A LoadBalancer typically implements three interfaces: *
{@link Helper Helper} is implemented by gRPC library and provided to {@link Factory * Factory}. It provides functionalities that a {@code LoadBalancer} implementation would typically * need. * *
All methods on the {@link LoadBalancer} interface are called from a Synchronization Context, * meaning they are serialized, thus the balancer implementation doesn't need to worry about * synchronization among them. {@link io.grpc.LoadBalancer.Helper#getSynchronizationContext} * allows implementations to schedule tasks to be run in the same Synchronization Context, with or * without a delay, thus those tasks don't need to worry about synchronizing with the balancer * methods. * *
However, the actual running thread may be the network thread, thus the following rules must be * followed to prevent blocking or even dead-locking in a network: * *
A {@link LoadBalancer} keeps states like the latest addresses from NameResolver, the * Subchannel(s) and their latest connectivity states. These states are mutated within the * Synchronization Context, * *
A typical {@link SubchannelPicker SubchannelPicker} holds a snapshot of these states. It may * have its own states, e.g., a picker from a round-robin load-balancer may keep a pointer to the * next Subchannel, which are typically mutated by multiple threads. The picker should only mutate * its own state, and should not mutate or re-acquire the states of the LoadBalancer. This way the * picker only needs to synchronize its own states, which is typically trivial to implement. * *
When the LoadBalancer states changes, e.g., Subchannels has become or stopped being READY, and * we want subsequent RPCs to use the latest list of READY Subchannels, LoadBalancer would create a * new picker, which holds a snapshot of the latest Subchannel list. Refer to the javadoc of {@link * io.grpc.LoadBalancer.SubchannelStateListener#onSubchannelState onSubchannelState()} how to do * this properly. * *
No synchronization should be necessary between LoadBalancer and its pickers if you follow
* the pattern above. It may be possible to implement in a different way, but that would usually
* result in more complicated threading.
*
* @since 1.2.0
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
@NotThreadSafe
public abstract class LoadBalancer {
@Internal
@NameResolver.ResolutionResultAttr
public static final Attributes.Key Implementations should not modify the given {@code servers}.
*
* @param resolvedAddresses the resolved server addresses, attributes, and config.
* @since 1.21.0
*/
public void handleResolvedAddresses(ResolvedAddresses resolvedAddresses) {
if (recursionCount++ == 0) {
// Note that the information about the addresses actually being accepted will be lost
// if you rely on this method for backward compatibility.
acceptResolvedAddresses(resolvedAddresses);
}
recursionCount = 0;
}
/**
* Accepts newly resolved addresses from the name resolution system. The {@link
* EquivalentAddressGroup} addresses should be considered equivalent but may be flattened into a
* single list if needed.
*
* Implementations can choose to reject the given addresses by returning {@code false}.
*
* Implementations should not modify the given {@code addresses}.
*
* @param resolvedAddresses the resolved server addresses, attributes, and config.
* @return {@code true} if the resolved addresses were accepted. {@code false} if rejected.
* @since 1.49.0
*/
public boolean acceptResolvedAddresses(ResolvedAddresses resolvedAddresses) {
if (resolvedAddresses.getAddresses().isEmpty()
&& !canHandleEmptyAddressListFromNameResolution()) {
handleNameResolutionError(Status.UNAVAILABLE.withDescription(
"NameResolver returned no usable address. addrs=" + resolvedAddresses.getAddresses()
+ ", attrs=" + resolvedAddresses.getAttributes()));
return false;
} else {
if (recursionCount++ == 0) {
handleResolvedAddresses(resolvedAddresses);
}
recursionCount = 0;
return true;
}
}
/**
* Represents a combination of the resolved server address, associated attributes and a load
* balancing policy config. The config is from the {@link
* LoadBalancerProvider#parseLoadBalancingPolicyConfig(Map)}.
*
* @since 1.21.0
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public static final class ResolvedAddresses {
private final List The initial state of a Subchannel is IDLE. You won't get a notification for the initial IDLE
* state.
