Skip to content

Latest commit

 

History

History
243 lines (203 loc) · 11 KB

A76-ring-hash-improvements.md

File metadata and controls

243 lines (203 loc) · 11 KB

A76: Improvements to the Ring Hash LB Policy

Abstract

This proposal describes two improvements to the ring_hash load balancing policy:

  1. The ability to use ring hash without xDS, by extending the policy configuration to define the request header to use as the request hash key.
  2. The ability to specify endpoint hash keys explicitly, instead of hashing the endpoint IP address.

Background

Terminology

  • The request hash key, after being hashed, defines where a given request is to be placed on the ring in order to find the closest endpoints.
  • The endpoint hash key, after being hashed, determines the locations of an endpoint on the ring.

Using ring hash without xDS by explicitly setting the request hash key

gRPC supports the ring_hash load balancing policy for consistent hashing. This policy currently requires using xDS for configuration because users have no other way to configure the hash for a request but to use the route configuration hash_policy field in the RouteAction route configuration. This makes the ring_hash policy unusable without an xDS infrastructure in place.

This proposal extends the configuration of ring_hash policy to specify a header to hash. This will make it possible to use ring_hash by configuring it entirely in the service config. If this configuration is omitted, we will preserve the current behavior of using the xDS hash policy.

Using an explicit endpoint hash key

Another limitation of the current ring_hash load balancing policy is that it always hashes the endpoint IP address to place the endpoints on the ring. In some scenario, this choice is not ideal: for example, Kubernetes Statefulsets offer a way to configure workloads with sticky identity such that endpoints keep storage and network identifier across restarts. However, the IP address may change across restarts. After a deployment, if all IP addresses have changed, then a completely new ring has to be constructed, even though it may have been desirable to keep the ring unchanged based on the Statefulsets identities, so that each instance stays at the same location on the ring.

Envoy offers a solution to control endpoint hashes independently of IP addresses. This mechanism uses the "envoy.lb" LbEndpoint.Metadata field hash_key value available in EDS instead of the endpoint IP address, as described in the Envoy documentation for ring hash. This proposal adds support for setting the endpoint hash key explicitly via EDS by reusing the configuration mechanism implemented in Envoy. To retain the advantage of being able to use ring_hash without xDS, custom gRPC name resolvers will be able to set this endpoint attribute through the language specific resolver attribute interface.

Related Proposals:

This proposal extends the following existing gRFCs:

Proposal

Explicitly setting the request hash key

A new field request_hash_header will be added to the ring_hash policy config:

    message RingHashLoadBalancingConfig {
      // (existing fields omitted)
      string request_hash_header = 3;
    }

Upon loading the load balancing config, if the request_hash_header field contains a value that is not a valid header name, then the configuration is rejected. If the request_hash_header refers to a binary header (suffixed with -bin), the configuration is also rejected.

At pick time:

  • If request_hash_header is empty, then the request hash key will be based on the xDS hash policy in RDS, which keeps the existing LB configuration for ring hash working as before with xDS. If the request hash key has not been set by xDS, then we will fail the pick.
  • If request_hash_header is not empty, and the header has a non-empty value, then the request hash key will be set to this value. If the header contains multiple values, then values are joined with a comma , character before hashing.
  • If request_hash_header is not empty, and the request has no value associated with the header, then the picker will generate a random hash for the request.

If the picker has generated a random hash, it will walk the ring from this hash, and pick the first READY endpoint. If no endpoint is currently in CONNECTING state, it will trigger a connection attempt on at most one endpoint that is in IDLE state along the way.

When a new picker is created, we will compute whether at least one of the endpoints is connecting, and store that information in the picker (picker_has_a_child_connecting in the pseudo code below). This avoids having to compute this information on every pick when using a random hash.

