Struct MetadataCache 

pub struct MetadataCache {
    pub applied_index: u64,
    pub last_applied_hlc: Hlc,
    pub leases: HashMap<(DescriptorId, u64), DescriptorLease>,
    pub topology_log: Vec<TopologyChange>,
    pub routing_log: Vec<RoutingChange>,
    pub cluster_version: u16,
    pub catalog_entries_applied: u64,
}

In-memory view of the committed metadata state.

Fields

applied_index: u64

last_applied_hlc: Hlc

leases: HashMap<(DescriptorId, u64), DescriptorLease>

(descriptor_id, node_id) -> lease.

topology_log: Vec<TopologyChange>

Topology mutations applied so far.

routing_log: Vec<RoutingChange>

cluster_version: u16

catalog_entries_applied: u64

Monotonically-increasing count of committed CatalogDdl entries. Exposed for tests and metrics — planners read catalog state through the host-side SystemCatalog, not this counter.

Implementations

impl MetadataCache

pub fn new() -> Self

pub fn advance_applied_index(&mut self, index: u64)

Advance the applied-index watermark past a committed Raft no-op (a leader-transition entry with no MetadataEntry payload).

A no-op mutates no cache state, but it IS applied, so the cache watermark must move in lockstep with the Raft applied index the tick loop reports. Otherwise the startup applied-index sanity check reads cache_applied < watcher_current and fails the boot with a spurious gap — which is exactly what happens on a fresh single-node start, where every group’s first committed entry is an election no-op at index 1. Monotonic: never regresses the watermark.

pub fn apply(&mut self, index: u64, entry: &MetadataEntry)

Apply a committed entry. Idempotent by applied_index: entries at or below the current watermark are ignored.

Trait Implementations

impl Debug for MetadataCache

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for MetadataCache

fn default() -> MetadataCache

Returns the “default value” for a type.