envoy.service.load_stats.v2.lrs.proto Maven / Gradle / Ivy
syntax = "proto3";
package envoy.service.load_stats.v2;
option java_outer_classname = "LrsProto";
option java_multiple_files = true;
option java_package = "io.envoyproxy.envoy.service.load_stats.v2";
option java_generic_services = true;
import "envoy/api/v2/core/base.proto";
import "envoy/api/v2/endpoint/load_report.proto";
import "google/protobuf/duration.proto";
import "validate/validate.proto";
// [#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 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.
envoy.api.v2.core.Node node = 1;
// A list of load stats to report.
repeated envoy.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;
}