src/rrdpush.c

   1 #include "common.h"
   2
   3 int rrdpush_enabled = 0;
   4 int rrdpush_exclusive = 1;
   5
   6 static char *central_netdata = NULL;
   7 static char *api_key = NULL;
   8
   9 #define CONNECTED_TO_SIZE 100
  10
  11 // data collection happens from multiple threads
  12 // each of these threads calls rrdset_done()
  13 // which in turn calls rrdset_done_push()
  14 // which uses this pipe to notify the streaming thread
  15 // that there are more data ready to be sent
  16 #define PIPE_READ 0
  17 #define PIPE_WRITE 1
  18 int rrdpush_pipe[2] = { -1, -1 };
  19
  20 // a buffer used to store data to be sent.
  21 // the format is the same as external plugins.
  22 static BUFFER *rrdpush_buffer = NULL;
  23
  24 // locking to get exclusive access to shared resources
  25 // (rrdpush_pipe[PIPE_WRITE], rrdpush_buffer
  26 static pthread_mutex_t rrdpush_mutex = PTHREAD_MUTEX_INITIALIZER;
  27
  28 // if the streaming thread is connected to a central netdata
  29 // this is set to 1, otherwise 0.
  30 static volatile int rrdpush_connected = 0;
  31
  32 // to have the remote netdata re-sync the charts
  33 // to its current clock, we send for this many
  34 // iterations a BEGIN line without microseconds
  35 // this is for the first iterations of each chart
  36 static unsigned int remote_clock_resync_iterations = 60;
  37
  38 #define rrdpush_lock() pthread_mutex_lock(&rrdpush_mutex)
  39 #define rrdpush_unlock() pthread_mutex_unlock(&rrdpush_mutex)
  40
  41 // checks if the current chart definition has been sent
  42 static inline int need_to_send_chart_definition(RRDSET *st) {
  43     RRDDIM *rd;
  44     rrddim_foreach_read(rd, st)
  45         if(!rrddim_flag_check(rd, RRDDIM_FLAG_EXPOSED))
  46             return 1;
  47
  48     return 0;
  49 }
  50
  51 // sends the current chart definition
  52 static inline void send_chart_definition(RRDSET *st) {
  53     buffer_sprintf(rrdpush_buffer, "CHART '%s' '%s' '%s' '%s' '%s' '%s' '%s' %ld %d\n"
  54                 , st->id
  55                 , st->name
  56                 , st->title
  57                 , st->units
  58                 , st->family
  59                 , st->context
  60                 , rrdset_type_name(st->chart_type)
  61                 , st->priority
  62                 , st->update_every
  63     );
  64
  65     RRDDIM *rd;
  66     rrddim_foreach_read(rd, st) {
  67         buffer_sprintf(rrdpush_buffer, "DIMENSION '%s' '%s' '%s' " COLLECTED_NUMBER_FORMAT " " COLLECTED_NUMBER_FORMAT " '%s %s'\n"
  68                        , rd->id
  69                        , rd->name
  70                        , rrd_algorithm_name(rd->algorithm)
  71                        , rd->multiplier
  72                        , rd->divisor
  73                        , rrddim_flag_check(rd, RRDDIM_FLAG_HIDDEN)?"hidden":""
  74                        , rrddim_flag_check(rd, RRDDIM_FLAG_DONT_DETECT_RESETS_OR_OVERFLOWS)?"noreset":""
  75         );
  76         rrddim_flag_set(rd, RRDDIM_FLAG_EXPOSED);
  77     }
  78 }
  79
  80 // sends the current chart dimensions
  81 static inline void send_chart_metrics(RRDSET *st) {
  82     buffer_sprintf(rrdpush_buffer, "BEGIN %s %llu\n", st->id, (st->counter_done > remote_clock_resync_iterations)?st->usec_since_last_update:0);
  83
  84     RRDDIM *rd;
  85     rrddim_foreach_read(rd, st) {
  86         if(rrddim_flag_check(rd, RRDDIM_FLAG_UPDATED) && rrddim_flag_check(rd, RRDDIM_FLAG_EXPOSED))
  87             buffer_sprintf(rrdpush_buffer, "SET %s = " COLLECTED_NUMBER_FORMAT "\n"
  88                        , rd->id
  89                        , rd->collected_value
  90         );
  91     }
  92
  93     buffer_strcat(rrdpush_buffer, "END\n");