*
* If the new state is not SHUTDOWN, this method should create a new picker and call {@link
* Helper#updateBalancingState Helper.updateBalancingState()}. Failing to do so may result in
* unnecessary delays of RPCs. Please refer to {@link PickResult#withSubchannel
* PickResult.withSubchannel()}'s javadoc for more information.
*
* SHUTDOWN can only happen in two cases. One is that LoadBalancer called {@link
* Subchannel#shutdown} earlier, thus it should have already discarded this Subchannel. The other
* is that Channel is doing a {@link ManagedChannel#shutdownNow forced shutdown} or has already
* terminated, thus there won't be further requests to LoadBalancer. Therefore, the LoadBalancer
* usually don't need to react to a SHUTDOWN state.
*
* @param subchannel the involved Subchannel
* @param stateInfo the new state
* @since 1.2.0
* @deprecated This method will be removed. Stop overriding it. Instead, pass {@link
* SubchannelStateListener} to {@link Subchannel#start} to receive Subchannel state
* updates
*/
@Deprecated
public void handleSubchannelState(
Subchannel subchannel, ConnectivityStateInfo stateInfo) {
// Do nothing. If the implementation doesn't implement this, it will get subchannel states from
// the new API. We don't throw because there may be forwarding LoadBalancers still plumb this.
}
/**
* The channel asks the load-balancer to shutdown. No more methods on this class will be called
* after this method. The implementation should shutdown all Subchannels and OOB channels, and do
* any other cleanup as necessary.
*
* @since 1.2.0
*/
public abstract void shutdown();
/**
* Whether this LoadBalancer can handle empty address group list to be passed to {@link
* #handleResolvedAddresses(ResolvedAddresses)}. The default implementation returns
* {@code false}, meaning that if the NameResolver returns an empty list, the Channel will turn
* that into an error and call {@link #handleNameResolutionError}. LoadBalancers that want to
* accept empty lists should override this method and return {@code true}.
*
* This method should always return a constant value. It's not specified when this will be
* called.
*/
public boolean canHandleEmptyAddressListFromNameResolution() {
return false;
}
/**
* The channel asks the LoadBalancer to establish connections now (if applicable) so that the
* upcoming RPC may then just pick a ready connection without waiting for connections. This
* is triggered by {@link ManagedChannel#getState ManagedChannel.getState(true)}.
*
* If LoadBalancer doesn't override it, this is no-op. If it infeasible to create connections
* given the current state, e.g. no Subchannel has been created yet, LoadBalancer can ignore this
* request.
*
* @since 1.22.0
*/
public void requestConnection() {}
/**
* The main balancing logic. It must be thread-safe. Typically it should only
* synchronize on its own state, and avoid synchronizing with the LoadBalancer's state.
*
* @since 1.2.0
*/
@ThreadSafe
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public abstract static class SubchannelPicker {
/**
* Make a balancing decision for a new RPC.
*
* @param args the pick arguments
* @since 1.3.0
*/
public abstract PickResult pickSubchannel(PickSubchannelArgs args);
/**
* Tries to establish connections now so that the upcoming RPC may then just pick a ready
* connection without having to connect first.
*
* No-op if unsupported.
*
* @deprecated override {@link LoadBalancer#requestConnection} instead.
* @since 1.11.0
*/
@Deprecated
public void requestConnection() {}
}
/**
* Provides arguments for a {@link SubchannelPicker#pickSubchannel(
* LoadBalancer.PickSubchannelArgs)}.
*
* @since 1.2.0
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public abstract static class PickSubchannelArgs {
/**
* Call options.
*
* @since 1.2.0
*/
public abstract CallOptions getCallOptions();
/**
* Headers of the call. {@link SubchannelPicker#pickSubchannel} may mutate it before before
* returning.
*
* @since 1.2.0
*/
public abstract Metadata getHeaders();
/**
* Call method.
*
* @since 1.2.0
*/
public abstract MethodDescriptor getMethodDescriptor();
}
/**
* A balancing decision made by {@link SubchannelPicker SubchannelPicker} for an RPC.
