stuhood commented on code in PR #161:
URL: https://github.com/apache/datafusion-site/pull/161#discussion_r2969858243
##########
content/blog/2026-03-20-writing-table-providers.md:
##########
@@ -0,0 +1,651 @@
+---
+layout: post
+title: Writing Custom Table Providers in Apache DataFusion
+date: 2026-03-20
+author: Tim Saucer (rerun.io)
+categories: [tutorial]
+---
+<!--
+{% comment %}
+Licensed to the Apache Software Foundation (ASF) under one or more
+contributor license agreements. See the NOTICE file distributed with
+this work for additional information regarding copyright ownership.
+The ASF licenses this file to you under the Apache License, Version 2.0
+(the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+{% endcomment %}
+-->
+
+[TOC]
+
+One of DataFusion's greatest strengths is its extensibility. If your data lives
+in a custom format, behind an API, or in a system that DataFusion does not
+natively support, you can teach DataFusion to read it by implementing a
+**custom table provider**. This post walks through the three layers you need to
+understand and explains where your work should actually happen.
+
+## The Three Layers
+
+---
+
+When DataFusion executes a query against a table, three abstractions
collaborate
+to produce results:
+
+1. **[`TableProvider`]** -- Describes the table (schema, capabilities) and
+ produces an execution plan when queried.
+2. **[`ExecutionPlan`]** -- Describes *how* to compute the result:
partitioning,
+ ordering, and child plan relationships.
+3. **[`SendableRecordBatchStream`]** -- The async stream that *actually does
the
+ work*, yielding `RecordBatch`es one at a time.
+
+Think of these as a funnel: `TableProvider::scan()` is called once during
+planning to create an `ExecutionPlan`, then `ExecutionPlan::execute()` is
called
+once per partition to create a stream, and those streams are where rows are
+actually produced during execution.
Review Comment:
Would it be useful to know how these types relate to physical vs logical
planning?
##########
content/blog/2026-03-20-writing-table-providers.md:
##########
@@ -0,0 +1,651 @@
+---
+layout: post
+title: Writing Custom Table Providers in Apache DataFusion
+date: 2026-03-20
+author: Tim Saucer (rerun.io)
+categories: [tutorial]
+---
+<!--
+{% comment %}
+Licensed to the Apache Software Foundation (ASF) under one or more
+contributor license agreements. See the NOTICE file distributed with
+this work for additional information regarding copyright ownership.
+The ASF licenses this file to you under the Apache License, Version 2.0
+(the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+{% endcomment %}
+-->
+
+[TOC]
+
+One of DataFusion's greatest strengths is its extensibility. If your data lives
+in a custom format, behind an API, or in a system that DataFusion does not
+natively support, you can teach DataFusion to read it by implementing a
+**custom table provider**. This post walks through the three layers you need to
+understand and explains where your work should actually happen.
+
+## The Three Layers
+
+---
+
+When DataFusion executes a query against a table, three abstractions
collaborate
+to produce results:
+
+1. **[`TableProvider`]** -- Describes the table (schema, capabilities) and
+ produces an execution plan when queried.
+2. **[`ExecutionPlan`]** -- Describes *how* to compute the result:
partitioning,
+ ordering, and child plan relationships.
+3. **[`SendableRecordBatchStream`]** -- The async stream that *actually does
the
+ work*, yielding `RecordBatch`es one at a time.
+
+Think of these as a funnel: `TableProvider::scan()` is called once during
+planning to create an `ExecutionPlan`, then `ExecutionPlan::execute()` is
called
+once per partition to create a stream, and those streams are where rows are
+actually produced during execution.
+
+[`TableProvider`]:
https://docs.rs/datafusion/latest/datafusion/catalog/trait.TableProvider.html
+[`ExecutionPlan`]:
https://docs.rs/datafusion/latest/datafusion/physical_plan/trait.ExecutionPlan.html
+[`SendableRecordBatchStream`]:
https://docs.rs/datafusion/latest/datafusion/execution/type.SendableRecordBatchStream.html
+
+## Layer 1: TableProvider
+
+---
+
+A [`TableProvider`] represents a queryable data source. For a minimal read-only
+table, you need four methods:
+
+```rust
+impl TableProvider for MyTable {
+ fn as_any(&self) -> &dyn Any { self }
+
+ fn schema(&self) -> SchemaRef {
+ Arc::clone(&self.schema)
+ }
+
+ fn table_type(&self) -> TableType {
+ TableType::Base
+ }
+
+ async fn scan(
+ &self,
+ state: &dyn Session,
+ projection: Option<&Vec<usize>>,
+ filters: &[Expr],
+ limit: Option<usize>,
+ ) -> Result<Arc<dyn ExecutionPlan>> {
+ // Build and return an ExecutionPlan -- keep this lightweight!
