src/rrdpush.c

   1 #include "common.h"
   2
   3 #define PIPE_READ 0
   4 #define PIPE_WRITE 1
   5
   6 int rrdpush_pipe[2];
   7
   8 static BUFFER *rrdpush_buffer = NULL;
   9 static pthread_mutex_t rrdpush_mutex = PTHREAD_MUTEX_INITIALIZER;
  10 static volatile RRDHOST *last_host = NULL;
  11 static volatile int rrdpush_connected = 0;
  12
  13 static inline void rrdpush_lock() {
  14     pthread_mutex_lock(&rrdpush_mutex);
  15 }
  16
  17 static inline void rrdpush_unlock() {
  18     pthread_mutex_unlock(&rrdpush_mutex);
  19 }
  20
  21 static inline int need_to_send_chart_definition(RRDSET *st) {
  22     RRDDIM *rd;
  23     rrddim_foreach_read(rd, st)
  24         if(!rrddim_flag_check(rd, RRDDIM_FLAG_EXPOSED))
  25             return 1;
  26
  27
  28     // fprintf(stderr, "NOT Sending CHART '%s' '%s'\n", st->id, st->name);
  29     return 0;
  30 }
  31
  32 static inline void send_chart_definition(RRDSET *st) {
  33     // fprintf(stderr, "Sending CHART '%s' '%s'\n", st->id, st->name);
  34
  35     buffer_sprintf(rrdpush_buffer, "CHART '%s' '%s' '%s' '%s' '%s' '%s' '%s' %ld %d\n"
  36                 , st->id
  37                 , st->name
  38                 , st->title
  39                 , st->units
  40                 , st->family
  41                 , st->context
  42                 , rrdset_type_name(st->chart_type)
  43                 , st->priority
  44                 , st->update_every
  45     );
  46
  47     RRDDIM *rd;
  48     rrddim_foreach_read(rd, st) {
  49         buffer_sprintf(rrdpush_buffer, "DIMENSION '%s' '%s' '%s' " COLLECTED_NUMBER_FORMAT " " COLLECTED_NUMBER_FORMAT " '%s %s'\n"
  50                        , rd->id
  51                        , rd->name
  52                        , rrd_algorithm_name(rd->algorithm)
  53                        , rd->multiplier
  54                        , rd->divisor
  55                        , rrddim_flag_check(rd, RRDDIM_FLAG_HIDDEN)?"hidden":""
  56                        , rrddim_flag_check(rd, RRDDIM_FLAG_DONT_DETECT_RESETS_OR_OVERFLOWS)?"noreset":""
  57         );
  58         rrddim_flag_set(rd, RRDDIM_FLAG_EXPOSED);
  59     }
  60 }
  61
  62 static inline void send_chart_metrics(RRDSET *st) {
  63     buffer_sprintf(rrdpush_buffer, "BEGIN %s %llu\n", st->id, st->usec_since_last_update);
  64
  65     RRDDIM *rd;
  66     rrddim_foreach_read(rd, st) {
  67         if(rrddim_flag_check(rd, RRDDIM_FLAG_UPDATED) && rrddim_flag_check(rd, RRDDIM_FLAG_EXPOSED))
  68             buffer_sprintf(rrdpush_buffer, "SET %s = " COLLECTED_NUMBER_FORMAT "\n"
  69                        , rd->id
  70                        , rd->collected_value
  71         );
  72     }
  73
  74     buffer_strcat(rrdpush_buffer, "END\n");
  75 }
  76
  77 static void reset_all_charts(void) {
  78     rrd_rdlock();
  79
  80     RRDHOST *host;
  81     rrdhost_foreach_read(host) {
  82         rrdhost_rdlock(host);
  83
  84         RRDSET *st;
  85         rrdset_foreach_read(st, host) {
  86             rrdset_rdlock(st);
  87
  88             RRDDIM *rd;
  89             rrddim_foreach_read(rd, st)
  90                 rrddim_flag_clear(rd, RRDDIM_FLAG_EXPOSED);
  91
  92             rrdset_unlock(st);
  93         }
  94         rrdhost_unlock(host);
  95     }
  96     rrd_unlock();
  97
  98     last_host = NULL;
  99 }
 100
 101 void rrdset_done_push(RRDSET *st) {
 102     static int error_shown = 0;
 103
 104     if(unlikely(!rrdset_flag_check(st, RRDSET_FLAG_ENABLED)))
 105         return;
 106
 107     if(unlikely(!rrdpush_buffer || !rrdpush_connected)) {
 108         if(!error_shown)
 109             error("PUSH: not ready - discarding collected metrics.");
 110
 111         error_shown = 1;
