| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "stream-sender-internals.h" |
| 4 | #include "stream-replication-sender.h" |
| 5 | |
| 6 | #define TIME_TO_CONSIDER_PARENTS_SIMILAR 120 |
| 7 | |
| 8 | struct stream_parent { |
| 9 | STRING *destination; // the parent destination |
| 10 | bool ssl; // the parent uses SSL |
| 11 | |
| 12 | bool banned_permanently; // when the parent is the origin of this host |
| 13 | bool banned_for_this_session; // when the parent is before us in the streaming path |
| 14 | bool banned_temporarily_erroneous; // when the parent is blocked by another node we host |
| 15 | STREAM_HANDSHAKE reason; |
| 16 | uint32_t attempts; // how many times we have tried to connect to this parent |
| 17 | usec_t since_ut; // the last time we tried to connect to it |
| 18 | usec_t postpone_until_ut; // based on the reason, a randomized time to wait for reconnection |
| 19 | |
| 20 | struct { |
| 21 | ND_UUID host_id; // the machine_guid of the agent |
| 22 | int status; // the response code of the stream_info call |
| 23 | uint32_t nonce; // a random 32-bit number |
| 24 | size_t nodes; // how many nodes the parent has |
| 25 | size_t receivers; // how many receivers the parent has |
| 26 | |
| 27 | // these are from RRDHOST_STATUS and can only be used when status == 200 |
| 28 | RRDHOST_DB_STATUS db_status; |
| 29 | RRDHOST_DB_LIVENESS db_liveness; |
| 30 | RRDHOST_INGEST_TYPE ingest_type; |
| 31 | RRDHOST_INGEST_STATUS ingest_status; |
| 32 | time_t db_first_time_s; // the oldest timestamp for us in the parent's database |
| 33 | time_t db_last_time_s; // the latest timestamp for us in the parent's database |
| 34 | } remote; |
| 35 | |
| 36 | struct { |
| 37 | size_t batch; // the batch priority (>= 1, 0 == excluded) |
| 38 | size_t order; // the final order of the parent (>= 1, 0 == excluded) |
| 39 | bool random; // this batch has more than 1 parents, so we flipped coins to select order |
| 40 | bool info; // we go stream info from the parent |
| 41 | bool skipped; // we skipped this parent for some reason |
| 42 | } selection; |
| 43 | |
| 44 | STREAM_PARENT *prev; |
| 45 | STREAM_PARENT *next; |
| 46 | }; |
| 47 | |
| 48 | static const char *stream_parent_effective_service(const char *definition, int default_port, char *service, size_t service_size) { |
| 49 | if(!service || !service_size) |
| 50 | return ""; |
| 51 | |
| 52 | if(!connect_to_definition_get_service(definition, default_port, service, service_size)) |
| 53 | snprintfz(service, service_size, "%d", default_port); |
| 54 | |
| 55 | return service; |
| 56 | } |
| 57 | |
| 58 | // -------------------------------------------------------------------------------------------------------------------- |
| 59 | // block unresponsive parents for some time, to allow speeding up the connection of the rest |
| 60 | |
| 61 | struct blocked_parent { |
| 62 | STRING *destination; |
| 63 | usec_t until; |
| 64 | }; |
| 65 | |
| 66 | DEFINE_JUDYL_TYPED(BLOCKED_PARENTS, struct blocked_parent *); |
| 67 | static BLOCKED_PARENTS_JudyLSet blocked_parents_set = { 0 }; |
| 68 | static RW_SPINLOCK blocked_parents_spinlock = RW_SPINLOCK_INITIALIZER; |
| 69 | |
| 70 | static void block_parent_for_all_nodes(STREAM_PARENT *d, time_t duration_s) { |
| 71 | rw_spinlock_write_lock(&blocked_parents_spinlock); |
| 72 | |
| 73 | struct blocked_parent *p = BLOCKED_PARENTS_GET(&blocked_parents_set, (Word_t)d->destination); |
| 74 | if(!p) { |
| 75 | p = callocz(1, sizeof(*p)); |
| 76 | p->destination = string_dup(d->destination); |
| 77 | BLOCKED_PARENTS_SET(&blocked_parents_set, (Word_t)p->destination, p); |
| 78 | } |
| 79 | p->until = now_monotonic_usec() + duration_s * USEC_PER_SEC; |
| 80 | |
| 81 | rw_spinlock_write_unlock(&blocked_parents_spinlock); |
| 82 | } |
| 83 | |
| 84 | static bool is_a_blocked_parent(STREAM_PARENT *d) { |
| 85 | rw_spinlock_read_lock(&blocked_parents_spinlock); |
| 86 | |
| 87 | struct blocked_parent *p = BLOCKED_PARENTS_GET(&blocked_parents_set, (Word_t)d->destination); |
| 88 | bool ret = p && p->until > now_monotonic_usec(); |
| 89 | |
| 90 | rw_spinlock_read_unlock(&blocked_parents_spinlock); |
| 91 | return ret; |
| 92 | } |
| 93 | |
| 94 | // -------------------------------------------------------------------------------------------------------------------- |
| 95 | |
| 96 | STREAM_HANDSHAKE stream_parent_get_disconnect_reason(STREAM_PARENT *d) { |
| 97 | if(!d) return STREAM_HANDSHAKE_PARENT_INTERNAL_ERROR; |
| 98 | return d->reason; |
| 99 | } |
| 100 | |
| 101 | void stream_parent_set_host_disconnect_reason(RRDHOST *host, STREAM_HANDSHAKE reason, time_t since) { |
| 102 | host->stream.snd.status.reason = reason; |
| 103 | struct stream_parent *d = host->stream.snd.parents.current; |
| 104 | if(!d) return; |
| 105 | d->since_ut = since * USEC_PER_SEC; |
| 106 | d->reason = reason; |
| 107 | } |
| 108 | |
| 109 | static inline usec_t randomize_wait_ut(time_t min, time_t max) { |
| 110 | min = (min < SENDER_MIN_RECONNECT_DELAY ? SENDER_MIN_RECONNECT_DELAY : min); |
| 111 | if(max < min) max = min; |
| 112 | |
| 113 | usec_t min_ut = min * USEC_PER_SEC; |
| 114 | usec_t max_ut = max * USEC_PER_SEC; |
| 115 | usec_t wait_ut = min_ut + os_random(max_ut - min_ut); |
| 116 | return now_realtime_usec() + wait_ut; |
| 117 | } |
| 118 | |
| 119 | void stream_parents_host_reset(RRDHOST *host, STREAM_HANDSHAKE reason) { |
