mule/peer

The leadership read surface, mirroring Oban.Peer. A record of closures, like engine.Engine and notifier.Notifier, so each backend swaps its own implementation in at startup: the Postgres backend wires these to the lease-election peer actor’s cell + a database read, and the in-memory / single-node backend uses always_leader.

The leader-only gates (cron, pruner, the stager) call is_leader on their hot path, so it must be a pure cell read — never a database round-trip, never a blocking call into the actor (which would panic on restart; see peer/cell). get_leader mirrors Oban.Peer.get_leader, which does query the database, and is never called from a gate.

Types

Peer

</>

pub type Peer {
  Peer(
    is_leader: fn() -> Bool,
    get_leader: fn() -> option.Option(String),
  )
}

Constructors

```
Peer(
  is_leader: fn() -> Bool,
  get_leader: fn() -> option.Option(String),
)
```
Arguments

is_leader

Is this node the leader right now? A pure in-memory cell read: never touches the database, never blocks, never panics; False when the cell has no entry (no election has run). Mirrors Oban.Peer.leader?.

get_leader

The current leader’s node string, or None. Mirrors Oban.Peer.get_leader, which queries the database — so this may do a DB round-trip and must not be used on any gate’s hot path.

Values

always_leader

</>

pub fn always_leader(node node: String) -> Peer

The single-node / in-memory backend: always leader, backed by no actor and no table. Used directly by mule.start for the in-memory engine.

put_is_leader

</>

pub fn put_is_leader(
  instance_name instance_name: String,
  is_leader is_leader: Bool,
) -> Nil

Cache this node’s leadership for instance_name. Called by the peer actor on every election so the gates always see fresh state.

read_is_leader

</>

pub fn read_is_leader(
  instance_name instance_name: String,
) -> Bool

Read this node’s cached leadership for instance_name (False when no election has populated the cell). The total read the is_leader closure is built from.