alamb commented on code in PR #66:
URL: https://github.com/apache/datafusion-site/pull/66#discussion_r2044440144
##########
content/blog/2025-04-04-datafusion-userdefined-window-functions.md:
##########
@@ -0,0 +1,346 @@
+---
+layout: post
+title: User defined Window Functions in DataFusion
+date: 2025-04-04
+author: Aditya Singh Rathore
+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 %}
+-->
+
+## Introduction
Review Comment:
I think e can forgo the initial heading as it ends up rendering strangely on
the blog page
```suggestion
```
##########
content/blog/2025-04-04-datafusion-userdefined-window-functions.md:
##########
@@ -0,0 +1,154 @@
+---
+layout: post
+title: User defined Window Functions in DataFusion
+date: 2025-04-04
+author: Aditya Singh Rathore
+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 %}
+-->
+
+## Introduction
+Window functions are a powerful feature in SQL, allowing for complex
analytical computations over a subset of data. However, efficiently
implementing them, especially sliding windows, can be quite challenging. With
DataFusion's recent support for user-defined window functions , developers now
have more flexibility and performance improvements at their disposal.
+
+In this post, we'll explore:
+
+- What window functions are and why they matter
+
+- Understanding sliding windows
+
+- The challenges of computing window aggregates efficiently
+
+- How DataFusion optimizes these computations
+
+- Alternative approaches and their trade-offs
+
+## Understanding Window Functions in SQL
+
+Imagine you're analyzing sales data and want insights without losing the finer
details. This is where **window functions** come into play. Unlike **GROUP
BY**, which condenses data, window functions let you retain each row while
performing calculations over a defined **range** —like having a moving lens
over your dataset.
+
+Picture a business tracking daily sales. They need a running total to
understand cumulative revenue trends without collapsing individual
transactions. SQL makes this easy:
+```sql
+SELECT id, value, SUM(value) OVER (ORDER BY id) AS running_total
+FROM sales;
+```
+This helps in analytical queries where we need cumulative sums, moving
averages, or ranking without losing individual records.
+
+
+## User Defined Window Functions
+
+Writing a user defined window function is slightly more complex than an
aggregate function due
+to the variety of ways that window functions are called. I recommend reviewing
the
+[online
documentation](https://datafusion.apache.org/library-user-guide/adding-udfs.html)
+for a description of which functions need to be implemented. The details of
how to implement
+these generally follow the same patterns as described above for aggregate
functions.
+
+## Understaing Sliding Window
+
+Sliding windows define a **moving range** of data over which aggregations are
computed. Unlike simple cumulative functions, these windows are dynamically
updated as new data arrives.
+
+For instance, if we want a 7-day moving average of sales:
+
+```sql
+SELECT date, sales,
+ AVG(sales) OVER (ORDER BY date ROWS BETWEEN 6 PRECEDING AND CURRENT
ROW) AS moving_avg
+FROM sales;
+```
+Here, each row’s result is computed based on the last 7 days, making it
computationally intensive as data grows.
+
+## Why Computing Sliding Windows Is Hard
+
+Imagine you’re at a café, and the barista is preparing coffee orders. If they
made each cup from scratch without using pre-prepared ingredients, the process
would be painfully slow. This is exactly the problem with naïve sliding window
computations.
+
+Computing sliding windows efficiently is tricky because:
+
+- **High Computation Costs:** Just like making coffee from scratch for each
customer, recalculating aggregates for every row is expensive.
+
+- **Data Shuffling:** In large distributed systems, data must often be
shuffled between nodes, causing delays—like passing orders between multiple
baristas who don’t communicate efficiently.
+
+- **State Management:** Keeping track of past computations is like remembering
previous orders without writing them down—error-prone and inefficient.
+
+Many traditional query engines struggle to optimize these computations
effectively, leading to sluggish performance.
+
+## How DataFusion Making it fast
+In the world of big data, every millisecond counts. Imagine you’re analyzing
stock market data, tracking sensor readings from millions of IoT devices, or
crunching through massive customer logs—speed matters. This is where
[DataFusion](https://datafusion.apache.org/) shines, making window function
computations blazing fast. Let’s break down how it achieves this remarkable
performance.
+
+DataFusion now supports [user-defined window aggregates
(UDWAs)](https://datafusion.apache.org/library-user-guide/adding-udfs.html),
meaning you can bring your own aggregation logic and use it within a window
function.
+
+```sql
+let my_udwa = create_my_custom_udwa();
+ctx.register_udaf("my_moving_avg", my_udwa);
+
+// Then use in SQL:
+SELECT
+ user_id,
+ my_moving_avg(score) OVER (
+ PARTITION BY user_id
+ ORDER BY game_time
+ ROWS BETWEEN 2 PRECEDING AND CURRENT ROW
+ ) AS moving_score
+FROM leaderboard;
+```
+This gives you full flexibility to build **domain-specific logic** that plugs
seamlessly into DataFusion’s engine — all without sacrificing performance.
+
+
+## Performance Gains
+
+To demonstrate efficiency, we benchmarked a 1-million row dataset with a
sliding window aggregate.
Review Comment:
Yeah, I think we should remove this section -- I commented it out in my
commit.
If you want to report performance numbers, I think you would have to setup a
benchmark.
The blog you cite is problematic because:
1. It compare normal (not window) aggregates
2. It does not mention postgres (it only compares DataFusion and Spark)
(this section feels like it was created by an LLM and has the results we
*WANT* to see, even though there is no actual evidence to support the
conclusion)
--
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: github-unsubscr...@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org
For additional commands, e-mail: github-h...@datafusion.apache.org