| 120 | usec_t until_ut = randomize_wait_ut(stream_send.parents.reconnect_delay_s / 2, stream_send.parents.reconnect_delay_s + 5); |
| 121 | rw_spinlock_write_lock(&host->stream.snd.parents.spinlock); |
| 122 | for (STREAM_PARENT *d = host->stream.snd.parents.all; d; d = d->next) { |
| 123 | d->postpone_until_ut = until_ut; |
| 124 | d->banned_for_this_session = false; |
| 125 | d->reason = reason; |
| 126 | } |
| 127 | rw_spinlock_write_unlock(&host->stream.snd.parents.spinlock); |
| 128 | } |
| 129 | |
| 130 | static void stream_parent_set_reconnect_delay(STREAM_PARENT *d, STREAM_HANDSHAKE reason, time_t secs) { |
| 131 | if(!d) return; |
| 132 | d->reason = reason; |
| 133 | d->postpone_until_ut = randomize_wait_ut(5, secs); |
| 134 | } |
| 135 | |
| 136 | void stream_parent_set_host_reconnect_delay(RRDHOST *host, STREAM_HANDSHAKE reason, time_t secs) { |
| 137 | stream_parent_set_reconnect_delay(host->stream.snd.parents.current, reason, secs); |
| 138 | } |
| 139 | |
| 140 | static void stream_parent_set_connect_failure_reason(RRDHOST *host, STREAM_PARENT *d, STREAM_HANDSHAKE reason, time_t secs) { |
| 141 | host->stream.snd.status.reason = reason; |
| 142 | pulse_host_status(host, PULSE_HOST_STATUS_SND_NO_DST_FAILED, reason); |
| 143 | pulse_sender_connection_failed(d ? string2str(d->destination) : NULL, reason); |
| 144 | stream_parent_set_reconnect_delay(d, reason, secs); |
| 145 | } |
| 146 | |
| 147 | void stream_parent_set_host_connect_failure_reason(RRDHOST *host, STREAM_HANDSHAKE reason, time_t secs) { |
| 148 | stream_parent_set_connect_failure_reason(host, host->stream.snd.parents.current, reason, secs); |
| 149 | } |
| 150 | |
| 151 | usec_t stream_parent_get_reconnection_ut(STREAM_PARENT *d) { |
| 152 | return d ? d->postpone_until_ut : 0; |
| 153 | } |
| 154 | |
| 155 | bool stream_parent_is_ssl(STREAM_PARENT *d) { |
| 156 | return d ? d->ssl : false; |
| 157 | } |
| 158 | |
| 159 | usec_t stream_parent_handshake_error_to_json(BUFFER *wb, RRDHOST *host) { |
| 160 | usec_t last_attempt = 0; |
| 161 | rw_spinlock_read_lock(&host->stream.snd.parents.spinlock); |
| 162 | for(STREAM_PARENT *d = host->stream.snd.parents.all; d ; d = d->next) { |
| 163 | if(d->since_ut > last_attempt) |
| 164 | last_attempt = d->since_ut; |
| 165 | |
| 166 | buffer_json_add_array_item_string(wb, stream_handshake_error_to_string(d->reason)); |
| 167 | } |
| 168 | rw_spinlock_read_unlock(&host->stream.snd.parents.spinlock); |
| 169 | return last_attempt; |
| 170 | } |
| 171 | |
| 172 | void rrdhost_stream_parents_to_json(BUFFER *wb, RRDHOST_STATUS *s) { |
| 173 | char buf[1024]; |
| 174 | |
| 175 | rw_spinlock_read_lock(&s->host->stream.snd.parents.spinlock); |
| 176 | |
| 177 | usec_t now_ut = now_realtime_usec(); |
| 178 | STREAM_PARENT *d; |
| 179 | for (d = s->host->stream.snd.parents.all; d; d = d->next) { |
| 180 | buffer_json_add_array_item_object(wb); |
| 181 | buffer_json_member_add_uint64(wb, "attempts", d->attempts + 1); |
| 182 | { |
| 183 | if (d->ssl) { |
| 184 | snprintfz(buf, sizeof(buf) - 1, "%s:SSL", string2str(d->destination)); |
| 185 | buffer_json_member_add_string(wb, "destination", buf); |
| 186 | } |
| 187 | else |
| 188 | buffer_json_member_add_string(wb, "destination", string2str(d->destination)); |
| 189 | |
| 190 | buffer_json_member_add_datetime_rfc3339(wb, "since", d->since_ut, false); |
| 191 | buffer_json_member_add_duration_ut(wb, "age", d->since_ut < now_ut ? (int64_t)(now_ut - d->since_ut) : 0); |
| 192 | |
| 193 | if(!d->banned_for_this_session && !d->banned_permanently && !d->banned_temporarily_erroneous) { |
| 194 | buffer_json_member_add_string(wb, "last_handshake", stream_handshake_error_to_string(d->reason)); |
| 195 | |
| 196 | if (d->postpone_until_ut > now_ut) { |
| 197 | buffer_json_member_add_datetime_rfc3339(wb, "next_check", d->postpone_until_ut, false); |
| 198 | buffer_json_member_add_duration_ut(wb, "next_in", (int64_t)(d->postpone_until_ut - now_ut)); |
| 199 | } |
| 200 | |
| 201 | if(d->selection.batch) { |
| 202 | buffer_json_member_add_uint64(wb, "batch", d->selection.batch); |
| 203 | buffer_json_member_add_uint64(wb, "order", d->selection.order); |
| 204 | buffer_json_member_add_boolean(wb, "random", d->selection.random); |
| 205 | } |
| 206 | |
| 207 | buffer_json_member_add_boolean(wb, "info", d->selection.info); |
| 208 | buffer_json_member_add_boolean(wb, "skipped", d->selection.skipped); |
| 209 | } |
| 210 | else { |
| 211 | if(d->banned_permanently) |
| 212 | buffer_json_member_add_string(wb, "ban", "it is the localhost"); |
| 213 | else if(d->banned_for_this_session) |
| 214 | buffer_json_member_add_string(wb, "ban", "it is our parent"); |
| 215 | else if(d->banned_temporarily_erroneous) |
| 216 | buffer_json_member_add_string(wb, "ban", "it is erroneous"); |
| 217 | } |
| 218 | } |
| 219 | buffer_json_object_close(wb); // each candidate |
| 220 | } |
| 221 | |
| 222 | rw_spinlock_read_unlock(&s->host->stream.snd.parents.spinlock); |
| 223 | } |
| 224 | |
| 225 | void rrdhost_stream_parent_ssl_init(struct sender_state *s) { |
| 226 | static SPINLOCK sp = SPINLOCK_INITIALIZER; |
| 227 | spinlock_lock(&sp); |
| 228 | |
| 229 | if(netdata_ssl_streaming_sender_ctx || !s->host) { |
| 230 | spinlock_unlock(&sp); |
| 231 | goto cleanup; |
| 232 | } |
| 233 | |
| 234 | rw_spinlock_read_lock(&s->host->stream.snd.parents.spinlock); |
| 235 | |
| 236 | for(STREAM_PARENT *d = s->host->stream.snd.parents.all; d ; d = d->next) { |
| 237 | if (d->ssl) { |
| 238 | // we need to initialize SSL |