  94 }
  95
  96 // resets all the chart, so that their definitions
  97 // will be resent to the central netdata
  98 static void reset_all_charts(void) {
  99     rrd_rdlock();
 100
 101     RRDHOST *host;
 102     rrdhost_foreach_read(host) {
 103         rrdhost_rdlock(host);
 104
 105         RRDSET *st;
 106         rrdset_foreach_read(st, host) {
 107
 108             // make it re-align the current time
 109             // on the remote host
 110             st->counter_done = 0;
 111
 112             rrdset_rdlock(st);
 113
 114             RRDDIM *rd;
 115             rrddim_foreach_read(rd, st)
 116                 rrddim_flag_clear(rd, RRDDIM_FLAG_EXPOSED);
 117
 118             rrdset_unlock(st);
 119         }
 120         rrdhost_unlock(host);
 121     }
 122     rrd_unlock();
 123 }
 124
 125 void rrdset_done_push(RRDSET *st) {
 126     static int error_shown = 0;
 127
 128     if(unlikely(!rrdset_flag_check(st, RRDSET_FLAG_ENABLED)))
 129         return;
 130
 131     rrdpush_lock();
 132
 133     if(unlikely(!rrdpush_buffer || !rrdpush_connected)) {
 134         if(!error_shown)
 135             error("STREAM: not ready - discarding collected metrics.");
 136
 137         error_shown = 1;
 138
 139         rrdpush_unlock();
 140         return;
 141     }
 142     error_shown = 0;
 143
 144     rrdset_rdlock(st);
 145     if(need_to_send_chart_definition(st))
 146         send_chart_definition(st);
 147
 148     send_chart_metrics(st);
 149     rrdset_unlock(st);
 150
 151     // signal the sender there are more data
 152     if(write(rrdpush_pipe[PIPE_WRITE], " ", 1) == -1)
 153         error("STREAM: cannot write to internal pipe");
 154
 155     rrdpush_unlock();
 156 }
 157
 158 static inline void rrdpush_flush(void) {
 159     rrdpush_lock();
 160     if(buffer_strlen(rrdpush_buffer))
 161         error("STREAM: discarding %zu bytes of metrics data already in the buffer.", buffer_strlen(rrdpush_buffer));
 162
 163     buffer_flush(rrdpush_buffer);
 164     reset_all_charts();
 165     rrdpush_unlock();
 166 }
 167
 168 int rrdpush_init() {
 169     rrdpush_enabled = config_get_boolean("stream", "enabled", rrdpush_enabled);
 170     rrdpush_exclusive = config_get_boolean("stream", "exclusive", rrdpush_exclusive);
 171     central_netdata = config_get("stream", "stream metrics to", "");
 172     api_key = config_get("stream", "api key", "");
 173
 174     if(!rrdpush_enabled || !central_netdata || !*central_netdata || !api_key || !*api_key) {
 175         rrdpush_enabled = 0;
 176         rrdpush_exclusive = 0;
 177     }
 178
 179     return rrdpush_enabled;
 180 }
 181
 182 void *central_netdata_push_thread(void *ptr) {
 183     struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
 184
 185     info("STREAM: central netdata push thread created with task id %d", gettid());
 186
 187     if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
 188         error("STREAM: cannot set pthread cancel type to DEFERRED.");
 189
 190     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 191         error("STREAM: cannot set pthread cancel state to ENABLE.");
 192
 193     int timeout = (int)config_get_number("stream", "timeout seconds", 60);
 194     int default_port = (int)config_get_number("stream", "default port", 19999);
 195     size_t max_size = (size_t)config_get_number("stream", "buffer size bytes", 1024 * 1024);
 196     unsigned int reconnect_delay = (unsigned int)config_get_number("stream", "reconnect delay seconds", 5);
 197     remote_clock_resync_iterations = (unsigned int)config_get_number("stream", "initial clock resync iterations", remote_clock_resync_iterations);
 198     int sock = -1;
 199
