Add redis cluster

api/docs/BUILD

@@ -33,6 +33,7 @@ proto_library(
         "//envoy/config/accesslog/v2:als",
         "//envoy/config/accesslog/v2:file",
         "//envoy/config/bootstrap/v2:bootstrap",
+        "//envoy/config/cluster/redis:redis_cluster",
         "//envoy/config/common/tap/v2alpha:common",
         "//envoy/config/filter/accesslog/v2:accesslog",
         "//envoy/config/filter/dubbo/router/v2alpha1:router",

api/envoy/config/cluster/redis/BUILD

@@ -0,0 +1,8 @@
+load("@envoy_api//bazel:api_build_system.bzl", "api_proto_library_internal")
+
+licenses(["notice"])  # Apache 2
+
+api_proto_library_internal(
+    name = "redis_cluster",
+    srcs = ["redis_cluster.proto"],
+)

api/envoy/config/cluster/redis/redis_cluster.proto

@@ -0,0 +1,54 @@
+syntax = "proto3";
+
+package envoy.config.cluster.redis;
+
+option java_outer_classname = "RedisClusterProto";
+option java_multiple_files = true;
+option java_package = "io.envoyproxy.envoy.config.cluster.redis";
+option go_package = "v2";
+
+import "google/protobuf/duration.proto";
+
+import "validate/validate.proto";
+import "gogoproto/gogo.proto";
+
+// [#protodoc-title: Redis Cluster Configuration]
+// This cluster adds support for `Redis Cluster <https://redis.io/topics/cluster-spec>`_, as part
+// of :ref:`Envoy's support for Redis Cluster <arch_overview_redis>`.
+//
+// Redis Cluster is an extension of Redis which supports sharding and high availability (where a
+// shard that loses its master fails over to a replica, and designates it as the new master).
+// However, as there is no unified frontend or proxy service in front of Redis Cluster, the client
+// (in this case Envoy) must locally maintain the state of the Redis Cluster, specifically the
+// topology. A random node in the cluster is queried for the topology using the `CLUSTER SLOTS
+// command <https://redis.io/commands/cluster-slots>`_. This result is then stored locally, and
+// updated at user-configured intervals.
+//
+// Example:
+//
+// .. code-block:: yaml
+//
+//     name: name
+//     connect_timeout: 0.25s
+//     dns_lookup_family: V4_ONLY
+//     hosts:
+//     - socket_address:
+//       address: foo.bar.com
+//       port_value: 22120
+//     cluster_type:
+//     name: envoy.clusters.redis
+//     typed_config:
+//       "@type": type.googleapis.com/google.protobuf.Struct
+//       value:
+//         cluster_refresh_rate: 30s
+//         cluster_refresh_timeout: 0.5s
+
+message RedisClusterConfig {
+  // Interval between successive topology refresh requests. If not set, this defaults to 5s.
+  google.protobuf.Duration cluster_refresh_rate = 1
+      [(validate.rules).duration.gt = {}, (gogoproto.stdduration) = true];
+
+  // Timeout for topology refresh request. If not set, this defaults to 3s.
+  google.protobuf.Duration cluster_refresh_timeout = 2
+      [(validate.rules).duration.gt = {}, (gogoproto.stdduration) = true];
+}

docs/build.sh

@@ -83,6 +83,7 @@ PROTO_RST="
   /envoy/config/accesslog/v2/als/envoy/config/accesslog/v2/als.proto.rst
   /envoy/config/accesslog/v2/file/envoy/config/accesslog/v2/file.proto.rst
   /envoy/config/bootstrap/v2/bootstrap/envoy/config/bootstrap/v2/bootstrap.proto.rst
+  /envoy/config/cluster/redis/redis_cluster/envoy/config/cluster/redis/redis_cluster.proto.rst
   /envoy/config/common/tap/v2alpha/common/envoy/config/common/tap/v2alpha/common.proto.rst
   /envoy/config/ratelimit/v2/rls/envoy/config/ratelimit/v2/rls.proto.rst
   /envoy/config/metrics/v2/metrics_service/envoy/config/metrics/v2/metrics_service.proto.rst

docs/root/api-v2/config/cluster/cluster.rst

@@ -0,0 +1,8 @@
+Cluster
+=======
+
+.. toctree::
+  :glob:
+  :maxdepth: 1
+
+  redis/*

docs/root/api-v2/config/config.rst

@@ -12,3 +12,4 @@ Extensions
   transport_socket/transport_socket
   resource_monitor/resource_monitor
   common/common
+  cluster/cluster

docs/root/intro/arch_overview/redis.rst

@@ -57,6 +57,31 @@ If passive healthchecking is desired, also configure
 For the purposes of passive healthchecking, connect timeouts, command timeouts, and connection
 close map to 5xx. All other responses from Redis are counted as a success.
 
