proto_files/orc8r/protos/metricsd.proto

// SPDX-FileCopyrightText: 2020 The Magma Authors.
// SPDX-FileCopyrightText: 2022 Open Networking Foundation <support@opennetworking.org>
//
// SPDX-License-Identifier: BSD-3-Clause

syntax = "proto3";

package magma.orc8r;

import "metrics.proto";
import "orc8r/protos/common.proto";

option go_package = "magma/orc8r/lib/go/protos";

// For each metric name tracked, we use an enum
enum MetricName {
  // Prometheus metrics
  process_start_time_seconds     = 0;
  process_virtual_memory_bytes   = 1;
  process_resident_memory_bytes  = 2;
  process_cpu_seconds_total      = 3;
  process_open_fds               = 4;
  process_max_fds                = 5;

  // subscriberdb metrics
  s6a_auth_success               = 6;
  s6a_auth_failure               = 7;
  s6a_location_update            = 8;
  diameter_capabilities_exchange = 9;
  diameter_watchdog              = 10;
  diameter_disconnect            = 11;

  // More prometheus metrics
  python_info                    = 50;

  // Metricsd metrics
  service_metrics_collected      = 60;
  process_uptime_seconds         = 61;

  // mobilityd metrics
  ip_address_allocated           = 100;
  ip_address_released            = 101;
  ip_already_allocated           = 102;
  ip_allocation_failure          = 103;

  // enodebd metrics
  enodeb_mgmt_connected          = 200;
  enodeb_opstate_enabled         = 201;
  enodeb_rf_tx_enabled           = 202;
  enodeb_gps_connected           = 203;
  enodeb_ptp_connected           = 204;
  enodeb_mme_connected           = 205;
  rrc_estab_attempts             = 206;
  rrc_estab_successes            = 207;
  pdcp_user_plane_bytes_ul       = 208;
  pdcp_user_plane_bytes_dl       = 209;
  rrc_reestab_attempts           = 210;
  rrc_reestab_attempts_reconf_fail = 211;
  rrc_reestab_attempts_ho_fail   = 212;
  rrc_reestab_attempts_other     = 213;
  rrc_reestab_successes          = 214;
  erab_estab_attempts            = 215;
  erab_estab_successes           = 216;
  erab_estab_failures            = 217;
  erab_release_requests          = 218;
  erab_release_requests_user_inactivity = 219;
  erab_release_requests_normal   = 220;
  erab_release_requests_radio_resources_not_available = 221;
  erab_release_requests_reduce_load = 222;
  erab_release_requests_fail_in_radio_proc = 223;
  erab_release_requests_eutran_reas = 224;
  erab_release_requests_radio_radio_conn_lost = 225;
  erab_release_requests_oam_intervention = 226;
  enodeb_mgmt_configured         = 227;
  enodeb_reboot_timer_active     = 228;
  enodeb_reboots                 = 229;
  enodeb_rf_tx_desired           = 230;

  // Kernsnoopd metrics
  magma_bytes_sent_total         = 260;
  linux_bytes_sent_total         = 261;

  // Magmad metrics
  magmad_ping_rtt_ms             = 300;
  cpu_percent                    = 301;
  swap_memory_percent            = 302;
  virtual_memory_percent         = 303;
  disk_percent                   = 304;
  bytes_sent                     = 305;
  bytes_received                 = 306;
  temperature                    = 307;
  checkin_status                 = 308;
  streamer_responses             = 309;
  mem_total                      = 310;
  mem_available                  = 311;
  mem_used                       = 312;
  mem_free                       = 313;
  bootstrap_exception            = 314;
  upgrader2_time_taken           = 315;
  upgrader2_error                = 316;
  upgrader2_prepared             = 317;
  upgrader2_downloaded           = 318;
  upgrader2_canary               = 319;
  upgrader2_stable               = 320;
  upgrader2_idle                 = 321;
  unexpected_service_restarts    = 322;
  unattended_upgrade_status      = 323;
  service_restart_status         = 324;

