabled = appconfig_get_boolean(&stream_config, key, "default proxy enabled", rrdpush_enabled);
 494     rrdpush_enabled = appconfig_get_boolean(&stream_config, machine_guid, "proxy enabled", rrdpush_enabled);
 495
 496     rrdpush_destination = appconfig_get(&stream_config, key, "default proxy destination", rrdpush_destination);
 497     rrdpush_destination = appconfig_get(&stream_config, machine_guid, "proxy destination", rrdpush_destination);
 498
 499     rrdpush_api_key = appconfig_get(&stream_config, key, "default proxy api key", rrdpush_api_key);
 500     rrdpush_api_key = appconfig_get(&stream_config, machine_guid, "proxy api key", rrdpush_api_key);
 501
 502     if(!strcmp(machine_guid, "localhost"))
 503         host = localhost;
 504     else
 505         host = rrdhost_find_or_create(
 506                 hostname
 507                 , machine_guid
 508                 , os
 509                 , update_every
 510                 , history
 511                 , mode
 512                 , (health_enabled != CONFIG_BOOLEAN_NO)
 513                 , (rrdpush_enabled && rrdpush_destination && *rrdpush_destination && rrdpush_api_key && *rrdpush_api_key)
 514                 , rrdpush_destination
 515                 , rrdpush_api_key
 516         );
 517
 518     if(!host) {
 519         close(fd);
 520         error("STREAM %s [receive from [%s]:%s]: failed to find/create host structure.", hostname, client_ip, client_port);
 521         return 1;
 522     }
 523
 524 #ifdef NETDATA_INTERNAL_CHECKS
 525     info("STREAM %s [receive from [%s]:%s]: client willing to stream metrics for host '%s' with machine_guid '%s': update every = %d, history = %d, memory mode = %s, health %s"
 526          , hostname
 527          , client_ip
 528          , client_port
 529          , host->hostname
 530          , host->machine_guid
 531          , host->rrd_update_every
 532          , host->rrd_history_entries
 533          , rrd_memory_mode_name(host->rrd_memory_mode)
 534          , (health_enabled == CONFIG_BOOLEAN_NO)?"disabled":((health_enabled == CONFIG_BOOLEAN_YES)?"enabled":"auto")
 535     );
 536 #endif // NETDATA_INTERNAL_CHECKS
 537
 538     struct plugind cd = {
 539             .enabled = 1,
 540             .update_every = default_rrd_update_every,
 541             .pid = 0,
 542             .serial_failures = 0,
 543             .successful_collections = 0,
 544             .obsolete = 0,
 545             .started_t = now_realtime_sec(),
 546             .next = NULL,
 547     };
 548
 549     // put the client IP and port into the buffers used by plugins.d
 550     snprintfz(cd.id,           CONFIG_MAX_NAME,  "%s:%s", client_ip, client_port);
 551     snprintfz(cd.filename,     FILENAME_MAX,     "%s:%s", client_ip, client_port);
 552     snprintfz(cd.fullfilename, FILENAME_MAX,     "%s:%s", client_ip, client_port);
 553     snprintfz(cd.cmd,          PLUGINSD_CMD_MAX, "%s:%s", client_ip, client_port);
 554
 555     info("STREAM %s [receive from [%s]:%s]: initializing communication...", host->hostname, client_ip, client_port);
 556     if(send_timeout(fd, START_STREAMING_PROMPT, strlen(START_STREAMING_PROMPT), 0, 60) != strlen(START_STREAMING_PROMPT)) {
 557         error("STREAM %s [receive from [%s]:%s]: cannot send ready command.", host->hostname, client_ip, client_port);
 558         close(fd);
 559         return 0;
 560     }
 561
 562     // remove the non-blocking flag from the socket
 563     if(fcntl(fd, F_SETFL, fcntl(fd, F_GETFL, 0) & ~O_NONBLOCK) == -1)
 564         error("STREAM %s [receive from [%s]:%s]: cannot remove the non-blocking flag from socket %d", host->hostname, client_ip, client_port, fd);
 565
 566     // convert the socket to a FILE *
 567     FILE *fp = fdopen(fd, "r");
 568     if(!fp) {
 569         error("STREAM %s [receive from [%s]:%s]: failed to get a FILE for FD %d.", host->hostname, client_ip, client_port, fd);
 570         close(fd);
 571         return 0;
 572     }
 573
