xDS REST and gRPC protocol

Envoy discovers its various dynamic resources via the filesystem or by querying one or more management servers. Collectively, these discovery services and their corresponding APIs are referred to as xDS. Resources are requested via subscriptions, by specifying a filesystem path to watch, initiating gRPC streams, or polling a REST-JSON URL. The latter two methods involve sending requests with a DiscoveryRequest proto payload. Resources are delivered in a DiscoveryResponse proto payload in all methods. We discuss each type of subscription below.

Resource Types

Every configuration resource in the xDS API has a type associated with it. Resource types follow a versioning scheme. Resource types are versioned independent of the transports described below.

The following v2 xDS resource types are supported:

The following v3 xdS resource types are supported:

The concept of type URLs appears below, and takes the form type.googleapis.com/ – e.g., type.googleapis.com/envoy.api.v2.Cluster for a Cluster resource. In various requests from Envoy and responses by the management server, the resource type URL is stated.

Filesystem subscriptions

The simplest approach to delivering dynamic configuration is to place it at a well known path specified in the ConfigSource. Envoy will use inotify (kqueue on macOS) to monitor the file for changes and parse the DiscoveryResponse proto in the file on update. Binary protobufs, JSON, YAML and proto text are supported formats for the DiscoveryResponse.

There is no mechanism available for filesystem subscriptions to ACK/NACK updates beyond stats counters and logs. The last valid configuration for an xDS API will continue to apply if an configuration update rejection occurs.

Streaming gRPC subscriptions

API flow

For typical HTTP routing scenarios, the core resource types for the client’s configuration are Listener, RouteConfiguration, Cluster, and ClusterLoadAssignment. Each Listener resource may point to a RouteConfiguration resource, which may point to one or more Cluster resources, and each Cluster resource may point to a ClusterLoadAssignment resource.

Envoy fetches all Listener and Cluster resources at startup. It then fetches whatever RouteConfiguration and ClusterLoadAssignment resources that are required by the Listener and Cluster resources. In effect, every Listener or Cluster resource is a root to part of Envoy’s configuration tree.

A non-proxy client such as gRPC might start by fetching only the specific Listener resources that it is interested in. It then fetches the RouteConfiguration resources required by those Listener resources, followed by whichever Cluster resources are required by those RouteConfiguration resources, followed by the ClusterLoadAssignment resources required by the Cluster resources. In effect, the original Listener resources are the roots to the client’s configuration tree.

Variants of the xDS Transport Protocol

Four Variants

There are four variants of the xDS transport protocol used via streaming gRPC, which cover all combinations of two dimensions.

The first dimension is State of the World (SotW) vs. incremental. The SotW approach was the original mechanism used by xDS, in which the client must specify all resource names it is interested in with each request (except when making a wildcard request in LDS/CDS), and the server must return all resources the client has subscribed to in each request (in LDS/CDS). This means that if the client is already subscribing to 99 resources and wants to add an additional one, it must send a request with all 100 resource names, rather than just the one new one. And the server must then respond by sending all 100 resources, even if the 99 that were already subscribed to have not changed (in LDS/CDS). This mechanism can be a scalability limitation, which is why the incremental protocol variant was introduced. The incremental approach allows both the client and server to indicate only deltas relative to their previous state – i.e., the client can say that it wants to add or remove its subscription to a particular resource name without resending those that have not changed, and the server can send updates only for those resources that have changed. The incremental protocol also provides a mechanism for lazy loading of resources. For details on the incremental protocol, see Incremental xDS below.

The second dimension is using a separate gRPC stream for each resource type vs. aggregating all resource types onto a single gRPC stream. The former approach was the original mechanism used by xDS, and it offers an eventual consistency model. The latter approach was added for environments in which explicit control of sequencing is required. For details, see Eventual consistency considerations below.

