Struct Src 

pub(crate) struct Src<'a> {
    pub node: u32,
    pub seq: u64,
    pub min_ts: i64,
    pub max_ts: i64,
    pub dir: Option<&'a Path>,
    tables: Option<&'a [(&'static str, RecordBatch)]>,
}

One candidate the read path may have to open: a published block directory, or the snapshot of a block that is still open (section 4).

Everything a scan needs before it opens anything — the range to prune on, the identity a cursor is built from, and where the tables are. The two cases differ in exactly two ways: a snapshot has no directory to read sidecars out of, and its tables are already in memory.

Fields

node: u32

seq: u64

min_ts: i64

max_ts: i64

dir: Option<&'a Path>

None for a snapshot. Skipping the sidecar probes is not a special case: “no filter file” already means “scan me” for every block published before that filter existed.

tables: Option<&'a [(&'static str, RecordBatch)]>

Implementations

impl<'a> Src<'a>

pub fn disk(b: &'a BlockRef) -> Src<'a>

pub fn open(o: &'a Open) -> Src<'a>

pub fn overlaps(&self, from: i64, to: i64) -> bool

pub fn load(&self, name: &str) -> Result<Option<RecordBatch>>

One named table, or None if this source does not have it.

A missing table on disk is not an error: retention unlinking an expired block underneath a scan is normal. An open block’s tables are already Arrow, under the same names the published files would carry — including the empty ones, which publish skips and which read identically to absent everywhere downstream.