 574     rrdhost_wrlock(host);
 575     host->connected_senders++;
 576     if(health_enabled != CONFIG_BOOLEAN_NO)
 577         host->health_delay_up_to = now_realtime_sec() + alarms_delay;
 578     rrdhost_unlock(host);
 579
 580     // call the plugins.d processor to receive the metrics
 581     info("STREAM %s [receive from [%s]:%s]: receiving metrics...", host->hostname, client_ip, client_port);
 582     size_t count = pluginsd_process(host, &cd, fp, 1);
 583     error("STREAM %s [receive from [%s]:%s]: disconnected (completed updates %zu).", host->hostname, client_ip, client_port, count);
 584
 585     rrdhost_wrlock(host);
 586     host->connected_senders--;
 587     if(!host->connected_senders) {
 588         if(health_enabled == CONFIG_BOOLEAN_AUTO)
 589             host->health_enabled = 0;
 590
 591         host->senders_disconnected_time = now_realtime_sec();
 592     }
 593     rrdhost_unlock(host);
 594
 595     rrdpush_sender_thread_stop(host);
 596
 597     // cleanup
 598     fclose(fp);
 599
 600     return (int)count;
 601 }
 602
 603 struct rrdpush_thread {
 604     int fd;
 605     char *key;
 606     char *hostname;
 607     char *machine_guid;
 608     char *os;
 609     char *client_ip;
 610     char *client_port;
 611     int update_every;
 612 };
 613
 614 void *rrdpush_receiver_thread(void *ptr) {
 615     struct rrdpush_thread *rpt = (struct rrdpush_thread *)ptr;
 616
 617     if (pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
 618         error("STREAM %s [receive]: cannot set pthread cancel type to DEFERRED.", rpt->hostname);
 619
 620     if (pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 621         error("STREAM %s [receive]: cannot set pthread cancel state to ENABLE.", rpt->hostname);
 622
 623
 624     info("STREAM %s [%s]:%s: receive thread created (task id %d)", rpt->hostname, rpt->client_ip, rpt->client_port, gettid());
 625     rrdpush_receive(rpt->fd, rpt->key, rpt->hostname, rpt->machine_guid, rpt->os, rpt->update_every, rpt->client_ip, rpt->client_port);
 626     info("STREAM %s [receive from [%s]:%s]: receive thread ended (task id %d)", rpt->hostname, rpt->client_ip, rpt->client_port, gettid());
 627
 628     freez(rpt->key);
 629     freez(rpt->hostname);
 630     freez(rpt->machine_guid);
 631     freez(rpt->os);
 632     freez(rpt->client_ip);
 633     freez(rpt->client_port);
 634     freez(rpt);
 635
 636     pthread_exit(NULL);
 637     return NULL;
 638 }
 639
 640 void rrdpush_sender_thread_spawn(RRDHOST *host) {
 641     rrdhost_wrlock(host);
 642
 643     if(!host->rrdpush_spawn) {
 644         if(pthread_create(&host->rrdpush_thread, NULL, rrdpush_sender_thread, (void *) host))
 645             error("STREAM %s [send]: failed to create new thread for client.", host->hostname);
 646
 647         else if(pthread_detach(host->rrdpush_thread))
 648             error("STREAM %s [send]: cannot request detach newly created thread.", host->hostname);
 649
 650         rrdhost_flag_clear(host, RRDHOST_ORPHAN);
 651         host->rrdpush_spawn = 1;
 652     }
 653
 654     rrdhost_unlock(host);
 655 }
 656
 657 int rrdpush_receiver_thread_spawn(RRDHOST *host, struct web_client *w, char *url) {
 658     (void)host;
 659
 660     info("STREAM [receive from [%s]:%s]: new client connection.", w->client_ip, w->client_port);
 661
 662     char *key = NULL, *hostname = NULL, *machine_guid = NULL, *os = "unknown";
 663     int update_every = default_rrd_update_every;
 664     char buf[GUID_LEN + 1];
 665
 666     while(url) {
 667         char *value = mystrsep(&url, "?&");
 668         if(!value || !*value) continue;
 669
 670         char *name = mystrsep(&value, "=");
 671         if(!name || !*name) continue;
 672         if(!value || !*value) continue;
 673
 674         if(!strcmp(name, "key"))
 675             key = value;
 676         else if(!strcmp(name, "hostname"))
 677             hostname = value;
 678         else if(!strcmp(name, "machine_guid"))
