upstream: add failure percentage-based outlier detection (envoyproxy#…

…8130)

Description: Add a new outlier detection mode which compares each host's rate of request failure to a configured fixed threshold.

Risk Level: Low
Testing: 2 new unit tests added.
Docs Changes: New mode and config options described.
Release Notes: white_check_mark
Fixes envoyproxy#8105

Signed-off-by: James Forcier <jforcier@grubhub.com>

api/envoy/api/v2/cluster/outlier_detection.proto

@@ -111,4 +111,34 @@ message OutlierDetection {
   // is set to true.
   google.protobuf.UInt32Value enforcing_local_origin_success_rate = 15
       [(validate.rules).uint32.lte = 100];
+
+  // The failure percentage to use when determining failure percentage-based outlier detection. If
+  // the failure percentage of a given host is greater than or equal to this value, it will be
+  // ejected. Defaults to 85.
+  google.protobuf.UInt32Value failure_percentage_threshold = 16 [(validate.rules).uint32.lte = 100];
+
+  // The % chance that a host will be actually ejected when an outlier status is detected through
+  // failure percentage statistics. This setting can be used to disable ejection or to ramp it up
+  // slowly. Defaults to 0.
+  //
+  // [#next-major-version: setting this without setting failure_percentage_threshold should be
+  // invalid in v4.]
+  google.protobuf.UInt32Value enforcing_failure_percentage = 17 [(validate.rules).uint32.lte = 100];
+
+  // The % chance that a host will be actually ejected when an outlier status is detected through
+  // local-origin failure percentage statistics. This setting can be used to disable ejection or to
+  // ramp it up slowly. Defaults to 0.
+  google.protobuf.UInt32Value enforcing_failure_percentage_local_origin = 18
+      [(validate.rules).uint32.lte = 100];
+
+  // The minimum number of hosts in a cluster in order to perform failure percentage-based ejection.
+  // If the total number of hosts in the cluster is less than this value, failure percentage-based
+  // ejection will not be performed. Defaults to 5.
+  google.protobuf.UInt32Value failure_percentage_minimum_hosts = 19;
+
+  // The minimum number of total requests that must be collected in one interval (as defined by the
+  // interval duration above) to perform failure percentage-based ejection for this host. If the
+  // volume is lower than this setting, failure percentage-based ejection will not be performed for
+  // this host. Defaults to 50.
+  google.protobuf.UInt32Value failure_percentage_request_volume = 20;
 }

api/envoy/data/cluster/v2alpha/outlier_detection_event.proto

@@ -39,6 +39,7 @@ message OutlierDetectionEvent {
     option (validate.required) = true;
     OutlierEjectSuccessRate eject_success_rate_event = 9;
     OutlierEjectConsecutive eject_consecutive_event = 10;
+    OutlierEjectFailurePercentage eject_failure_percentage_event = 11;
   }
 }
 
@@ -75,6 +76,12 @@ enum OutlierEjectionType {
   // is set to *true*.
   // See :ref:`Cluster outlier detection <arch_overview_outlier_detection>` documentation for
   SUCCESS_RATE_LOCAL_ORIGIN = 4;
+  // Runs over aggregated success rate statistics from every host in cluster and selects hosts for
+  // which ratio of failed replies is above configured value.
+  FAILURE_PERCENTAGE = 5;
+  // Runs over aggregated success rate statistics for local origin failures from every host in
+  // cluster and selects hosts for which ratio of failed replies is above configured value.
+  FAILURE_PERCENTAGE_LOCAL_ORIGIN = 6;
 }
 
 // Represents possible action applied to upstream host
@@ -97,3 +104,8 @@ message OutlierEjectSuccessRate {
 
 message OutlierEjectConsecutive {
 }
+
+message OutlierEjectFailurePercentage {
+  // Host's success rate at the time of the ejection event on a 0-100 range.
+  uint32 host_success_rate = 1 [(validate.rules).uint32.lte = 100];
+}

docs/root/configuration/upstream/cluster_manager/cluster_runtime.rst

@@ -102,6 +102,31 @@ outlier_detection.success_rate_stdev_factor
   <envoy_api_field_cluster.OutlierDetection.success_rate_stdev_factor>`
   setting in outlier detection
 
+outlier_detection.enforcing_failure_percentage
+  :ref:`enforcing_failure_percentage
+  <envoy_api_field_cluster.OutlierDetection.enforcing_failure_percentage>`
+  setting in outlier detection
+
+outlier_detection.enforcing_failure_percentage_local_origin
+  :ref:`enforcing_failure_percentage_local_origin
+  <envoy_api_field_cluster.OutlierDetection.enforcing_failure_percentage_local_origin>`
+  setting in outlier detection
+
+outlier_detection.failure_percentage_request_volume
+  :ref:`failure_percentage_request_volume
+  <envoy_api_field_cluster.OutlierDetection.failure_percentage_request_volume>`
+  setting in outlier detection
+
+outlier_detection.failure_percentage_minimum_hosts
+  :ref:`failure_percentage_minimum_hosts
+  <envoy_api_field_cluster.OutlierDetection.failure_percentage_minimum_hosts>`
+  setting in outlier detection
+
+outlier_detection.failure_percentage_threshold
+  :ref:`failure_percentage_threshold
+  <envoy_api_field_cluster.OutlierDetection.failure_percentage_threshold>`
+  setting in outlier detection
+
 Core
 ----
 