*
* The outcome of the decision will be one of the following:
* The Subchannel should either be an original Subchannel returned by {@link
* Helper#createSubchannel Helper.createSubchannel()}, or a wrapper of it preferably based on
* {@code ForwardingSubchannel}. At the very least its {@link Subchannel#getInternalSubchannel
* getInternalSubchannel()} must return the same object as the one returned by the original.
* Otherwise the Channel cannot use it for the RPC.
*
* When the RPC tries to use the return Subchannel, which is briefly after this method
* returns, the state of the Subchannel will decide where the RPC would go:
*
* All buffered RPCs will stay buffered until the next call of {@link
* Helper#updateBalancingState Helper.updateBalancingState()}, which will trigger a new picking
* process.
*
* Note that Subchannel's state may change at the same time the picker is making the
* decision, which means the decision may be made with (to-be) outdated information. For
* example, a picker may return a Subchannel known to be READY, but it has become IDLE when is
* about to be used by the RPC, which makes the RPC to be buffered. The LoadBalancer will soon
* learn about the Subchannels' transition from READY to IDLE, create a new picker and allow the
* RPC to use another READY transport if there is any.
*
* You will want to avoid running into a situation where there are READY Subchannels out
* there but some RPCs are still buffered for longer than a brief time.
* In order to prevent unnecessary delay of RPCs, the rules of thumb are:
* This is an optional property.
*
* @param key the option key
* @param value the option value
*/
public This is a required property.
*
* @throws IllegalArgumentException if {@code addrs} is empty
*/
public Builder setAddresses(List This is an optional property. Default is empty if not set.
*/
public Builder setAttributes(Attributes attrs) {
this.attrs = checkNotNull(attrs, "attrs");
return this;
}
/**
* Creates a new args object.
*/
public CreateSubchannelArgs build() {
return new CreateSubchannelArgs(addrs, attrs, customOptions);
}
}
/**
* Key for a key-value pair. Uses reference equality.
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public static final class Key The LoadBalancer is responsible for closing unused Subchannels, and closing all
* Subchannels within {@link #shutdown}.
*
* It must be called from {@link #getSynchronizationContext the Synchronization Context}
*
* @since 1.22.0
*/
public Subchannel createSubchannel(CreateSubchannelArgs args) {
throw new UnsupportedOperationException();
}
/**
* Out-of-band channel for LoadBalancer’s own RPC needs, e.g., talking to an external
* load-balancer service.
*
* The LoadBalancer is responsible for closing unused OOB channels, and closing all OOB
* channels within {@link #shutdown}.
*
* @since 1.4.0
*/
public abstract ManagedChannel createOobChannel(EquivalentAddressGroup eag, String authority);
/**
* Accept a list of EAG for multiple authorities: https://github.com/grpc/grpc-java/issues/4618
* */
public ManagedChannel createOobChannel(List The target string will be resolved by a {@link NameResolver} created according to the
* target string.
*
* The LoadBalancer is responsible for closing unused OOB channels, and closing all OOB
* channels within {@link #shutdown}.
*
* @since 1.20.0
*/
public ManagedChannel createResolvingOobChannel(String target) {
return createResolvingOobChannelBuilder(target).build();
}
/**
* Creates an out-of-band channel builder for LoadBalancer's own RPC needs, e.g., talking to an
* external load-balancer service, that is specified by a target string. See the documentation
* on {@link ManagedChannelBuilder#forTarget} for the format of a target string.
*
* The target string will be resolved by a {@link NameResolver} created according to the
* target string.
*
* The returned oob-channel builder defaults to use the same authority and ChannelCredentials
* (without bearer tokens) as the parent channel's for authentication. This is different from
* {@link #createResolvingOobChannelBuilder(String, ChannelCredentials)}.
*
* The LoadBalancer is responsible for closing unused OOB channels, and closing all OOB
* channels within {@link #shutdown}.
*
* @deprecated Use {@link #createResolvingOobChannelBuilder(String, ChannelCredentials)}
* instead.