+Redis Cluster Support (Experimental)
+----------------------------------------
+
+Envoy currently offers experimental support for `Redis Cluster <https://redis.io/topics/cluster-spec>`_.
+This requires Envoy to track the topology of the cluster- which nodes exist, and which are the current 
+masters for each shard, and which shards and nodes enter or leave the cluster.
+
+The user may be familiar with `Twemproxy <https://github.com/twitter/twemproxy>`_, another proxy
+that interfaces with Redis. However, Twemproxy support for Redis differs in several important ways 
+from Envoy's. First, it does not support Redis Cluster. Second, it acts as service in front of 
+Redis, as opposed to being a sidecar proxy. Finally, it is un-maintained. 
+
+Why do we need a new system, aside from the last point? Beside the expected benefits of a sidecar 
+proxy such as unified retries and so on, it allows us to essentially build a common client for 
+all services to talk to Redis. This is important because while Redis non-cluster clients are 
+performant and have feature parity across various implementations, Redis Cluster clients do not! 
+Notably, support for spreading load across read replicas differs between various clients, 
+something very important if one wants to scale a Redis fleet to handle heavy read traffic coming 
+from services implemented in different languages, each with their own redis client. Hence, these 
+services can use a Redis non-cluster client and get all the benefits of Redis Cluster without 
+worrying about different or unexpected behaviors due to client differences.
+
+For topology configuration details, see the Redis Cluster
+:ref:`v2 API reference <envoy_api_msg_config.cluster.redis.RedisClusterConfig>`
+
 Supported commands
 ------------------
 

docs/root/intro/version_history.rst

@@ -11,6 +11,7 @@ Version history
 * ext_authz: added option to `ext_authz` that allows the filter clearing route cache.
 * http: mitigated a race condition with the :ref:`delayed_close_timeout<envoy_api_field_config.filter.network.http_connection_manager.v2.HttpConnectionManager.delayed_close_timeout>` where it could trigger while actively flushing a pending write buffer for a downstream connection.
 * jwt_authn: make filter's parsing of JWT more flexible, allowing syntax like ``jwt=eyJhbGciOiJS...ZFnFIw,extra=7,realm=123``
+* redis: add support for Redis cluster custom cluster type.
 * redis: added :ref:`prefix routing <envoy_api_field_config.filter.network.redis_proxy.v2.RedisProxy.prefix_routes>` to enable routing commands based on their key's prefix to different upstream.
 * redis: add support for zpopmax and zpopmin commands.
 * redis: added 

source/common/upstream/logical_dns_cluster.cc

@@ -42,19 +42,7 @@ LogicalDnsCluster::LogicalDnsCluster(
     }
   }
 
-  switch (cluster.dns_lookup_family()) {
-  case envoy::api::v2::Cluster::V6_ONLY:
-    dns_lookup_family_ = Network::DnsLookupFamily::V6Only;
-    break;
-  case envoy::api::v2::Cluster::V4_ONLY:
-    dns_lookup_family_ = Network::DnsLookupFamily::V4Only;
-    break;
-  case envoy::api::v2::Cluster::AUTO:
-    dns_lookup_family_ = Network::DnsLookupFamily::Auto;
-    break;
-  default:
-    NOT_REACHED_GCOVR_EXCL_LINE;
-  }
+  dns_lookup_family_ = getDnsLookupFamilyFromCluster(cluster);
 
   const envoy::api::v2::core::SocketAddress& socket_address =
       lbEndpoint().endpoint().address().socket_address();