 200     if(!rrdpush_enabled || !central_netdata || !*central_netdata || !api_key || !*api_key)
 201         goto cleanup;
 202
 203     // initialize rrdpush globals
 204     rrdpush_buffer = buffer_create(1);
 205     rrdpush_connected = 0;
 206     if(pipe(rrdpush_pipe) == -1) fatal("STREAM: cannot create required pipe.");
 207
 208     // initialize local variables
 209     size_t begin = 0;
 210     size_t reconnects_counter = 0;
 211     size_t sent_bytes = 0;
 212     size_t sent_connection = 0;
 213
 214     struct timeval tv = {
 215             .tv_sec = timeout,
 216             .tv_usec = 0
 217     };
 218
 219     struct pollfd fds[2], *ifd, *ofd;
 220     nfds_t fdmax;
 221
 222     ifd = &fds[0];
 223     ofd = &fds[1];
 224
 225     char connected_to[CONNECTED_TO_SIZE + 1];
 226
 227     for(;;) {
 228         if(netdata_exit) break;
 229
 230         if(unlikely(sock == -1)) {
 231             // stop appending data into rrdpush_buffer
 232             // they will be lost, so there is no point to do it
 233             rrdpush_connected = 0;
 234
 235             info("STREAM: connecting to central netdata at: %s", central_netdata);
 236             sock = connect_to_one_of(central_netdata, default_port, &tv, &reconnects_counter, connected_to, CONNECTED_TO_SIZE);
 237
 238             if(unlikely(sock == -1)) {
 239                 error("STREAM: failed to connect to central netdata at: %s", central_netdata);
 240                 sleep(reconnect_delay);
 241                 continue;
 242             }
 243
 244             info("STREAM: initializing communication to central netdata at: %s", connected_to);
 245
 246             char http[1000 + 1];
 247             snprintfz(http, 1000,
 248                     "STREAM key=%s&hostname=%s&machine_guid=%s&os=%s&update_every=%d HTTP/1.1\r\n"
 249                     "User-Agent: netdata-push-service/%s\r\n"
 250                     "Accept: */*\r\n\r\n"
 251                       , api_key
 252                       , localhost->hostname
 253                       , localhost->machine_guid
 254                       , localhost->os
 255                       , default_rrd_update_every
 256                       , program_version
 257             );
 258
 259             if(send_timeout(sock, http, strlen(http), 0, timeout) == -1) {
 260                 close(sock);
 261                 sock = -1;
 262                 error("STREAM: failed to send http header to netdata at: %s", connected_to);
 263                 sleep(reconnect_delay);
 264                 continue;
 265             }
 266
 267             info("STREAM: Waiting for STREAM from central netdata at: %s", connected_to);
 268
 269             if(recv_timeout(sock, http, 1000, 0, timeout) == -1) {
 270                 close(sock);
 271                 sock = -1;
 272                 error("STREAM: failed to receive STREAM from netdata at: %s", connected_to);
 273                 sleep(reconnect_delay);
 274                 continue;
 275             }
 276
 277             if(strncmp(http, "STREAM", 6)) {
 278                 close(sock);
 279                 sock = -1;
 280                 error("STREAM: server at %s, did not send STREAM", connected_to);
 281                 sleep(reconnect_delay);
 282                 continue;
 283             }
 284
 285             info("STREAM: Established communication with central netdata at: %s - sending metrics...", connected_to);
 286
 287             if(fcntl(sock, F_SETFL, O_NONBLOCK) < 0)
 288                 error("STREAM: cannot set non-blocking mode for socket.");
 289
 290             rrdpush_flush();
 291             sent_connection = 0;
 292
 293             // allow appending data into rrdpush_buffer
 294             rrdpush_connected = 1;