| 239 | |
| 240 | netdata_ssl_initialize_ctx(NETDATA_SSL_STREAMING_SENDER_CTX); |
| 241 | |
| 242 | ssl_security_location_for_context( |
| 243 | netdata_ssl_streaming_sender_ctx, |
| 244 | string2str(stream_send.parents.ssl_ca_file), |
| 245 | string2str(stream_send.parents.ssl_ca_path)); |
| 246 | |
| 247 | // stop the loop |
| 248 | break; |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | rw_spinlock_read_unlock(&s->host->stream.snd.parents.spinlock); |
| 253 | spinlock_unlock(&sp); |
| 254 | |
| 255 | cleanup: |
| 256 | s->sock.ctx = netdata_ssl_streaming_sender_ctx; |
| 257 | s->sock.verify_certificate = netdata_ssl_validate_certificate_sender; |
| 258 | } |
| 259 | |
| 260 | static void stream_parent_nd_sock_error_to_reason(STREAM_PARENT *d, ND_SOCK *sock) { |
| 261 | switch (sock->error) { |
| 262 | case ND_SOCK_ERR_CONNECTION_REFUSED: |
| 263 | d->reason = STREAM_HANDSHAKE_SP_CONNECTION_REFUSED; |
| 264 | d->postpone_until_ut = randomize_wait_ut(30, 60); |
| 265 | block_parent_for_all_nodes(d, 30); |
| 266 | break; |
| 267 | |
| 268 | case ND_SOCK_ERR_CANNOT_RESOLVE_HOSTNAME: |
| 269 | d->reason = STREAM_HANDSHAKE_SP_CANT_RESOLVE_HOSTNAME; |
| 270 | d->postpone_until_ut = randomize_wait_ut(30, 60); |
| 271 | block_parent_for_all_nodes(d, 30); |
| 272 | break; |
| 273 | |
| 274 | case ND_SOCK_ERR_NO_HOST_IN_DEFINITION: |
| 275 | d->reason = STREAM_HANDSHAKE_SP_NO_HOST_IN_DESTINATION; |
| 276 | d->banned_for_this_session = true; |
| 277 | d->postpone_until_ut = randomize_wait_ut(30, 60); |
| 278 | block_parent_for_all_nodes(d, 30); |
| 279 | break; |
| 280 | |
| 281 | case ND_SOCK_ERR_TIMEOUT: |
| 282 | d->reason = STREAM_HANDSHAKE_SP_CONNECT_TIMEOUT; |
| 283 | d->postpone_until_ut = randomize_wait_ut(300, d->remote.nodes < 10 ? 600 : 900); |
| 284 | block_parent_for_all_nodes(d, 300); |
| 285 | break; |
| 286 | |
| 287 | case ND_SOCK_ERR_SSL_INVALID_CERTIFICATE: |
| 288 | d->reason = STREAM_HANDSHAKE_CONNECT_INVALID_CERTIFICATE; |
| 289 | d->postpone_until_ut = randomize_wait_ut(300, 600); |
| 290 | block_parent_for_all_nodes(d, 300); |
| 291 | break; |
| 292 | |
| 293 | case ND_SOCK_ERR_SSL_CANT_ESTABLISH_SSL_CONNECTION: |
| 294 | case ND_SOCK_ERR_SSL_FAILED_TO_OPEN: |
| 295 | d->reason = STREAM_HANDSHAKE_CONNECT_SSL_ERROR; |
| 296 | d->postpone_until_ut = randomize_wait_ut(60, 180); |
| 297 | block_parent_for_all_nodes(d, 60); |
| 298 | break; |
| 299 | |
| 300 | default: |
| 301 | case ND_SOCK_ERR_POLL_ERROR: |
| 302 | case ND_SOCK_ERR_FAILED_TO_CREATE_SOCKET: |
| 303 | case ND_SOCK_ERR_UNKNOWN_ERROR: |
| 304 | d->reason = STREAM_HANDSHAKE_PARENT_INTERNAL_ERROR; |
| 305 | d->postpone_until_ut = randomize_wait_ut(30, 60); |
| 306 | break; |
| 307 | |
| 308 | case ND_SOCK_ERR_THREAD_CANCELLED: |
| 309 | case ND_SOCK_ERR_NO_DESTINATION_AVAILABLE: |
| 310 | d->reason = STREAM_HANDSHAKE_PARENT_INTERNAL_ERROR; |
| 311 | d->postpone_until_ut = randomize_wait_ut(30, 60); |
| 312 | break; |
| 313 | } |
| 314 | } |
| 315 | |
| 316 | int stream_info_to_json_v1(BUFFER *wb, const char *machine_guid) { |
| 317 | pulse_parent_stream_info_received_request(); |
| 318 | |
| 319 | buffer_reset(wb); |
| 320 | buffer_json_initialize(wb, "\"", "\"", 0, true, BUFFER_JSON_OPTIONS_DEFAULT); |
| 321 | |
| 322 | RRDHOST_STATUS status = { 0 }; |
| 323 | int ret = HTTP_RESP_OK; |
| 324 | RRDHOST *host; |
| 325 | if(!machine_guid || !*machine_guid || !(host = rrdhost_find_by_guid(machine_guid))) |
| 326 | ret = HTTP_RESP_NOT_FOUND; |
| 327 | else |
| 328 | rrdhost_status(host, now_realtime_sec(), &status, RRDHOST_STATUS_BASIC); |
| 329 | |
| 330 | buffer_json_member_add_uint64(wb, "version", 1); |
| 331 | buffer_json_member_add_uint64(wb, "status", ret); |
| 332 | buffer_json_member_add_uuid(wb, "host_id", localhost->host_id.uuid); |
| 333 | buffer_json_member_add_uint64(wb, "nodes", dictionary_entries(rrdhost_root_index)); |
| 334 | buffer_json_member_add_uint64(wb, "receivers", stream_receivers_currently_connected()); |
| 335 | buffer_json_member_add_uint64(wb, "nonce", os_random32()); |
| 336 | |
| 337 | if(ret == HTTP_RESP_OK) { |
| 338 | if((status.ingest.status == RRDHOST_INGEST_STATUS_ARCHIVED || status.ingest.status == RRDHOST_INGEST_STATUS_OFFLINE) && |
| 339 | !stream_control_children_should_be_accepted()) |
| 340 | status.ingest.status = RRDHOST_INGEST_STATUS_INITIALIZING; |
| 341 | |
| 342 | buffer_json_member_add_string(wb, "db_status", rrdhost_db_status_to_string(status.db.status)); |
| 343 | buffer_json_member_add_string(wb, "db_liveness", rrdhost_db_liveness_to_string(status.db.liveness)); |
| 344 | buffer_json_member_add_string(wb, "ingest_type", rrdhost_ingest_type_to_string(status.ingest.type)); |
| 345 | buffer_json_member_add_string(wb, "ingest_status", rrdhost_ingest_status_to_string(status.ingest.status)); |
| 346 | buffer_json_member_add_uint64(wb, "first_time_s", status.db.first_time_s); |
| 347 | buffer_json_member_add_uint64(wb, "last_time_s", status.db.last_time_s); |
| 348 | } |
| 349 | |
| 350 | buffer_json_finalize(wb); |
| 351 | return ret; |
| 352 | } |
| 353 | |
| 354 | static bool stream_info_json_parse_v1(struct json_object *jobj, const char *path, STREAM_PARENT *d, BUFFER *error) { |
| 355 | uint32_t version = 0; (void)version; |
| 356 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "version", version, error, JSONC_REQUIRED); |
| 357 | |
| 358 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "status", d->remote.status, error, JSONC_REQUIRED); |