source/common/upstream/upstream_impl.cc

@@ -1273,20 +1273,7 @@ StrictDnsClusterImpl::StrictDnsClusterImpl(
       local_info_(factory_context.localInfo()), dns_resolver_(dns_resolver),
       dns_refresh_rate_ms_(
           std::chrono::milliseconds(PROTOBUF_GET_MS_OR_DEFAULT(cluster, dns_refresh_rate, 5000))) {
-  switch (cluster.dns_lookup_family()) {
-  case envoy::api::v2::Cluster::V6_ONLY:
-    dns_lookup_family_ = Network::DnsLookupFamily::V6Only;
-    break;
-  case envoy::api::v2::Cluster::V4_ONLY:
-    dns_lookup_family_ = Network::DnsLookupFamily::V4Only;
-    break;
-  case envoy::api::v2::Cluster::AUTO:
-    dns_lookup_family_ = Network::DnsLookupFamily::Auto;
-    break;
-  default:
-    NOT_REACHED_GCOVR_EXCL_LINE;
-  }
-
+  dns_lookup_family_ = getDnsLookupFamilyFromCluster(cluster);
   const envoy::api::v2::ClusterLoadAssignment load_assignment(
       cluster.has_load_assignment() ? cluster.load_assignment()
                                     : Config::Utility::translateClusterHosts(cluster.hosts()));
@@ -1305,6 +1292,19 @@ StrictDnsClusterImpl::StrictDnsClusterImpl(
       load_assignment.policy(), overprovisioning_factor, kDefaultOverProvisioningFactor);
 }
 
+Network::DnsLookupFamily getDnsLookupFamilyFromCluster(const envoy::api::v2::Cluster& cluster) {
+  switch (cluster.dns_lookup_family()) {
+  case envoy::api::v2::Cluster::V6_ONLY:
+    return Network::DnsLookupFamily::V6Only;
+  case envoy::api::v2::Cluster::V4_ONLY:
+    return Network::DnsLookupFamily::V4Only;
+  case envoy::api::v2::Cluster::AUTO:
+    return Network::DnsLookupFamily::Auto;
+  default:
+    NOT_REACHED_GCOVR_EXCL_LINE;
+  }
+}
+
 void StrictDnsClusterImpl::startPreInit() {
   for (const ResolveTargetPtr& target : resolve_targets_) {
     target->startResolve();

source/common/upstream/upstream_impl.h

@@ -807,6 +807,11 @@ class BaseDynamicClusterImpl : public ClusterImplBase {
                              HostMap& updated_hosts, const HostMap& all_hosts);
 };
 
+/**
+ * Utility function to get Dns from cluster
+ */
+Network::DnsLookupFamily getDnsLookupFamilyFromCluster(const envoy::api::v2::Cluster& cluster);
+
 /**
  * Implementation of Upstream::Cluster that does periodic DNS resolution and updates the host
  * member set if the DNS members change.

source/extensions/clusters/redis/BUILD

@@ -0,0 +1,40 @@
+licenses(["notice"])  # Apache 2
+
+load(
+    "//bazel:envoy_build_system.bzl",
+    "envoy_cc_library",
+    "envoy_package",
+)
+
+envoy_package()
+
+envoy_cc_library(
+    name = "redis_cluster",
+    srcs = [
+        "redis_cluster.cc",
+        "redis_cluster.h",
+    ],
+    deps = [
+        "//include/envoy/api:api_interface",
+        "//include/envoy/http:codec_interface",
+        "//include/envoy/upstream:cluster_factory_interface",
+        "//include/envoy/upstream:cluster_manager_interface",
+        "//include/envoy/upstream:upstream_interface",
+        "//source/common/config:metadata_lib",
+        "//source/common/event:dispatcher_lib",
+        "//source/common/json:config_schemas_lib",
+        "//source/common/json:json_loader_lib",
+        "//source/common/network:utility_lib",
+        "//source/common/singleton:manager_impl_lib",
+        "//source/common/upstream:cluster_factory_lib",
+        "//source/common/upstream:upstream_includes",
+        "//source/common/upstream:upstream_lib",
+        "//source/extensions/clusters:well_known_names",
+        "//source/extensions/filters/network/common/redis:client_interface",
+        "//source/extensions/filters/network/common/redis:client_lib",
+        "//source/extensions/filters/network/common/redis:codec_interface",
+        "//source/extensions/transport_sockets/raw_buffer:config",
+        "//source/server:transport_socket_config_lib",
+        "@envoy_api//envoy/config/cluster/redis:redis_cluster_cc",
+    ],
+)