The following pseudo code describes the updated picker logic:

// Determine request hash.
using_random_hash = false;
if (config.request_hash_header.empty()) {
  // Set by the xDS config selector.
  request_hash = call_attributes.hash;
  if request_hash.empty() {
    // Something is wrong, since the xDS config selector is responsible
    // for generating a random hash is this case.
    return PICK_FAILED
  }
} else {
  header = headers.find(config.request_hash_header);
  if (header != null) {
    request_hash = ComputeHash(header);
  } else {
    request_hash = ComputeRandomHash();
    using_random_hash = true;
  }
}

first_index = ring.FindIndexForHash(request_hash);
if !using_random_hash {
    // Return based on A61 unchanged.
    // ...
} else {
  requested_connection = picker_has_endpoint_in_connecting_state;
  for (i = 0; i < ring.size(); ++i) {
    index = (first_index + i) % ring.size();
    if (ring[index].state == READY) {
      return ring[index].picker->Pick(...);
    }
    if (!requested_connection && ring[index].state == IDLE) {
      ring[index].endpoint.TriggerConnectionAttemptInControlPlane();
      requested_connection = true;
    }
  }
  if (requested_connection) return PICK_QUEUE;
}
// All children are in transient failure. Return the first failure.
return ring[first_index].picker->Pick(...);

This behavior ensures that a single RPC does not cause more than one endpoint to exit IDLE state at a time, and that a request missing the header does not incur extra latency in the common case where there is already at least one endpoint in READY state. It converges to picking a random endpoint, since each request may eventually cause a random endpoint to go from IDLE to READY.

Explicitly setting the endpoint hash key

The ring_hash policy will be changed such that the hash key used for determining the locations of each endpoint on the ring will be extracted from a pre-defined endpoint attribute called hash_key. If this attribute is set, then the endpoint is placed on the ring by hashing its value. If this attribute is not set or empty, then the endpoint's first address is hashed, matching the current behavior. The locations of an existing endpoint on the ring is updated if its hash_key endpoint attribute changes.

The xDS resolver, described in A74, will be changed to set the hash_key endpoint attribute to the value of LbEndpoint.Metadata envoy.lb hash_key field, as described in Envoy's documentation for the ring hash load balancer.

Temporary environment variable protection

Explicitly setting the request hash key will be gated by the GRPC_EXPERIMENTAL_RING_HASH_SET_REQUEST_HASH_KEY environment variable until sufficiently tested.

Adding support for the hash_key in xDS endpoint metadata could potentially break existing clients whose control plane is setting this key, because upgrading the client to a new version of gRPC would automatically cause the key to start being used. We expect that this change will not cause problems for most users, but just in case there is a problem, we will provide a migration story by supporting a temporary mechanism to tell gRPC to ignore the hash_key endpoint metadata. This will be enabled by setting the GRPC_XDS_ENDPOINT_HASH_KEY_BACKWARD_COMPAT environment variable to true. The first release will have the environment variable assume true if not set. A subsequent release will set it to false if not set, enabling the new behavior. A final release will remove the opt-out environment variable and leave the new behavior enabled.

Rationale

We originally proposed using language specific interfaces to set the request hash key. The advantage would have been that the request hash key would not have to be exposed through gRPC outgoing headers. However, this would have required defining language specific APIs, which would increase the complexity of this change.

We also discussed the option of exposing all LbEndpoint.metadata from EDS through name resolver attributes, instead of only extracting the specific hash_key attribute, so as to make them available to custom LB policies. We decided to keep only extract hash_key to limit the scope of this gRFC.

We discussed various option to handle requests that are missing a hash key in the non-xDS case. When using ring hash with xDS, the hash is assigned a random value in the xDS config selector, which ensure all picks for this request can trigger a connection to at most one endpoint. However, without xDS, there is no place in the code to assign the hash such that it retains this property. We considered the following alternative solutions:

  1. Add a config selector or filter to pick the hash. There currently is no public API to do so from the service config, so we would have had to define one.
  2. Using an internal request attribute to set the hash. Again, there is no cross-language public API for this.
  3. Failing the pick. We generally prefer the lack of a header to affect load balancing but not correctness, so this option was not ideal.
  4. Treating a missing header as being present but having the empty string for value. All instances of the channel would end up picking the same endpoint to send requests with a missing header, which could overload this endpoint if a lot of requests do not have a request hash key.

Implementation

Implemented in C-core in grpc/grpc#38312.