pub struct QueryRows { /* private fields */ }Expand description
A lazy, resumable cursor over the quads of an open .rete — the engine
side of for await (const [s, p, o, g] of graph.dump()) in the JS client.
§Why a cursor and not a callback
Rete::dump_each already streams in constant memory, but a Rust callback
cannot be paused to hand control back to JavaScript: to feed a JS iterator
it would have to buffer every quad first, which is exactly the Vec that
Rete::dump builds and that OOMs on a large file. This wraps
Rete::query_batch instead, so the scan can be suspended between calls and
resumed in place — the whole resume state is one opaque u64, never a
whole-graph materialization anywhere in the pipeline.
§Why batched (and not one call per quad)
Each wasm→JS call costs far more than decoding a triple, and every returned
String becomes a fresh JS string. Pulling one quad per call would make the
boundary the bottleneck; pulling all of them would reintroduce the Vec.
So the JS wrapper asks for DUMP_BATCH quads at a time and yields them one
by one — bounded, amortized, and lazy. Memory is O(batch), not O(graph).
§Cost model
The dictionary is not prefetched whole: each batch faults only the chunks its own terms live in, so taking five quads off the front costs five quads’ worth of dictionary rather than all of it. Index tiles fault in as the scan advances and stay resident, and so do dictionary chunks, so an unfiltered dump driven to the end still ends up fetching essentially the whole file — that is inherent in exporting a graph.
A filtered cursor (a bound s / p / o) is a different shape: the
scan routes to the one permutation that sorts on the bound prefix and drops
every tile whose synopsis proves it cannot match, from the tile directory,
without fetching it. On cordis.rete (801 MB, six named graphs) dumping
one predicate of one graph reads 16 MB where the unfiltered dump of that
graph reads 376 MB. Peak memory is O(faulted dictionary + index), never
O(quads), either way.
A SELECT answered to dictionary ids: the result handle of
Graph::query_rows and RemoteGraph::query_rows.
query evaluates a SELECT and decodes every term of every row to JSON; a
page of 25 rows of a 650,000-row result still costs the whole result’s
dictionary chunks (181 MB for a class of the nanopub dump). The handle keeps
the rows as ids, with the exact row count, and page(offset, limit) decodes
one window, as one batch of chunk reads. What the query itself needs is
decoded as before (the keys of an ORDER BY, the inputs of a FILTER); the
other columns wait for the page. free() it when done.
Implementations§
Source§impl QueryRows
impl QueryRows
Sourcepub fn vars(&self) -> String
pub fn vars(&self) -> String
The projected variables, a JSON array (for SELECT *, the variables
some row binds, in name order).
Sourcepub fn page(&self, offset: usize, limit: usize) -> Result<String, JsValue>
pub fn page(&self, offset: usize, limit: usize) -> Result<String, JsValue>
Rows offset..offset + limit (cut at the end), in the envelope of
query(..., "json"): {"kind":"select","vars":[…],"rows":[{var:term}]}
plus "offset" and "total". Calling it again, or for another window,
reuses the dictionary chunks already read.
Trait Implementations§
Source§impl FromWasmAbi for QueryRows
impl FromWasmAbi for QueryRows
Source§impl IntoWasmAbi for QueryRows
impl IntoWasmAbi for QueryRows
Source§impl LongRefFromWasmAbi for QueryRows
impl LongRefFromWasmAbi for QueryRows
Source§impl OptionFromWasmAbi for QueryRows
impl OptionFromWasmAbi for QueryRows
Source§impl OptionIntoWasmAbi for QueryRows
impl OptionIntoWasmAbi for QueryRows
Source§impl RefFromWasmAbi for QueryRows
impl RefFromWasmAbi for QueryRows
Source§type Abi = WasmPtr<WasmRefCell<QueryRows>>
type Abi = WasmPtr<WasmRefCell<QueryRows>>
Self are recovered from.Source§impl RefMutFromWasmAbi for QueryRows
impl RefMutFromWasmAbi for QueryRows
impl SupportsConstructor for QueryRows
impl SupportsInstanceProperty for QueryRows
impl SupportsStaticProperty for QueryRows
Source§impl TryFromJsValue for QueryRows
impl TryFromJsValue for QueryRows
Source§impl VectorFromWasmAbi for QueryRows
impl VectorFromWasmAbi for QueryRows
Source§impl VectorIntoWasmAbi for QueryRows
impl VectorIntoWasmAbi for QueryRows
Auto Trait Implementations§
impl !RefUnwindSafe for QueryRows
impl !Send for QueryRows
impl !Sync for QueryRows
impl !UnwindSafe for QueryRows
impl Forget for QueryRows
impl Freeze for QueryRows
impl Move for QueryRows
impl Unpin for QueryRows
impl UnsafeUnpin for QueryRows
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
impl<ST, DT> CastableFrom<ST, Initialized, Initialized> for DT
impl<ST, DT> CastableFrom<ST, Uninit, Uninit> for DT
impl<T> Read<Exclusive, BecauseExclusive> for Twhere
T: ?Sized,
Source§impl<T> ReturnWasmAbi for Twhere
T: IntoWasmAbi,
impl<T> ReturnWasmAbi for Twhere
T: IntoWasmAbi,
Source§type Abi = <T as IntoWasmAbi>::Abi
type Abi = <T as IntoWasmAbi>::Abi
IntoWasmAbi::AbiSource§fn return_abi(self) -> <T as ReturnWasmAbi>::Abi
fn return_abi(self) -> <T as ReturnWasmAbi>::Abi
IntoWasmAbi::into_abi, except that it may throw and never
return in the case of Err.