Struct OpenSlot 

pub struct OpenSlot {
    shards: Arc<[Shard]>,
}

Every shard of one signal’s open blocks, asked together.

The interesting half is OpenSlot::fresh, and what makes it fresh rather than merely recent is the order the queues already enforce. submit acknowledges an export only after the job is in some shard’s channel, so every acknowledged export is queued before a request issued after it — and if each shard answers only once its own channel is empty, the answers together necessarily contain them all. That is read-your-writes, for the price of FIFOs Mira was already paying, with no shared counter and no clock.

Sharding does not weaken it, because the argument never depended on there being one queue: an acknowledged export is in exactly one shard’s channel until that shard appends it. It does mean the read path has to ask all of them, which is what fresh does.

Fields

shards: Arc<[Shard]>

Empty for a slot nobody serves: fresh returns nothing, forever. That is exactly what a unit test that only wants an Api wants.

Implementations

impl OpenSlot

pub async fn fresh(&self) -> Vec<Arc<Open>>

Everything acknowledged before this call, as one readable block per shard that has anything open.

Falls back to a shard’s last snapshot when its flusher cannot be reached: the request queue is full, or the task is gone. Both are overload or shutdown, and a query that waits its turn behind an overloaded ingest path is a worse answer than one that is a few milliseconds stale.

Every ask goes out before any answer is awaited, so the shards work concurrently; awaiting them in turn would put a whole flusher’s backlog between one shard’s answer and the next one’s question.

Trait Implementations

impl Clone for OpenSlot

fn clone(&self) -> OpenSlot

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Default for OpenSlot

fn default() -> OpenSlot

Returns the “default value” for a type.