* @since 1.31.0
*/
@Deprecated
public ManagedChannelBuilder createResolvingOobChannelBuilder(String target) {
throw new UnsupportedOperationException("Not implemented");
}
/**
* Creates an out-of-band channel builder for LoadBalancer's own RPC needs, e.g., talking to an
* external load-balancer service, that is specified by a target string and credentials. See
* the documentation on {@link Grpc#newChannelBuilder} for the format of a target string.
*
* The target string will be resolved by a {@link NameResolver} created according to the
* target string.
*
* The LoadBalancer is responsible for closing unused OOB channels, and closing all OOB
* channels within {@link #shutdown}.
*
* @since 1.35.0
*/
public ManagedChannelBuilder createResolvingOobChannelBuilder(
String target, ChannelCredentials creds) {
throw new UnsupportedOperationException();
}
/**
* Set a new state with a new picker to the channel.
*
* When a new picker is provided via {@code updateBalancingState()}, the channel will apply
* the picker on all buffered RPCs, by calling {@link SubchannelPicker#pickSubchannel(
* LoadBalancer.PickSubchannelArgs)}.
*
* The channel will hold the picker and use it for all RPCs, until {@code
* updateBalancingState()} is called again and a new picker replaces the old one. If {@code
* updateBalancingState()} has never been called, the channel will buffer all RPCs until a
* picker is provided.
*
* It should be called from the Synchronization Context. Currently will log a warning if
* violated. It will become an exception eventually. See #5015 for the background.
*
* The passed state will be the channel's new state. The SHUTDOWN state should not be passed
* and its behavior is undefined.
*
* @since 1.6.0
*/
public abstract void updateBalancingState(
@Nonnull ConnectivityState newState, @Nonnull SubchannelPicker newPicker);
/**
* Call {@link NameResolver#refresh} on the channel's resolver.
*
* It should be called from the Synchronization Context. Currently will log a warning if
* violated. It will become an exception eventually. See #5015 for the background.
*
* @since 1.18.0
*/
public void refreshNameResolution() {
throw new UnsupportedOperationException();
}
/**
* Historically the channel automatically refreshes name resolution if any subchannel
* connection is broken. It's transitioning to let load balancers make the decision. To
* avoid silent breakages, the channel checks if {@link #refreshNameResolution} is called
* by the load balancer. If not, it will do it and log a warning. This will be removed in
* the future and load balancers are completely responsible for triggering the refresh.
* See #8088 for the background.
*
* This should rarely be used, but sometimes the address for the subchannel wasn't
* provided by the name resolver and a refresh needs to be directed somewhere else instead.
* Then you can call this method to disable the short-tem check for detecting LoadBalancers
* that need to be updated for the new expected behavior.
*
* @since 1.38.0
* @deprecated Warning has been removed
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/8088")
@Deprecated
public void ignoreRefreshNameResolutionCheck() {
// no-op
}
/**
* Returns a {@link SynchronizationContext} that runs tasks in the same Synchronization Context
* as that the callback methods on the {@link LoadBalancer} interface are run in.
*
* Pro-tip: in order to call {@link SynchronizationContext#schedule}, you need to provide a
* {@link ScheduledExecutorService}. {@link #getScheduledExecutorService} is provided for your
* convenience.
*
* @since 1.17.0
*/
public SynchronizationContext getSynchronizationContext() {
// TODO(zhangkun): make getSynchronizationContext() abstract after runSerialized() is deleted
throw new UnsupportedOperationException();
}
/**
* Returns a {@link ScheduledExecutorService} for scheduling delayed tasks.
*
* This service is a shared resource and is only meant for quick tasks. DO NOT block or run
* time-consuming tasks.
*
* The returned service doesn't support {@link ScheduledExecutorService#shutdown shutdown()}
* and {@link ScheduledExecutorService#shutdownNow shutdownNow()}. They will throw if called.
*
* @since 1.17.0
*/
public ScheduledExecutorService getScheduledExecutorService() {
throw new UnsupportedOperationException();
}
/**
* Returns the authority string of the channel, which is derived from the DNS-style target name.