So, the four variants of the xDS transport protocol are:

  1. State of the World (Basic xDS): SotW, separate gRPC stream for each resource type

  2. Incremental xDS: incremental, separate gRPC stream for each resource type

  3. Aggregated Discovery Service (ADS): SotW, aggregate stream for all resource types

  4. Incremental ADS: incremental, aggregate stream for all resource types

RPC Services and Methods for Each Variant

For the non-aggregated protocol variants, there is a separate RPC service for each resource type. Each of these RPC services can provide a method for each of the SotW and Incremental protocol variants. Here are the RPC services and methods for each resource type:

  • Listener: Listener Discovery Service (LDS) - SotW: ListenerDiscoveryService.StreamListeners - Incremental: ListenerDiscoveryService.DeltaListeners

  • RouteConfiguration: Route Discovery Service (RDS) - SotW: RouteDiscoveryService.StreamRoutes - Incremental: RouteDiscoveryService.DeltaRoutes

  • ScopedRouteConfiguration: Scoped Route Discovery Service (SRDS) - SotW: ScopedRouteDiscoveryService.StreamScopedRoutes - Incremental: ScopedRouteDiscoveryService.DeltaScopedRoutes

  • VirtualHost: Virtual Host Discovery Service (VHDS) - SotW: N/A - Incremental: VirtualHostDiscoveryService.DeltaVirtualHosts

  • Cluster: Cluster Discovery Service (CDS) - SotW: ClusterDiscoveryService.StreamClusters - Incremental: ClusterDiscoveryService.DeltaClusters

  • ClusterLoadAssignment: Endpoint Discovery Service (EDS) - SotW: EndpointDiscoveryService.StreamEndpoints - Incremental: EndpointDiscoveryService.DeltaEndpoints

  • Secret: Secret Discovery Service (SDS) - SotW: SecretDiscoveryService.StreamSecrets - Incremental: SecretDiscoveryService.DeltaSecrets

  • Runtime: Runtime Discovery Service (RTDS) - SotW: RuntimeDiscoveryService.StreamRuntime - Incremental: RuntimeDiscoveryService.DeltaRuntime

In the aggregated protocol variants, all resource types are multiplexed on a single gRPC stream, where each resource type is treated as a separate logical stream within the aggregated stream. In effect, it simply combines all of the above separate APIs into a single stream by treating requests and responses for each resource type as a separate sub-stream on the single aggregated stream. The RPC service and methods for the aggregated protocol variants are:

  • SotW: AggregatedDiscoveryService.StreamAggregatedResources

  • Incremental: AggregatedDiscoveryService.DeltaAggregatedResources

For all of the SotW methods, the request type is DiscoveryRequest and the response type is DiscoveryResponse.

For all of the incremental methods, the request type is DeltaDiscoveryRequest and the response type is DeltaDiscoveryResponse.

Configuring Which Variant to Use

In the xDS API, the ConfigSource message indicates how to obtain resources of a particular type. If the ConfigSource contains a gRPC ApiConfigSource, it points to an upstream cluster for the management server; this will initiate an independent bidirectional gRPC stream for each xDS resource type, potentially to distinct management servers. If the ConfigSource contains a AggregatedConfigSource, it tells the client to use ADS.

Currently, the client is expected to be given some local configuration that tells it how to obtain the Listener and Cluster resources. Listener resources may include a ConfigSource that indicates how the RouteConfiguration resources are obtained, and Cluster resources may include a ConfigSource that indicates how the ClusterLoadAssignment resources are obtained.

Client Configuration

In Envoy, the bootstrap file contains two ConfigSource messages, one indicating how Listener resources are obtained and another indicating how Cluster resources are obtained. It also contains a separate ApiConfigSource message indicating how to contact the ADS server, which will be used whenever a ConfigSource message (either in the bootstrap file or in a Listener or Cluster resource obtained from a management server) contains an AggregatedConfigSource message.

In a gRPC client that uses xDS, only ADS is supported, and the bootstrap file contains the name of the ADS server, which will be used for all resources. The ConfigSource messages in the Listener and Cluster resources must contain AggregatedConfigSource messages.

The xDS transport Protocol

Transport API version

In addition the resource type version described above, the xDS wire protocol has a transport version associated with it. This provides type versioning for messages such as DiscoveryRequest and DiscoveryResponse.

ACK/NACK and resource instance versioning

Each xDS stream begins with a DiscoveryRequest from the client, specifying the list of resources to subscribe to, the type URL corresponding to the subscribed resources, the node identifier and an empty version_info.

In addition to resource and transport type versioning schemes above, which operate at the type level, Envoy has a resource instance version. Unlike the resource/transport types, this is not a property of the API but is instead a reflection of the specific revision of a named resource delivered over xDS.