| 359 | JSONC_PARSE_TXT2UUID_OR_ERROR_AND_RETURN(jobj, path, "host_id", d->remote.host_id.uuid, error, JSONC_REQUIRED); |
| 360 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "nodes", d->remote.nodes, error, JSONC_REQUIRED); |
| 361 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "receivers", d->remote.receivers, error, JSONC_REQUIRED); |
| 362 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "nonce", d->remote.nonce, error, JSONC_REQUIRED); |
| 363 | |
| 364 | if(d->remote.status == HTTP_RESP_OK) { |
| 365 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "first_time_s", d->remote.db_first_time_s, error, JSONC_REQUIRED); |
| 366 | JSONC_PARSE_UINT64_OR_ERROR_AND_RETURN(jobj, path, "last_time_s", d->remote.db_last_time_s, error, JSONC_REQUIRED); |
| 367 | JSONC_PARSE_TXT2ENUM_OR_ERROR_AND_RETURN(jobj, path, "db_status", RRDHOST_DB_STATUS_2id, d->remote.db_status, error, JSONC_REQUIRED); |
| 368 | JSONC_PARSE_TXT2ENUM_OR_ERROR_AND_RETURN(jobj, path, "db_liveness", RRDHOST_DB_LIVENESS_2id, d->remote.db_liveness, error, JSONC_REQUIRED); |
| 369 | JSONC_PARSE_TXT2ENUM_OR_ERROR_AND_RETURN(jobj, path, "ingest_type", RRDHOST_INGEST_TYPE_2id, d->remote.ingest_type, error, JSONC_REQUIRED); |
| 370 | JSONC_PARSE_TXT2ENUM_OR_ERROR_AND_RETURN(jobj, path, "ingest_status", RRDHOST_INGEST_STATUS_2id, d->remote.ingest_status, error, JSONC_REQUIRED); |
| 371 | return true; |
| 372 | } |
| 373 | |
| 374 | buffer_sprintf(error, "status reported (%d) is not OK (%d)", d->remote.status, HTTP_RESP_OK); |
| 375 | |
| 376 | d->remote.db_first_time_s = 0; |
| 377 | d->remote.db_last_time_s = 0; |
| 378 | d->remote.db_status = 0; |
| 379 | d->remote.db_liveness = 0; |
| 380 | d->remote.ingest_type = 0; |
| 381 | d->remote.ingest_status = 0; |
| 382 | |
| 383 | return false; |
| 384 | } |
| 385 | |
| 386 | static bool stream_info_fetch(STREAM_PARENT *d, const char *uuid, int default_port, ND_SOCK *sender_sock, bool ssl, const char *hostname) { |
| 387 | char effective_service[NI_MAXSERV + 1]; |
| 388 | stream_parent_effective_service(string2str(d->destination), default_port, effective_service, sizeof(effective_service)); |
| 389 | |
| 390 | ND_LOG_STACK lgs[] = { |
| 391 | ND_LOG_FIELD_STR(NDF_DST_IP, d->destination), |
| 392 | ND_LOG_FIELD_TXT(NDF_DST_PORT, effective_service), |
| 393 | ND_LOG_FIELD_TXT(NDF_REQUEST_METHOD, "GET"), |
| 394 | ND_LOG_FIELD_END(), |
| 395 | }; |
| 396 | ND_LOG_STACK_PUSH(lgs); |
| 397 | |
| 398 | char buf[HTTP_HEADER_SIZE]; |
| 399 | CLEAN_ND_SOCK sock = ND_SOCK_INIT(sender_sock->ctx, sender_sock->verify_certificate); |
| 400 | |
| 401 | // Build HTTP request |
| 402 | snprintf(buf, sizeof(buf), |
| 403 | "GET /api/v3/stream_info?machine_guid=%s" HTTP_1_1 HTTP_ENDL |
| 404 | "Host: %s" HTTP_ENDL |
| 405 | "User-Agent: %s/%s" HTTP_ENDL |
| 406 | "Accept: */*" HTTP_ENDL |
| 407 | "Accept-Encoding: identity" HTTP_ENDL // disable chunked encoding |
| 408 | "TE: identity" HTTP_ENDL // disable chunked encoding |
| 409 | "Pragma: no-cache" HTTP_ENDL |
| 410 | "Cache-Control: no-cache" HTTP_ENDL |
| 411 | "Connection: close" HTTP_HDR_END, |
| 412 | uuid, |
| 413 | string2str(d->destination), |
| 414 | rrdhost_program_name(localhost), |
| 415 | rrdhost_program_version(localhost)); |
| 416 | |
| 417 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 418 | "STREAM PARENTS '%s': fetching stream info from '%s'...", |
| 419 | hostname, string2str(d->destination)); |
| 420 | |
| 421 | pulse_stream_info_sent_request(); |
| 422 | |
| 423 | // Establish connection |
| 424 | d->reason = STREAM_HANDSHAKE_SP_CONNECTING; |
| 425 | if (!nd_sock_connect_to_this(&sock, string2str(d->destination), default_port, 5, ssl)) { |
| 426 | d->selection.info = false; |
| 427 | stream_parent_nd_sock_error_to_reason(d, &sock); |
| 428 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 429 | "STREAM PARENTS '%s': failed to connect for stream info to '%s': %s", |
| 430 | hostname, string2str(d->destination), |
| 431 | ND_SOCK_ERROR_2str(sock.error)); |
| 432 | return false; |
| 433 | } |
| 434 | |
| 435 | // Send HTTP request |
| 436 | ssize_t sent = nd_sock_send_timeout(&sock, buf, strlen(buf), 0, 5); |
| 437 | if (sent <= 0) { |
| 438 | d->selection.info = false; |
| 439 | stream_parent_nd_sock_error_to_reason(d, &sock); |
| 440 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 441 | "STREAM PARENTS '%s': failed to send stream info request to '%s': %s", |
| 442 | hostname, string2str(d->destination), |
| 443 | ND_SOCK_ERROR_2str(sock.error)); |
| 444 | return false; |
| 445 | } |
| 446 | |
| 447 | // Receive HTTP response |
| 448 | size_t total_received = 0; |
| 449 | size_t payload_received = 0; |
| 450 | size_t content_length = 0; |
| 451 | char *payload_start = NULL; |
| 452 | |
| 453 | while (!payload_received || content_length < payload_received) { |
| 454 | size_t remaining = sizeof(buf) - total_received; |
| 455 | |
| 456 | if (remaining <= 1) { |
| 457 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 458 | "STREAM PARENTS '%s': stream info receive buffer is full while receiving response from '%s'", |
| 459 | hostname, string2str(d->destination)); |
| 460 | d->selection.info = false; |
| 461 | d->reason = STREAM_HANDSHAKE_PARENT_INTERNAL_ERROR; |
| 462 | return false; |
| 463 | } |
| 464 | |
| 465 | ssize_t received = nd_sock_recv_timeout(&sock, buf + total_received, remaining - 1, 0, 5); |
| 466 | if (received <= 0) { |