 112         return;
 113     }
 114     error_shown = 0;
 115
 116     rrdpush_lock();
 117     rrdset_rdlock(st);
 118
 119     if(st->rrdhost != last_host) {
 120         buffer_sprintf(rrdpush_buffer, "HOST '%s' '%s'\n", st->rrdhost->machine_guid, st->rrdhost->hostname);
 121         last_host = st->rrdhost;
 122     }
 123
 124     if(need_to_send_chart_definition(st))
 125         send_chart_definition(st);
 126
 127     send_chart_metrics(st);
 128
 129     // signal the sender there are more data
 130     if(write(rrdpush_pipe[PIPE_WRITE], " ", 1) == -1)
 131         error("Cannot write to internal pipe");
 132
 133     rrdset_unlock(st);
 134     rrdpush_unlock();
 135 }
 136
 137 static inline void rrdpush_flush(void) {
 138     rrdpush_lock();
 139     if(buffer_strlen(rrdpush_buffer))
 140         error("PUSH: discarding %zu bytes of metrics data already in the buffer.", buffer_strlen(rrdpush_buffer));
 141
 142     buffer_flush(rrdpush_buffer);
 143     reset_all_charts();
 144     last_host = NULL;
 145     rrdpush_unlock();
 146 }
 147
 148 void *central_netdata_push_thread(void *ptr) {
 149     struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
 150
 151     info("Central netdata push thread created with task id %d", gettid());
 152
 153     if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
 154         error("Cannot set pthread cancel type to DEFERRED.");
 155
 156     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 157         error("Cannot set pthread cancel state to ENABLE.");
 158
 159
 160     rrdpush_buffer = buffer_create(1);
 161
 162     if(pipe(rrdpush_pipe) == -1)
 163         fatal("Cannot create required pipe.");
 164
 165     struct timeval tv = {
 166             .tv_sec = 60,
 167             .tv_usec = 0
 168     };
 169
 170     rrdpush_connected = 0;
 171     size_t begin = 0;
 172     size_t max_size = 1024 * 1024;
 173     size_t reconnects_counter = 0;
 174     size_t sent_bytes = 0;
 175     size_t sent_connection = 0;
 176     int sock = -1;
 177
 178     struct pollfd fds[2], *ifd, *ofd;
 179
 180     ifd = &fds[0];
 181     ofd = &fds[1];
 182
 183     for(;;) {
 184         if(netdata_exit) break;
 185
 186         if(unlikely(sock == -1)) {
 187             rrdpush_connected = 0;
 188
 189             info("PUSH: connecting to central netdata at: %s", central_netdata_to_push_data);
 190             sock = connect_to_one_of(central_netdata_to_push_data, 19999, &tv, &reconnects_counter);
 191
 192             if(unlikely(sock == -1)) {
 193                 error("PUSH: failed to connect to central netdata at: %s", central_netdata_to_push_data);
 194                 continue;
 195             }
 196
 197             info("PUSH: connected to central netdata at: %s", central_netdata_to_push_data);
 198
 199             char http[1000 + 1];
 200             snprintfz(http, 1000, "GET /stream?key=%s HTTP/1.1\r\nUser-Agent: netdata-push-service/%s\r\nAccept: */*\r\n\r\n", config_get("global", "central netdata api key", ""), program_version);
 201             if(send_timeout(sock, http, strlen(http), 0, 60) == -1) {
 202                 close(sock);
 203                 sock = -1;
 204                 error("PUSH: failed to send http header to netdata at: %s", central_netdata_to_push_data);
 205                 sleep(5);
 206                 continue;
 207             }
 208
 209             if(recv_timeout(sock, http, 1000, 0, 60) == -1) {
 210                 close(sock);
 211                 sock = -1;
 212                 error("PUSH: failed to receive OK from netdata at: %s", central_netdata_to_push_data);
 213                 sleep(5);
 214                 continue;
 215             }
 216
