ShardedDaemonProcess push mode
#7229
Conversation
|
Thanks for fixing my dumb mistake in the test suite @Arkatufus - much appreciated lol |
| it is recommended to limit the nodes the sharded daemon process will run on using a role. | ||
| This cluster tool is intended for small numbers of consumers and will not scale well to a large set. In large clusters it is recommended to limit the nodes the sharded daemon process will run on using a role. | ||
|
|
||
| ## Push-Based Communication |
There was a problem hiding this comment.
Documented the two new pieces of functionality introduced here, which I'll review further down in the code itself
| @@ -126,7 +132,9 @@ public void ShardedDaemonProcess_must_not_run_if_the_role_does_not_match_node_ro | |||
|
|
|||
| var probe = CreateTestProbe(); | |||
| var settings = ShardedDaemonProcessSettings.Create(Sys).WithRole("workers"); | |||
| ShardedDaemonProcess.Get(Sys).Init("roles", 3, id => MyActor.Props(id, probe.Ref), settings, PoisonPill.Instance); | |||
| var actorRef = ShardedDaemonProcess.Get(Sys).Init("roles", 3, id => MyActor.Props(id, probe.Ref), settings, | |||
There was a problem hiding this comment.
Assert that we get a null IActorRef back here.
| @@ -5,34 +5,41 @@ | |||
| // </copyright> | |||
| //----------------------------------------------------------------------- | |||
|
|
|||
| #nullable enable | |||
There was a problem hiding this comment.
Every time I make API updates to an existing area of Akka.NET I always try to enable nullable so it'll be less jarring when we make it the default in the future.
| } | ||
|
|
||
| public string Name { get; } | ||
| public string[] Identities { get; } | ||
| public IActorRef ShardingRef { get; } | ||
| public ShardedDaemonProcessSettings Settings { get; } | ||
|
|
||
| public ITimerScheduler Timers { get; set; } | ||
| public ITimerScheduler Timers { get; set; } = null!; // gets set by Akka.NET |
There was a problem hiding this comment.
Made compiler warning go away.
| /// <see cref="ShardRegion"/> mechanism in order to piggyback off of its reliability and message buffering | ||
| /// features, which <see cref="Router"/> and <see cref="ActorSelection"/> do not support. | ||
| /// </remarks> | ||
| internal sealed class DaemonMessageRouter : UntypedActor |
There was a problem hiding this comment.
new class - DaemonMessageRouter. Performs round-robin message routing (according to the sending node - we don't care what other nodes who might be sending messages are doing) to all of the individual worker actors.
|
|
||
| // have to remember to always allow the sharding envelope to be forwarded | ||
| _shardingRef.Forward(new ShardingEnvelope(nextId, message)); | ||
| if (_index == int.MaxValue) _index = 0; |
There was a problem hiding this comment.
Wrap-around if we're about to integer overflow.
| public void Init(string name, int numberOfInstances, Func<int, Props> propsFactory) => | ||
| /// <returns>A reference to a router actor that will distribute all messages evenly across the workers | ||
| /// using round-robin message routing. <c>null</c> if the ShardedDaemonProcess is misconfigured.</returns> | ||
| public IActorRef? Init(string name, int numberOfInstances, Func<int, Props> propsFactory) => |
There was a problem hiding this comment.
Modified all of these methods to return an IActorRef? - will return null if we're the wrong role type to be hosting a worker and we'll also now log a WARNING accordingly if that happens.
| { | ||
| var sharding = ClusterSharding.Get(_system); | ||
| var shardingRef = sharding.Start( | ||
| typeName: $"sharded-daemon-process-{name}", | ||
| typeName: FormatWorkerProcessName(name), |
There was a problem hiding this comment.
Use a shared function to name ShardedDaemonProcesses and proxies - uses the exact same lexical structure as before so it won't break any existing deployments.
| /// <param name="role">The role where the worker actors are hosted.</param> | ||
| /// <returns>A reference to a router actor that will distribute all messages evenly across the workers | ||
| /// using round-robin message routing.</returns> | ||
| public IActorRef InitProxy(string name, int numberOfInstances, string role) |
There was a problem hiding this comment.
This is new - since we added push-based communication to interact with ShardedDaemonProcesses I thought it also made sense to leverage ShardRegionProxy to allow non-hosting nodes to communicate with the worker processes. The only issue here is that if the user gets the number of entities wrong, then our routing logic will be skewed or broken. Not too different than if the user gets the message extractor, role, or entity type wrong so it's not a big deal IMHO.
| @@ -4,7 +4,7 @@ | |||
| // Copyright (C) 2013-2023 .NET Foundation <https://github.com/akkadotnet/akka.net> | |||
| // </copyright> | |||
| //----------------------------------------------------------------------- | |||
|
|
|||
| #nullable enable | |||
There was a problem hiding this comment.
Enable nullable on the settings class too
Changes
close #7195 - adds "push" mode to communicating with sharded daemon processes.
Checklist
For significant changes, please ensure that the following have been completed (delete if not relevant):
ShardedDaemonProcesss #7195