* If overridden by a load balancer, {@link #getUnsafeChannelCredentials} must also be
* overridden to call {@link #getChannelCredentials} or provide appropriate credentials.
*
* @since 1.2.0
*/
public abstract String getAuthority();
/**
* Returns the ChannelCredentials used to construct the channel, without bearer tokens.
*
* @since 1.35.0
*/
public ChannelCredentials getChannelCredentials() {
return getUnsafeChannelCredentials().withoutBearerTokens();
}
/**
* Returns the UNSAFE ChannelCredentials used to construct the channel,
* including bearer tokens. Load balancers should generally have no use for
* these credentials and use of them is heavily discouraged. These must be used
* very carefully to avoid sending bearer tokens to untrusted servers
* as the server could then impersonate the client. Generally it is only safe
* to use these credentials when communicating with the backend.
*
* @since 1.35.0
*/
public ChannelCredentials getUnsafeChannelCredentials() {
throw new UnsupportedOperationException();
}
/**
* Returns the {@link ChannelLogger} for the Channel served by this LoadBalancer.
*
* @since 1.17.0
*/
public ChannelLogger getChannelLogger() {
throw new UnsupportedOperationException();
}
/**
* Returns the {@link NameResolver.Args} that the Channel uses to create {@link NameResolver}s.
*
* @since 1.22.0
*/
public NameResolver.Args getNameResolverArgs() {
throw new UnsupportedOperationException();
}
/**
* Returns the {@link NameResolverRegistry} that the Channel uses to look for {@link
* NameResolver}s.
*
* @since 1.22.0
*/
public NameResolverRegistry getNameResolverRegistry() {
throw new UnsupportedOperationException();
}
}
/**
* A logical connection to a server, or a group of equivalent servers represented by an {@link
* EquivalentAddressGroup}.
*
* It maintains at most one physical connection (aka transport) for sending new RPCs, while
* also keeps track of previous transports that has been shut down but not terminated yet.
*
* If there isn't an active transport yet, and an RPC is assigned to the Subchannel, it will
* create a new transport. It won't actively create transports otherwise. {@link
* #requestConnection requestConnection()} can be used to ask Subchannel to create a transport if
* there isn't any.
*
* {@link #start} must be called prior to calling any other methods, with the exception of
* {@link #shutdown}, which can be called at any time.
*
* @since 1.2.0
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public abstract static class Subchannel {
/**
* Starts the Subchannel. Can only be called once.
*
* Must be called prior to any other method on this class, except for {@link #shutdown} which
* may be called at any time.
*
* Must be called from the {@link Helper#getSynchronizationContext Synchronization Context},
* otherwise it may throw. See
* #5015 for more discussions.
*
* @param listener receives state updates for this Subchannel.
*/
public void start(SubchannelStateListener listener) {
throw new UnsupportedOperationException("Not implemented");
}
/**
* Shuts down the Subchannel. After this method is called, this Subchannel should no longer
* be returned by the latest {@link SubchannelPicker picker}, and can be safely discarded.
*
* Calling it on an already shut-down Subchannel has no effect.
*
* It should be called from the Synchronization Context. Currently will log a warning if
* violated. It will become an exception eventually. See #5015 for the background.
*
* @since 1.2.0
*/
public abstract void shutdown();
/**
* Asks the Subchannel to create a connection (aka transport), if there isn't an active one.
*
* It should be called from the Synchronization Context. Currently will log a warning if
* violated. It will become an exception eventually. See #5015 for the background.
*
* @since 1.2.0
*/
public abstract void requestConnection();
/**
* Returns the addresses that this Subchannel is bound to. This can be called only if
* the Subchannel has only one {@link EquivalentAddressGroup}. Under the hood it calls
* {@link #getAllAddresses}.
*
* It should be called from the Synchronization Context. Currently will log a warning if
* violated. It will become an exception eventually. See #5015 for the background.
*
* @throws IllegalStateException if this subchannel has more than one EquivalentAddressGroup.
