syntax = "proto3";

package envoy.service.load_stats.v2;

import "envoy/api/v2/core/base.proto";
import "envoy/api/v2/endpoint/load_report.proto";

import "google/protobuf/duration.proto";

import "validate/validate.proto";

option java_package = "io.envoyproxy.envoy.service.load_stats.v2";
option java_outer_classname = "LrsProto";
option java_multiple_files = true;
option java_generic_services = true;

// [#protodoc-title: Load reporting service]

service LoadReportingService {
  // Advanced API to allow for multi-dimensional load balancing by remote
  // server. For receiving LB assignments, the steps are:
  // 1, The management server is configured with per cluster/zone/load metric
  //    capacity configuration. The capacity configuration definition is
  //    outside of the scope of this document.
  // 2. Envoy issues a standard {Stream,Fetch}Endpoints request for the clusters
  //    to balance.
  //
  // Independently, Envoy will initiate a StreamLoadStats bidi stream with a
  // management server:
  // 1. Once a connection establishes, the management server publishes a
  //    LoadStatsResponse for all clusters it is interested in learning load
  //    stats about.
  // 2. For each cluster, Envoy load balances incoming traffic to upstream hosts
  //    based on per-zone weights and/or per-instance weights (if specified)
  //    based on intra-zone LbPolicy. This information comes from the above
  //    {Stream,Fetch}Endpoints.
  // 3. When upstream hosts reply, they optionally add header <define header
  //    name> with ASCII representation of EndpointLoadMetricStats.
  // 4. Envoy aggregates load reports over the period of time given to it in
  //    LoadStatsResponse.load_reporting_interval. This includes aggregation
  //    stats Envoy maintains by itself (total_requests, rpc_errors etc.) as
  //    well as load metrics from upstream hosts.
  // 5. When the timer of load_reporting_interval expires, Envoy sends new
  //    LoadStatsRequest filled with load reports for each cluster.
  // 6. The management server uses the load reports from all reported Envoys
  //    from around the world, computes global assignment and prepares traffic
  //    assignment destined for each zone Envoys are located in. Goto 2.
  rpc StreamLoadStats(stream LoadStatsRequest) returns (stream LoadStatsResponse) {
  }
}

// A load report Envoy sends to the management server.
// [#not-implemented-hide:] Not configuration. TBD how to doc proto APIs.
message LoadStatsRequest {
  // Node identifier for Envoy instance.
  api.v2.core.Node node = 1;

  // A list of load stats to report.
  repeated api.v2.endpoint.ClusterStats cluster_stats = 2;
}

// The management server sends envoy a LoadStatsResponse with all clusters it
// is interested in learning load stats about.
// [#not-implemented-hide:] Not configuration. TBD how to doc proto APIs.
message LoadStatsResponse {
  // Clusters to report stats for.
  repeated string clusters = 1 [(validate.rules).repeated = {min_items: 1}];

  // The minimum interval of time to collect stats over. This is only a minimum for two reasons:
  // 1. There may be some delay from when the timer fires until stats sampling occurs.
  // 2. For clusters that were already feature in the previous *LoadStatsResponse*, any traffic
  //    that is observed in between the corresponding previous *LoadStatsRequest* and this
  //    *LoadStatsResponse* will also be accumulated and billed to the cluster. This avoids a period
  //    of inobservability that might otherwise exists between the messages. New clusters are not
  //    subject to this consideration.
  google.protobuf.Duration load_reporting_interval = 2;

  // Set to *true* if the management server supports endpoint granularity
  // report.
  bool report_endpoint_granularity = 3;
}