  // Pipelined metrics
  dp_send_msg_error              = 350;
  arp_default_gw_mac_error       = 351;
  openflow_error_msg             = 352;
  unknown_pkt_direction          = 353;
  network_iface_status           = 354;
  enforcement_rule_install_fail  = 355;
  enforcement_stats_rule_install_fail = 356;

  ///////////////////////////////
  // GATEWAY FAILURES & ALERTS //
  ///////////////////////////////
  mme_restarted                  = 400;
  s1_reset_from_enb              = 401;

  // SCTP
  sctp_reset                     = 410;
  sctp_shutdown                  = 411;

  // s6a
  s6a_subscriberdb_connection_failure = 420;
  s6a_auth_info_response_timer_expired = 421;

  // UE attach, Serivce Request
  duplicate_attach_request       = 430;
  initial_context_setup_failure_received = 431;
  initial_context_setup_request_timer_expired = 432;
  nas_attach_accept_timer_expired = 433;
  nas_auth_rsp_timer_expired = 434;
  nas_security_mode_command_timer_expired = 435;
  security_mode_reject_received  = 436;
  authentication_failure         = 437; // cause=mac_failure,resync,amf
  mme_spgw_create_session_req    = 438;
  mme_spgw_create_session_rsp    = 439;
  mme_spgw_delete_session_req    = 440;
  mme_spgw_delete_session_rsp    = 441;

  // UE S1AP context
  ue_context_release_request     = 450;
  ue_context_release_command_timer_expired = 451;

  // Detach
  implicit_detach_timer_expired  = 452;

  // eNB SCTP shutdown context clean up time
  enb_sctp_shutdown_ue_clean_up_timer_expired  = 453;

  // Other errors
  s1ap_error_ind_received        = 480;
  gtpu_error_ind_received        = 481;
  nas_non_delivery_indication_received = 482;
  emm_status_received            = 483;
  emm_status_sent                = 484;

  ////////////////////////
  // GATEWAY PROCEDURES //
  ////////////////////////
  mme_new_association            = 500;
  ue_attach                      = 501; // result=failure,accept,complete
  service_request                = 502;
  ue_detach                      = 503; // cause=network_initiated,
                                        // deregister_ue,implicit_detach,
                                        // ue_initiated
  spgw_create_session            = 504;
  spgw_delete_session            = 505;
  ue_pdn_connection              = 506;
  tracking_area_update           = 507;
  s1_setup                       = 508;

  // Generic service metrics
  service_errors                 = 550;
}

// Possible labels, used as metric_name{label_name=label_value}
enum MetricLabelName {
  result                         = 0;
  cause                          = 1;
  tau_type                       = 2;
  pdn_type                       = 3;
  action                         = 4;
  host                           = 5; // magmad_ping_rtt_ms
  metric                         = 6; // magmad_ping_rtt_ms
  patchlevel                     = 7; // python_info
  minor                          = 8; // python_info
  major                          = 9; // python_info
  version                        = 10; // python_info
  implementation                 = 11; // python_info
  sensor                         = 12; // magmad_temperature
  error_code                     = 13;
  error_type                     = 14;
  iface_name                     = 15; // pipelined network_iface_status
  service_name                   = 16; // magmad unexpected_service_restarts
  rule_id                        = 17; // pipelined rule install failures
  imsi                           = 18; // pipelined rule install failures
  dest_service                   = 19; // kernsnoopd name of destination service
  binary_name                    = 20; // kernsnoopd name of linux binary
}

message MetricsContainer {
  string gatewayId = 1;
  // TODO: we should strip this out of metricsd (or duplicate it ourselves)
  // Prometheus 2 no longer supports protobuf client model
  repeated io.prometheus.client.MetricFamily family = 2;
}

message PushedMetric {
  string   metricName        = 1;
  double   value             = 2;
  int64    timestampMS       = 3;
  repeated LabelPair labels  = 4;
}

message LabelPair {
  string name  = 1;
  string value = 2;
}

message PushedMetricsContainer {
  string   networkId = 1;
  repeated PushedMetric metrics = 2;
}

service MetricsController {
  // Report a collection of metrics from a service
  rpc Collect(MetricsContainer) returns (Void) {}

  // Push a collection of metrics to metricsd
  rpc Push(PushedMetricsContainer) returns (Void) {}
}