* Use {@link #getAllAddresses} instead
* @since 1.2.0
*/
public final EquivalentAddressGroup getAddresses() {
List It should be called from the Synchronization Context. Currently will log a warning if
* violated. It will become an exception eventually. See #5015 for the background.
*
* @since 1.14.0
*/
public List RPCs made on this Channel is not counted when determining ManagedChannel's {@link
* ManagedChannelBuilder#idleTimeout idle mode}. In other words, they won't prevent
* ManagedChannel from entering idle mode.
*
* Warning: RPCs made on this channel will prevent a shut-down transport from terminating. If
* you make long-running RPCs, you need to make sure they will finish in time after the
* Subchannel has transitioned away from {@code READY} state
* (notified through {@link #handleSubchannelState}).
*
* Warning: this is INTERNAL API, is not supposed to be used by external users, and may
* change without notice. If you think you must use it, please file an issue.
*/
@Internal
public Channel asChannel() {
throw new UnsupportedOperationException();
}
/**
* Returns a {@link ChannelLogger} for this Subchannel.
*
* @since 1.17.0
*/
public ChannelLogger getChannelLogger() {
throw new UnsupportedOperationException();
}
/**
* Replaces the existing addresses used with this {@code Subchannel}. If the new and old
* addresses overlap, the Subchannel can continue using an existing connection.
*
* It must be called from the Synchronization Context or will throw.
*
* @throws IllegalArgumentException if {@code addrs} is empty
* @since 1.22.0
*/
public void updateAddresses(List Warning: this is INTERNAL API, is not supposed to be used by external users, and may
* change without notice. If you think you must use it, please file an issue and we can consider
* removing its "internal" status.
*/
@Internal
public Object getInternalSubchannel() {
throw new UnsupportedOperationException();
}
}
/**
* Receives state changes for one {@link Subchannel}. All methods are run under {@link
* Helper#getSynchronizationContext}.
*
* @since 1.22.0
*/
public interface SubchannelStateListener {
/**
* Handles a state change on a Subchannel.
*
* The initial state of a Subchannel is IDLE. You won't get a notification for the initial
* IDLE state.
*
* If the new state is not SHUTDOWN, this method should create a new picker and call {@link
* Helper#updateBalancingState Helper.updateBalancingState()}. Failing to do so may result in
* unnecessary delays of RPCs. Please refer to {@link PickResult#withSubchannel
* PickResult.withSubchannel()}'s javadoc for more information.
*
* When a subchannel's state is IDLE or TRANSIENT_FAILURE and the address for the subchannel
* was received in {@link LoadBalancer#handleResolvedAddresses}, load balancers should call
* {@link Helper#refreshNameResolution} to inform polling name resolvers that it is an
* appropriate time to refresh the addresses. Without the refresh, changes to the addresses may
* never be detected.
*
* SHUTDOWN can only happen in two cases. One is that LoadBalancer called {@link
* Subchannel#shutdown} earlier, thus it should have already discarded this Subchannel. The
* other is that Channel is doing a {@link ManagedChannel#shutdownNow forced shutdown} or has
* already terminated, thus there won't be further requests to LoadBalancer. Therefore, the
* LoadBalancer usually don't need to react to a SHUTDOWN state.
*
* @param newState the new state
* @since 1.22.0
*/
void onSubchannelState(ConnectivityStateInfo newState);
}
/**
* Factory to create {@link LoadBalancer} instance.
*
* @since 1.2.0
*/
@ThreadSafe
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public abstract static class Factory {
/**
* Creates a {@link LoadBalancer} that will be used inside a channel.
*
* @since 1.2.0
*/
public abstract LoadBalancer newLoadBalancer(Helper helper);
}
}
*
*
* @since 1.2.0
*/
@Immutable
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public static final class PickResult {
private static final PickResult NO_RESULT = new PickResult(null, null, Status.OK, false);
@Nullable private final Subchannel subchannel;
@Nullable private final ClientStreamTracer.Factory streamTracerFactory;
// An error to be propagated to the application if subchannel == null
// Or OK if there is no error.
