Charmed Zookeeper

ZooKeeperCluster class and methods.

ZooKeeperCluster is a general-purpose handler for managing the ZooKeeper peer-relation data, and scale-up/down orchestration.

Exposed attributes include a reference to the cluster peer relation for ZooKeeper, as well as wrappers for grabbing units that have started the ZooKeeper service, found from the peer unit data. Exposed methods include the setting of passwords for the cluster, ensuring units start up in increasing order with the correct servers config, and handling the adding/removing of members from the ZooKeeper quorum.

Instances of ZooKeeperCluster are to be created during the __init__ for the target Charm. or from another library. It does not set any relation data itself, and as such is reliant on the calling Charm to set required relation data where necessary during life-cycle events. It maintains a status attribute for passing success/failure of called methods back to the calling Charm.

Example minimal usage for ZooKeeperCluster:

class ZooKeeperCharm(CharmBase):
    def __init__(self, *args):
        super().__init__(*args)
        self.cluster =  ZooKeeperCluster(self)

        self.framework.observe(getattr(self.on, "install"), self._on_install)
        self.framework.observe(getattr(self.on, "start"), self._on_start)
        self.framework.observe(
            getattr(self.on, "leader_elected"), self._on_cluster_relation_updated
        )
        self.framework.observe(
            getattr(self.on, "cluster_relation_changed"), self._on_cluster_relation_updated
        )
        self.framework.observe(
            getattr(self.on, "cluster_relation_joined"), self._on_cluster_relation_updated
        )
        self.framework.observe(
            getattr(self.on, "cluster_relation_departed"), self._on_cluster_relation_updated
        )

    def _on_install(self, event):
        unit_myid = self.cluster.get_unit_id(self.unit) + 1 

        # write_unit_id(id=unit_myid)

    def _on_start(self, event):
        # setting cluster passwords during first deployment
        if self.unit.is_leader():
            self.cluster.relation.data[self.app].update({"super_password": self.cluster.generate_password()})
            self.cluster.relation.data[self.app].update({"sync_password": self.cluster.generate_password()})

        # checking passwords have been set by leader on other units, in case they started first
        if not self.cluster.passwords_set():
            event.defer()
            return
        
        # units must be started in increasing unit.id order
        try:
            servers, unit_config = self.cluster.ready_to_start(self.unit)
        except (NotUnitTurnError, UnitNotFoundError, NoPasswordError) as e:
            self.unit.status = self.cluster.status
            event.defer()
            return

        # start_zookeeper_service()


        # set useful metadata for each unit
        self.cluster.relation.data[self.unit].update(unit_config)
        
        # NECESSARY - set data so that subsequent `self.cluster` commands pick up this unit as a 'started' one
        self.cluster.relation.data[self.unit].update({"state": "started"})
    
    def _on_cluster_relation_updated(self, event):
        if not self.unit.is_leader():
            return

        # units need to exist in the app data to be iterated through for next_turn
        for unit in self.cluster.started_units:
            unit_id = self.cluster.get_unit_id(unit)
            current_value = self.cluster.relation.data[self.app].get(str(unit_id), None)

            # sets to "added" for init quorum leader, if not already exists
            # may already exist if during the case of a failover of unit 0
            if unit_id == 0:
                self.cluster.relation.data[self.app].update(
                    {str(unit_id): current_value or "added"}
                )

        if not self.cluster.passwords_set:
            event.defer()
            return

        # adds + removes members for all self-confirmed started units
        updated_servers = self.cluster.update_cluster()

        # either Active if successful, else Maintenance
        self.unit.status = self.cluster.status

        if self.cluster.status == ActiveStatus():
            self.cluster.relation.data[self.app].update(updated_servers)
        else:
            # in the event some unit wasn't started/ready
            event.defer()
            return