An example EDS request might be:

  1. version_info:
  2. node: { id: envoy }
  3. resource_names:
  4. - foo
  5. - bar
  6. type_url: type.googleapis.com/envoy.api.v2.ClusterLoadAssignment
  7. response_nonce:

The management server may reply either immediately or when the requested resources are available with a DiscoveryResponse, e.g.:

  1. version_info: X
  2. resources:
  3. - foo ClusterLoadAssignment proto encoding
  4. - bar ClusterLoadAssignment proto encoding
  5. type_url: type.googleapis.com/envoy.api.v2.ClusterLoadAssignment
  6. nonce: A

After processing the DiscoveryResponse, Envoy will send a new request on the stream, specifying the last version successfully applied and the nonce provided by the management server. The version provides Envoy and the management server a shared notion of the currently applied configuration, as well as a mechanism to ACK/NACK configuration updates.

ACK

If the update was successfully applied, the version_info will be X, as indicated in the sequence diagram:

Version update after ACK

NACK

If Envoy had instead rejected configuration update X, it would reply with error_detail populated and its previous version, which in this case was the empty initial version. The error_detail has more details around the exact error message populated in the message field:

No version update after NACK

In the sequence diagrams, the following format is used to abbreviate messages:

  • DiscoveryRequest: (V=version_info,R=resource_names,N=response_nonce,T=type_url)

  • DiscoveryResponse: (V=version_info,R=resources,N=nonce,T=type_url)

After a NACK, an API update may succeed at a new version Y:

ACK after NACK

ACK and NACK semantics summary

Versioning and Node Identifier

Each stream has its own notion of versioning, there is no shared versioning across resource types. When ADS is not used, even each resource of a given resource type may have a distinct version, since the Envoy API allows distinct EDS/RDS resources to point at different ConfigSources.

Only the first request on a stream is guaranteed to carry the node identifier. The subsequent discovery requests on the same stream may carry an empty node identifier. This holds true regardless of the acceptance of the discovery responses on the same stream. The node identifier should always be identical if present more than once on the stream. It is sufficient to only check the first message for the node identifier as a result.

When to send an update

The management server should only send updates to the Envoy client when the resources in the DiscoveryResponse have changed. Envoy replies to any DiscoveryResponse with a DiscoveryRequest containing the ACK/NACK immediately after it has been either accepted or rejected. If the management server provides the same set of resources rather than waiting for a change to occur, it will cause needless work on both the client and the management server, which could have a severe performance impact.

Within a stream, new DiscoveryRequests supersede any prior DiscoveryRequests having the same resource type. This means that the management server only needs to respond to the latest DiscoveryRequest on each stream for any given resource type.

How the client specifies what resources to return

xDS requests allow the client to specify a set of resource names as a hint to the server about which resources the client is interested in. In the SotW protocol variants, this is done via the resource_names specified in the DiscoveryRequest; in the incremental protocol variants, this is done via the resource_names_subscribe and resource_names_unsubscribe fields in the DeltaDiscoveryRequest.

Normally (see below for exceptions), requests must specify the set of resource names that the client is interested in. The management server must supply the requested resources if they exist. The client will silently ignore any supplied resources that were not explicitly requested. When the client sends a new request that changes the set of resources being requested, the server must resend any newly requested resources, even if it previously sent those resources without having been asked for them and the resources have not changed since that time. If the list of resource names becomes empty, that means that the client is no longer interested in any resources of the specified type.

For Listener and Cluster resource types, there is also a “wildcard” mode, which is triggered when the initial request on the stream for that resource type contains no resource names. In this case, the server should use site-specific business logic to determine the full set of resources that the client is interested in, typically based on the client’s node identification. Note that once a stream has entered wildcard mode for a given resource type, there is no way to change the stream out of wildcard mode; resource names specified in any subsequent request on the stream will be ignored.

Client Behavior

Envoy will always use wildcard mode for Listener and Cluster resources. However, other xDS clients (such as gRPC clients that use xDS) may specify explicit resource names for these resource types, for example if they only have a singleton listener and already know its name from some out-of-band configuration.

Grouping Resources into Responses

In the incremental protocol variants, the server sends each resource in its own response. This means that if the server has previously sent 100 resources and only one of them has changed, it may send a response containing only the changed resource; it does not need to resend the 99 resources that have not changed, and the client must not delete the unchanged resources.

