[
https://issues.apache.org/jira/browse/ARROW-10368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17219142#comment-17219142
]
Remi Dettai commented on ARROW-10368:
-------------------------------------
bq. I wonder if we could not go one step further and try to add a new logical
plan that makes it possible to add custom sources
This is definitively outside the scope of the original ticket, but here is what
I came up with for the custom source logical plan:
{code:java}
/// Produces rows from a custom user implementation
LogicalPlan::CustomScan {
/// A shared reference to the custom implementation
scanner: Arc<dyn CustomScanner>,
},
{code}
with:
{code:java}
#[async_trait]
/// A user implemented scanner that can be used by datafusion
pub trait CustomScanner: Send + Sync + fmt::Debug {
/// reference to the schema of the data as it will be read by this scanner
fn projected_schema(&self) -> &SchemaRef;
/// string display of this scanner
fn format(&self) -> &str;
/// apply projection on this scanner
fn project(
&self,
required_columns: &HashSet<String>,
has_projection: bool,
) -> Result<Arc<dyn CustomScanner>>;
/// get scanner partitioning
fn output_partitioning(&self) -> Partitioning;
/// get iterator for a given partition
async fn execute(&self, partition: usize) -> Result<Box<dyn RecordBatchReader
+ Send>>;
}
{code}
I am now wondering if we shouldn't make this the common interface for all
LogicalPlan::XxxScan rather than having them in the enum.
> [Rust][DataFusion] Make InMemoryScan work on iterators of RecordBatch
> ---------------------------------------------------------------------
>
> Key: ARROW-10368
> URL: https://issues.apache.org/jira/browse/ARROW-10368
> Project: Apache Arrow
> Issue Type: Improvement
> Components: Rust, Rust - DataFusion
> Reporter: Remi Dettai
> Priority: Major
>
> Currently, InMemoryScan takes a Vec<Vec<RecordBatch>> as data.
> - the outer Vec separates the partitions
> - the inner Vec contains all the RecordBatch for one partition
> The inner Vec is then converted into an iterator when the LogicalPlan is
> turned into a PhysicalPlan.
> I suggest that InMemoryScan should take Vec<Iter<RecordBatch>>. This would
> make it possible to plug custom Scan implementations into datafusion without
> the need to read them entirely into memory. It would still work pretty
> seamlessly with Vec<Vec<RecordBatch>> that would just need a to be converted
> with data.map(|x| x.iter()) first.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