 217             if(strncmp(http, "STREAM", 6)) {
 218                 close(sock);
 219                 sock = -1;
 220                 error("PUSH: netdata servers at  %s, did not send STREAM", central_netdata_to_push_data);
 221                 sleep(5);
 222                 continue;
 223             }
 224
 225             if(fcntl(sock, F_SETFL, O_NONBLOCK) < 0)
 226                 error("PUSH: cannot set non-blocking mode for socket.");
 227
 228             rrdpush_flush();
 229             sent_connection = 0;
 230             rrdpush_connected = 1;
 231         }
 232
 233         ifd->fd = rrdpush_pipe[PIPE_READ];
 234         ifd->events = POLLIN;
 235         ifd->revents = 0;
 236
 237         ofd->fd = sock;
 238         ofd->events = POLLOUT;
 239         ofd->revents = 0;
 240
 241         nfds_t fdmax = 2;
 242
 243         if(begin < buffer_strlen(rrdpush_buffer))
 244             ofd->events = POLLOUT;
 245         else
 246             ofd->events = 0;
 247
 248         if(netdata_exit) break;
 249         int retval = poll(fds, fdmax, 60 * 1000);
 250         if(netdata_exit) break;
 251
 252         if(unlikely(retval == -1)) {
 253             if(errno == EAGAIN || errno == EINTR)
 254                 continue;
 255
 256             error("PUSH: Failed to poll().");
 257             close(sock);
 258             sock = -1;
 259             break;
 260         }
 261         else if(unlikely(!retval)) {
 262             // timeout
 263             continue;
 264         }
 265
 266         if(ifd->revents & POLLIN) {
 267             char buffer[1000 + 1];
 268             if(read(rrdpush_pipe[PIPE_READ], buffer, 1000) == -1)
 269                 error("PUSH: Cannot read from internal pipe.");
 270         }
 271
 272         if(ofd->revents & POLLOUT && begin < buffer_strlen(rrdpush_buffer)) {
 273             // info("PUSH: send buffer is ready, sending %zu bytes starting at %zu", buffer_strlen(rrdpush_buffer) - begin, begin);
 274
 275             // fprintf(stderr, "PUSH BEGIN\n");
 276             // fwrite(&rrdpush_buffer->buffer[begin], 1, buffer_strlen(rrdpush_buffer) - begin, stderr);
 277             // fprintf(stderr, "\nPUSH END\n");
 278
 279             rrdpush_lock();
 280             ssize_t ret = send(sock, &rrdpush_buffer->buffer[begin], buffer_strlen(rrdpush_buffer) - begin, MSG_DONTWAIT);
 281             if(ret == -1) {
 282                 if(errno != EAGAIN && errno != EINTR) {
 283                     error("PUSH: failed to send metrics to central netdata at %s. We have sent %zu bytes on this connection.", central_netdata_to_push_data, sent_connection);
 284                     close(sock);
 285                     sock = -1;
 286                 }
 287             }
 288             else {
 289                 sent_connection += ret;
 290                 sent_bytes += ret;
 291                 begin += ret;
 292                 if(begin == buffer_strlen(rrdpush_buffer)) {
 293                     buffer_flush(rrdpush_buffer);
 294                     begin = 0;
 295                 }
 296             }
 297             rrdpush_unlock();
 298         }
 299
 300         // protection from overflow
 301         if(rrdpush_buffer->len > max_size) {
 302             errno = 0;
 303             error("PUSH: 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);
 304             if(sock != -1) {
 305                 close(sock);
 306                 sock = -1;
 307             }
 308         }
 309     }
 310
 311     debug(D_WEB_CLIENT, "Central netdata push thread exits.");
 312     if(sock != -1) {
 313         close(sock);
 314     }
 315
 316     static_thread->enabled = 0;
 317     pthread_exit(NULL);
 318     return NULL;
 319 }