| 467 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 468 | "STREAM PARENTS '%s': socket receive error while querying stream info on '%s' " |
| 469 | "(total received %zu, payload received %zu, content length %zu): %s", |
| 470 | hostname, string2str(d->destination), |
| 471 | total_received, payload_received, content_length, |
| 472 | ND_SOCK_ERROR_2str(sock.error)); |
| 473 | |
| 474 | d->selection.info = false; |
| 475 | stream_parent_nd_sock_error_to_reason(d, &sock); |
| 476 | return false; |
| 477 | } |
| 478 | |
| 479 | total_received += received; |
| 480 | buf[total_received] = '\0'; |
| 481 | |
| 482 | if(!payload_start) { |
| 483 | char *headers_end = strstr(buf, HTTP_HDR_END); |
| 484 | if (!headers_end) |
| 485 | // we have not received the whole header yet |
| 486 | continue; |
| 487 | |
| 488 | payload_start = headers_end + sizeof(HTTP_HDR_END) - 1; |
| 489 | } |
| 490 | |
| 491 | // the payload size so far |
| 492 | payload_received = total_received - (payload_start - buf); |
| 493 | |
| 494 | if(!content_length) { |
| 495 | char *content_length_ptr = strstr(buf, "Content-Length: "); |
| 496 | if (!content_length_ptr) { |
| 497 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 498 | "STREAM PARENTS '%s': stream info response from '%s' does not have a Content-Length", |
| 499 | hostname, string2str(d->destination)); |
| 500 | |
| 501 | d->selection.info = false; |
| 502 | d->reason = STREAM_HANDSHAKE_PARENT_INTERNAL_ERROR; |
| 503 | return false; |
| 504 | } |
| 505 | content_length = strtoul(content_length_ptr + strlen("Content-Length: "), NULL, 10); |
| 506 | if (!content_length) { |
| 507 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 508 | "STREAM PARENTS '%s': stream info response from '%s' has invalid Content-Length", |
| 509 | hostname, string2str(d->destination)); |
| 510 | |
| 511 | d->selection.info = false; |
| 512 | d->reason = STREAM_HANDSHAKE_PARENT_INTERNAL_ERROR; |
| 513 | return false; |
| 514 | } |
| 515 | } |
| 516 | } |
| 517 | |
| 518 | // Parse HTTP response and extract JSON |
| 519 | CLEAN_JSON_OBJECT *jobj = json_tokener_parse(payload_start); |
| 520 | if (!jobj) { |
| 521 | d->selection.info = false; |
| 522 | d->reason = STREAM_HANDSHAKE_SP_NO_STREAM_INFO; |
| 523 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 524 | "STREAM PARENTS '%s': failed to parse stream info response from '%s', JSON data: %s", |
| 525 | hostname, string2str(d->destination), payload_start); |
| 526 | return false; |
| 527 | } |
| 528 | |
| 529 | CLEAN_BUFFER *error = buffer_create(0, NULL); |
| 530 | |
| 531 | if(!stream_info_json_parse_v1(jobj, "", d, error)) { |
| 532 | d->selection.info = false; |
| 533 | d->reason = STREAM_HANDSHAKE_SP_NO_STREAM_INFO; |
| 534 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 535 | "STREAM PARENTS '%s': failed to extract fields from JSON stream info response from '%s': %s" |
| 536 | " - JSON data: %s", |
| 537 | hostname, string2str(d->destination), |
| 538 | buffer_tostring(error), |
| 539 | payload_start); |
| 540 | return false; |
| 541 | } |
| 542 | |
| 543 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 544 | "STREAM PARENTS '%s': received stream_info data from '%s': " |
| 545 | "status: %d, nodes: %zu, receivers: %zu, first_time_s: %ld, last_time_s: %ld, " |
| 546 | "db status: %s, db liveness: %s, ingest type: %s, ingest status: %s", |
| 547 | hostname, string2str(d->destination), |
| 548 | d->remote.status, d->remote.nodes, d->remote.receivers, |
| 549 | d->remote.db_first_time_s, d->remote.db_last_time_s, |
| 550 | RRDHOST_DB_STATUS_2str(d->remote.db_status), |
| 551 | RRDHOST_DB_LIVENESS_2str(d->remote.db_liveness), |
| 552 | RRDHOST_INGEST_TYPE_2str(d->remote.ingest_type), |
| 553 | RRDHOST_INGEST_STATUS_2str(d->remote.ingest_status)); |
| 554 | |
| 555 | d->selection.info = true; |
| 556 | d->reason = STREAM_HANDSHAKE_NEVER; |
| 557 | return true; |
| 558 | } |
| 559 | |
| 560 | static int compare_last_time(const void *a, const void *b) { |
| 561 | STREAM_PARENT *parent_a = *(STREAM_PARENT **)a; |
| 562 | STREAM_PARENT *parent_b = *(STREAM_PARENT **)b; |
| 563 | |
| 564 | if (parent_a->remote.db_last_time_s < parent_b->remote.db_last_time_s) return 1; |
| 565 | else if (parent_a->remote.db_last_time_s > parent_b->remote.db_last_time_s) return -1; |
| 566 | else { |
| 567 | if(parent_a->since_ut < parent_b->since_ut) return -1; |
| 568 | else if(parent_a->since_ut > parent_b->since_ut) return 1; |
| 569 | else { |
| 570 | if(parent_a->attempts < parent_b->attempts) return -1; |
| 571 | else if(parent_a->attempts > parent_b->attempts) return 1; |
| 572 | else return 0; |
| 573 | } |
| 574 | } |
| 575 | } |
| 576 | |
| 577 | bool stream_parent_connect_to_one_unsafe( |
| 578 | ND_SOCK *sender_sock, |
| 579 | RRDHOST *host, |
| 580 | int default_port, |
| 581 | time_t timeout, |
| 582 | char *connected_to, |
| 583 | size_t connected_to_size, |
| 584 | STREAM_PARENT **destination) |
| 585 | { |
| 586 | sender_sock->error = ND_SOCK_ERR_NO_DESTINATION_AVAILABLE; |
| 587 | |
| 588 | // count the parents |
| 589 | size_t size = 0; |
| 590 | for (STREAM_PARENT *d = host->stream.snd.parents.all; d; d = d->next) { |
| 591 | d->selection.order = 0; |
| 592 | d->selection.batch = 0; |
| 593 | d->selection.random = false; |
| 594 | d->selection.info = false; |
| 595 | d->selection.skipped = true; |
| 596 | size++; |
| 597 | } |
| 598 | |
| 599 | // do we have any parents? |
| 600 | if(!size) { |