docs/root/configuration/upstream/cluster_manager/cluster_stats.rst

@@ -141,6 +141,10 @@ statistics will be rooted at *cluster.<name>.outlier_detection.* and contain the
   ejections_detected_consecutive_local_origin_failure, Counter, Number of detected consecutive local origin failure ejections (even if unenforced)
   ejections_enforced_local_origin_success_rate, Counter, Number of enforced success rate outlier ejections for locally originated failures
   ejections_detected_local_origin_success_rate, Counter, Number of detected success rate outlier ejections for locally originated failures (even if unenforced)
+  ejections_enforced_failure_percentage, Counter, Number of enforced failure percentage outlier ejections. Exact meaning of this counter depends on :ref:`outlier_detection.split_external_local_origin_errors<envoy_api_field_cluster.OutlierDetection.split_external_local_origin_errors>` config item. Refer to :ref:`Outlier Detection documentation<arch_overview_outlier_detection>` for details.
+  ejections_detected_failure_percentage, Counter, Number of detected failure percentage outlier ejections (even if unenforced). Exact meaning of this counter depends on :ref:`outlier_detection.split_external_local_origin_errors<envoy_api_field_cluster.OutlierDetection.split_external_local_origin_errors>` config item. Refer to :ref:`Outlier Detection documentation<arch_overview_outlier_detection>` for details.
+  ejections_enforced_failure_percentage_local_origin, Counter, Number of enforced failure percentage outlier ejections for locally originated failures
+  ejections_detected_failure_percentage_local_origin, Counter, Number of detected failure percentage outlier ejections for locally originated failures (even if unenforced)
   ejections_total, Counter, Deprecated. Number of ejections due to any outlier type (even if unenforced)
   ejections_consecutive_5xx, Counter, Deprecated. Number of consecutive 5xx ejections (even if unenforced)
 

docs/root/intro/arch_overview/upstream/outlier.rst

@@ -145,6 +145,33 @@ Most configuration items, namely
 types of errors, but :ref:`outlier_detection.enforcing_success_rate<envoy_api_field_cluster.OutlierDetection.enforcing_success_rate>` applies
 to externally originated errors only and :ref:`outlier_detection.enforcing_local_origin_success_rate<envoy_api_field_cluster.OutlierDetection.enforcing_local_origin_success_rate>`  applies to locally originated errors only.
 
+.. _arch_overview_outlier_detection_failure_percentage:
+
+Failure Percentage
+^^^^^^^^^^^^^^^^^^
+
+Failure Percentage based outlier ejection functions similarly to the success rate detecion type, in
+that it relies on success rate data from each host in a cluster. However, rather than compare those
+values to the mean success rate of the cluster as a whole, they are compared to a flat
+user-configured threshold. This threshold is configured via the
+:ref:`outlier_detection.failure_percentage_threshold<envoy_api_field_cluster.OutlierDetection.failure_percentage_threshold>`
+field.
+
+The other configuration fields for failure percentage based ejection are similar to the fields for
+success rate ejection. Failure percentage based ejection also obeys
+:ref:`outlier_detection.split_external_local_origin_errors<envoy_api_field_cluster.OutlierDetection.split_external_local_origin_errors>`;
+the enforcement percentages for externally- and locally-originated errors are controlled by
+:ref:`outlier_detection.enforcing_failure_percentage<envoy_api_field_cluster.OutlierDetection.enforcing_failure_percentage>`
+and
+:ref:`outlier_detection.enforcing_failure_percentage_local_origin<envoy_api_field_cluster.OutlierDetection.enforcing_failure_percentage_local_origin>`,
+respectively. As with success rate detection, detection will not be performed for a host if its
+request volume over the aggregation interval is less than the
+:ref:`outlier_detection.failure_percentage_request_volume<envoy_api_field_cluster.OutlierDetection.failure_percentage_request_volume>`
+value. Detection also will not be performed for a cluster if the number of hosts with the minimum
+required request volume in an interval is less than the
+:ref:`outlier_detection.failure_percentage_minimum_hosts<envoy_api_field_cluster.OutlierDetection.failure_percentage_minimum_hosts>`
+value.
+
 .. _arch_overview_outlier_detection_grpc:
 
 gRPC

docs/root/intro/version_history.rst

@@ -61,6 +61,7 @@ Version history
 * tracing: added :ref:`max_path_tag_length <envoy_api_field_config.filter.network.http_connection_manager.v2.HttpConnectionManager.tracing>` to support customizing the length of the request path included in the extracted `http.url <https://github.com/opentracing/specification/blob/master/semantic_conventions.md#standard-span-tags-and-log-fields>` tag.
 * upstream: added :ref:`an option <envoy_api_field_Cluster.CommonLbConfig.close_connections_on_host_set_change>` that allows draining HTTP, TCP connection pools on cluster membership change.
 * upstream: added network filter chains to upstream connections, see :ref:`filters<envoy_api_field_Cluster.filters>`.
+* upstream: added new :ref:`failure-percentage based outlier detection<arch_overview_outlier_detection_failure_percentage>` mode.
 * upstream: use p2c to select hosts for least-requests load balancers if all host weights are the same, even in cases where weights are not equal to 1.
 * upstream: added :ref:`fail_traffic_on_panic <envoy_api_field_Cluster.CommonLbConfig.ZoneAwareLbConfig.fail_traffic_on_panic>` to allow failing all requests to a cluster during panic state.
 * zookeeper: parse responses and emit latency stats.