Add ClusterClient initial contact discovery feature #7260
Conversation
| 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 | ||
| } |
There was a problem hiding this comment.
New discovery options
|
|
||
| discovery | ||
| { | ||
| method = <method> |
There was a problem hiding this comment.
If method is null, empty, whitespace, or "<method>" (not set), we'll use the default akka.discovery.method value. If that value also not set, we fall back to "config" method.
| discovery | ||
| { | ||
| method = <method> | ||
| actor-system-name = null |
There was a problem hiding this comment.
if "actor-system-name" is null or whitespace, we fall back to ActorSystem.Name
There was a problem hiding this comment.
Do we want to log a warning when this happens? Kind of seems like the user should have to specify this value given that these are two separate applications communicating typically.
There was a problem hiding this comment.
Sounds like a good idea, I'll do that
| _log.Debug($"Config discovery serving: {string.Join(", ", _resolvedServices.Values)}"); | ||
| } | ||
|
|
||
| public bool TryRemoveEndpoint(string serviceName, ResolvedTarget target) |
There was a problem hiding this comment.
Added this so we can add and remove new discovery entry during runtime, used for MNTR testing.
| return true; | ||
| } | ||
|
|
||
| public bool TryAddEndpoint(string serviceName, ResolvedTarget target) |
There was a problem hiding this comment.
Added this so we can add and remove new discovery entry during runtime, used for MNTR testing.
| @@ -419,7 +607,8 @@ private bool Establishing(object message) | |||
| } | |||
| else | |||
| { | |||
| // ok, use another instead | |||
| // prune out actors that failed to be identified | |||
| PruneContacts((string) actorIdentify.MessageId); | |||
There was a problem hiding this comment.
Dead contact detection, remove the dead contact if it doesn't respond to Identify
| @@ -390,16 +578,16 @@ private bool Establishing(object message) | |||
|
|
|||
| switch (message) | |||
| { | |||
| case DeadLetter { Message: ClusterReceptionist.GetContacts } dl: | |||
| PruneContacts(dl.Recipient.Path.Address.ToString()); | |||
There was a problem hiding this comment.
Dead contact detection, remove contact from contact list if GetContacts were redirected to DeadLetters
| case ActorIdentity actorIdentify: | ||
| // prune out actors that failed to be identified | ||
| if (actorIdentify.Subject is null) | ||
| PruneContacts((string) actorIdentify.MessageId); |
There was a problem hiding this comment.
Continue removal of dead contacts if it doesn't respond to Identify
| return true; | ||
|
|
||
| case ClusterReceptionist.ReceptionistShutdown: | ||
| { | ||
| if (receptionist.Equals(Sender)) | ||
| { | ||
| _log.Info("Receptionist [{0}] is shutting down, reestablishing connection", receptionist); | ||
| PruneContacts(Sender.Path.Address.ToString()); |
There was a problem hiding this comment.
Remove the receptionist from the contact list if it is shut down
| if(_initialContactsSelections.Count == 0 && _contacts.Count == 0) | ||
| Rediscover(); |
There was a problem hiding this comment.
Switch to discovery state if we ran out of contacts in the contact list
...ntrib/cluster/Akka.Cluster.Tools.Tests.MultiNode/ClusterClient/ClusterClientDiscoverySpec.cs
Outdated
Show resolved
Hide resolved
src/contrib/cluster/Akka.Cluster.Tools.Tests/ClusterClient/ClusterClientConfigSpec.cs
Show resolved
Hide resolved
| return true; | ||
|
|
||
| case ClusterReceptionist.Contacts: | ||
| // ignored, we're re-discovering |
There was a problem hiding this comment.
Why would we ignore this - does this message contain any valid contacts?
There was a problem hiding this comment.
I guess we could incorporate the result with the possible future discovery result, but this is risky because we're already inside the discovery phase. If there's suddenly a Contacts message coming when we're in this state, that means something is seriously wrong with the network / configuration and there could be a possible split-brain happening
There was a problem hiding this comment.
We don't care about split brains in this scenario though - the client just needs 1 node to talk to.
There was a problem hiding this comment.
As long as someone is answering the phone, that's good enough.
| discovery | ||
| { | ||
| method = <method> | ||
| actor-system-name = null |
There was a problem hiding this comment.
Do we want to log a warning when this happens? Kind of seems like the user should have to specify this value given that these are two separate applications communicating typically.
|
Superceeded by #7261 |
Fixes #7243
Changes
Add ClusterClient initial contact discovery feature