ShardLifecycle

Remarks

Requested → Starting → Ready → Draining → Stopping → Stopped, with Failed and Lost off the side. Each transition is one method, and the ones that are not edges are refused — a shard going Stopped → Ready would be a process the cluster had already written off answering a heartbeat, which is a real thing that happens when a supervisor restarts something it should not have.

⚠ Draining is one-way, and that is what makes a rollout finish. Doc 27 § Upgrades drains old-version shards and waits; a shard that could be talked out of draining would make the rollout's completion a race rather than an invariant.

Fields and properties (2)

• public ShardId Shard

  Which shard.

• public ShardState State

  Where it is in its life.

Methods (10)

• public ShardLifecycle(ShardId shard, HealthOptions? options = null)

  Stands one up.

• public void Requested(ShardKey shardKey, ShardCapacity shardCapacity)

  A decision that this shard should exist.

• public void Starting(RealmInstanceId realmInstance, RealmEndpoint at)

  The placement backend created something.

• public void Ready(RealmEndpoint at)

  The realm loaded its map and is accepting sessions.

• public ShardState Heartbeat(ShardHeartbeat sample)

  Records a heartbeat and answers with the state the realm should be in.

• public void Drain()

  Stop taking arrivals and move everyone out at safe moments.

• public void Stopped()

  It ended the way it was asked to.

• public void Lost()

  It went away without being asked.

• public bool Expire()

  Declares a silent shard lost, if it has been silent long enough.

• public ShardReport Report()

  What this shard is, for anybody who asks.

Used by (5)

• FakeClusterVixen.Live.Realm.Cluster.Tests
• FakeShardVixen.Live.Realm.Cluster.Tests
• RealmClusterTestsVixen.Live.Realm.Cluster.Tests
• ShardGrainVixen.Live.Orchestrator
• ShardLifecycleTestsVixen.Live.Orchestrator.Tests