Add ClusterClient initial contact discovery feature #7261
Conversation
| #nullable enable | ||
| namespace Akka.Cluster.Tools.Client; | ||
|
|
||
| public class ClusterClientDiscovery: UntypedActor, IWithUnboundedStash, IWithTimers |
There was a problem hiding this comment.
All new discovery logic are isolated into this new actor
| return new RootActorPath(address) / "system" / _discoverySettings.ReceptionistName; | ||
| } | ||
|
|
||
| private static async Task<ResolveResult> ResolveContact(Contact contact, TimeSpan timeout, CancellationToken ct) |
There was a problem hiding this comment.
This method mimics ActorSelection.ResolveOne() but instead of throwing on failure, it wraps the result in an envelope instead.
| Self.Tell(DiscoverTick.Instance); | ||
| } | ||
|
|
||
| private ActorPath ResolvedTargetToReceptionistActorPath(ServiceDiscovery.ResolvedTarget target) |
There was a problem hiding this comment.
This method converts ServiceDiscovery.ResolvedTarget to remote ActorPath
| _targetActorSystemName = _discoverySettings.ActorSystemName!; | ||
| } | ||
|
|
||
| _transportProtocol = ((ExtendedActorSystem)Context.System).Provider.DefaultAddress.Protocol; |
There was a problem hiding this comment.
Best case guess on what transport protocol we're using right now. Should handle custom protocols, such as the one used in Remote.TestKit
| case DiscoverTick: | ||
| if(_verboseLogging && _log.IsDebugEnabled) | ||
| _log.Debug("Discovering initial contacts"); | ||
|
|
||
| _serviceDiscovery!.Lookup(_lookup, _discoveryTimeout) | ||
| .PipeTo(Self, Self, failure: cause => new DiscoveryFailure(cause)); |
There was a problem hiding this comment.
Actual Akka.Discovery is done here.
| if (terminated.ActorRef.Equals(clusterClient)) | ||
| { | ||
| if(_verboseLogging && _log.IsInfoEnabled) | ||
| _log.Info("Cluster client failed to reconnect to all receptionists, rediscovering."); | ||
|
|
||
| // Kickoff discovery lookup | ||
| Self.Tell(DiscoverTick.Instance); | ||
| Become(Discovering); | ||
| } |
There was a problem hiding this comment.
If cluster client dies, return to Discovering state
There was a problem hiding this comment.
So this proxying guarantees that the caller's code doesn't get disrupted if we can to kill / recreate the underlying ClusterClient - LGTM.
| use-initial-contacts-discovery = false | ||
|
|
||
| discovery | ||
| { | ||
| method = <method> | ||
| actor-system-name = null | ||
| receptionist-name = receptionist | ||
| service-name = null | ||
| port-name = null | ||
| discovery-retry-interval = 1s | ||
| discovery-timeout = 60s | ||
| } | ||
|
|
||
| verbose-logging = false |
There was a problem hiding this comment.
New discovery settings.
If method is null, empty, whitespace, or "" (not set), use the discovery method declared in "akka.discovery.method". If that also isn't set, use "config" method as default fallback. Warn the user when these happened.
If actor-system-name is null or whitespace, use the current actor system name instead. Warn the user when these happened.
| } | ||
|
|
||
| [InternalStableApi] | ||
| public bool TryRemoveEndpoint(string serviceName, ResolvedTarget target) |
There was a problem hiding this comment.
New Discovery.Config method for MNTR testing
| } | ||
|
|
||
| [InternalStableApi] | ||
| public bool TryAddEndpoint(string serviceName, ResolvedTarget target) |
There was a problem hiding this comment.
New Discovery.Config method for MNTR testing
| public ClusterClientSettings( | ||
| IImmutableSet<ActorPath> initialContacts, | ||
| TimeSpan establishingGetContactsInterval, | ||
| TimeSpan refreshContactsInterval, | ||
| TimeSpan heartbeatInterval, | ||
| TimeSpan acceptableHeartbeatPause, | ||
| int bufferSize, | ||
| bool useLegacySerialization, | ||
| bool useInitialContactsDiscovery, | ||
| ClusterClientDiscoverySettings? discoverySettings = null, | ||
| TimeSpan? reconnectTimeout = null, | ||
| bool verboseLogging = false) |
There was a problem hiding this comment.
New ClusterClientSettings ctor, backward compatible.
| akka.remote.dot-netty.tcp.hostname = localhost | ||
| akka.actor.provider = cluster | ||
| akka.remote.log-remote-lifecycle-events = off | ||
| akka.cluster.client { |
There was a problem hiding this comment.
Are we disabling heartbeats on purpose here? 1d is an eternity in the context of this spec.
| cluster.client { | ||
| heartbeat-interval = 1s | ||
| acceptable-heartbeat-pause = 2s | ||
| use-initial-contacts-discovery = true |
| discovery | ||
| { | ||
| method = config | ||
| config.services.test-cluster.endpoints = [] |
There was a problem hiding this comment.
I assume this gets populated later using the new internal methods you added for dynamically updating config-based discovery
| _discoveryService = | ||
| (ConfigServiceDiscovery)Discovery.Discovery.Get(Sys).LoadServiceDiscovery("config"); | ||
| var address = GetAddress(_config.First); | ||
| _discoveryService.TryAddEndpoint("test-cluster", new ServiceDiscovery.ResolvedTarget(address.Host, address.Port)); |
There was a problem hiding this comment.
Populates the discovery list here
| { | ||
| RunOn(() => | ||
| { | ||
| Cluster.Get(Sys).Leave(Node(_config.First).Address); |
There was a problem hiding this comment.
The entire cluster goes down here, gracefully
...ntrib/cluster/Akka.Cluster.Tools.Tests.MultiNode/ClusterClient/ClusterClientDiscoverySpec.cs
Show resolved
Hide resolved
| Self.Tell(DiscoverTick.Instance); | ||
| } | ||
|
|
||
| private ActorPath ResolvedTargetToReceptionistActorPath(ServiceDiscovery.ResolvedTarget target) |
| if (terminated.ActorRef.Equals(clusterClient)) | ||
| { | ||
| if(_verboseLogging && _log.IsInfoEnabled) | ||
| _log.Info("Cluster client failed to reconnect to all receptionists, rediscovering."); | ||
|
|
||
| // Kickoff discovery lookup | ||
| Self.Tell(DiscoverTick.Instance); | ||
| Become(Discovering); | ||
| } |
There was a problem hiding this comment.
So this proxying guarantees that the caller's code doesn't get disrupted if we can to kill / recreate the underlying ClusterClient - LGTM.
Fixes #7243
This PR is superior compared to #7260, no underlying logic inside ClusterClient actor were changed at all.
Changes
Checklist
For significant changes, please ensure that the following have been completed (delete if not relevant):