+ Ok(Arc::new(MyExecPlan::new(
+ Arc::clone(&self.schema),
+ projection,
+ limit,
+ )))
+ }
+}
+```
+
+The `scan` method is the heart of `TableProvider`. It receives three pushdown
+hints from the optimizer, each reducing the amount of data your source needs
+to produce:
+
+- **`projection`** -- Which columns are needed. This reduces the **width** of
+ the output. If your source supports it, read only these columns rather than
+ the full schema.
+- **`filters`** -- Predicates the engine would like you to apply during the
+ scan. This reduces the **number of rows** by skipping data that does not
+ match. Implement `supports_filters_pushdown` to advertise which filters you
+ can handle.
+- **`limit`** -- A row count cap. This also reduces the **number of rows** --
+ if you can stop reading early once you have produced enough rows, this avoids
+ unnecessary work.
+
+### Keep `scan()` Lightweight
+
+This is a critical point: **`scan()` runs during planning, not execution.** It
+should return quickly. Best practices are to avoid performing I/O, network
+calls, or heavy computation here. The `scan` method's job is to *describe* how
+the data will be produced, not to produce it. All the real work belongs in the
+stream (Layer 3).
+
+A common pitfall is to fetch data or open connections in `scan()`. This blocks
+the planning thread and can cause timeouts or deadlocks, especially if the
query
+involves multiple tables or subqueries that all need to be planned before
+execution begins.
+
+### Existing Implementations to Learn From
+
+DataFusion ships several `TableProvider` implementations that are excellent
+references:
+
+- **[`MemTable`]** -- Holds data in memory as `Vec<RecordBatch>`. The simplest
+ possible provider; great for tests and small datasets.
+- **[`StreamTable`]** -- Wraps a user-provided stream factory. Useful when your
+ data arrives as a continuous stream (e.g., from Kafka or a socket).
+- **[`SortedTableProvider`]** -- Wraps another `TableProvider` and advertises a
+ known sort order, enabling the optimizer to skip redundant sorts.
+
+[`MemTable`]:
https://docs.rs/datafusion/latest/datafusion/datasource/memory/struct.MemTable.html
+[`StreamTable`]:
https://docs.rs/datafusion/latest/datafusion/datasource/stream/struct.StreamTable.html
+[`SortedTableProvider`]:
https://docs.rs/datafusion/latest/datafusion/datasource/struct.SortedTableProvider.html
+
+## Layer 2: ExecutionPlan
+
+---
+
+An [`ExecutionPlan`] is a node in the physical query plan tree. Your table
+provider's `scan()` method returns one. The required methods are:
+
+```rust
+impl ExecutionPlan for MyExecPlan {
+ fn name(&self) -> &str { "MyExecPlan" }
+
+ fn as_any(&self) -> &dyn Any { self }
+
+ fn properties(&self) -> &PlanProperties {
+ &self.properties
+ }
+
+ fn children(&self) -> Vec<&Arc<dyn ExecutionPlan>> {
+ vec![] // Leaf node -- no children
+ }
+
+ fn with_new_children(
+ self: Arc<Self>,
+ children: Vec<Arc<dyn ExecutionPlan>>,
+ ) -> Result<Arc<dyn ExecutionPlan>> {
+ assert!(children.is_empty());
+ Ok(self)
+ }
+
+ fn execute(
+ &self,
+ partition: usize,
+ context: Arc<TaskContext>,
+ ) -> Result<SendableRecordBatchStream> {
+ // This is where you build and return your stream
+ // ...
+ }
+}
+```
+
+The key properties to set correctly in [`PlanProperties`] are **output
+partitioning** and **output ordering**.
+
+**Output partitioning** tells the engine how many partitions your data has,
+which determines parallelism. If your source naturally partitions data (e.g.,
+by file or by shard), expose that here.
+
+**Output ordering** declares whether your data is naturally sorted. This
+enables the optimizer to avoid inserting a `SortExec` when a query requires
+ordered data. Getting this right can be a significant performance win.
+
+### Partitioning Strategies
+
+Since `execute()` is called once per partition, partitioning directly controls
+the parallelism of your table scan. Each partition runs on its own task, so
+more partitions means more concurrent work -- up to the number of available
+cores.
Review Comment:
It would probably be good to briefly explain the relationship between tasks
and threads. For example: is it ok to have a lot more tasks than cores, or
should you cap the number of partitions (and thus tasks) you expose?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