 679             machine_guid = value;
 680         else if(!strcmp(name, "update_every"))
 681             update_every = (int)strtoul(value, NULL, 0);
 682         else if(!strcmp(name, "os"))
 683             os = value;
 684     }
 685
 686     if(!key || !*key) {
 687         error("STREAM [receive from [%s]:%s]: request without an API key. Forbidding access.", w->client_ip, w->client_port);
 688         buffer_flush(w->response.data);
 689         buffer_sprintf(w->response.data, "You need an API key for this request.");
 690         return 401;
 691     }
 692
 693     if(!hostname || !*hostname) {
 694         error("STREAM [receive from [%s]:%s]: request without a hostname. Forbidding access.", w->client_ip, w->client_port);
 695         buffer_flush(w->response.data);
 696         buffer_sprintf(w->response.data, "You need to send a hostname too.");
 697         return 400;
 698     }
 699
 700     if(!machine_guid || !*machine_guid) {
 701         error("STREAM [receive from [%s]:%s]: request without a machine GUID. Forbidding access.", w->client_ip, w->client_port);
 702         buffer_flush(w->response.data);
 703         buffer_sprintf(w->response.data, "You need to send a machine GUID too.");
 704         return 400;
 705     }
 706
 707     if(regenerate_guid(key, buf) == -1) {
 708         error("STREAM [receive from [%s]:%s]: API key '%s' is not valid GUID. Forbidding access.", w->client_ip, w->client_port, key);
 709         buffer_flush(w->response.data);
 710         buffer_sprintf(w->response.data, "Your API key is invalid.");
 711         return 401;
 712     }
 713
 714     if(regenerate_guid(machine_guid, buf) == -1) {
 715         error("STREAM [receive from [%s]:%s]: machine GUID '%s' is not GUID. Forbidding access.", w->client_ip, w->client_port, key);
 716         buffer_flush(w->response.data);
 717         buffer_sprintf(w->response.data, "Your machine GUID is invalid.");
 718         return 404;
 719     }
 720
 721     if(!appconfig_get_boolean(&stream_config, key, "enabled", 0)) {
 722         error("STREAM [receive from [%s]:%s]: API key '%s' is not allowed. Forbidding access.", w->client_ip, w->client_port, machine_guid);
 723         buffer_flush(w->response.data);
 724         buffer_sprintf(w->response.data, "Your API key is not permitted access.");
 725         return 401;
 726     }
 727
 728     if(!appconfig_get_boolean(&stream_config, machine_guid, "enabled", 1)) {
 729         error("STREAM [receive from [%s]:%s]: machine GUID '%s' is not allowed. Forbidding access.", w->client_ip, w->client_port, machine_guid);
 730         buffer_flush(w->response.data);
 731         buffer_sprintf(w->response.data, "Your machine guide is not permitted access.");
 732         return 404;
 733     }
 734
 735     struct rrdpush_thread *rpt = mallocz(sizeof(struct rrdpush_thread));
 736     rpt->fd           = w->ifd;
 737     rpt->key          = strdupz(key);
 738     rpt->hostname     = strdupz(hostname);
 739     rpt->machine_guid = strdupz(machine_guid);
 740     rpt->os           = strdupz(os);
 741     rpt->client_ip    = strdupz(w->client_ip);
 742     rpt->client_port  = strdupz(w->client_port);
 743     rpt->update_every = update_every;
 744     pthread_t thread;
 745
 746     debug(D_SYSTEM, "STREAM [receive from [%s]:%s]: starting receiving thread.", w->client_ip, w->client_port);
 747
 748     if(pthread_create(&thread, NULL, rrdpush_receiver_thread, (void *)rpt))
 749         error("STREAM [receive from [%s]:%s]: failed to create new thread for client.", w->client_ip, w->client_port);
 750
 751     else if(pthread_detach(thread))
 752         error("STREAM [receive from [%s]:%s]: cannot request detach newly created thread.", w->client_ip, w->client_port);
 753
 754     // prevent the caller from closing the streaming socket
 755     if(w->ifd == w->ofd)
 756         w->ifd = w->ofd = -1;
 757     else
 758         w->ifd = -1;
 759
 760     buffer_flush(w->response.data);
 761     return 200;
 762 }