In the SotW protocol variants, all resource types except for Listener and Cluster are grouped into responses in the same way as in the incremental protocol variants. However, Listener and Cluster resource types are handled differently: the server must include the complete state of the world, meaning that all resources of the relevant type that are needed by the client must be included, even if they did not change since the last response. This means that if the server has previously sent 100 resources and only one of them has changed, it must resend all 100 of them, even the 99 that were not modified.

Note that all of the protocol variants operate on units of whole named resources. There is no mechanism for providing incremental updates of repeated fields within a named resource. Most notably, there is currently no mechanism for incrementally updating individual endpoints within an EDS response.

Deleting Resources

In the incremental proocol variants, the server signals the client that a resource should be deleted via the removed_resources field of the response. This tells the client to remove the resource from its local cache.

In the SotW protocol variants, the criteria for deleting resources is more complex. For Listener and Cluster resource types, if a previously seen resource is not present in a new response, that indicates that the resource has been removed, and the client must delete it; a response containing no resources means to delete all resources of that type. However, for other resource types, the API provides no mechanism for the server to tell the client that resources have been deleted; instead, deletions are indicated implicitly by parent resources being changed to no longer refer to a child resource. For example, when the client receives an LDS update removing a Listener that was previously pointing to RouteConfiguration A, if no other Listener is pointing to RouteConfiguration A, then the client may delete A. For those resource types, an empty DiscoveryResponse is effectively a no-op from the client’s perspective.

Knowing When a Requested Resource Does Not Exist

The SotW protocol variants do not provide any explicit mechanism to determine when a requested resource does not exist.

Responses for Listener and Cluster resource types must include all resources requested by the client. However, it may not be possible for the client to know that a resource does not exist based solely on its absence in a response, because the delivery of the updates is eventually consistent: if the client initially sends a request for resource A, then sends a request for resources A and B, and then sees a response containing only resource A, the client cannot conclude that resource B does not exist, because the response may have been sent on the basis of the first request, before the server saw the second request.

For other resource types, because each resource can be sent in its own response, there is no way to know from the next response whether the newly requested resource exists, because the next response could be an unrelated update for another resource that had already been subscribed to previously.

As a result, clients are expected to use a timeout (recommended duration is 15 seconds) after sending a request for a new resource, after which they will consider the requested resource to not exist if they have not received the resource. In Envoy, this is done for RouteConfiguration and ClusterLoadAssignment resources during resource warming.

Note that this timeout is not strictly necessary when using wildcard mode for Listener and Cluster resource types, because in that case every response will contain all existing resources that are relevant to the client, so the client can know that a resource does not exist by its absence in the next response it sees. However, using a timeout is still recommended in this case, since it protects against the case where the management server fails to send a response in a timely manner.

Note that even if a requested resource does not exist at the moment when the client requests it, that resource could be created at any time. Management servers must remember the set of resources being requested by the client, and if one of those resources springs into existence later, the server must send an update to the client informing it of the new resource. Clients that initially see a resource that does not exist must be prepared for the resource to be created at any time.

Unsubscribing From Resources

In the incremental protocol variants, resources can be unsubscribed to via the resource_names_unsubscribe field.

In the SotW protocol variants, each request must contain the full list of resource names being subscribed to in the resource_names field, so unsubscribing to a set of resources is done by sending a new request containing all resource names that are still being subscribed to but not containing the resource names being unsubscribed to. For example, if the client had previously been subscribed to resources A and B but wishes to unsubscribe from B, it must send a new request containing only resource A.

Note that for Listener and Cluster resource types where the stream is in “wildcard” mode (see How the client specifies what resources to return for details), the set of resources being subscribed to is determined by the server instead of the client, so there is no mechanism for the client to unsubscribe from resources.

Requesting Multiple Resources on a Single Stream

For EDS/RDS, Envoy may either generate a distinct stream for each resource of a given type (e.g. if each ConfigSource has its own distinct upstream cluster for a management server), or may combine together multiple resource requests for a given resource type when they are destined for the same management server. While this is left to implementation specifics, management servers should be capable of handling one or more resource_names for a given resource type in each request. Both sequence diagrams below are valid for fetching two EDS resources {foo, bar}:

Multiple EDS requests on the same stream Multiple EDS requests on distinct streams

Resource updates

As discussed above, Envoy may update the list of resource_names it presents to the management server in each DiscoveryRequest that ACK/NACKs a specific DiscoveryResponse. In addition, Envoy may later issue additional DiscoveryRequests at a given version_info to update the management server with new resource hints. For example, if Envoy is at EDS version X and knows only about cluster foo, but then receives a CDS update and learns about bar in addition, it may issue an additional DiscoveryRequest for X with {foo,bar} as resource_names.