 295         }
 296
 297         ifd->fd = rrdpush_pipe[PIPE_READ];
 298         ifd->events = POLLIN;
 299         ifd->revents = 0;
 300
 301         ofd->fd = sock;
 302         ofd->revents = 0;
 303         if(begin < buffer_strlen(rrdpush_buffer)) {
 304             ofd->events = POLLOUT;
 305             fdmax = 2;
 306         }
 307         else {
 308             ofd->events = 0;
 309             fdmax = 1;
 310         }
 311
 312         if(netdata_exit) break;
 313         int retval = poll(fds, fdmax, timeout * 1000);
 314         if(netdata_exit) break;
 315
 316         if(unlikely(retval == -1)) {
 317             if(errno == EAGAIN || errno == EINTR)
 318                 continue;
 319
 320             error("STREAM: Failed to poll().");
 321             close(sock);
 322             sock = -1;
 323             break;
 324         }
 325         else if(unlikely(!retval)) {
 326             // timeout
 327             continue;
 328         }
 329
 330         if(ifd->revents & POLLIN) {
 331             char buffer[1000 + 1];
 332             if(read(rrdpush_pipe[PIPE_READ], buffer, 1000) == -1)
 333                 error("STREAM: Cannot read from internal pipe.");
 334         }
 335
 336         if(ofd->revents & POLLOUT && begin < buffer_strlen(rrdpush_buffer)) {
 337             // info("STREAM: send buffer is ready, sending %zu bytes starting at %zu", buffer_strlen(rrdpush_buffer) - begin, begin);
 338
 339             // fprintf(stderr, "PUSH BEGIN\n");
 340             // fwrite(&rrdpush_buffer->buffer[begin], 1, buffer_strlen(rrdpush_buffer) - begin, stderr);
 341             // fprintf(stderr, "\nPUSH END\n");
 342
 343             rrdpush_lock();
 344             ssize_t ret = send(sock, &rrdpush_buffer->buffer[begin], buffer_strlen(rrdpush_buffer) - begin, MSG_DONTWAIT);
 345             if(ret == -1) {
 346                 if(errno != EAGAIN && errno != EINTR) {
 347                     error("STREAM: failed to send metrics to central netdata at %s. We have sent %zu bytes on this connection.", connected_to, sent_connection);
 348                     close(sock);
 349                     sock = -1;
 350                 }
 351             }
 352             else {
 353                 sent_connection += ret;
 354                 sent_bytes += ret;
 355                 begin += ret;
 356                 if(begin == buffer_strlen(rrdpush_buffer)) {
 357                     buffer_flush(rrdpush_buffer);
 358                     begin = 0;
 359                 }
 360             }
 361             rrdpush_unlock();
 362         }
 363
 364         // protection from overflow
 365         if(rrdpush_buffer->len > max_size) {
 366             errno = 0;
 367             error("STREAM: 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.", rrdpush_buffer->len, rrdpush_buffer->len - begin, sent_bytes, sent_connection);
 368             if(sock != -1) {
 369                 close(sock);
 370                 sock = -1;
 371             }
 372         }
 373     }
 374
 375 cleanup:
 376     debug(D_WEB_CLIENT, "STREAM: central netdata push thread exits.");
 377
 378     // make sure the data collection threads do not write data
 379     rrdpush_connected = 0;
 380
 381     // close the pipe
 382     if(rrdpush_pipe[PIPE_READ] != -1)  close(rrdpush_pipe[PIPE_READ]);
 383     if(rrdpush_pipe[PIPE_WRITE] != -1) close(rrdpush_pipe[PIPE_WRITE]);
 384
 385     // close the socket
 386     if(sock != -1) close(sock);
 387
 388     rrdpush_lock();
 389     buffer_free(rrdpush_buffer);
 390     rrdpush_buffer = NULL;
 391     rrdpush_unlock();
 392
 393     static_thread->enabled = 0;
 394     pthread_exit(NULL);
 395     return NULL;
 396 }