| 601 | nd_log(NDLS_DAEMON, NDLP_DEBUG, "STREAM PARENTS '%s': no parents configured", rrdhost_hostname(host)); |
| 602 | return false; |
| 603 | } |
| 604 | |
| 605 | STREAM_PARENT **array = callocz(size, sizeof(*array)); |
| 606 | usec_t now_ut = now_realtime_usec(); |
| 607 | bool rc = false; |
| 608 | |
| 609 | // fetch stream info for all of them and put them in the array |
| 610 | size_t count = 0, skipped_but_useful = 0, skipped_not_useful = 0, potential = 0; |
| 611 | for (STREAM_PARENT *d = host->stream.snd.parents.all; d && count < size ; d = d->next) { |
| 612 | if (nd_thread_signaled_to_cancel()) { |
| 613 | sender_sock->error = ND_SOCK_ERR_THREAD_CANCELLED; |
| 614 | goto cleanup; |
| 615 | } |
| 616 | |
| 617 | // make sure they all have a random number |
| 618 | // this is taken from the parent, but if the stream_info call fails, |
| 619 | // we generate a random number for every parent here |
| 620 | d->remote.nonce = os_random32(); |
| 621 | d->banned_temporarily_erroneous = is_a_blocked_parent(d); |
| 622 | |
| 623 | if (d->banned_permanently || d->banned_for_this_session) |
| 624 | continue; |
| 625 | |
| 626 | if (d->banned_temporarily_erroneous) { |
| 627 | potential++; |
| 628 | host->stream.snd.status.reason = d->reason; |
| 629 | continue; |
| 630 | } |
| 631 | |
| 632 | if (d->postpone_until_ut > now_ut) { |
| 633 | skipped_but_useful++; |
| 634 | potential++; |
| 635 | host->stream.snd.status.reason = d->reason; |
| 636 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 637 | "STREAM PARENTS '%s': skipping useful parent '%s': POSTPONED FOR %ld SECS MORE: %s", |
| 638 | rrdhost_hostname(host), |
| 639 | string2str(d->destination), |
| 640 | (time_t)((d->postpone_until_ut - now_ut) / USEC_PER_SEC), |
| 641 | stream_handshake_error_to_string(d->reason)); |
| 642 | continue; |
| 643 | } |
| 644 | |
| 645 | if(stream_info_fetch(d, host->machine_guid, default_port, |
| 646 | sender_sock, stream_parent_is_ssl(d), rrdhost_hostname(host))) { |
| 647 | switch(d->remote.ingest_type) { |
| 648 | case RRDHOST_INGEST_TYPE_VIRTUAL: |
| 649 | case RRDHOST_INGEST_TYPE_LOCALHOST: |
| 650 | d->reason = STREAM_HANDSHAKE_PARENT_IS_LOCALHOST; |
| 651 | d->since_ut = now_ut; |
| 652 | d->postpone_until_ut = randomize_wait_ut(3600, 7200); |
| 653 | d->banned_permanently = true; |
| 654 | skipped_not_useful++; |
| 655 | |
| 656 | if(rrdhost_is_host_in_stream_path_before_us(host, d->remote.host_id, 1)) { |
| 657 | // we passed hops == 1, to make sure this succeeds only when the parent |
| 658 | // is the origin child of this node |
| 659 | nd_log(NDLS_DAEMON, NDLP_INFO, |
| 660 | "STREAM PARENTS '%s': destination '%s' is banned permanently because it is the origin server", |
| 661 | rrdhost_hostname(host), string2str(d->destination)); |
| 662 | } |
| 663 | else { |
| 664 | nd_log(NDLS_DAEMON, NDLP_WARNING, |
| 665 | "STREAM PARENTS '%s': destination '%s' is banned permanently because it is the origin server, " |
| 666 | "but it is not in the stream path before us!", |
| 667 | rrdhost_hostname(host), string2str(d->destination)); |
| 668 | } |
| 669 | continue; |
| 670 | |
| 671 | default: |
| 672 | case RRDHOST_INGEST_TYPE_CHILD: |
| 673 | case RRDHOST_INGEST_TYPE_ARCHIVED: |
| 674 | break; |
| 675 | } |
| 676 | |
| 677 | switch(d->remote.ingest_status) { |
| 678 | case RRDHOST_INGEST_STATUS_INITIALIZING: |
| 679 | d->reason = STREAM_HANDSHAKE_PARENT_IS_INITIALIZING; |
| 680 | d->since_ut = now_ut; |
| 681 | d->postpone_until_ut = randomize_wait_ut(30, 60); |
| 682 | pulse_sender_stream_info_failed(string2str(d->destination), d->reason); |
| 683 | skipped_but_useful++; |
| 684 | potential++; |
| 685 | host->stream.snd.status.reason = d->reason; |
| 686 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 687 | "STREAM PARENTS '%s': skipping useful parent '%s': %s", |
| 688 | rrdhost_hostname(host), string2str(d->destination), |
| 689 | stream_handshake_error_to_string(d->reason)); |
| 690 | continue; |
| 691 | |
| 692 | case RRDHOST_INGEST_STATUS_REPLICATING: |
| 693 | case RRDHOST_INGEST_STATUS_ONLINE: |
| 694 | if(rrdhost_is_host_in_stream_path_before_us(host, d->remote.host_id, host->sender->hops)) { |
| 695 | d->reason = STREAM_HANDSHAKE_PARENT_NODE_ALREADY_CONNECTED; |
| 696 | d->since_ut = now_ut; |
| 697 | d->postpone_until_ut = randomize_wait_ut(3600, 7200); |
| 698 | d->banned_for_this_session = true; |
| 699 | skipped_not_useful++; |
| 700 | nd_log(NDLS_DAEMON, NDLP_INFO, |
| 701 | "STREAM PARENTS '%s': destination '%s' is banned for this session, because it is in our path before us.", |
| 702 | rrdhost_hostname(host), string2str(d->destination)); |
| 703 | pulse_sender_stream_info_failed(string2str(d->destination), d->reason); |
| 704 | continue; |
| 705 | } |
| 706 | // else { |
| 707 | // skip = true; |
| 708 | // if(!netdata_conf_is_parent()) { |
| 709 | // nd_log(NDLS_DAEMON, NDLP_INFO, |
| 710 | // "STREAM PARENTS '%s': destination '%s' reports I am already connected.", |
| 711 | // rrdhost_hostname(host), string2str(d->destination)); |
| 712 | // } |
| 713 | // } |
| 714 | break; |
| 715 | |
| 716 | default: |
| 717 | case RRDHOST_INGEST_STATUS_OFFLINE: |
| 718 | break; |
| 719 | } |
| 720 | } |
| 721 | else |
| 722 | pulse_sender_stream_info_failed(string2str(d->destination), d->reason); |
| 723 | |
| 724 | d->selection.skipped = false; |
| 725 | d->selection.batch = count + 1; |
| 726 | d->selection.order = count + 1; |
| 727 | array[count++] = d; |