CDS response leads to EDS resource hint update

There is a race condition that may arise here; if after a resource hint update is issued by Envoy at X, but before the management server processes the update it replies with a new version Y, the resource hint update may be interpreted as a rejection of Y by presenting an X version_info. To avoid this, the management server provides a nonce that Envoy uses to indicate the specific DiscoveryResponse each DiscoveryRequest corresponds to:

EDS update race motivates nonces

The management server should not send a DiscoveryResponse for any DiscoveryRequest that has a stale nonce. A nonce becomes stale following a newer nonce being presented to Envoy in a DiscoveryResponse. A management server does not need to send an update until it determines a new version is available. Earlier requests at a version then also become stale. It may process multiple DiscoveryRequests at a version until a new version is ready.

Requests become stale

An implication of the above resource update sequencing is that Envoy does not expect a DiscoveryResponse for every DiscoveryRequests it issues.

Resource warming

Clusters and Listeners go through warming before they can serve requests. This process happens both during Envoy initialization and when the Cluster or Listener is updated. Warming of Cluster is completed only when a ClusterLoadAssignment response is supplied by management server. Similarly, warming of Listener is completed only when a RouteConfiguration is supplied by management server if the listener refers to an RDS configuration. Management server is expected to provide the EDS/RDS updates during warming. If management server does not provide EDS/RDS responses, Envoy will not initialize itself during the initialization phase and the updates sent via CDS/LDS will not take effect until EDS/RDS responses are supplied.

Eventual consistency considerations

Since Envoy’s xDS APIs are eventually consistent, traffic may drop briefly during updates. For example, if only cluster X is known via CDS/EDS, a RouteConfiguration references cluster X and is then adjusted to cluster Y just before the CDS/EDS update providing Y, traffic will be blackholed until Y is known about by the Envoy instance.

For some applications, a temporary drop of traffic is acceptable, retries at the client or by other Envoy sidecars will hide this drop. For other scenarios where drop can’t be tolerated, traffic drop could have been avoided by providing a CDS/EDS update with both X and Y, then the RDS update repointing from X to Y and then a CDS/EDS update dropping X.

In general, to avoid traffic drop, sequencing of updates should follow a make before break model, wherein:

  • CDS updates (if any) must always be pushed first.

  • EDS updates (if any) must arrive after CDS updates for the respective clusters.

  • LDS updates must arrive after corresponding CDS/EDS updates.

  • RDS updates related to the newly added listeners must arrive after CDS/EDS/LDS updates.

  • VHDS updates (if any) related to the newly added RouteConfigurations must arrive after RDS updates.

  • Stale CDS clusters and related EDS endpoints (ones no longer being referenced) can then be removed.

xDS updates can be pushed independently if no new clusters/routes/listeners are added or if it’s acceptable to temporarily drop traffic during updates. Note that in case of LDS updates, the listeners will be warmed before they receive traffic, i.e. the dependent routes are fetched through RDS if configured. Clusters are warmed when adding/removing/updating clusters. On the other hand, routes are not warmed, i.e., the management plane must ensure that clusters referenced by a route are in place, before pushing the updates for a route.

Aggregated Discovery Service

It’s challenging to provide the above guarantees on sequencing to avoid traffic drop when management servers are distributed. ADS allow a single management server, via a single gRPC stream, to deliver all API updates. This provides the ability to carefully sequence updates to avoid traffic drop. With ADS, a single stream is used with multiple independent DiscoveryRequest/DiscoveryResponse sequences multiplexed via the type URL. For any given type URL, the above sequencing of DiscoveryRequest and DiscoveryResponse messages applies. An example update sequence might look like:

EDS/CDS multiplexed on an ADS stream

A single ADS stream is available per Envoy instance.

