24 posts tagged with "Data Lakehouse"

When you query an Iceberg table with WHERE user_id = 'abc-123', Spark reads every Parquet file that could contain that value. It first checks partition pruning — does this file belong to the right partition? Then it checks column statistics — does the min/max range for user_id in this file include 'abc-123'? But for high-cardinality columns like UUIDs, user IDs, session IDs, or trace IDs, min/max statistics are nearly useless. The min might be 'aaa...' and the max might be 'zzz...', so every file passes the min/max check even though only one file actually contains the value.

This is where bloom filters come in. A bloom filter is a compact probabilistic data structure embedded in each Parquet file that can definitively say "this value is NOT in this file" — allowing Spark to skip the file entirely. For point lookups on high-cardinality columns, bloom filters can reduce I/O by 80-90%.

This post covers everything you need to know: how bloom filters work internally, when to use them, how to configure them on Iceberg tables, how to validate they are present in your Parquet files, and what false positives mean for your data correctness.

Every Spark join starts the same way: read both sides, shuffle the data across the network so matching keys end up on the same executor, then join. That shuffle is the single most expensive operation in most Spark jobs — it moves data across the network, writes temporary files to disk, and consumes memory on every executor in the cluster.

But what if both tables are already organized by the join key on disk? If the left table's customer_id=42 rows are in bucket 42 and the right table's customer_id=42 rows are also in bucket 42, there is nothing to shuffle. Each executor can join its local partitions independently.

That is exactly what Storage Partitioned Join (SPJ) does. Introduced in Spark 3.3 and matured in Spark 3.4+, SPJ is the most impactful — and least understood — optimization available for Iceberg+Spark workloads. This post shows you how it works, how to set it up, how to verify it, and where it breaks.

If you are running Apache Iceberg on AWS, the single most impactful configuration decision you will make is your choice of FileIO implementation. Most teams start with HadoopFileIO and s3a:// paths because that is what their existing Hadoop-based stack already uses. It works, but it leaves significant performance on the table.

Iceberg's native S3FileIO was built from the ground up for object storage. It uses the AWS SDK v2 directly, skips the Hadoop filesystem abstraction entirely, and implements optimizations that s3a cannot — progressive multipart uploads, native bulk deletes, and zero serialization overhead. Teams that switch typically see faster writes, faster commits, and lower memory usage across the board.

This post covers everything you need to run Iceberg on AWS efficiently: why S3FileIO outperforms s3a, how to configure every critical property, how to avoid S3 throttling, how to set up Glue catalog correctly, and how to secure your tables with encryption and credential vending.

You have an operational database — PostgreSQL, MySQL, or DynamoDB — and you need its data in your Iceberg lakehouse. Not a daily snapshot dump. Not a nightly batch export. You need changes replicated continuously so that your analytics, ML models, and dashboards reflect reality within minutes.

This is Change Data Capture (CDC) on Iceberg, and it is one of the most common — and most operationally challenging — data engineering patterns in production today. The ingestion part is straightforward. The hard parts are handling deletes efficiently, keeping read performance from degrading, managing schema changes, and operating the pipeline at scale without it falling over at 3 AM.

This guide covers the two primary CDC architectures (direct materialization and the bronze-silver pattern), table design for CDC workloads, Iceberg's built-in CDC capabilities, compaction strategies, and the operational patterns that keep CDC pipelines healthy in production.

MERGE INTO is the most powerful and the most misused operation in Apache Iceberg. It handles upserts, conditional deletes, SCD Type-2 updates, and CDC application — all in a single atomic statement. But it is also the operation most likely to trigger a full table scan, blow up your compute costs, and produce thousands of small files if you do not write it carefully.

The difference between a well-written and a poorly-written MERGE INTO on the same table can be the difference between 30 seconds and 30 minutes — and between $2 and $200 in compute cost. This post shows you exactly how to write it right.

Your Iceberg tables are created with the right properties. Your partitions are well-designed. But your queries are still slower than you expected. The dashboard that should load in 3 seconds takes 45. The data scientist's notebook times out. The problem is not your table design — it is that you have not tuned the layers between the query and the data.

Apache Iceberg has a sophisticated query planning pipeline that can skip entire partitions, skip individual files within a partition, and even skip row groups within a file. But each of these layers only works if you configure it correctly. This post walks through every pruning layer, explains exactly how Iceberg uses metadata to skip work, and gives you the Spark configurations to control it all.

You have just migrated to Apache Iceberg — or you are about to create your first Iceberg table. You open the documentation and find dozens of table properties, multiple partition transforms, and configuration knobs that interact with each other in non-obvious ways. Where do you start? Which properties actually matter? How many buckets should you use? What happens when two jobs write to the same table at the same time?

This guide answers all of those questions. We will walk through every table property that matters for production Iceberg tables, explain how to design partition specs that balance read and write performance, cover commit conflict resolution, and give you concrete recommendations for both partitioned and non-partitioned tables.

Every AI project starts with the same bottleneck: data. Not the volume of data — most organizations have plenty of that. The bottleneck is data quality, data versioning, and data reproducibility. Can you guarantee that the dataset you trained on last month has not changed? Can you trace exactly which features went into a model prediction? Can you roll back a corrupted training set in minutes instead of days?

These are data engineering problems, not machine learning problems. And Apache Iceberg — originally built for large-scale analytics — turns out to solve them remarkably well.

This post covers four concrete patterns for using Iceberg as the data foundation for AI workloads: feature stores, training data versioning, LLM fine-tuning pipelines, and agentic AI data access.

If you are reading this, you probably fall into one of two camps. Either your Hive tables are already on cloud object storage (S3, GCS, ADLS) and you want to convert them to Iceberg format. Or — and this is the harder problem — your Hive tables are sitting on an on-premises Hadoop cluster with HDFS, and you need to move everything to a cloud-based lakehouse with Iceberg.

This guide covers both scenarios. We start with the harder one — migrating from on-prem Hadoop HDFS to a cloud data lake with Iceberg — because that is where most teams get stuck. Then we cover the table format conversion for data already on cloud storage. Both paths converge at the same destination: a modern, open lakehouse built on Apache Iceberg.

Every Apache Iceberg overview mentions time travel. "Query your data as it existed at any point in time." It sounds impressive, gets a mention in the feature list, and then most teams never use it beyond the occasional ad-hoc debugging query.

That is a missed opportunity. Iceberg's snapshot system is not just a convenience feature — it is a production-grade capability that can replace custom auditing infrastructure, eliminate data recovery anxiety, and solve one of machine learning's hardest problems: dataset reproducibility.

This post goes beyond the basics. We will cover the snapshot architecture, the practical query patterns, branching and tagging, the Write-Audit-Publish pattern, and real-world use cases that make time travel indispensable.