| 728 | } |
| 729 | |
| 730 | // can we use any parent? |
| 731 | if(!count) { |
| 732 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 733 | "STREAM PARENTS '%s': no parents available (%zu skipped but useful, %zu skipped not useful, %zu potential)", |
| 734 | rrdhost_hostname(host), skipped_but_useful, skipped_not_useful, potential); |
| 735 | |
| 736 | if(!potential) { |
| 737 | if(host->stream.snd.status.reason != STREAM_HANDSHAKE_SP_NO_DESTINATION) { |
| 738 | host->stream.snd.status.reason = STREAM_HANDSHAKE_SP_NO_DESTINATION; |
| 739 | pulse_sender_connection_failed(NULL, host->stream.snd.status.reason); |
| 740 | } |
| 741 | pulse_host_status(host, PULSE_HOST_STATUS_SND_NO_DST, 0); |
| 742 | } |
| 743 | |
| 744 | goto cleanup; |
| 745 | } |
| 746 | |
| 747 | // order the parents in the array the way we want to connect |
| 748 | if(count > 1) { |
| 749 | qsort(array, count, sizeof(STREAM_PARENT *), compare_last_time); |
| 750 | |
| 751 | size_t base = 0, batch = 0; |
| 752 | while (base < count) { |
| 753 | // find how many have similar db_last_time_s; |
| 754 | size_t similar = 1; |
| 755 | if(!array[base]->remote.nonce) array[base]->remote.nonce = os_random32(); |
| 756 | time_t tB = array[base]->remote.db_last_time_s; |
| 757 | for (size_t i = base + 1; i < count; i++) { |
| 758 | time_t tN = array[i]->remote.db_last_time_s; |
| 759 | if ((tN > tB && tN - tB <= TIME_TO_CONSIDER_PARENTS_SIMILAR) || |
| 760 | (tB - tN <= TIME_TO_CONSIDER_PARENTS_SIMILAR)) |
| 761 | similar++; |
| 762 | else |
| 763 | break; |
| 764 | } |
| 765 | |
| 766 | // if we have only 1 similar, move on |
| 767 | if (similar == 1) { |
| 768 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 769 | "STREAM PARENTS '%s': reordering keeps parent No %zu, '%s'", |
| 770 | rrdhost_hostname(host), base, string2str(array[base]->destination)); |
| 771 | array[base]->selection.order = base + 1; |
| 772 | array[base]->selection.batch = batch + 1; |
| 773 | array[base]->selection.random = false; |
| 774 | base++; |
| 775 | batch++; |
| 776 | continue; |
| 777 | } |
| 778 | else { |
| 779 | // reorder the parents who have similar db_last_time |
| 780 | |
| 781 | while (similar > 1) { |
| 782 | size_t chosen = base; |
| 783 | for(size_t i = base + 1 ; i < base + similar ;i++) { |
| 784 | uint32_t i_nonce = array[i]->remote.nonce | os_random32(); |
| 785 | uint32_t chosen_nonce = array[chosen]->remote.nonce | os_random32(); |
| 786 | if(i_nonce > chosen_nonce) chosen = i; |
| 787 | } |
| 788 | |
| 789 | if (chosen != base) |
| 790 | SWAP(array[base], array[chosen]); |
| 791 | |
| 792 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 793 | "STREAM PARENTS '%s': random reordering of %zu similar parents (slots %zu to %zu), No %zu is '%s'", |
| 794 | rrdhost_hostname(host), |
| 795 | similar, base, base + similar, |
| 796 | base, string2str(array[base]->destination)); |
| 797 | |
| 798 | array[base]->selection.order = base + 1; |
| 799 | array[base]->selection.batch = batch + 1; |
| 800 | array[base]->selection.random = true; |
| 801 | base++; |
| 802 | similar--; |
| 803 | } |
| 804 | |
| 805 | // the last one of the similar |
| 806 | array[base]->selection.order = base + 1; |
| 807 | array[base]->selection.batch = batch + 1; |
| 808 | array[base]->selection.random = true; |
| 809 | base++; |
| 810 | batch++; |
| 811 | } |
| 812 | } |
| 813 | } |
| 814 | else { |
| 815 | array[0]->selection.order = 1; |
| 816 | array[0]->selection.batch = 1; |
| 817 | array[0]->selection.random = false; |
| 818 | |
| 819 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 820 | "STREAM PARENTS '%s': only 1 parent is available: '%s'", |
| 821 | rrdhost_hostname(host), string2str(array[0]->destination)); |
| 822 | } |
| 823 | |
| 824 | // now the parents are sorted based on preference of connection |
| 825 | for(size_t i = 0; i < count ;i++) { |
| 826 | STREAM_PARENT *d = array[i]; |
| 827 | |
| 828 | if(d->postpone_until_ut > now_ut) |
| 829 | continue; |
| 830 | |
| 831 | if(nd_thread_signaled_to_cancel()) { |
| 832 | sender_sock->error = ND_SOCK_ERR_THREAD_CANCELLED; |
| 833 | host->stream.snd.status.reason = STREAM_HANDSHAKE_DISCONNECT_SIGNALED_TO_STOP; |
| 834 | pulse_host_status(host, PULSE_HOST_STATUS_SND_OFFLINE, host->stream.snd.status.reason); |
| 835 | goto cleanup; |
| 836 | } |
| 837 | |
| 838 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 839 | "STREAM PARENTS '%s': connecting to '%s' (default port: %d, parent %zu of %zu)...", |
| 840 | rrdhost_hostname(host), string2str(d->destination), default_port, |
| 841 | i + 1, count); |
| 842 | |
| 843 | char effective_service[NI_MAXSERV + 1]; |
| 844 | stream_parent_effective_service(string2str(d->destination), default_port, effective_service, sizeof(effective_service)); |
| 845 | ND_LOG_STACK lgs[] = { |
| 846 | ND_LOG_FIELD_STR(NDF_DST_IP, d->destination), |
| 847 | ND_LOG_FIELD_TXT(NDF_DST_PORT, effective_service), |
| 848 | ND_LOG_FIELD_END(), |
| 849 | }; |
| 850 | ND_LOG_STACK_PUSH(lgs); |
| 851 | |
| 852 | d->since_ut = now_ut; |
| 853 | d->attempts++; |
| 854 | pulse_host_status(host, PULSE_HOST_STATUS_SND_CONNECTING, 0); |
| 855 | if (nd_sock_connect_to_this(sender_sock, string2str(d->destination), |
| 856 | default_port, timeout, stream_parent_is_ssl(d))) { |
| 857 | |
| 858 | if (connected_to && connected_to_size) |
| 859 | strncpyz(connected_to, string2str(d->destination), connected_to_size); |
| 860 | |
| 861 | *destination = d; |