An example minimal bootstrap.yaml fragment for ADS configuration is:

  1. node:
  2. id: <node identifier>
  3. dynamic_resources:
  4. cds_config: {ads: {}}
  5. lds_config: {ads: {}}
  6. ads_config:
  7. api_type: GRPC
  8. grpc_services:
  9. envoy_grpc:
  10. cluster_name: ads_cluster
  11. static_resources:
  12. clusters:
  13. - name: ads_cluster
  14. connect_timeout: { seconds: 5 }
  15. type: STATIC
  16. hosts:
  17. - socket_address:
  18. address: <ADS management server IP address>
  19. port_value: <ADS management server port>
  20. lb_policy: ROUND_ROBIN
  21. http2_protocol_options: {}
  22. upstream_connection_options:
  23. # configure a TCP keep-alive to detect and reconnect to the admin
  24. # server in the event of a TCP socket disconnection
  25. tcp_keepalive:
  26. ...
  27. admin:
  28. ...

Incremental xDS

Incremental xDS is a separate xDS endpoint that:

  • Allows the protocol to communicate on the wire in terms of resource/resource name deltas (“Delta xDS”). This supports the goal of scalability of xDS resources. Rather than deliver all 100k clusters when a single cluster is modified, the management server only needs to deliver the single cluster that changed.

  • Allows the Envoy to on-demand / lazily request additional resources. For example, requesting a cluster only when a request for that cluster arrives.

An Incremental xDS session is always in the context of a gRPC bidirectional stream. This allows the xDS server to keep track of the state of xDS clients connected to it. There is no REST version of Incremental xDS yet.

In the delta xDS wire protocol, the nonce field is required and used to pair a DeltaDiscoveryResponse to a DeltaDiscoveryRequest ACK or NACK. Optionally, a response message level system_version_info is present for debugging purposes only.

DeltaDiscoveryRequest can be sent in the following situations:

In this first example the client connects and receives a first update that it ACKs. The second update fails and the client NACKs the update. Later the xDS client spontaneously requests the “wc” resource.

Incremental session example

On reconnect the Incremental xDS client may tell the server of its known resources to avoid resending them over the network. Because no state is assumed to be preserved from the previous stream, the reconnecting client must provide the server with all resource names it is interested in.

Incremental reconnect example

Resource names

Resources are identified by a resource name or an alias. Aliases of a resource, if present, can be identified by the alias field in the resource of a DeltaDiscoveryResponse. The resource name will be returned in the name field in the resource of a DeltaDiscoveryResponse.

Subscribing to Resources

The client can send either an alias or the name of a resource in the resource_names_subscribe field of a DeltaDiscoveryRequest in order to subscribe to a resource. Both the names and aliases of resources should be checked in order to determine whether the entity in question has been subscribed to.

A resource_names_subscribe field may contain resource names that the server believes the client is already subscribed to, and furthermore has the most recent versions of. However, the server must still provide those resources in the response; due to implementation details hidden from the server, the client may have “forgotten” those resources despite apparently remaining subscribed.

Unsubscribing from Resources

When a client loses interest in some resources, it will indicate that with the resource_names_unsubscribe field of a DeltaDiscoveryRequest. As with resource_names_subscribe, these may be resource names or aliases.

A resource_names_unsubscribe field may contain superfluous resource names, which the server thought the client was already not subscribed to. The server must cleanly process such a request; it can simply ignore these phantom unsubscriptions.

Knowing When a Requested Resource Does Not Exist

When a resource subscribed to by a client does not exist, the server will send a Resource whose name field matches the name that the client subscribed to and whose resource field is unset. This allows the client to quickly determine when a resource does not exist without waiting for a timeout, as would be done in the SotW protocol variants. However, clients are still encouraged to use a timeout to protect against the case where the management server fails to send a response in a timely manner.

REST-JSON polling subscriptions

Synchronous (long) polling via REST endpoints is also available for the xDS singleton APIs. The above sequencing of messages is similar, except no persistent stream is maintained to the management server. It is expected that there is only a single outstanding request at any point in time, and as a result the response nonce is optional in REST-JSON. The JSON canonical transform of proto3 is used to encode DiscoveryRequest and DiscoveryResponse messages. ADS is not available for REST-JSON polling.

When the poll period is set to a small value, with the intention of long polling, then there is also a requirement to avoid sending a DiscoveryResponse unless a change to the underlying resources has occurred via a resource update.