// subchannel being null and error being OK means RPC needs to wait
private final Status status;
// True if the result is created by withDrop()
private final boolean drop;
private PickResult(
@Nullable Subchannel subchannel, @Nullable ClientStreamTracer.Factory streamTracerFactory,
Status status, boolean drop) {
this.subchannel = subchannel;
this.streamTracerFactory = streamTracerFactory;
this.status = checkNotNull(status, "status");
this.drop = drop;
}
/**
* A decision to proceed the RPC on a Subchannel.
*
*
*
*
*
*
*
*
*
*
* @param subchannel the picked Subchannel. It must have been {@link Subchannel#start started}
* @param streamTracerFactory if not null, will be used to trace the activities of the stream
* created as a result of this pick. Note it's possible that no
* stream is created at all in some cases.
* @since 1.3.0
*/
public static PickResult withSubchannel(
Subchannel subchannel, @Nullable ClientStreamTracer.Factory streamTracerFactory) {
return new PickResult(
checkNotNull(subchannel, "subchannel"), streamTracerFactory, Status.OK,
false);
}
/**
* Equivalent to {@code withSubchannel(subchannel, null)}.
*
* @since 1.2.0
*/
public static PickResult withSubchannel(Subchannel subchannel) {
return withSubchannel(subchannel, null);
}
/**
* A decision to report a connectivity error to the RPC. If the RPC is {@link
* CallOptions#withWaitForReady wait-for-ready}, it will stay buffered. Otherwise, it will fail
* with the given error.
*
* @param error the error status. Must not be OK.
* @since 1.2.0
*/
public static PickResult withError(Status error) {
Preconditions.checkArgument(!error.isOk(), "error status shouldn't be OK");
return new PickResult(null, null, error, false);
}
/**
* A decision to fail an RPC immediately. This is a final decision and will ignore retry
* policy.
*
* @param status the status with which the RPC will fail. Must not be OK.
* @since 1.8.0
*/
public static PickResult withDrop(Status status) {
Preconditions.checkArgument(!status.isOk(), "drop status shouldn't be OK");
return new PickResult(null, null, status, true);
}
/**
* No decision could be made. The RPC will stay buffered.
*
* @since 1.2.0
*/
public static PickResult withNoResult() {
return NO_RESULT;
}
/**
* The Subchannel if this result was created by {@link #withSubchannel withSubchannel()}, or
* null otherwise.
*
* @since 1.2.0
*/
@Nullable
public Subchannel getSubchannel() {
return subchannel;
}
/**
* The stream tracer factory this result was created with.
*
* @since 1.3.0
*/
@Nullable
public ClientStreamTracer.Factory getStreamTracerFactory() {
return streamTracerFactory;
}
/**
* The status associated with this result. Non-{@code OK} if created with {@link #withError
* withError}, or {@code OK} otherwise.
*
* @since 1.2.0
*/
public Status getStatus() {
return status;
}
/**
* Returns {@code true} if this result was created by {@link #withDrop withDrop()}.
*
* @since 1.8.0
*/
public boolean isDrop() {
return drop;
}
@Override
public String toString() {
return MoreObjects.toStringHelper(this)
.add("subchannel", subchannel)
.add("streamTracerFactory", streamTracerFactory)
.add("status", status)
.add("drop", drop)
.toString();
}
@Override
public int hashCode() {
return Objects.hashCode(subchannel, status, streamTracerFactory, drop);
}
/**
* Returns true if the {@link Subchannel}, {@link Status}, and
* {@link ClientStreamTracer.Factory} all match.
*/
@Override
public boolean equals(Object other) {
if (!(other instanceof PickResult)) {
return false;
}
PickResult that = (PickResult) other;
return Objects.equal(subchannel, that.subchannel) && Objects.equal(status, that.status)
&& Objects.equal(streamTracerFactory, that.streamTracerFactory)
&& drop == that.drop;
}
}
/**
* Arguments for creating a {@link Subchannel}.
*
* @since 1.22.0
*/
@ExperimentalApi("https://github.com/grpc/grpc-java/issues/1771")
public static final class CreateSubchannelArgs {
private final List
*
*
*
*