| 862 | |
| 863 | // move the current item to the end of the list |
| 864 | // without this, this destination will break the loop again and again |
| 865 | // not advancing the destinations to find one that may work |
| 866 | DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(host->stream.snd.parents.all, d, prev, next); |
| 867 | DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(host->stream.snd.parents.all, d, prev, next); |
| 868 | |
| 869 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 870 | "STREAM PARENTS '%s': connected to '%s' (default port: %d, fd %d)...", |
| 871 | rrdhost_hostname(host), string2str(d->destination), default_port, |
| 872 | sender_sock->fd); |
| 873 | |
| 874 | sender_sock->error = ND_SOCK_ERR_NONE; |
| 875 | host->stream.snd.status.reason = STREAM_HANDSHAKE_SP_CONNECTED; |
| 876 | pulse_host_status(host, PULSE_HOST_STATUS_SND_CONNECTING, host->stream.snd.status.reason); |
| 877 | rc = true; |
| 878 | goto cleanup; |
| 879 | } |
| 880 | else { |
| 881 | stream_parent_nd_sock_error_to_reason(d, sender_sock); |
| 882 | host->stream.snd.status.reason = d->reason; |
| 883 | pulse_sender_connection_failed(string2str(d->destination), d->reason); |
| 884 | pulse_host_status(host, PULSE_HOST_STATUS_SND_CONNECTING, host->stream.snd.status.reason); |
| 885 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 886 | "STREAM PARENTS '%s': stream connection to '%s' failed (default port: %d): %s", |
| 887 | rrdhost_hostname(host), |
| 888 | string2str(d->destination), default_port, |
| 889 | ND_SOCK_ERROR_2str(sender_sock->error)); |
| 890 | } |
| 891 | } |
| 892 | |
| 893 | pulse_host_status(host, PULSE_HOST_STATUS_SND_OFFLINE, 0); |
| 894 | |
| 895 | cleanup: |
| 896 | freez(array); |
| 897 | return rc; |
| 898 | } |
| 899 | |
| 900 | bool stream_parent_connect_to_one( |
| 901 | ND_SOCK *sender_sock, |
| 902 | RRDHOST *host, |
| 903 | int default_port, |
| 904 | time_t timeout, |
| 905 | char *connected_to, |
| 906 | size_t connected_to_size, |
| 907 | STREAM_PARENT **destination) { |
| 908 | |
| 909 | rw_spinlock_read_lock(&host->stream.snd.parents.spinlock); |
| 910 | bool rc = stream_parent_connect_to_one_unsafe( |
| 911 | sender_sock, host, default_port, timeout, connected_to, connected_to_size, destination); |
| 912 | rw_spinlock_read_unlock(&host->stream.snd.parents.spinlock); |
| 913 | return rc; |
| 914 | } |
| 915 | |
| 916 | // -------------------------------------------------------------------------------------------------------------------- |
| 917 | // create stream parents linked list |
| 918 | |
| 919 | struct stream_parent_init_tmp { |
| 920 | RRDHOST *host; |
| 921 | STREAM_PARENT *list; |
| 922 | int count; |
| 923 | }; |
| 924 | |
| 925 | static bool stream_parent_add_one_unsafe(char *entry, void *data) { |
| 926 | struct stream_parent_init_tmp *t = data; |
| 927 | |
| 928 | STREAM_PARENT *d = callocz(1, sizeof(STREAM_PARENT)); |
| 929 | char *colon_ssl = strstr(entry, ":SSL"); |
| 930 | if(colon_ssl) { |
| 931 | *colon_ssl = '\0'; |
| 932 | d->ssl = true; |
| 933 | } |
| 934 | else |
| 935 | d->ssl = false; |
| 936 | |
| 937 | d->destination = string_strdupz(entry); |
| 938 | d->since_ut = now_realtime_usec(); |
| 939 | |
| 940 | __atomic_add_fetch(&netdata_buffers_statistics.rrdhost_senders, sizeof(STREAM_PARENT), __ATOMIC_RELAXED); |
| 941 | |
| 942 | DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(t->list, d, prev, next); |
| 943 | |
| 944 | t->count++; |
| 945 | nd_log(NDLS_DAEMON, NDLP_DEBUG, |
| 946 | "STREAM PARENTS '%s': added streaming destination No %d: '%s'", |
| 947 | rrdhost_hostname(t->host), t->count, string2str(d->destination)); |
| 948 | |
| 949 | return false; // we return false, so that we will get all defined destinations |
| 950 | } |
| 951 | |
| 952 | void rrdhost_stream_parents_update_from_destination(RRDHOST *host) { |
| 953 | rw_spinlock_write_lock(&host->stream.snd.parents.spinlock); |
| 954 | rrdhost_stream_parents_free(host, true); |
| 955 | |
| 956 | if(host->stream.snd.destination) { |
| 957 | struct stream_parent_init_tmp t = { |
| 958 | .host = host, |
| 959 | .list = NULL, |
| 960 | .count = 0, |
| 961 | }; |
| 962 | foreach_entry_in_connection_string(string2str(host->stream.snd.destination), stream_parent_add_one_unsafe, &t); |
| 963 | host->stream.snd.parents.all = t.list; |
| 964 | } |
| 965 | |
| 966 | rw_spinlock_write_unlock(&host->stream.snd.parents.spinlock); |
| 967 | } |
| 968 | |
| 969 | void rrdhost_stream_parents_free(RRDHOST *host, bool having_write_lock) { |
| 970 | if(!having_write_lock) |
| 971 | rw_spinlock_write_lock(&host->stream.snd.parents.spinlock); |
| 972 | |
| 973 | while (host->stream.snd.parents.all) { |
| 974 | STREAM_PARENT *tmp = host->stream.snd.parents.all; |
| 975 | DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(host->stream.snd.parents.all, tmp, prev, next); |
| 976 | string_freez(tmp->destination); |
| 977 | freez(tmp); |
| 978 | __atomic_sub_fetch(&netdata_buffers_statistics.rrdhost_senders, sizeof(STREAM_PARENT), __ATOMIC_RELAXED); |
| 979 | } |
| 980 | |
| 981 | host->stream.snd.parents.all = NULL; |
| 982 | host->stream.snd.parents.current = NULL; |
| 983 | |
| 984 | if(!having_write_lock) |
| 985 | rw_spinlock_write_unlock(&host->stream.snd.parents.spinlock); |
| 986 | } |
| 987 | |
| 988 | void rrdhost_stream_parents_init(RRDHOST *host) { |
| 989 | rw_spinlock_init(&host->stream.snd.parents.spinlock); |
| 990 | } |